AI & Machine Learning

Architecting autonomy: A design pattern for high-compute, governed agents in biotech

Fri, 16 Oct 2026 16:00:00 +0000

In production systems today, most AI agents operate as lightweight dispatchers. They parse a prompt, make an API call, retrieve a record, and hand back control. Their execution is fast, stateless and cheap. Biotechnology does not work this way.

Folding a protein target, screening a 50,000 compound ligand library, or evaluating structure-activity relationships (SAR) are not quick API lookups. They are sustained, stateful, GPU-bound simulations that run for minutes or even hours across specialized silicon. When you ask a general-purpose large language model (LLM) endpoint to coordinate a complex drug discovery workflow, the standard paradigm breaks: Timeouts occur, context windows overflow with atomic coordinates, and proprietary chemical Intellectual Property (IP) risks exposure to external networks.

Building enterprise grade research and development (R&D) agents requires moving past the monolithic assistant model. In this blog post, we will discuss how to use a three-layer architecture: separating cognitive orchestration (the brain) from computational execution (the muscle), and anchoring both inside an enterprise governance plane to build scalable agents for drug discovery use cases in biotech.

A three-layer pattern

The reference architecture below shows how to architect, optimize, and govern high-compute autonomous agents in biotechnology, specifically Drug Discovery.

Figure 1 — The three-layer pattern: a cognitive orchestrator (brain), a scalable GPU execution layer (muscle), and an enterprise governance plane.

The execution layer: scaling the muscle - high-compute execution

The execution layer is responsible for the heavy scientific computations. The first architectural decision is determining where workloads should run:

Strategic, proprietary pipelines require dedicated GPU infrastructure within your security perimeter
Routine or standardized workloads can safely be delegated to managed cloud endpoints

In this architecture, protein folding is driven by ESMFold and generative design by GenMol. Both run as NVIDIA Inference Microservices or NIMs deployed on Google Kubernetes Engine (GKE) using G4 VMs backed by NVIDIA RTX PRO 6000 Blackwell GPUs.

A critical hurdle in containerized GPU workloads is the cold-start penalty, the latency delay required to pull multi-gigabyte model weights from an image registry and stage them into GPU memory (VRAM).

To ensure low latency responsiveness, we build the following as a recommendation into the reference architecture:

Maintain warm pools: Pin the GKE node pool to a minimum of one active node.
Preload weights: Model weights are loaded directly into VRAM at pod initialization, ensuring containers are immediately ready for inference.
Resilient scheduling: GKE manages pod lifecycles and restarts failed instances without tearing down the underlying VM infrastructure.

This approach trades custom, manual CUDA kernel tuning for NVIDIA's pre-compiled, containerized runtimes, dramatically reducing infrastructure engineering overhead while ensuring deterministic throughput.

In biotech, chemical structures represent millions of dollars in proprietary IP. A docking request transmits SMILE strings or 3D coordinate matrices that must never leave company custody.

Containerized NIMs eliminate the friction between rapid prototyping and rigorous data isolation. During early exploration, researchers can query NVIDIA hosted endpoints. When transitioning to proprietary lead candidates, the exact same container is deployed to GKE inside a private VPC. The agent’s orchestration logic remains identical; only the target URI updates from an external proxy to an internal VPC endpoint. No structural data leaves the enterprise perimeter.

The cognitive layer: Multi-agent reasoning

With compute execution isolated, the cognitive layer manages agentic workflow sequencing. Built on Agent Runtime, within the Gemini platform, this reference architecture shows a system that deploys a hierarchical multi-agent structure, one lead orchestrator, and two specialized domain sub-agents (Lead Optimization and Safety Screening) . The multi-agent design can be modified to suit the use case at hand.

Based on this design, when an end-user submits a high-level task such as “ Analyze Candidate 4b against Target X and produce a SAR report”, the system applies four core orchestration patterns discussed below;

In-process native agent handoffs: The lead orchestrator delegates tasks to specialist sub-agents using Agent Runtime’s native SDK rather than network-serialized agent-to-agent HTTP protocols. This eliminates intermediate network hops and keeps internal handoffs instantaneous. Only the outermost boundary communicates outward to user-facing surfaces like the Gemini agent.
Passing bulky payloads by reference: LLMs degrade quickly when forced to ingest massive arrays of 3D spatial coordinates. In this architecture, raw structural data (PDB files, docking poses, trajectory logs) is written directly to Cloud Storage. Agents pass lightweight, signed object URIs and high-level structural summaries, keeping context windows compact and inference costs down.
Decoupling workflow logic from biochemical reasoning: General-purpose LLMs excel at task planning and synthesis, but are prone to hallucinating biochemical nuances. To control this behavior, this architecture maintains strong separation of concerns, namely, workflow and tasks decomposition are handled by Gemini. Domain chemistry and pharmacological reasoning are handled by specialized Nemotron models hosted on dedicated GPU nodes.
Context-driven evaluation over rigid cutoffs: Traditional computational screening enforces arbitrary numerical cutoffs (e.g., discard if binding affinity ,-7.5 kcal/mol) However, therapeutic context fundamentally alters risk tolerance. Preventive cardiovascular therapy taken daily by healthy patients requires extremely strict safety margins, or a novel oncology candidate targeting an intractable cancer may accept higher toxicity if efficacy is demonstrated.

By passing disease context directly into the specialized reasoning models, drug candidates are evaluated against dynamic, target-specific profiles rather than static thresholds, preventing the premature culling of viable drug leads.

Performance best practices for high-compute pipelines

Optimizing an agentic system that interfaces with heavy GPU compute requires challenging common engineering instincts:

Common Assumption	Architectural Reality	Recommended Best Practice
"Swap in a faster driver model to cut latency."	The primary orchestrator's token generation is rarely the bottleneck in multi-step R&D pipelines.	Optimize I/O, not token speed: Focus on reducing payload sizes and intermediate network hops rather than compromising orchestrator reasoning depth.
"Parallelize tool calls across workers."	External scientific services and containerized models often serialize requests internally on dedicated hardware.	Respect execution topology: Align client-side concurrency with backend queue mechanics to prevent pipeline lockouts and connection timeouts.
"Pass full simulation states between tools."	Large JSON bodies cause schema validation failures, prompt truncation, and token bloat.	Pass heavy payloads by reference: Store raw coordinates in object storage and pass lightweight URIs between agent boundaries.
"Return outputs only when the pipeline finishes."	Long multi-minute runs leave scientists blind to whether jobs are proceeding or stalled.	Stream intermediate stage completions: Emit progress updates at each completed milestone (e.g., folding complete, docking running) for real-time visibility.
"Chain multiple granular micro-tools."	Deep tool-chains increase the probability of hallucinated arguments and schema mismatches.	Collapse compute pipelines: Wrap end-to-end simulation pipelines behind a single, clean tool interface that takes clear parameters and returns a consolidated object.

The enterprise governance and asymmetric autonomy

Deploying high-compute agents in enterprise environments requires robust data sovereignty, audit trails, and process control. The Gemini platform establishes these safeguards natively with the following components:

Agent Identity: Every autonomous agent operates with a cryptographically verifiable identity. Thus, for a biotech use case, when a candidate molecule is advanced or rejected, the event is permanently logged to that specific agent principal.
Agent Gateway: This serves the single ingress/egress point for every interaction within the perimeter. Centralized IAM and regional residency rules help ensure sensitive molecular data remains within designated cloud regions.
Agent Registry: This feature maintains a curated catalog of approved agents, validated tool schemas, and authorized Model Context Protocol (MCP) servers, preventing rogue tools from executing in production.
Model Armor: This functionality inspects incoming prompts and outgoing payloads in real time to prevent prompt injection and data exfiltration.

The payoff is clear. Identity, RBAC, auditing, residency, and data protection are native properties of the Gemini platform; this allows enterprise customers to focus on science rather than having to build a compliance stack for every project.

In high-stakes research and development, full autonomy introduces unacceptable liability, while requiring approval for trivial steps can create frictions. The solution addressing this to use asymmetric gating in the following ways;

Automate reversible operations: For example, allow the agent to resolve chemical naming synonyms, pull structural PDBs, and run preliminary Lipinski screens autonomously without interrupting the researcher.

Gate irreversible operations: For instance, initiating multi-GPU docking run, advancing a lead compound to wet-lab synthesis, or archiving a therapeutic program must require explicit human sign-off.

The pattern, and the path forward

High-compute autonomous agents succeed when the boundaries between cognitive planning, numerical computation, and system governance are clean in the following ways:

Keep the orchestrator focused on decomposition and synthesis, leaving intensive biophysical simulation to specialized, containerized microservices and models.
Eliminate context bloat by passing complex chemical data by reference through cloud storage.
Anchor the workflow in enterprise governance, pairing cryptographic agent identity with asymmetric human oversight.

As generative and structural biology models expand toward whole-cell modeling and complex dynamic simulations, this decoupled architecture ensures research systems scale efficiently while keeping proprietary IP secure and auditable.

Reach out to your Google sales representative if you have any requirements to build a use case like this and we will be happy to talk and support you in your agent development journey on Google Cloud.

Four evaluation engineering lessons learned from building agent plugins

Fri, 16 Oct 2026 16:00:00 +0000

The real challenge of building an agentic system isn't the construction, it's the measurement.

Evaluation engineering is a critical foundation for building reliable AI systems. Setting up systematic, repeatable tests to measure your agent’s true behavior can be the difference between shipping a tool you can trust and one that silently breaks, fails to call APIs, or makes mistakes.

I’ve run evaluation engineering initiatives for years at Google and developing the newly released Google Cloud Developer plugin served as a live testbed to stress-test these key principles across real-world developer workflows.

Here are four eval lessons we learned from building the new plugin that you can apply to your own builds.

1. Grade the execution steps instead of the summary

When we first evaluated agents, we set up an LLM-as-judge to inspect the agent’s final text reply and scored it against a checklist. However, we quickly realized it’s fundamentally flawed because the grader only sees what the agent wrote, not what it actually did. When we ran our initial 11-prompt evaluation suite for the Google Cloud Developer plugin, the automated grader marked it as failing three prompts. But when we opened the raw tool-call logs and read the transcripts, the plugin had actually passed all three. On a Cloud Run cleanup prompt, for instance, our rubric required checking command syntax before running anything. The agent ran gcloud help run services list as its very first step. It did the work, but its final message was focused on warning the user about permanent data loss rather than reciting that it had checked the help docs.

The grader penalized the agent for doing the right thing simply because it didn’t brag about it in the summary. If your rubric line says "did it run X?", a LLM judge is guessing unless you evaluate the actual tool-call trajectory.

Takeaway: Don't grade an agent on its closing pitch. Score the raw execution trajectory to verify what commands were actually executed, validated, and scoped.

2. Design evals for asynchronous operations

When you start reading tool-call trajectories, how you extract that telemetry matters just as much as what you're reading. Extracting agent telemetry into a flattened, sequential list of events obscures concurrent tool dispatch, which is one of the critical causes of behavioral failure. Modern agent harnesses often batch multiple tool invocations into a single assistant turn using an array of calls.

In our evaluations, when an agent was asked to inspect an environment and look up documentation, the harness dispatched the skill-loading call and two documentation searches simultaneously within a single tool_calls payload. Because activating a skill is a multi-stage process, where the initial call returns a short launch stub and the subsequent turn injects the full instruction body, the instructions meant to govern the search arrived in the prompt context after the system had already executed the searches.

The system never gave the skill a dedicated turn to let its guidance influence the agent's actions. If telemetry pipelines strip out turn structures and merge calls into a flat timeline, engineers will waste days wondering why an agent disregarded unambiguous instructions that a developer embedded in a skill.

Takeaway: Preserve turn structure in telemetry extractors so you can identify when client-side parallel dispatch issues tool calls before injected guidance enters the context window.

3. Test live runs and "dry runs" side-by-side

You need to inspect real API calls to catch harness quirks, like parallel dispatch. But if you only test your agent in one environment condition, you're flying half-blind. From day one, you need to run two conditions side-by-side: a live run and a plan-only run. In the live condition, the agent has real, sandboxed credentials and actually executes commands. This is how you see if syntax works, parameters serialize cleanly, and the agent exercises restraint before doing something destructive.

Live runs frequently hit an unexpected permission error on step two and abort before you ever reach the behavior you’re trying to test. That's why you need the plan-only condition, where you strip credentials and tell the agent to explain what it would do without running anything. This strips out permission roadblocks and shows you the model’s pure reasoning and routing intent. Roughly half our most important findings came from the delta between the two.

For example, on one of our onboarding prompts, a live run was blocked immediately on a project list call. The plan-only run sailed right through and produced an 8,000-character plan that completely ignored organization policies, admin approvals, and governance. The live run masked a major cognitive flaw that the plan-only run exposed in seconds.

Takeaway: Run live and plan-only evaluations side by side. Live runs test actual execution and safety restraint; plan-only runs expose the model's raw reasoning and routing without getting tripped up by permission errors.

4. Use “golden sets” as durable baselines

Frontier model architectures change every few months, agent harnesses improve, and plugin manifest standards will inevitably evolve. When building the Google Cloud Developer plugin, we treated evaluations as continuous integration testing for these non-deterministic environments. That means we stress-tested the composite stack, including the harness, the live model, and the tools running together on realistic tasks.

Instead of guessing from the final text output, this approach captures the whole trajectory: what the agent thought, what commands it actually ran, and whether it solved the problem without breaking anything. Then you score that run against golden sets curated by real human experts who know what the ideal path looks like.

Takeaway: Build golden sets based on what the entire system should actually do in the wild, including your harness, model, and plugins. Models change, but those baseline runs are your safety net.

Reminder: Use evaluation suites over “one-shot” demos

A common mistake I see among engineers is what I like to call the one-shot demo. This is when someone feeds a prompt into a LLM, watches it build a complete 3D game or solves a Rubik’s cube in a single pass and calls it an eval. But in reality, this is just a vanity metric.

To do quality work with AI, I truly believe that you need evaluation suites. Without evaluations, teams are like literally running into a dark cave blind. Engineering an enterprise-grade agent means testing the system's ability to handle multi-step interactions, manage permissions, and avoid destructive outcomes like accidentally deleting a project because you asked it to "find opportunities to save costs."

Where to go from here

Building an agent plugin or MCP server is the beginning. Shipping something you can actually trust with real cloud infrastructure requires significantly more work in evaluations to ensure it’s doing what you expect it to do. At the end of the day, the only thing that outlasts model churn is a stress-tested eval suite. Check out the Google Cloud Developer plugin documentation or explore this codelab to get started with the plugin in Antigravity.

What the Gemini agent means for startups

Thu, 15 Oct 2026 16:00:00 +0000

Small teams, short runway, and constant pressure to ship faster — it’s the reality we see every day among the founders we work with at Google for Startups. When you are building a company from the ground up, every hour spent on operational overhead or infrastructure plumbing is an hour taken away from your core product.

Last week at Gemini at Work 2026, we introduced the Gemini agent: a single, universal agent designed to work both as your personal assistant and as a dedicated member of your team. And this week, we made our new frontier model, Gemini 4 Argon, widely available — trained to maintain context across long, multi-step tasks.

Here is a look at how the Gemini agent works, five practical takeaways for founders, and how startups are already using it to punch above their weight.

What is the Gemini agent?

Instead of prompting separate chatbots for writing, coding, and data analysis, the Gemini agent is a single agent that takes high-level objectives and returns finished work. Three principles define it:

Unified: A single agent can now handle knowledge work, data analysis, creative asset generation, and writing and running code. For complex jobs, it dynamically spins up temporary, job-specific sub-agents and coordinates their work in parallel or in sequence.
Omnipresent access: It works across the surfaces you use every day — on the web, command line, desktop, and mobile, as well as natively inside Gmail, Docs, Sheets, Slides, Drive, Chat, and Calendar.
Persistent: It runs in the cloud with continuous memory across four dimensions (session, semantic, procedural, and episodic). It onboards itself like a new hire — learning your team, tools, and business context so you don't have to re-explain your startup with every prompt, and keeping multi-day tasks moving even after you close your laptop.

For founders wearing five hats at once, Gemini works like a digital chief of staff that's already briefed on your calendar, projects and documents. Ask it to research market trends, build a financial model in Sheets and turn both into a pitch deck, without repeating what the project or goals are at each step. It also works ahead of you. When an investor emails asking for your latest product update as a deck, Gemini spots the request and offers to handle it; one click and it comes back with completed work, ready for sharing or review.

Five takeaways for startups

1. Coworker agents and skills help you scale your output

Beyond personal assistance, you can create persistent coworker agents that work with your entire team. Each agent comes with its own Workspace account, email address, Drive storage, and directory presence. You can define a specific role — such as a launch coordinator, QA triager, or financial analyst — and your teammates can add agents to a Chat space or @mention it in a Google Doc comment to delegate work.

You can also equip agents with reusable skills — or let Gemini write skills for itself as it learns your processes. For example, with operational reporting skills connected to BigQuery and the Knowledge Catalog, anyone on your team can ask a plain-language question to build and save a live report. Once saved, those reports run on demand without incurring additional token costs. Gemini also offers pre-built specializations, starting with legal and financial services.

What it means for startups: Coworker agents can immediately increase the capacity of your teams, and skills can standardize repeatable processes like investor and operational reporting without burning tokens on every run.

2. Open and third-party model choice

Gemini is the agent, and the model underneath is your choice. As startup architectures mature, sending every task to a single frontier model quickly eats into gross margins. With the Gemini agent, you can dynamically route workloads across Gemini models, third-party models like Anthropic’s Claude, and lightweight open models like Gemma.

What it means for startups: You get flexibility without lock-in. You can optimize your cost per task — using compact models for high-volume, deterministic jobs and frontier models for complex reasoning — and swap models as your product evolves without rebuilding your workflows.

3. Connected to your services and tools

An agent is only as useful as the systems it can reach. In addition to Google Workspace, the Gemini agent connects directly to your engineering, data, and business stack, including partner solutions:

Development and collaboration: Git, Slack, Jira and more
Data and local files: BigQuery, Postgres, Snowflake, Databricks, and more alongside files on your desktop
Custom tools via MCP: Any Model Context Protocol (MCP) server inside or outside your network so your engineers can build and share internal tools.

What it means for startups: Instead of spending engineering cycles maintaining custom connectors, your team can plug the agent straight into your existing codebase, databases, and MCP tools so it can act on live data wherever it lives.

4. Built-in security and governance

For B2B startups, the enterprise security audit is often where deals stall. Under the hood, every agent is assigned a cryptographic Agent Identity for clear audit trails, runs inside an isolated sandbox, and protected with built-in prompt injection defenses. External traffic is routed through Agent Gateway, allowing you to enforce governance policies. And managing coworker agents is as straightforward as managing users today, since they all have a Workspace account.

What it means for startups: When an enterprise buyer’s CISO asks how your AI workflows handle data privacy, prompt injection, and access control, you have an enterprise-grade architecture ready — helping you clear procurement hurdles.

5. Protect your runway with real-time cost controls

Autonomous agents that run sub-agents for hours or days can quickly run up a bill if a loop goes unchecked. You can now set hard, real-time spend caps directly in the Cloud Billing Console; if a cap is reached, the project's agent automatically pauses so a bug never turns into a billing surprise. Underneath, Gemini runs on Google’s specialized AI infrastructure, including our inference-optimized TPU 8i, which delivers 80% better performance-per-dollar than the prior generation.

What it means for startups: You and your developers can experiment with long-running, parallel agentic loops with a built-in financial safety net that protects your cash reserves and builds confidence with your backers.

Founders already building

Startups and AI-native innovators are already putting Gemini to work. Immersive storytelling platform Whenere used Dreamlands, an agentic 3D worldbuilding platform built on Gemini platform, to bring Pride and Prejudice to Fortnite — with Gemini generating textures for each new object in about 7 seconds for under 10 cents; one artist created nearly 250 custom assets in a single day and completed a historically accurate estate in under a week. They join application-building platform Lovable and fellow startups and innovators including OpenEvidence, Recursion, and Replit and more.

Build with Google for Startups

The agentic era gives lean teams the leverage to punch far above their weight — whether you’re deploying coworker agents to cover internal operations or building agentic workflows directly into your product.

While these new Gemini agent capabilities are not publicly available yet, startups can contact our sales team to learn more and discuss how to get ready. In the meantime, you can explore the Google for Startups program to access credits, hands-on technical guidance, mentorship, global founder community programs and more.

How Visko streams real-time world models on Google Cloud TPUs

Thu, 15 Oct 2026 16:00:00 +0000

Editor's note: Generative AI startups are moving rapidly beyond static clip generation to build interactive, real-time worlds. In this technical deep dive, we explore how AI startup Visko collaborated with Google Cloud TPU engineers to deploy Orbis — its continuously streaming Live World Model—on Cloud TPUs, going from initial kickoff to production handover in just 30 days.

What makes a live video world model so challenging to serve in production? The bottleneck isn't average throughput — it's beating a relentless real-time clock. Unlike offline video generators where a user submits a prompt and waits in a queue, a live world model streams continuously. Every new chunk of video and synchronized audio must finish rendering before the viewer finishes watching the previous one, or the world freezes.

Today, we're excited to share how Visko, an AI startup pioneering interactive world models, teamed up with Google Cloud to solve this exact challenge. Working side by side, Visko and Google Cloud engineers deployed Visko's end-to-end live inference pipeline — combining video generation, super-resolution upscaling, and synchronized audio—onto a single Cloud TPU slice. Together, the teams brought end-to-end generation down, rendering at 1.07x playback speed.

Here is an inside look at how we engineered the pipeline to hit that milestone in a single month, the architectural optimizations that moved the needle, and practical lessons for other startups building ultra-low-latency generative AI on Google Cloud.

Redefining interactive media with Live World Models

While most generative video platforms focus on rendering fixed, pre-baked clips, Visko is building what it calls Live Models: video world models that generate continuously and respond on the fly.

Visko's flagship model, Orbis, streams live video starting from a single text or image prompt. As the stream plays, viewers can enter new steering prompts in real time, dynamically evolving the scene without restarting or breaking continuity. For instance, in one test session, the livestream opens on a cat napping beside a crackling fireplace; when the viewer steers the scene to introduce a bowl of food, the bowl materializes seamlessly, and the cat stretches, stands up, and meows.

Since launching its public demo on September 1, 2026, Orbis has unlocked new possibilities across interactive entertainment, live commerce, AI companions, immersive education, and robotics simulation. To scale these interactive sessions reliably, Visko partnered with Google Cloud TPU engineers in August 2026 to port the Orbis live pipeline to Google's Tensor Processing Unit (TPU). Within 30 days, the full multimodal pipeline was running in real time on TPUs.

Visko’s ambition is to power a new generation of interactive worlds that people can step into and shape in real time. Streaming continuous video and audio under a strict clock is a fundamentally different system challenge than offline clip generation. Working side by side with Google Cloud's TPU engineers, we brought the full Orbis pipeline to TPUs in just 30 days — giving us the low-latency compute and memory bandwidth required to bring Live Models to creators and developers at scale.

"Visko is bringing an exciting new class of real-time, steerable workloads to the Cloud TPU ecosystem. By co-engineering their live multimodal pipeline on a singleTPU slice using JAX, torchax, and custom Pallas kernels, we proved how quickly a lean startup team can go from PyTorch prototype to render at 1.07x playback speed on Google Cloud's AI Hypercomputer."
– Dave Elliott, Head of Developer Advocacy for AI, Google Cloud

Live video is a hard-deadline problem

When you serve an offline clip generator, latency is forgiving — the user waits once while the backend processes frames. A live world model has no waiting room.

Every single chunk must be denoised, decoded from latents, upscaled to 1664x936 (at 18 fps), and paired with generated audio faster than realtime playback speed, for as long as the session runs.

Two structural constraints make live world models far more demanding than standard batch inference:

Batch size of one: Each live stream represents an individual viewer's interactive session. You cannot mask latency by grouping dozens of concurrent requests into a massive batch — the standard playbook for keeping AI accelerators saturated.
Strict sequential dependency: Every chunk depends directly on the one before it. Because the video model maintains a rolling memory of recent video latents so that chunk N+1 smoothly continues the action from chunk N, you cannot pre-render chunks out of order or parallelize across future timestamps.

For startup infrastructure engineers coming from LLM serving, it helps to think of live world modeling not as a tokens-per-second throughput benchmark, but as an unforgiving per-step latency SLO.

Architecting the multimodal pipeline on a single TPU slice

To keep infrastructure lean and cost-effective as Visko scales, the team designed the live pipeline to chain three distinct generative models on a single Cloud TPU slice running in BF16 precision.

When a viewer submits an initial prompt or a live steering update, all three generative models — the core video diffusion model, the super-resolution upscaler, and the audio generator — share the same chips, taking turns on the hardware for each chunk.

Figure 1: Visko's end-to-end Orbis live pipeline chains video diffusion, latent decoding, super-resolution upscaling, and pure-JAX audio generation sequentially on a single Cloud TPU slice, completing each chunk faster than the playback speed.

To combine rapid startup iteration with bare-metal accelerator speed, the software stack is built on three open-source pillars:

JAX: Google's high-performance Python array library for accelerators, compiled into optimized machine code by the XLA compiler.
torchax: An interoperability layer that runs PyTorch models directly on Cloud TPUs via JAX, enabling startup teams to bring existing PyTorch components over without an upfront, ground-up rewrite.
Pallas: The custom kernel programming language for TPUs in JAX.

Real-time performance breakdown: Hitting the clock

By profiling and optimizing each stage on a single slice in BF16, the engineering team achieved a total per-chunk latency of under the target while — maintaining roughly 7% headroom to absorb variance during live streaming:

Figure 2: Chart of per-chunk latency budget on one TPU slice.

Because video denoising accounts for nearly two-thirds of the total compute budget, that stage became the primary focus of our kernel and parallelism optimizations.

Inside the stack: Three optimizations that unlocked real-time streaming

1. Ulysses context parallelism across all TPU chips

A video chunk expands into a massive sequence of spatiotemporal tokens inside the video transformer. To distribute that workload across the TPU chips at batch size 1, the team implemented Ulysses context parallelism.

Instead of splitting model weights, Ulysses splits the token sequence across chips. Immediately before the attention layer, a high-speed all-to-all interconnect exchange regroups the tensors so that each chip holds the entire video sequence, but only for one-quarter of the attention heads. After each chip computes attention over the full sequence for its assigned heads, a second all-to-all exchange restores the sequence-split layout for the feed-forward layers.

Figure 3: Ulysses context parallelism across TPU chips at batch size 1. High-speed all-to-all Inter-Chip Interconnect (ICI) exchanges transition tensors between a sequence-split layout for feed-forward layers and a head-split layout for Pallas splash attention.

2. Custom Pallas splash attention from MaxDiffusion

To accelerate the attention computation itself, the team integrated a custom Pallas splash attention kernel from MaxDiffusion, Google Cloud's open-source reference library for diffusion models on Cloud TPU.

Splash attention brings FlashAttention-style tiling to TPU architecture: rather than materializing the full quadratic attention matrix in High Bandwidth Memory (HBM), it computes attention in compact tiles that remain inside ultra-fast on-chip VMEM. Startup teams can drop the exact kernel used in this port straight into their own pipelines via custom_splash_attention.py.

3. A 5.8x audio speedup by moving from torchax to pure JAX

We first brought all three Orbis models onto Cloud TPU with torchax, which runs PyTorch code on TPU through JAX and XLA without a rewrite. For the video and upscaling models, torchax paired with Pallas attention kernels met our latency target directly, and both stay in torchax in the final pipeline.

Only the audio model needed a rewrite. Moving the audio stage from torchax to native JAX made it 5.8x faster and brought the full pipeline under the playback deadline.

Field notes for AI builders: Cloud TPU lessons

Takeaways from tuning live multimodal inference on Cloud TPU:

Ulysses context parallelism for single stream video: At batch size 1, splitting the token sequence across chips with all-to-all interconnect exchanges was faster than both tensor parallelism and ring attention.
Native BF16 for continuous streams: FP8 quantization cuts memory bandwidth pressure, but small quantization errors build up across chunks in a continuous stream. Native BF16 kept visual quality intact within the latency budget.
Selective JAX ports: torchax kept the video and upscaling models in PyTorch, so the team only rewrote the audio bottleneck in native JAX.
XLA compilation caching: The first boot spends a few minutes compiling XLA graphs. Caching compiled artifacts brings worker restarts under a minute.

Building real-time generative models on Cloud TPU? Start here

If your startup is pushing the boundaries of live video, world models, or ultra-low-latency multimodal inference, you can build with the same toolkit that powered Visko's port:

Accelerate diffusion attention out of the box: Grab the open-source splash attention kernel used in this collaboration from MaxDiffusion, and explore the Pallas documentation to write custom TPU kernels.
Migrate PyTorch fast, then optimize selectively: Bring your existing PyTorch models onto TPU quickly with torchax, pinpoint bottlenecks with the Cloud TPU Profiler, and port latency-critical stages to native JAX.
Spin up TPUs on demand: Use TPU Flex-start VMs to provision TPU capacity for short development and benchmarking runs without a long-term reservation.
Scale with Google for Startups: Early-stage AI startups looking for cloud credits, technical mentorship, and access to Google Cloud's AI Hypercomputer stack can apply to the Google for Startups Cloud Program.

^{Special thanks to the Visko.ai team for the close technical partnership, and to the Google Cloud TPU engineers who lead the project including: Yixin Shi, James Huang, and Elisa Tsai.}

Predictable AI spend in BigQuery: Introducing cost management for generative AI functions

Wed, 14 Oct 2026 16:00:00 +0000

The generative AI functions in BigQuery (AI.GENERATE, AI.IF, AI.CLASSIFY, AI.SCORE, and more) bring large language models (LLMs) directly into your SQL, so a single query can classify millions of support tickets or score each product review for sentiment. However, because these functions are billed by the token, scaling them across massive datasets highlights the importance of managing your budget proactively.

We recently announced the general availability (GA) of token-based cost management for BigQuery generative AI functions: daily quotas that let administrators set self-serve guardrails on AI token consumption, so teams can innovate while staying within your budget. In this blog, we provide an overview of cost management for BigQuery generative AI functions, walk through a few practical examples, and share resources to help you get started. Let’s jump in.

Quotas for managing costs

Four daily quotas, managed through the Google Cloud Identity and Access Management (IAM) quota system, provide controls at the project level and the per-user level. Per-user limits apply to each user within a project, not across your organization.

Quota name	Metric	Scope	Default value
AI input tokens per day	Input tokens used by the LLM	Per day, per project	200 billion
AI input tokens per day per user	Input tokens used by the LLM	Per day, per project, per user	150 billion
AI output tokens per day	Output and thinking tokens used by the LLM	Per day, per project, per project	20 billion
AI output tokens per day per user	Output and thinking tokens used by the LLM	Per day, per project, per user	15 billion

Key details:

The quotas are global. Usage is summed across all regions, so costs can't pile up unnoticed in a region you weren't watching.
Output quotas include "thinking" tokens. Reasoning tokens count toward your output quota, just as they count toward your Gemini platform bill.

Which functions are covered?

The quotas apply to these BigQuery functions that perform LLM-backed generation:

AI.GENERATE
AI.GENERATE_TABLE
AI.GENERATE_TEXT (and its alias ML.GENERATE_TEXT)
AI.CLASSIFY
AI.IF
AI.SCORE
AI.AGG

What's not covered: Embedding functions such as AI.GENERATE_EMBEDDING (also available as ML.GENERATE_EMBEDDING) and AI.EMBED are not subject to these quotas, because they run on different, substantially cheaper models.

Managing your quotas

Administrators can view and adjust the quotas directly in the Google Cloud console as workloads change, with no support case required.

Navigate to IAM & Admin → Quotas & System Limits.
Filter by service and select BigQuery API.
Sort by the Name column to bring the four AI quotas to the top.

Locating the gen AI token quotas in IAM & Admin → Quotas & System Limits.

4. Find the quota, for example AI input tokens per day.

5. Open the three-dot menu and choose Edit quota.

From there:

To tighten cost control, lower a quota. This takes effect immediately with no approval needed.
To support higher usage, request an increase from the same page. Many requests are auto-approved; larger increases may go through a quick review.

Editing a gen AI token quota in IAM & Admin → Quotas & System Limits.

You'll need the right permissions. You need one of: Owner, Editor, Quota Administrator (roles/servicemanagement.quotaAdmin), or Service Usage Admin (roles/serviceusage.serviceUsageAdmin).

Prefer automation? You can also manage these quotas programmatically through the Cloud Quotas API, the gcloud CLI, and Terraform to apply consistent guardrails across projects.

Monitoring usage

Beyond setting limits, the Quotas & System Limits page displays your historical token consumption so you can quickly detect spikes and right-size your budgets. Use these trends to see how new generative AI workloads and data pipelines affect your overall token usage.

Daily project-level token usage for a gen AI token quota in IAM & Admin → Quotas & System Limits.

Longer-term usage graph with selectable 7-day, 14-day, 30-day, and custom ranges.

For a step-by-step walkthrough, see the BigQuery documentation on controlling costs with token quotas.

Token quotas vs. billing spend caps

Token quotas aren't the only cost guardrail in Google Cloud.

Token quotas (this feature) are preventive and AI-specific. BigQuery tracks token usage during query execution and stops work close to the limit, with separate limits per project and per user.
Cloud Billing spend caps operate on dollars at the project and service level and can automatically block usage when a threshold is reached. They're a broad, financial backstop rather than a precise usage control, and because enforcement is not instant, some overages are incurred.

Use token quotas as your first line of defense for AI spend, and a spend cap as a project-wide safety net.

How enforcement works

Quotas are enforced by stopping work rather than letting a query run past your limit. A query is rejected up front with a QuotaExceeded error only if a quota it needs is already exhausted. Otherwise, it starts, and if a quota runs out mid-query, BigQuery stops sending new LLM calls:

Functions that support max_error_ratio (such as AI.IF) return partial results if unprocessed rows stay within the ratio, and otherwise fail.
General-purpose functions (such as AI.GENERATE) still succeed. Skipped rows have no generated value, and their status field contains the quota-exceeded error.

For details, see quota enforcement behavior in the documentation.

Sizing your input data: `AI.COUNT_TOKENS`

To choose quota values, it helps to know how much text your workloads send to a model. The AI.COUNT_TOKENS function helps with this.

AI.COUNT_TOKENS returns the number of tokens a model's tokenizer produces for a text string, without making a billable generation call. Run it over a column to measure your input data:

code_block: <ListValue: [StructValue([('code', "-- Count the input text tokens for each product review\r\nSELECT\r\n SUM(\r\n AI.COUNT_TOKENS(\r\n CONCAT('Classify the sentiment of this review as positive, negative, or neutral: ', review_text),\r\n\r\n endpoint => 'gemini-3.1-flash-lite').result) AS total_input_tokens\r\nFROM\r\n `my_project.my_dataset.product_reviews`;"), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7fd61b424a10>)])]>

The count is a lower bound on input tokens. It doesn't include output or thinking tokens, the instructions each AI function adds around your input, or non-text inputs such as images. Use it to:

Gauge the input side of a workload before you launch it.
Right-size your quotas together with a small sample run that measures actual input and output usage.
Translate tokens into dollars using the published Gemini platform pricing for your model.

From tokens to budget: a worked example

Suppose you want to keep a project's daily AI spend around $100 with the Gemini 3.1 Flash-Lite model. As of today's published pricing (BigQuery processes in batch which provides a 50% discount):

Input tokens: $0.125 per 1 million tokens
Output tokens (response + reasoning): $0.75 per 1 million tokens

If a small sample run shows an average 100:1 input-to-output token ratio, you can solve for a daily budget:

code_block: <ListValue: [StructValue([('code', 'Cost: 0.125 × I + 0.75 × O = 100 (I, O in millions of tokens)\r\nRatio: I = 100 × O'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7fd6204bc7d0>)])]>

Solving gives about 754 million input tokens and 7.54 million output tokens per day. Set AI input tokens per day and AI output tokens per day to these values to hold the project near $100/day. If a project uses models with different prices, budget per model or blend the rates.

Best practices for managing AI cost

Start with per-user quotas. They stop one user or runaway experiment from draining the project's budget.
Measure before you scale. Use AI.COUNT_TOKENS to size your input, then check a small sample run in the usage chart before setting limits.
Mind the output side. Output tokens cost more. Where a function accepts max_output_tokens, set it to bound generation length.
Monitor and iterate. Adjust limits as real usage and budgets change.

Get started

Token-based cost management is available now. To begin:

Open IAM & Admin → Quotas & System Limits in the Google Cloud Console and filter by BigQuery API.
Review the four AI-related token quotas and set limits that match your budget.
Use AI.COUNT_TOKENS to size your input data.

For the full details, see the documentation on controlling generative AI costs.

With token-based guardrails in place, your teams can use generative AI in BigQuery with the cost predictability your organization needs.

Gemini 4 Argon available on Google Cloud, built to help solve your business’ hardest problems

Tue, 13 Oct 2026 16:00:00 +0000

Last week we announced that Gemini Enterprise has evolved to make it even simpler to get work done, and introduced the Gemini agent, your new universal agent for work. It's always on and ready to answer your questions. It will write code, complete longer and more complex tasks, or perform whatever you need to get the job done, all from a single prompt box. And model intelligence underpins it all.

To enable a new frontier of work, we are making our new frontier model, Gemini 4 Argon, available on Google Cloud.

We trained Argon to work with both depth and precision, which allows it to maintain context across long, multi-step tasks which are reinforced with strong domain expertise along the way. Whether your team is running complex software builds, handling detailed financial and legal work, or deploying cybersecurity tools like CodeMender, Argon helps your teams maintain focus from start to finish.

Customers and partners using Gemini 4 Argon

Four ways Argon changes how your business works

1. Hand off your hardest work

Your teams already use Gemini for business to answer questions and draft content. With Argon, they can hand off the deep, multi-step work that used to take hours or days, and get back finished work they can use.

Argon reads long documents that mix text, tables, charts and images, and turns them into one clear output grounded in those sources. It can pull findings from many sources into a single structured answer, analyze data over time, and write stronger legal, finance, and executive documents. On long tasks, it keeps going until the job is done. We trained Argon to respect system boundaries, report negative findings honestly, and pause for confirmation before taking irreversible steps, and it tells you when it couldn't find something or finish a task.

Customers who tested Argon tell us it excels at turning huge, unstructured inputs into cohesive outputs.

"As software development becomes more agentic, teams need models that can support long-running, multi-step work across all stages of the development lifecycle. Gemini 4 Argon is designed to help organizations bring AI into development, security, and operations while keeping it within established enterprise workflows and controls.” – Stuart Moncada, VP, GitLab.

2. Precise, verifiable results on the work that drives your business

In finance, legal, tax, and other regulated work, calculations and data extraction need to be precise, repeatable, and verifiable.

For financial quantitative work, Argon can help automate multi-step ledger reconciliations, spreadsheet formulas, and variance analyses. It delivers high calculation integrity, citation fidelity, and formula accuracy across quantitative models. The proof is in the numbers: Argon leads on the Vals Index, which measures economic impact across finance, coding, legal, and tax work.

In legal research, Argon can help legal teams review case law and regulatory filings to pull out verified citations and key precedents. It leads on domain-specific evaluations like Harvey's Legal Agent Benchmark for legal research and drafting, with strong citation fidelity and direct answers when helping teams evaluate complex legal precedents.

"Gemini 4 Argon shows how much progress Google has made on agentic legal reasoning. We saw substantial improvements over Gemini 3.1 Pro across our legal agent evaluations, particularly in corporate M&A, data privacy and cybersecurity, and EC/VC. Accuracy also held up on long-document analysis, an encouraging result for lawyers working through detailed transactional matters." – Niko Grupen, Head of Applied Research, Harvey.

“We collaborated with Google on NowAI-Bench to help shape Gemini 4 for key enterprise work. In early access, it delivered roughly 20% improved task success than Gemini 3.8 Flash, with standout gains in HR and customer service. On the ServiceNow AI Platform, our L1 Service Desk AI specialist brings Google’s most advanced intelligence, governed and secure, turning tasks into fully autonomous work.” - Nenshad Bardoliwalla, Senior Vice President, AI Products, ServiceNow.

3. Build and ship products faster

Argon is built to handle complex, multi-step projects from start to finish. Our customers consistently ask for intelligence that can break down goals, track progress, and update multiple files at the same time.

"Gemini 4 Argon handles complex tasks reliably, from 3D games to rich animations. In Figma Make, it helps teams turn their most ambitious, multi-step ideas into impressive prototypes.” – Loredana Crisan, Chief Design Officer, Figma.

To see how well Argon handles this, we put it to work on our own tools. The team behind our open-source Agent Development Kit (ADK) set out to reduce framework overhead and improve code performance while maintaining full compatibility with existing applications. Working autonomously overnight within automated testing guardrails, Argon surfaced optimizations, now live in google/adk-python, that cut in-memory session prep time by 40% in long sessions and speed up cold imports by 21%. To learn more, read the engineering deep dive here.

Across Google, engineering teams using Argon in Google Antigravity have increased agentic code submissions by more than 35% while cutting code rollbacks by nearly a third. The Chrome team used Argon to finish an architectural refactor across hundreds of files in Chromium's 40 million-line codebase, completing months of work in weeks. Argon's improvements in spatial reasoning, mathematical precision and visual coding also make it faster to go from idea to working prototype.

For more advanced algorithms, our code optimization and discovery agent AlphaEvolve uses Argon to deliver strong performance improvements across real-world enterprise workloads, including factory scheduling, logistics, game code speed-ups, and chip layout — as well as complex mathematical benchmarks, and more.

To learn how to best use Argon, check out the developer's guide.

4. Stronger defense for your code and business

While Argon has cyber mitigations that block offensive tasks such as pentesting and exploitation, you can still take full advantage of Argon’s defensive capabilities using CodeMender to find and fix vulnerabilities, or use Argon with your own security harness. It ties for first place on CWE-Bench v1 with a top score of 68% for remediating security vulnerabilities.

"We are integrating Gemini 4 Argon into MarketingX for automated codebase security audits. In our evaluations, the model successfully identified 100% of top-priority vulnerabilities with 0% fabricated vulnerabilities across verified claims. This precision unlocks proactive, code-level security for our customers before deployment.” - Yunus Emre, Chief AI Officer of AI Lab, HubX.

Security also means being able to trust Argon inside your own environment. Argon is our most resilient model yet against indirect prompt injections, where malicious instructions or context are used to hijack a model’s behavior. It leads the Gray Swan Indirect Prompt Injection benchmark, cutting attack success across 15 adaptive attempts to 0.7%. That matters when Argon reads email, documents and webpages that may contain hidden malicious instructions.

We are also releasing Argon without cyber guardrails to a select set of trusted defenders, which allows for testing a full breadth of cybersecurity capabilities. Trusted defenders are vetted through our Fairwind Program, a limited access initiative that provides government, critical infrastructure, and cybersecurity partners exclusive, unmitigated access to advanced models.

Enterprise customers and partners in our early access program are already seeing strong wins with Argon.

^{Leaders from Amplitude, Deloitte, and Rakuten share Gemini 4 Argon results}

Get started

Starting today, Gemini 4 Argon is available in Gemini for business on the web, desktop, and mobile. Enterprise developers and builders can access Argon via API in Gemini platform under model ID gemini-4-argon, in Google AI Studio, and in Google Antigravity.

Argon launches at an introductory rate of $2 per million input tokens and $10 per million output tokens through November 30, 2026, with cached input tokens priced at 95% off. After November 30, standard pricing of $4 per million input tokens and $20 per million output tokens will apply. Review the model card and developer guide for complete technical specifications, token limits, and migration instructions.

Cyber defenders can join us at Security Summit to learn how to use Gemini 4 Argon to defend against machine-speed threats and automate remediation.

Welcome to Gemini at Work 2026: Introducing the Gemini agent

Thu, 08 Oct 2026 12:00:00 +0000

Editor’s note: This article is adapted from Thomas Kurian’s keynote address at Gemini at Work 2026. It includes a number of announcements:

Gemini, a single, universal agent for work that answers your questions, handles your knowledge work, creates your images and media, and writes and runs code.
Inline in Workspace, Gemini works directly inside Gmail, Drive, Docs, Slides, Sheets, Chat, and Calendar, carrying the same memory, skills, and controls it has everywhere else.
New data and analytics skills that allow both technical teams and everyday business users to use plain-language questions to get to actionable operational insights in minutes.
Industry-specific specialization with tools, skills, connectors and knowledge specific for financial services and legal teams.
How we secure and govern agents through identity and policy management, authorization and permission controls, secure sandboxing, network gateways, and more.
Leading cost controls through multi-model orchestration, Smart Routing, real-time spend caps, and more.
Customer use cases that showcase the breadth and depth of our customers around the world.

Bringing the best technology to all of you is what drives us at Google Cloud. In the last year, nearly 500 Google Cloud customers each processed more than one trillion tokens. Today, nearly 80% of all Google Cloud customers are using our AI products, and nearly 90% of the Fortune 100 use Gemini Enterprise. At that kind of scale, organizations have moved past experimentation and are running their business on it.

Work now starts in the prompt window.

To meet this moment, today we are excited to announce the Gemini agent: your new single, universal agent for work. It has all of your business context and can be used for everything from knowledge work to answering questions, and from content creation to coding, all from a single prompt box. It plans the work, uses skills and tools, connects to your systems, and brings back something finished — inside the documents, the inbox, and the developer environments you already work in. It chooses the best model for the job, has built-in cost controls, and most importantly, has the security, administration, and governance required by your company.

The Gemini agent is your new single, universal agent for work.

Introducing the Gemini agent

Gemini answers your questions, handles your knowledge work, creates your images and media, and writes and runs code — all in a single agent and a single API. You give it objectives, not instructions. You delegate an outcome and come back to finished work. For an agent to do that, it has to be connected to your personal workflows, your systems of record, and your enterprise controls. Gemini can function as your personal assistant or as a team member, where it works on behalf of a group of people, like a project manager within a team, or on behalf of a specific role in an organization, like an analyst in your finance department.

The agentic capabilities in Gemini are built around core architectural principles:

Unified agent: Gemini is an agent that can answer your questions when you chat with it, work autonomously to complete objectives you assign it, and generate code — all from a single interface. You can assign it work, schedule tasks that need to be completed, or have it respond to events.
Omnipresent access: Gemini can be accessed from any device including web, iOS and Android mobile devices, Windows and Mac desktops, and any channel (command line, Google Workspace, Microsoft 365, or Slack). It can also be integrated into third-party applications and surfaces and operates as a headless agent, meaning it does not need a dedicated user interface.
Persistent execution: It runs in the cloud, meaning it maintains a single set of memories, context, and one personalization graph no matter what device or channel you access it on. You never have to re-brief it or wonder which machine you told it to do something on. Work that takes hours or days keeps running after you close your laptop, and it is still there when you come back.
Multi-agent orchestration: Gemini doesn’t work alone. It can dynamically create a roster of sub-agents — temporary, job-specific agents, each with their own identity — to tackle multi-step tasks. Gemini can communicate with these sub-agents to coordinate workflows, including parallel and sequential steps that can run for hours or days. In addition, Gemini can act as a coworker agent — which is more like a team member, with a persistent, defined role and operational presence across days, sessions, and multiple changing responsibilities — and delegate and coordinate tasks. Coworker agents have dedicated identities including their own @agents.company.com emails, their own persistent storage, and only have access to the context that you or your team members provide.
Deeply contextual: Gemini arrives already knowing your tools, data, and work history. Whether you ask it a question, assign it an objective, or have it generate code, Gemini learns from every interaction with you. The longer you collaborate with it, the better it understands you and how you work. Teams can create dedicated projects to further optimize the context, skills, and tools Gemini uses when it works with you.
Model choice flexibility: Gemini is the agent, and the model underneath it is a separate choice. It runs each job on the model that fits best — orchestrating across our Gemini family of models and Claude models from Anthropic today, and other leading private and open models in the future — to deliver optimal quality and lower your costs. The best model for the task is not always the largest one. Matching the model to the work raises accuracy on the difficult jobs and lowers cost on the simple ones. And the leading model changes every few months, so keeping that choice open means your context, your skills, and your data stay put when it does.

Top customers across industries were early testers and are already realizing its benefits: Premium sportswear brand On tested this new dynamic selection capability to accelerate its speed-to-market. This builds on the proven, large-scale multi-model strategies already used by leaders: Shopify blends frontier models for millions of merchants to unlock data and drive sales, and PayPal routes 10 million multi-model requests every week.

Powering Gemini: Skills, tools, and context

Gemini understands your business — everything from pricing to product portfolios to departmental norms — and all the institutional knowledge that makes your company unique. Three capabilities make that possible: tools that connect it to your systems, skills that teach it how your work gets done, and context that lets it remember.

Tools and tools registry: Gemini can securely connect to the software your company already runs, including collaboration tools like Confluence, Microsoft Office, Teams, Slack, and Workspace; development tools like Git and Jira; enterprise platforms like Salesforce and ServiceNow; databases like BigQuery, Databricks, Postgres, and Snowflake; and files on your own desktop. It can also connect and work securely with any Model Context Protocol (MCP) server inside or outside your company network. It also offers an enterprise tools registry, so teams can build and publish tools for the rest of the company.
Skills and skills registry: Skills are reusable sets of instructions, knowledge, or workflows stored as modular prompts that teach agents how to perform specific, multi-step tasks. Gemini ships with a robust global library of skills. Teams or departments can build and publish custom skills to a shared company registry, and you can build your own personal skills. Gemini chooses the right skills and tools to perform a task and continuously learns from each execution, improving its consistency, while saving token costs.
Context and memory: Gemini learns from each question you ask it, each objective you assign it, the people you work with, and uses your personal context and your organization's context to personalize and improve the quality of its responses. It keeps four kinds of memory: session memory for the task in front of it, even when it runs for days; semantic memory, a structured knowledge base it builds as it reads documents, talks to people, and works with other agents; procedural memory for how a job gets done, including skills it writes for itself; and episodic memory of everything it has done before. Gemini onboards itself the way a new hire would, learning you, your tools, and your team before it starts.

Customers seeing significant benefits from Gemini Enterprise include:

BNP Paribas, a European leader in banking and financial services, is transforming Corporate & Institutional Banking by deploying Gemini Enterprise into LLM@CIB, its internal generative AI assistant available to more than 65,000 employees, to streamline workflows like preparing credit memos and boosting the productivity of customer advisers while maintaining strict enterprise-grade security and data governance.
Bradesco, one of Brazil’s largest banking and financial services conglomerates, is streamlining heavy operational workflows by deploying Gemini Enterprise for automated fiscal, accounting, and contractual analysis. The finance team has accelerated proposal reviews, service classification, and budgetary workflows, cutting document review time from 1 hour to 5 minutes while reducing risk inconsistencies by 60% and achieving over 10% in financial efficiencies.
Merck is exploring new frontiers in molecular research with the potential to significantly compress discovery timelines while maintaining complete data privacy and security.

Orange Spain, a leading telecommunications operator in Spain, is speeding up its digital transformation by deploying more than 1,000 custom Gemini Enterprise agents. Employees across HR, IT, sales, and customer service have built custom agents to automate incident triage, analyze network issues, and retrieve corporate knowledge, boosting operational agility and response times while helping non-technical staff automate daily workflows with zero IT bottlenecks.
Santee Cooper, South Carolina’s state-owned electric and water utility, is modernizing its financial planning and grid reliability by deploying a specialized AI model built on Gemini Enterprise. Financial planning teams and utility operators are automating complex financial scenario modeling and query execution using simple natural language and expect to boost modeling speed by up to 75% while mitigating market risk across a $10 billion grid expansion budget.
SOMPO, a leading Japanese insurance provider, built over 10,000 custom AI agents across its 34,000 employees to automate tasks like document search and summarization for employee efficiency. Their engineering team used Gemini Enterprise to cut the development time for new models from one week to a single day, freeing up engineers for more important work.
Ulta Beauty is helping guests shop 30,000 products by deploying the Ulta AI shopping assistant built with Gemini Enterprise. Online shoppers have translated conversational questions into personalized beauty recommendations, boosting digital sales conversion rates three-fold while skipping traditional search filters to buy the products they love instantly.
Wesfarmers, the parent company of Australian retail giants Bunnings, Kmart, and Officeworks, is deploying agents across its portfolio. An internal agent at Bunnings saved staff half a million hours of administrative work, while shopping agents — that help customers discover products and build their carts — have increased conversion rates up to three times higher for Kmart and Officeworks.

Organizations worldwide are driving measurable impact.

Gemini in Google Workspace

Gemini not only answers your questions, handles your knowledge work, and writes your code — it works directly inside Gmail, Drive, Docs, Slides, Sheets, Chat, and Calendar, carrying the same memory, skills, and controls it has everywhere else. It works inline: in the email thread, in the document, and in the chat space.

Inside Workspace, Gemini works three ways:

Personal assistance: Gemini works as your personal agent and arrives already knowing your calendar, your team, your projects, and how your documents relate to one another. It is briefed before you start. For example, you can ask Gemini to set up a meeting with the usual team of regional event leads next week, without supplying a single name or email address. Gemini determines who those people are from the membership of your chat space and the thread from your last event, checks their calendars, and starts an email thread to coordinate a time that works, even with external participants. It handles work that spans applications the same way: researching market trends, building a financial model in Sheets, then creating a deck that presents both, without you re-explaining the project at each step.
Proactive delegation: Gemini uses this same intelligence to proactively suggest tasks you can delegate to it. For example, if your manager emails you asking for the latest project update as a slide deck, Workspace Intelligence recognizes that as a delegatable task and gives you a single-click option to pass it to Gemini. It applies the same reasoning to your inbox, surfacing the message that matters most rather than the one that arrived last, and explaining why.
A member of your team: You can also use Gemini to create a coworker agent that works with your entire team. You simply describe the role you need, and Gemini creates it. The agent receives its own Workspace account, including an email address, calendar, Drive, and presence in your company directory. Your colleagues work with it the way they work with anyone else, by adding it to a Chat space or @mentioning it. For example, a marketing manager can ask an events coordinator agent in a chat group to draft a launch readiness document, which it posts back to the group when complete. The marketing manager could also tag the agent in a document comment, where it can suggest an edit in the Doc and reply in the comment thread, appearing under its own name in version history. A coworker agent acts under its own identity rather than yours, and it sees only what you share with it. Access follows the sharing and membership your team already uses, and no outside connector holds your data.

Data analysis, data engineering and machine learning

We are also giving Gemini skills built for specific domains, starting with data.

Data and analytics represent a foundational domain where Gemini transforms business workflows, enabling organizations to move from plain-language questions to actionable operational insights in minutes. We are introducing purpose-built capabilities tailored to both technical teams and everyday business users:

For Data and ML engineers: We are extending Gemini with machine learning skills and tools that back every data scientist and data engineer with a team of agents. Engineers describe the outcome they want in plain language, and Gemini generates PySpark code, provides notebooks to edit and test it, trains models, and troubleshoots and fixes pipeline issues on its own.
For business users: Anyone in your organization can generate reliable, real-time operational reports simply by asking. Gemini uses operational reporting skills integrated with BigQuery and the Knowledge Catalog to construct and save the query. Once saved, teams can run reports on demand without incurring token costs, guaranteeing consistent, verified answers every time.

Three Google Cloud capabilities keep those answers grounded:

Knowledge Catalog: Gemini integrates directly with your Knowledge Catalog to map business definitions once so all agents use them. This deep understanding of your organization’s schemas and business rules for terms like “net margin” and “addressable market” dramatically improves accuracy. Whether your metrics live in Databricks, dbt, LookML, or SAP, Gemini reads them directly where they sit to ensure factual responses.
Smart Storage: Ninety percent of enterprise data is unstructured — PDFs, images, scans, and audio recordings — which often remain "dark" and unreadable. Smart Storage changes this by enriching unstructured objects in place and writing context back directly onto the object itself. The intelligence stays where the bytes reside and inherits your existing security posture.
Borderless Lakehouse: Organizations should not be forced to move or duplicate their data to use AI. Our borderless Lakehouse allows Gemini to query Amazon S3 and Azure Data Lake with no variable egress fees, read directly from Salesforce Data 360, SAP, and Workday without copying data, and federate open Apache Iceberg tables across Databricks Unity, Snowflake Horizon, and AWS Glue.

You can start by registering data sets that you discover in your lakehouse with the Knowledge Catalog, and then assign Gemini simple or complex analysis to perform. It identifies the necessary data sets from the Knowledge Catalog, generates the necessary SQL, Spark or Python code to calculate the results, and runs the code on our Managed Spark Service with Lightning Engine or in BigQuery. It can also generate charts and build dashboards in the tools your analysts already use.

By grounding its data agents in the Knowledge Catalog, Bloomberg Media lifted its SQL query accuracy by 63% during initial development. As William Anderson, Bloomberg Media’s CTO, noted: “...by grounding our AI in a trusted institutional context, we ensure confidence in the accuracy and quality of every insight generated.”

Additional customers are seeing tremendous value from our data solutions:

Deutsche Telekom used BigQuery, Cloud Storage, and Knowledge Catalog to build a unified data lakehouse to overcome a fragmented and slow data infrastructure spread across 40+ legacy on-premise systems, allowing their teams to now move ten times faster while addressing European data sovereignty and security requirements.
Etsy migrated three petabytes of data to a unified lakehouse. By using BigQuery and open-source Apache Iceberg, they can now perform critical data joins up to 60% faster, allowing them to train models on much larger datasets and better connect buyers with the perfect items.
Snap connected its Prism agent directly to its storage archives, cutting technical diagnostic troubleshooting times down from 30 minutes to just 30 seconds.

Industry-specific specializations

Every industry runs on knowledge a general model does not have. We are specializing Gemini for individual industries, with the skills, tools, and knowledge specific to each one. It is now in preview for Financial Services and Legal, and coming soon to Government, Healthcare, and Retail.

Financial Services: Gemini for Financial Services automates what is usually pieced together across many workstreams: investment research, credit analysis, and risk modeling. It draws on trusted financial data from FactSet, LSEG, S&P Global, SEC filings, and your own proprietary repositories. Built with more than 50 foundational skills and designed to satisfy rigorous compliance, it shows its reasoning through confidence scores, explicit methodologies, full data lineage for easy auditing, and source citations you can check. Gemini for Financial Services is already being used by global financial institutions like CME Group and Deutsche Bank.
Legal: Gemini inherits matter-level permissions and ethical walls directly from document management platforms like NetDocuments and iManage. Partners like Harvey automate complex legal work while meeting the firm's confidentiality standards, while Onit streamlines contract lifecycle management without exposing privilege. Cooley is building a confidential information redaction agent on Gemini Enterprise for Legal that automates redactions, including personally identifiable information and confidential material, before legal filings go public. In high-stakes litigation, this can cut days of document review while maintaining the firm's rigorous standards for client confidentiality.

In addition, we’ve seen our ecosystem and customers create new vertical applications as well. For example, Deloitte has created a multi-agentic solution to help organizations identify the highest-value opportunities for AI applications for their business. Their Agentic AI blueprint develops operational heat maps to pinpoint areas ready for agentic redesign, accelerating the deployment of custom agentic workflows and achieving up to a 3x faster path to measurable business impact.

Securing and governing agents

Two factors determine whether an enterprise agent program succeeds or stalls: whether you can govern it, and whether you can afford it.

When deploying autonomous agents across an organization, governance comes down to answering four fundamental questions:

Who is the agent, and what identity does it have?
What is it allowed to do or what permissions is it granted?
What did it do and where can I see what it did?
What should it never touch?

Identity, policy, and observability answer the first three. The fourth is Agent Gateway.

Identity: Every agent gets its own identity, cryptographically attested and governed like an employee, with least-privilege permissions. That identity is stamped into the logs that capture its work, and into any virtual machine spun up to run code on its behalf.
Authorization and permissions: You give each agent fine-grained, role-based access permissions, approved by your organization’s security administrators. When the Gemini agent connects to an external system, its identity is mapped and propagated through industry standards such as OAuth.
Auditing: In addition, every action Gemini takes is written to an audit trail and attributed to the agent rather than to a person. Since the identity travels into any virtual machine the agent spins up to run code, you can monitor those logs with our observability tools in real time and catch anomalous behavior before it matters.
Policy management and control: Gemini gives you identity, discovery, governance, and security by default. All Gemini agents execute tasks safely inside an Agent Sandbox with its own network boundary. All traffic — in, out, and between agents — passes through Agent Gateway, an AI network firewall enforcing your organization’s policies in real time. You write the policy once, such as: “agents may not open documents classified Need to Know.” Gemini applies it to every agent in your company instead of checking one agent at a time.

Leading cost controls and spending options

While per-token prices have dropped 98% since 2024, enterprise AI volume has exploded. Running every simple loop through a premium model quickly breaks corporate budgets. We announced new flexible spending options recently and are adding to those today:

Multi-model orchestration: As mentioned earlier, Gemini allows you to choose the right model for the job, and even combine different models for specific tasks in larger projects to deliver optimal quality and lower your cost.
Smart routing: Our routing tool automatically triages enterprise workloads so each one runs on the model that delivers maximum performance at the lowest possible cost.
Real-time spend caps: You can set a hard limit on a project's AI spend in the Cloud Billing Console, and Gemini enforces it. Gemini monitors token usage and sandbox costs, and if a spend cap is triggered, that project's agent pauses, and you can choose to resume work with a single click in the console. Because the tracking is per project, companies can charge AI costs back to specific departments.

Infrastructure: 80% better price-performance

All of this innovation — the agent, the data, the identity, and everything else — runs on our AI Infrastructure and our models. They are the reason the economics work.

First, our AI infrastructure. We co-design the chips, the network, the data center, and the software that connects it all as one highly-optimized system — our AI Hypercomputer. This vertical integration delivers significantly better price-performance than the generic setups, and we are continuing to push the state of the art here. For example, our latest TPU 8i system delivers 80% better price-performance than the prior generation, making complex, high-value agent tasks both fast and highly affordable.

On top of this silicon sit our models: Argon for frontier reasoning, Flash for speed and volume, Omni for generative media, and Gemma for lightweight, open-weights edge workloads.

The edge capabilities are already real. NASA’s Jet Propulsion Laboratory is running Gemma directly on a satellite in orbit, a first for a vision-language model. JPL’s NAVI software analyzes Earth imagery on board the satellite and sends a summary with key information to ground operators.

Supercharged by our ecosystem

We remain committed to choice at every layer of the stack, ensuring you have the freedom to choose the best tech stack for your unique needs.

Inside Gemini, you have a broad choice of third-party agents from leading SaaS providers as well as connectors that are ready to use. They are even easier to discover as collections in the new AI Solution Finder.

Around Gemini, our services ecosystem continues to rapidly scale to meet enterprise demand. Together, in the last month alone, our global consulting partners have put more than a hundred thousand of their developers and consultants through Gemini hackathons.

In addition, Accenture has made a significant investment by establishing a dedicated Gemini Enterprise Business Group to accelerate customer adoption and innovation. Several other partners have also expanded their commitments to support the growing demand for Gemini, including establishing specialized centers of excellence and training Gemini credentialed practitioners to help customers build and deploy production-grade agents faster.

Gemini Enterprise in action: Global momentum

Organizations worldwide are driving measurable impact across every major region:

Europe and Middle East

Arden University, a UK provider of flexible online and blended learning, is preparing students for the global job market by rolling out Gemini Enterprise. Over 30,000 students and staff will now have access to advanced Al tools to personalize their studies and gain essential skills for the workplace.
Commerzbank, one of Germany’s largest private banks, is using Gemini Enterprise to automate document quality assurance, reducing manual work on document reviews from 20 hours to just one. They are expanding this with a multi-agent system for AI-driven QA.
Lloyds Banking Group is scaling agentic AI across the bank. Envoy, a secure, Gemini-powered pipeline, has already onboarded nearly a thousand engineers building production-ready AI use cases. From fraud detection to commercial client onboarding, Envoy gives teams a reusable agent marketplace, a "golden path" for building agents, and a CI/CD pipeline — all wrapped in the guardrails and security a highly regulated bank demands.

Nokia is tackling the complexity of modern telecom networks with an ecosystem of specialized agents built on Gemini Enterprise. This allows operators to fix issues before subscribers are impacted by automating troubleshooting and cutting issue resolution time by up to 80%.
On, a premium sportswear brand, is scaling workforce productivity and efficiencies by deploying Gemini Enterprise to every employee to drive personal productivity and foster an AI-ready culture. Team members have launched no code agents that help with efficient knowledge extraction, executive reporting, market research, scheduling, onboarding and more.
Ooredoo Qatar, the leading telecommunications provider in Qatar, is integrating Gemini Enterprise into their AI transformation strategy to drive advanced automation and intelligence. Its customer support, marketing, security, and field operations teams have deployed AI agents to optimize internal operations management, while accelerating rapid issue resolution and marketing campaigns to deliver around the clock, highly personalized, proactive customer experiences.
Qatar University, the nation's premier institution of higher education, is transforming higher education operations and academic research by deploying Gemini Enterprise campus-wide. Thousands of faculty and staff are deploying more than 2,000 custom AI agents to modernize institutional workflows, automate complex administrative operations, and enhance academic productivity.
Ryanair, Europe’s largest airline, is modernizing its collaborative operations and infrastructure by deploying Gemini Enterprise and Google Workspace to 35,000 employees. Flight crews, operations planners, and corporate staff will benefit from automated decision-making, optimized crew logistics, and enhanced corporate productivity, while a resilient dual-cloud strategy strengthens infrastructure resilience and supports the airline's target growth to 300 million passengers by 2034.
Starling, one of the UK's top-rated digital banks, built its Scam Intelligence tool on Gemini Enterprise. It analyzes online marketplace ads for suspicious content and has quadrupled the rate at which customers cancel potentially fraudulent payments.
Other major European customers include Berenberg in banking, Bouygues Telecom, the application building platform Lovable, and Siemens in automation.

Asia Pacific

Airwallex, the global financial platform processing over $200 billion annually, has deployed Gemini Enterprise across its infrastructure to resolve customer issues 10x faster, automate dispute handling to cut case time by 40%, and accelerate software time-to-market by over 95% while building specialized agentic finance workflows for automated treasury, payables, and payments.
DBS Bank, Southeast Asia’s largest bank, is moving past pilots into enterprise production by deploying deterministic chains of 70 to 80 specialized agents on Google Cloud to synthesize complex financial, regulatory, and market intelligence for corporate credit memos, shifting relationship managers from manual research to high-value client advisory while keeping humans firmly in the loop.
Hitachi is transforming global power and utilities by deploying agents directly to frontline workers via HMAX, Hitachi’s portfolio of AI-powered solutions. Field technicians have built agents for over 200 use cases — such as using image-recognition agents to verify photos of every tool and wire before and after a job — boosting productivity by up to 30% while keeping workers safer.
NTT DOCOMO, Japan’s leading mobile operator, built data agents that reduced time-to-insight from two weeks to instant, which gave four hundred fifty thousand hours a year back to the business. Building on this success, they are now deploying Gemini Enterprise to drive their next wave of productivity.
Tata Steel, a global steel manufacturing leader, successfully deployed more than 300 specialized AI agents in nine months to enhance manufacturing safety, optimize predictive supply chains, and cut customer complaint turnaround times by 50%.
Zip, an Australia-based digital financial services company, is transforming its product-development lifecycle in the U.S. by deploying Gemini Enterprise to build an AI-Native Product Factory. Cross-functional teams will use this unified agentic environment to connect customer research, design, engineering, and compliance, boosting product development velocity and parallel experimentation while lowering the cost and complexity of bringing new financial products and experiences to market.
Other APAC leaders include LG in electronics, Marubeni in trading, and TCS in technology services.

Latin America

Banco BV, a leader in the Brazilian financial sector, is using Gemini Enterprise to create personalized scripts for its customer relationship managers, accelerating client communications by 80%. The new AI platform allows the bank to quickly engage customers with tailored financing and investment opportunities.
CNA/SENAR, Brazil’s national agriculture and livestock federation, is expanding access to critical agricultural information by deploying JoIA, a conversational AI assistant on WhatsApp built on Gemini Enterprise. 40,000 active users, with the potential to support 5 millions producers across the country, have accessed localized input pricing, weather forecasts, and financial health evaluations to facilitate in-field decision-making.
Farmacias del Ahorro, one of Mexico’s leading drugstore and pharmacy retail chains, is accelerating corporate operational efficiency by deploying PotencIA, an internal program built on Gemini Enterprise. 1,000 corporate employees across IT, e-commerce, and marketing have adopted the tools for daily business workflows, boosting active user adoption to 95% while generating an estimated 140,000 hours in annual productivity gains.
Globo, Latin America’s largest media company, is optimizing early-stage software planning by deploying PlanejaAI, an AI agent solution built on Gemini Enterprise. Product Owners, managers, and analysts have accelerated epic refinement and quarterly planning, boosting pre-development plan accuracy by 33% and slashing validation time from 20 minutes to 6 seconds while saving 127 hours in a single quarter.
Grupo Bafar, a prominent Mexican food producer and consumer retail group, is accelerating national retail expansion by deploying Gemini Enterprise for automated contract management. Legal and retail operations teams have streamlined the preparation of operational annexes for store openings, boosting processing capacity tenfold while slashing contract preparation time from 140 to 17 minutes per branch.
Laureate Education, a major higher education network serving nearly 500,000 students across Mexico and Peru, is transforming academic and administrative operations by deploying Gemini Enterprise and Gemini for Education. Faculty and staff have personalized learning paths, enhanced teacher support resources, and modernised campus administration, boosting student engagement and operational readiness while driving long-term career employability for graduates.
Other LATAM leaders include iFood delivery services, Livelo in customer loyalty, SEBRAE in public services, and Yduqs in education.

North America

Albertsons, a leading U.S. food and drug retailer, is ensuring customers get the freshest produce with its patent-pending Intelligent Quality Control tool built with Gemini Enterprise. Distribution center inspectors are using computer vision and Gemini models to instantly evaluate fresh fruit and vegetable quality against company standards, boosting rating consistency across shifts and locations while speeding up inspections to get perishable food onto grocery shelves faster.
Avid, a leading provider of media technology for film and television, is streamlining video post-production through browser-based editorial experiences and agentic creative workflows built with Gemini Enterprise. Editors and directors can use conversational content discovery, AI-assisted transcription, and intelligent media organization within their creative workflows — including within a Google Chrome browser — helping teams spend less time on manual administrative tasks and more time focused on storytelling.
Constellation Energy Company, a leading energy supplier, will build Gemini Enterprise agentic workflows directly into its core operations. Power generation teams will have access to advanced AI capabilities to accelerate capacity delivery, optimize asset dispatch, and protect critical infrastructure.
Honeywell Technologies Forge platform includes predictive maintenance capabilities built on Gemini Enterprise, such as contextualizing operational data at scale and providing proactive maintenance recommendations. This helps customers minimize costly, unplanned outages, protecting and extending the lifespan of their mission-critical equipment.

PayPal built its AI platform on the Gemini Enterprise to create new commerce tools, cutting the time it takes to deploy new models from weeks to just minutes.
The Home Depot, the world’s largest home improvement retailer, is speeding up customer support and product discovery by deploying AI voice and digital assistants built on Gemini Enterprise. DIY shoppers and contractors calling stores or using Magic Apron have skipped rigid phone menus, searched local inventory with natural speech, and resolved inquiries in real time, boosting phone resolution speeds by four-fold (understanding customer needs in under 10 seconds) while allowing in-store associates more time to focus on face-to-face shoppers.
The RealReal, the leading online marketplace for authenticated luxury resale, is transforming luxury consignment by deploying Ask TRR, a conversational shopping concierge built with Gemini Enterprise. Resale shoppers have navigated and searched a massive, shifting catalog using natural language instead of rigid filters, which will boost user engagement across all 45 million members while making it effortless to discover and secure rare, "one-of-one" luxury pieces before they sell out.
Verizon is modernizing its customer and enterprise operations with Gemini Enterprise and agentic Data Cloud. The technology has helped resolve inbound customer calls and chats automatically, helped predict and resolve network anomalies, and automated marketing campaign orchestration, boosting customer satisfaction and automated resolution rates while unifying legacy data lakes into a single source of truth.
Whenere, an immersive storytelling platform, is revolutionizing interactive 3D worldbuilding, starting with bringing Pride and Prejudice to Fortnite. To build the Netherfield estate, the team used Dreamlands, an agentic 3D worldbuilding platform built on Gemini Enterprise. With Gemini generating each new texture in about seven seconds for under ten cents, one artist created nearly 250 custom assets in a single day and completed a historically accurate estate in under a week, all while keeping full creative control.
Leading hospital systems and healthcare services companies like Highmark Health and Seattle Children's rely on Gemini Enterprise to run secure administrative and clinical support workflows.
In public sector, Gemini Enterprise is being used across the states of Arizona and Missouri, the State University of New York (SUNY), and the Chief Digital and Artificial Intelligence Office, which has put Gemini Enterprise in the hands of 3 million uniformed and non-uniformed members of the US armed services. They have built more than a hundred thousand custom agents to streamline everyday tasks.
Small and midsize businesses including Aerotech, EDSpan Global,, Pinch Creative, Quadrant Health Group, Salt Productions, Strategem, and more are using Gemini Enterprise to empower their teams with AI agents that can streamline their operations, automate administrative workloads, help find information, and even support customer engagement. In fact, usage of Google Cloud’s AI tools by these customers has grown more than 5x year over year.
Other North American startups and innovators like Every Cure, Harvey, OpenEvidence, Recursion, and Replit are using Gemini Enterprise.

Our commitments to you

As you build the future on Gemini, we make three firm commitments:

We take the friction out: A future-proof platform that runs anywhere, executes any task, supports your choice of models, and applies enterprise controls.
We eliminate complexity: Google builds and integrates the entire stack: silicon, models, skills, tools, and security.
What you build stays yours: Your people and your data do not need to move, and your proprietary inputs and outputs remain entirely your own.

The technology is ready, our global partner ecosystem is scaling, and Gemini is ready to work for you.

Innovation in Ireland: How Irish brands scale with Gemini Enterprise

Thu, 08 Oct 2026 12:00:00 +0000

In recent decades, Ireland has grown into a vibrant hub for global technology. As modernization accelerates, Irish organizations — spanning government agencies, established enterprise brands, and growing startups in AI and other areas — are all advancing their technical foundations. Today, leaders across the country are injecting agentic AI into every layer of their business and operational workflows.

We are seeing organizations move from early prototypes to production-ready deployments using Gemini Enterprise. This progress supports Ireland's competitive position within the European market. According to a recent Implement Consulting Group report assessing Google’s local contribution, public and private leaders estimate between 40 billion to 45 billion euros of economic potential in global AI markets.

This ambition aligns with broader government initiatives. Since assuming the presidency of the Council of the European Union for the second half of 2026, Ireland has steered higher-level technology and policy discussions, outlining a National Digital and AI Strategy to improve productivity and innovation across sectors.

To support this strategy, we are focused on making these capabilities more accessible. Our local teams are collaborating with industry leaders to implement Gemini Enterprise which offers capabilities that range from conversational AI agents and natural language data analysis, to optimizing operations with agentic ecosystems.

The economic impact is already taking shape. In 2025 alone, Google operations — including our AI and cloud services — added an estimated 10 billion euros to the Irish gross domestic product. And this momentum is shared: Organizations across Ireland are using Google Cloud to modernize infrastructure and put AI to work.

Here are several ways Irish organizations are modernizing with Google Cloud and Gemini Enterprise:

Ryanair supports millions of passengers through AI-backed operations

Ryanair, Europe’s largest airline, is modernizing its collaborative operations and infrastructure by deploying Gemini Enterprise and Google Workspace for some 35,000 employees. Flight crews, operations planners, and corporate staff will benefit from streamlined decision-making, optimized crew logistics, and enhanced corporate productivity. At the same time, a dual-cloud strategy strengthens infrastructure resilience as the airline targets growth to 300 million passengers by 2034.

Smyths Toys rethinks the customer experience with agentic support

As one the world’s largest specialty toy retailers, Smyths Toys is reshaping its customer support with Google Cloud and Gemini Enterprise. By introducing an AI agent named Codie, the retailer has resolved more than 60% of its web inquiries and reduced holiday email backlogs. Customer care teams that previously handled up to 1,000 peak-season emails a day can now focus on the roughly 50 cases that need more complex judgment.

The Irish Revenue Commissioners: Advancing public sector data

Public sector agencies are also making strides in Ireland. The Irish Revenue Commissioners — responsible for taxes, customs, and keeping Ireland’s economy running smoothly — is working with Google Cloud to migrate from legacy on-premise systems to a unified AI-lakehouse architecture designed to break down data silos and establish a single source of truth. Strict data sovereignty controls safeguard taxpayer information while Gemini Enterprise powers agentic data analysis, helping Revenue deliver secure, responsive public services.

Virgin Media Ireland: Accelerating workflows

Supported by Liberty Global's partnership with Google Cloud, Virgin Media Ireland is moving AI from exploration to production. Through modern infrastructure and automation, the team now deploys workloads three times faster, is in the process of rolling out multiple AI agents, and is launching new customer services designed to drive business growth.

Irish startups scaling with enterprise tools

While established enterprises are modernizing legacy systems, Irish startups are building on modern technology from day one. Backed by a strong local tech ecosystem, these agile teams are creating services that compete on an international stage.

Hexis: Scaling personalized sports nutrition

Hexis builds technology solutions that help athletes refine their performance and recovery. By storing and processing that real-time biometric and behavioral data in Google Cloud, Hexis uses Gemini Enterprise to power natural language conversations that explain fueling adjustments to users throughout the day. Today, the platform supports 40% of the Tour de France peloton and half of Premier League clubs during the 2026–2027 season.

Kitman Labs: Unifying performance analytics

Founded in Dublin in 2012, Kitman Labs provides sports science and performance intelligence to more than 4,000 sporting organizations across 26 countries. By combining its intelligence platform with advanced analytics and Gemini Enterprise, Kitman Labs helps teams make data-informed decisions that support athlete health, lower injury risk, and extend player career longevity.

Spryt: Automating healthcare coordination

Spryt streamlines healthcare coordination across the U.K. and U.S., using AI built on Google Cloud and Gemini Enterprise to optimize patients’ experiences, reduce missed appointments, and cut administrative burden. The company collaborates with the NHS, operates in more than 70 US clinics, and has analyzed 224 million appointment records as part of a multiyear initiative with Mayo Clinic.

IMPT: Automating carbon offsetting

By connecting AI, blockchain, and Google Cloud, IMPT embeds verified carbon offsetting into everyday hotel bookings and online purchases. Following a 15-week migration to Google Cloud, IMPT supports 8 million listings across 195 countries. With Gemini Enterprise, the company’s lean team handles about 400 daily customer chats with a six-second median response time, allowing staff to focus on growing its business rather than managing systems.

Transform ideas from planning to production

As Ireland continues to expand as a global technology hub, local organizations are demonstrating how to translate AI potential into practical results. Explore our customer stories to see how other organizations are building with Google Cloud, or reach out to our team in Ireland to accelerate your journey.

What’s new in AI infrastructure and orchestration in September

Wed, 30 Sep 2026 16:00:00 +0000

We hereby declare September to be scalability month! As the world prepares for a surge of agentic fleets, we are shoring up our AI infrastructure and orchestration offerings to gracefully — and quickly — respond to that demand, all while maintaining workload isolation and security, and keeping costs in check. Read on to learn how these enhancements manifest across Google Cloud’s compute, network, storage, and orchestration offerings, plus new ways customers are using Google Cloud AI infrastructure, and third-party industry validation of our strategy.

Product, technology, and tools updates

Google Kubernetes Engine updates: The GKE team is all about improving the scalability of the platform, and in September, those improvements came in many shapes and sizes:

New feature: Need an execution runtime with higher density for your agentic workloads? We engineered the new open-source GKE Agent Substrate to run millions of sandboxes with 10x higher density than standard container runtimes. Agent Substrate also delivers sub-500ms resume operations at over 500 suspend/resume activations per second with a native zero-trust kernel and network isolation.
Product update: GKE now has scale-to-zero capabilities built-in. No need to configure complex components to scale your workloads down, thanks to the HPA with the Autoscaling Metric and support for KEP-2021, which do the job for you, out of the box. Read the blog to learn more.
Product update: Further, the GKE HPA (with the above-mentioned Autoscaling Metric) now lets you scale up and down based on custom PromQL metrics, in addition to standard metrics, allowing you to trigger workloads according to conditions that are meaningful and unique to your business. Read more here.
New feature: Yet another scalability feature is GKE Pod snapshots, which lets you save the running state of your workload, including CPU and GPU memory, and restore it on demand. According to internal tests, GKE Pod snapshots can reduce AI inference start-up by as much as 89%. Learn more here.
New migration tool: Finally, if you’ve always wanted to migrate your container workloads from AWS EKS to GKE but feared a daunting, high-friction engineering endeavor, we’ve just launched GKE agentic migration, a purpose-built agent plugin that replaces brittle, ad-hoc prompting with an AI-assisted migration pipeline protected by deterministic guardrails. Designed as a compilation of agent skills and a local Model Context Protocol (MCP) server, it uses AI to translate complex AWS EKS IaC and Kubernetes manifests directly into GKE landing zones. Get started with the onboarding guide.
Feature updates: Reinforcement learning (RL) and evaluation workloads are a beast: In a standard agentic RL loop, an LLM policy generates actions like code snippets on GPUs and executes them inside isolated CPU sandboxes to observe a reward signal. However, when scaling up this loop to support tens of thousands of parallel rollouts, infrastructure bottlenecks emerge, for instance idle accelerators, image cardinality, and a saturated control plane. To help, we developed GKE Agent Sandbox optimized for RL, plus an Agent Sandbox RL orchestration SDK and native integrations for popular RL gyms and harnesses. All are now generally available, and you can learn more here.

Storage updates: AI trains and creates lots of data, and that data has to live somewhere — in block storage systems, file systems, object stores and databases. We announced enhancements to our storage portfolio to help this critical layer roll with the agentic punches:

Product update: Filestore agent volumes offer high-performance, elastic, persistent file storage for agentic workloads. Thanks to its tight integration with GKE Agent Substrate and GKE Agent Sandbox, Filestore agent volumes automatically allocates and attaches a dedicated, isolated file workspace to GKE agent sandboxes in milliseconds. Request access to the preview here.
New product: If you run generative AI and RAG data layers — think Milvus, Pinecone, Qdrant, Vespa, Redis, and in-memory context caching — you may want to take a look at the M4N family of VMs, now GA, which offers the highest per-core IOPS and throughput of leading hyperscalers. Paired with Google Cloud's custom Titanium offload architecture and paired with Hyperdisk Extreme, M4N instances deliver up to 25,000 MiB/s (25 GiB/s) of aggregate host storage performance and up to 1 million IOPS.
New product: Another new Compute Engine product, Z4D, is GA, and a strong storage solution for AI/ML training and inference workloads. When configured as a bare metal instance, Z4D provides both the high local SSD (LSSD) capacity and low latency required by agentic microVMs, so you can run thousands of isolated sandboxes per host with native performance and efficiency. Learn about Z4D machines here.
Product update: Today’s AI training and inference pipelines create data faster than most storage management systems can keep up, creating challenges for teams trying to understand their storage estates. A new version of Storage Intelligence advisor makes it easier to answer the question: "What’s in my buckets?" and quickly identify unexpected changes. Then, enhanced batch operations let you automate bulk changes across your buckets — say, move storage classes, mass-delete stale or temporary data, or apply metadata, tagging, retention, or encryption changes. Learn about the latest in Storage Intelligence advisor here.
New product: Last but not least, a new version of AlloyDB for PostgreSQL brings together pioneering Google infrastructure — Colossus distributed file system, and Jupiter network — to power a new, no-compromises database architecture for the agentic era.

Practitioner guides and how-tos

How-to: Wish you could make GPUs and TPUs scattered around the globe behave as a single pool behind a single entry point? In this blog, we show you how to do just that. At the edge, the multi-cluster GKE Inference Gateway focuses on global, multi-region traffic distribution and high availability. Beneath that, the LLM-d router handles complex, memory-aware scheduling algorithms to keep utilization high. This architecture is deliberately runtime-, model-, and accelerator-agnostic, and in tests, routing traffic through the multi-cluster GKE Inference Gateway added less than 1% overhead.
Guide: Using or planning to use GKE on TPUs for AI model training or inference? Training massive Large Language Models (LLMs) or running high-throughput inference serving represents a significant investment in specialized AI hardware, such as Cloud TPUs and GPUs. To get the most out of every dollar spent, you need to understand workload lifecycle metrics in GKE. This detailed guide explains how to turn opaque cluster behaviors into actionable telemetry.

Customer and partner updates

GKE customer: Learn why gaming startup SeaVerse relies on GKE Agent Sandbox for its multi-tenant workloads, and how the platform helped it decrease its infrastructure costs by 60%.

Research, reports and deep-dives

In case you missed it, we’re also thrilled to share that Google has been named a leader, including achieving the highest score on either product or strategy, in three key analyst reports from Gartner and Forrester.

Google is a leader in The Forrester Wave™: Public Cloud Platforms, Q3 2026: Highest overall score of any cloud provider!
Google named a Leader in 2026 Gartner® Magic Quadrant™ for Strategic Cloud Platform Services: Positioned furthest for “Completeness of Vision” of all vendors evaluated.
Google is a Leader in the 2026 Gartner Magic Quadrant for Container Management: Positioned highest in “Ability to Execute” of all vendors evaluated.
Google Cloud achieved a Gold rating in the latest SemiAnalysis ClusterMax 3.0 report, which evaluates the reliability, performance, support, pricing, and security of GPU providers globally. See the full report for more.

August 2026

Product, technology, and tools updates

Product update: Filestore, Google Cloud’s first-party, secure, scalable NFS file service, has emerged as a popular storage platform for AI and agentic workflows, and now, it’s even better suited to the task, with a new backend storage layer built directly on Colossus, Google’s foundational distributed storage system. This new backend lets you provision IOPS independently from storage capacity, and is deeply integrated with GKE. In AI environments, this can help you service so-called agentic swarms — large groups of agents that need to read and write to a common dataset — without a drop off in performance. For more, check out the blog post.
New feature: gVisor sandboxes are now available in distributed Ray clusters on GKE. In partnership with Anyscale, we introduced an experimental library for Ray that brings gVisor, Google’s open-source application kernel, directly into distributed Ray clusters. gVisor provides lightweight environments with stronger isolation than ordinary containers, plus fast startup times and low memory overhead. To try out these sandboxing capabilities on GKE, head over to the Ray sandboxing User Guide.
Product update: Looking for high-performance, easy-to-use infrastructure on which to run a personal AI agent, but don’t want to spend a lot of money? New Cloud Run instances are dedicated, singleton compute runtimes on Cloud Run that won’t shut down when the agent is idle. Better yet, the cost to run a Cloud Run instance with 1 vCPU and 1 GiB of memory continuously for 30 days is just $5.70.

Practitioner guides, documentation and how-tos

How-to guide: Big news in Model Context Protocol (MCP) land: As of the 2026-07-28 specification, the protocol core is “completely stateless. The handshake is gone. The initialize / initialized handshake (SEP-2575) and the logical Mcp-Session-Id header (SEP-2567) have been removed entirely. Instead, every request is now self-describing and independent.” Whoa. Learn more about the changes that the latest MCP specification brings, and more importantly, how to implement them, in this Google Developers blog.
Guide: Real-time AI systems make a mess of traditional network load balancing techniques. “Instead of handling isolated requests, the backend has to manage a continuous, live bidirectional stream. You’re dealing with a constant stream of audio chunks, transcripts, model outputs, and synthesized speech flowing back and forth simultaneously.” Things only get worse when the user gets involved. “The server has to immediately halt its current speech generation, pivot to update the context, maybe trigger a new tool, and start drafting a different response; this must be done without dropping the connection.” For a new approach to managing load in the AI era, read Scaling real-time AI agents with session-aware load balancing.
How-to: Learn how to build an elastic, scalable LLM inference platform on GKE, even with a mix of different GPU accelerators. The proposed architecture combines Capacity Advisor and Compute Advisor, plus high-performance storage like RunAI:model streamer or GCPFuse with parallel downloads. Get all the details here.
Documentation: The thing about hosts with GPUs or TPUs is that you can’t use live migration to update them, setting up a maintenance challenge. In this new docs page, learn how to update accelerator-equipped hosts according to your tolerance for downtime for your training and inference workloads.
Documentation: Advanced Compute Images, or ACIs, are standardized image stacks for AI/ML and HPC infrastructure, so you don’t need to manually build your own custom images. In this new docs page, learn how to create an ACI image using the Google Cloud CLI, console, or SchedMD's Slurm workload manager.
Guide: AI workloads are notoriously difficult to architect, resource-intensive, and bursty, which can also lead to scaling bottlenecks and large pools of underutilized — or misutilized — compute resources. A new blog outlines the three main ways to achieve dynamic capacity management in Google Cloud: 1) scheduling capacity for planned downtime; 2) maintaining automated fallback capacity for unplanned downtime; and 3) relying on GKE’s core orchestration capabilities to automate resource allocation.

Customer and partner updates

Business orchestration software provider UiPath was dealing with spiky workloads, and wanted more predictable costs. To get there, it re-architected its infrastructure, moving from isolated clusters to a shared Google Cloud GPU fleet that included both A3 VM instances (NVIDIA H100 GPUs) for training with G4 VM instances (NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs) for inference. You can read more about their architecture here.
Mirendil, an frontier AI lab focused on accelerating AI development, announced that it is using AI Hypercomputer with both TPUs and NVIDIA GPUs to support its model pre-training and post-training applications.
Replenit, a retail CRM provider, built its AI decision engine in Google Cloud, using BigQuery, Gemini Enterprise Agent Platform, and open-source Gemma models that it runs on Cloud TPUs. This latter combination provided Replenit with 90% lower pipeline costs than their previous cloud provider, the company reports. Read the full case study for more.
Malachyte architected its AI-powered e-commerce recommendation platform on top of Bigtable, Managed Service for Apache Kafka, Pub/Sub, Compute Engine, and last but not least, GKE. See how it all comes together in this blog.

July 2026

Product, technology, and tools updates

Product update: Google Cloud Managed Lustre is now GA, and available in four distinct performance tiers that deliver throughput ranging from 125 MB/s, 250 MB/s, 500 MB/s, to 1000 MB/s per TiB of capacity — with the ability to scale up to 8 PB of storage capacity. The Managed Lustre solution is powered by DDN’s EXAScaler, combining DDN's decades of leadership in high-performance storage with Google Cloud's expertise in cloud infrastructure.
Product update: C4N network and storage optimized VMs are now GA. C4N is our first network- and block-storage-optimized VM series built to eliminate data-transfer bottlenecks. Powered by 5th Gen Intel Xeon Scalable processors and built on Google's Titanium offloading hardware, it achieves 400 Gbps network bandwidth, 95 million packets per second (MPPS), and up to 25 GiB/s of block storage throughput when paired with Hyperdisk Extreme.
New feature: GKE Dataplane V2 up to 15K Nodes with Network Policies (GA). This capability enables standard GKE clusters to scale up to 15,000 nodes while maintaining full active Network Policy enforcement, supporting the massive infrastructure needs of large enterprise and AI/ML customers.
New feature: Co-operative time-slicing in llm-d. If you’re running reinforcement learning (RL) workloads, you can now interleave independent RL jobs onto shared physical hardware, increasing aggregate accelerator duty cycles from a ~40% baseline up to 70% without impacting model convergence or accuracy.
New AI security tool: Looking to secure your AI supply chain on GKE, deploy AI workloads safely, and cut down on shadow AI? We open-sourced k8s-aibom, a lightweight, unprivileged Kubernetes controller that continuously monitors container clusters to automatically detect running AI runtimes (like vLLM and Triton) and generate standard CycloneDX Machine Learning Bill of Materials (ML-BOMs). Check out the k8s-aibom project and get involved.

Practitioner guides and how-tos

How-to guide: On July 27, Google announced Day 0 support for Moonshot AI’s Kimi K3 2.8-trillion-parameter open-weight model, the day weights were released. Whichever your preferred deployment path — via Model Garden, custom orchestration, or GKE with llm-d recipes — this guide offers detailed step-by-step instructions to help you evaluate and pilot Kimi K3 in Google Cloud.
How-to guide: Google Kubernetes Engine (GKE) managed DRANET supports both GPUs and TPUs. There are several configurations to use this implementation, including standard cluster (where you have full control) and autopilot cluster (where Google does the heavy configs for you). Take a deeper dive in the hands-on lab, GKE Autopilot clusters with TPUs, GKE managed DRANET and Gemma 4.
How-to guide: Learn to run Ray on TPUs, not GPUs. In Part 1 of this two-part series, we discuss TPU slices (hint: Ray thinks of them as just another accelerator on which to schedule), then walk through Ray’s various AI libraries (Part 2).
How-to guide: Evaluate TPUs for sample workloads using a new microbenchmark suite that helps you accurately assess whether a device is achieving its theoretical performance specifications, and to identify specific performance gaps or architecture-specific bottlenecks. Dive in here.
How-to guide: Scale your agents without killing your budget. Learn how GKE orchestration can help you safely pack more agents onto a fixed compute footprint with GKE Agent Sandbox and Pod snapshots. Whether your goal is performance or cost optimization, we teach you how to turn the right dials for optimal agent efficiency.
Technical blueprint: Inside the optimization of Mistral 3 large inference on Ironwood. This blog outlines how one Google team optimized Mistral 3 large MoE model inference on Google’s Ironwood (TPU v7x), achieving a 1.5x performance gain. They did so with hybrid sharding, replacing linear VPU summations with tree reductions, optimizing GMM/MLA kernels, and adopting asynchronous scheduling. As a result, they boosted throughput by up to 48% while maintaining benchmark accuracy neutrality. Read the full blog here.

Research, reports and deep-dives

Report: Google was named a Leader in the inaugural GartnerⓇ Magic Quadrant™ for AI Infrastructure, positioned highest for ‘Ability to Execute’ and furthest for ‘Completeness of Vision’. Gartner called out Google’s proprietary scalable compute, integrated AI Hypercomputer architecture, and the scale of our AI compute capacity as key strengths. Download a copy here.
Report: We recently surveyed more than 1,400 senior IT leaders for our State of AI Infrastructure report, and a resounding pattern emerged: The gap between AI ambition and infrastructure reality is widening. In fact, 83% of organizations say they require infrastructure upgrades to support production-grade agentic AI. Read the accompanying blog to understand how adapting your infrastructure to meet the demands that agentic applications place on your systems will help you move from pilot to production.

June 2026

Product, technology and tool updates

Product update: Protecting sensitive data used with AI is a critical part of advanced and secure cloud infrastructure. Confidential Computing cryptographically protects data in use in hardware-based Trusted Execution Environments (TEEs) with verifiable data integrity, and is now available on the accelerator-optimized G4 machine series, featuring NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. Get started with Confidential G4 VMs and Confidential G4 GKE Nodes.
Developer resource: The new TPU Developer Hub is the place to go for model builders, optimizers, and developers to learn to unlock the full performance of Google Cloud TPUs. Read more in this blog.
New product: Scale your AI workloads with the new OpenTelemetry-Based TPU AI Telemetry Collector Agent. For the first time, you can route high-fidelity TPU hardware telemetry to Google Cloud Monitoring, Google Managed Prometheus, or your own self-hosted Grafana stack.

Practitioner guides and how-tos

How-to guide: Learn how to build high availability into an AI inference workload running on GKE Inference Gateway with TPUs, Cloud Storage FUSE and Dynamic Resource Allocation (DRA). This blog provides an overview, or you can get all the technical details in the hands-on codelab.
How-to guide: Did you know you can connect your AI agents to unstructured data in Cloud Storage via Model Context Protocol (MCP)? In this blog, learn about why would want to do that from three customer examples, then how to do it, choosing either a fully managed service, or a self-managed local server for more customization and control.

Research, reports and deep-dives

Report: According to an independent benchmark report, GKE Inference Gateway outperforms the next leading managed Kubernetes service with 15.7% higher throughput, 92.8% shorter wait times, and 62.6% lower inter-token latency. This performance can be attributed to its use of prefix caching, which optimizes LLM performance by storing the KV cache (activation states) of long, repetitive prompt prefixes. Learn more in the blog.
Architecture deep dive: A closer look at the cold start problem, this time for TPUs and GKE, and how the Run:ai Model Streamer can help change the dynamic.

Customer and partner updates

Customer win: Leveraging GKE, BigQuery, Cloud SQL, and Gemini Enterprise Agent Platform, Pager Health is eliminating operational fragmentation to deliver a simplified, personalized U.S. healthcare experience that transforms lives.
Customer win: Trustpilot, the customer review platform, built a high-volume streaming pipeline using fine-tuned Gemma models with Dataflow and Gemini Enterprise Agent Platform running on cost-optimized A2 VMs using A100 GPUs, as well as optimized version of vLLM maintained by Gemini Enterprise Agent Platform.

May 2026

Product, technology and tool updates

Product update: GKE Agent Sandbox is now generally available.
New open-source project: Agent Substrate is a new open-source project aimed at continuing to push the limits of agentic infrastructure density
New feature: Google AI Edge Portal, a solution for testing and benchmarking on-device machine learning (ML) at scale, now supports benchmarking and debugging on-device LLMs. Read more here.
Product deep dive: We went into depth about Cloud Storage Rapid, a new family of high-performance storage offerings for AI workloads. At launch, offerings include Rapid Bucket (formerly Rapid Storage), a high-performance zonal object storage offering, and Rapid Cache (formerly Anywhere Cache), which accelerates reads on-demand and colocates compute and data for workloads in existing buckets.

Research, reports and deep dives

Architecture deep dive: Google Global Infrastructure VP Bikash Koley and Engineering Fellow Arjun Singh provide a high-level overview of the challenges that AI workloads pose to network infrastructure, and discuss the deep enhancements we’ve made to our data center fabrics, WAN, and global networks to better support them.
Architecture deep dive: We unveiled a new cluster-level reliability model for developing frontier AI models on TPUs, ditching instance-level reliability

Customer and partner updates

Customer win: Visual media provider Imgix serves more than 8 billion images and videos from AI Hypercomputer equipped with G4 VMs powered by NVIDIA RTX PRO 6000 Blackwell GPUs.

Empower your agents with the Google Cloud CLI remote MCP server

Wed, 30 Sep 2026 16:00:00 +0000

Today, we’re expanding our ecosystem of managed remote MCP servers by introducing the Google Cloud CLI remote MCP server in preview.

Powered by the popular gcloud and bq (BigQuery) command-line tools, this new server gives AI agents immediate, broad access to command-line operations for managing Google Cloud infrastructure and working with advanced BigQuery workflows securely and seamlessly.

Why CLI matters for AI agents

Agents are increasingly performing complex cloud operations, but standardizing how they interact with backend systems remains a challenge. The Google Cloud CLI remote MCP server bridges this gap by packaging the versatility of hundreds of gcloud and bq commands into one single MCP server. This results in two strong benefits for the agent:

Higher-level abstractions: CLI commands package complex multi-step workflows, validation checks, and high-level operations into unified commands rather than requiring multi-step API orchestration.
Leverages model training: LLMs are heavily pre-trained on public command-line documentation, syntaxes, and usage examples, making CLI invocation intuitive and highly accurate for models.

The benefits of putting CLI behind remote MCP

Managing cloud infrastructure with AI agents traditionally requires installing and maintaining Google Cloud CLI binaries inside agent execution environments. The Cloud CLI remote MCP server bridges CLI capabilities with MCP benefits by providing an isolated execution sandbox on Google Cloud infrastructure. This solves key infrastructure challenges:

Simplified dependency and runtime management: For teams building custom agents, maintaining local CLI versions and dependencies across dev, test, and production environments creates operational overhead. Remote MCP eliminates local installations and runtime maintenance.
Access for web-based agent endpoints: Web-hosted agent platforms and web interfaces (such as Gemini Enterprise and other hosted enterprise agent platforms) run in environments where users cannot control or install local packages. Remote MCP enables secure, managed access to Google Cloud CLI operations directly from these surfaces.

Enterprise-grade security and governance

Connecting an AI agent to your infrastructure requires strict, enterprise-ready safeguards. This remote server leverages Google Cloud's standard identity and governance frameworks to keep your environments secure:

Zero ambient credentials: The server isolates execution in a network-restricted proxy boundary with no ambient credentials. Authentication and authorization are handled through Agent Identity, OAuth 2.0, and Identity and Access Management (IAM).
Strict policy enforcement: Every command executed through the remote MCP server is run with the permissions of the authenticated caller identity. Both standard IAM permissions and organization policy service constraints are strictly enforced against downstream target resources.
Advanced protection with Model Armor: To minimize the risks associated with AI tool calling, the Cloud CLI remote MCP server integrates with Model Armor. You can proactively screen LLM prompts and responses to protect against risks like prompt injection and malicious inputs.
Cloud audit logging: The Cloud CLI remote MCP server can be configured to log every tool invocation to Audit Logs (Data Access logs under cloudcli.googleapis.com/mcp). Security teams can gain full visibility into caller identities, OAuth clients, and IAM authorization decisions (mcp.googleapis.com/tools.call) without exposing sensitive command payloads or personally identifiable information (PII).

Connecting to the Google Cloud CLI Remote MCP Server

Integrating cloud management into your agents no longer requires packaging Google Cloud CLI binaries, managing local execution runtimes, or maintaining dependencies inside agent container images. Because the Google Cloud CLI remote MCP server implements the standard Model Context Protocol, any MCP-compatible agent platform or orchestration runtime can connect immediately via standard configuration:

code_block: <ListValue: [StructValue([('code', '{\r\n "mcpServers": {\r\n "google-cloud-cli": {\r\n "uri": "https://cloudcli.googleapis.com/mcp",\r\n ...\r\n }\r\n }\r\n}'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7fd61b49ca10>)])]>

Your agent immediately gains access to execute gcloud and bq commands in a secure, network-isolated cloud sandbox.

Authentication is handled via keyless Agent Identity for hosted Google Cloud platforms, or standard OAuth 2.0 for external runtimes. For authentication options, see the MCP Authentication Guide.

Bringing infrastructure management to agents

The Cloud CLI remote MCP server exposes two powerful tools, run_gcloud_command and run_bq_command, giving your AI agents broad, immediate access to Google Cloud operations through natural language.

Managing cloud infrastructure with run_gcloud_command

With run_gcloud_command, agents can execute the full breadth of gcloud operations to manage, diagnose, and secure your Google Cloud environment. An example follows:

Observability and incident diagnostics: An agent streamlines incident diagnostics by automating command execution and reducing context-switching across tools.

Extending BigQuery operations with the run_bq_command tool

While the BigQuery MCP server already helps organizations analyze and explore data using AI agents, with the introduction of run_bq_command, agents can now tackle advanced BigQuery tasks such as resource allocation, job monitoring, and task scheduling by unlocking the full scope of bq CLI functionality. Key capabilities include:

Automating scheduled queries: An agent utilizes BigQuery Data Transfer Service configurations to schedule queries automatically.
Job and resource management: Gain deep insight into query execution details, including processed data volume, slot usage, and execution plans, as well as managing reservations.
Access and permissions control: Data administrators and owners can inspect and update table permissions directly through the agent.

Pricing and availability

The Google Cloud CLI MCP server is available today in public preview. There is no additional charge to use the MCP server itself. You pay only for the GCP resources you create and any applicable data transfer costs.

To get started, enable the Cloud CLI Execution API (`cloudcli.googleapis.com`) in your Google Cloud project, grant the required MCP Tool User (`roles/mcp.toolUser`) IAM role to your agent or user identity, and configure your MCP client to connect to `cloudcli.googleapis.com/mcp`.

Why your startup needs open models alongside frontier APIs

Mon, 28 Sep 2026 16:00:00 +0000

Every week, I talk with founders who are building at an unbelievable pace. Teams are moving from inception to product-market fit faster than ever, with foundation models wired deeply into their core product workflows.

Yet as startup architectures mature, a clear divide has emerged between teams struggling with margins and those scaling sustainably. The most effective engineering teams have abandoned the one-size-fits-all model strategy.

In the early days of LLMs the default architecture was simple: send every interaction to the largest model available. But as applications move into production, serving millions of people and running autonomous multi-agent workflows, relying on a single frontier model starts to strain in three places:

Latency penalties: Relying entirely on cloud round trips makes it difficult to deliver the sub-second responsiveness that interactive mobile and desktop apps require.
Infrastructure overhead: Self-hosting large open models with more than 70 billion parameters forces early-stage teams to act like infrastructure providers, pulling senior engineers on cluster provisioning and multi-GPU orchestration.
Margin erosion: Sending high-frequency, structured tasks (like intent routing, JSON extraction, or status validation) to general-purpose frontier endpoints spends capital that could be funding product differentiation.

Great engineering teams pick the right tool for each job. Most production requests don’t require a frontier generalist, and routing every call to one can actually slow your product down. Instead, the winning pattern is a compound AI stack: pairing frontier models for complex synthesis with compact, open-weight models that you can tune, control, and run anywhere.

It’s for these reasons that an open model like Gemma belongs in your model lineup. With more than one billion downloads across the developer community, Gemma 4 is our most capable open model family to date, using the same foundational research and technology behind the Gemini models.

Built under one roof

Gemma is built by Google DeepMind using the same foundational research and architecture advances behind the Gemini models. Because they share common DNA and developer tooling, your team can prototype in Google AI Studio and design hybrid architectures where Gemini and Gemma work together.

Released under a commercially permissive Apache 2.0 license, Gemma 4 is engineered for parameter and token efficiency. Rather than forcing a single model architecture onto every hardware target, Gemma 4 spans five sizes across four specialized architectures: compact E2B and E4B models with native audio and vision for mobile and edge devices; an encoder-free 12B Unified multimodal model; a 26B A4B Mixture-of-Experts (MoE) model that activates only 4B parameters per token for high-throughput serving; and a dense 31B model that fits on a single GPU for maximum reasoning quality and fine-tuning. Every model includes configurable thinking modes, native function calling, up to 256K context, and built-in Multi-Token Prediction (MTP) draft models for speculative decoding.

Real proof: How startups are winning with Gemma

Founders are using Gemma to solve urgent problems around unit economics, output accuracy, and responsiveness.

Flipping the architecture: Cue is a voice-activated desktop assistant that runs natively on a user's machine to automate everyday tasks. They integrated Gemma 4 E4B via Ollama on local hardware to handle real-time transcript formatting. While they originally planned for Gemma to be a weak offline fallback, benchmarking proved it was so fast and precise that they made it their default engine—driving a 44% latency drop (from 876 ms to 488 ms).
True edge independence: Mobile development studio HubX built BetterSpeak, a voice-based interactive mobile English-learning tutor that simulates immersive, real-time voice conversations. To bypass cellular network lag and avoid charging users expensive subscription fees to cover cloud hosting, they packaged a 4-bit quantized Gemma 4 E2B model (~2.9 GB) natively on-device. The result is an offline, speech-to-speech mobile tutor that costs them $0 in server bills.
Scientific discovery and air-gapped security: K-Dense has built Faraday, an AI-powered scientific collaborator optimized end-to-end across hardware, software, and sensor suites, powered by Gemma 4 together with K-Dense's Scientific Agent Skills. Faraday runs fully air-gapped, making it suitable for secure, proprietary scientific work in pharma and biotech. Deployed on an NVIDIA DGX Spark, Gemma 4 can also be fine-tuned locally on a user's own proprietary datasets.
Unlocking infinite gameplay and retention: Gaming company Latitude integrated Gemma across their AI-native game products. By swapping in Gemma for AI Dungeon, they significantly improved player retention, while their new AI RPG platform Voyage leverages Gemma to deliver high intelligence at a cost that enables unlimited user gameplay with ultra-fast latency.

Four workloads where Gemma wins for startups

If you’re evaluating where Gemma fits into your stack today, start with these four jobs:

1. Edge and local execution (low latency, true privacy)

If you’re building mobile apps, developer desktop tools, robotics, or offline-first experiences, every cloud round-trip adds latency that people can feel. Gemma can run directly on laptops (including Apple silicon), smartphones, and local appliances. Your users get immediate feedback, and sensitive data never has to leave their device.

You can handle many local interactions on-device for zero incremental cost, and keep a bridge to frontier models in the cloud for the requests that need it. When a local workflow calls for large-scale multimodal reasoning, long-context data synthesis, or complex planning, your application can route that specific request to Gemini.

2. High-throughput triage and agent routing

In multi-agent architectures, agents spend a surprising amount of tokens on simple tasks like checking statuses, classifying intent, and routing tickets. With Gemma as your front-line gatekeeper, those high-volume background tasks run on a compact model and your team can save frontier reasoning for the requests where it creates product value.

3. Task-specific fine-tuning for real moats

Adapting a model to your proprietary data is one way to build a competitive moat. Because Gemma gives you full access to model weights and has a compact memory footprint, your team can run parameter-efficient fine-tuning (LoRA or QLoRA) on a single GPU in hours rather than days.

4. Turnkey vertical starting lines

DeepMind releases domain-specific variants of Gemma, so you don’t have to start from scratch. One example is MedGemma. MedGemma scores 87.7% on the MedQA benchmark, matching the clinical accuracy of frontier models at roughly one-tenth the inference cost. In a blind clinical study, board-certified radiologists judged that 81% of chest X-ray reports generated by the lightweight MedGemma 1.5 4B were accurate enough to result in equivalent patient management compared to reports written by human experts.

Beyond healthcare, biotech startups use C2S Scale to model virtual cellular responses and accelerate oncology research. Meanwhile, DataGemma cross-references more than 240 billion public data points to help reduce numerical hallucinations. If you’re operating in a specialized market, starting with a model that already speaks your industry's language can save you engineering time and compute budget.

Deploy wherever your business lives

Gemma is designed to fit into your existing engineering stack without lock-in:

Apache 2.0 licensing: Gemma 4 ships under the Apache 2.0 license, giving startups the freedom to fine-tune, quantize, redistribute, and deploy commercial products on-premises or at the edge with full ownership of their custom weights.
Day-zero open tooling: Run and fine-tune Gemma with the tools your engineers already use, including vLLM, Ollama, llama.cpp, LM Studio, MLX, Unsloth, Hugging Face, Kaggle, Keras, PyTorch, JAX, and LiteRT-LM.
Serverless and managed cloud deployment: Prototype immediately in Google AI Studio, scale to zero on serverless GPUs with Cloud Run, or deploy dedicated endpoints from Model Garden on Gemini Enterprise Agent Platform when traffic surges and you don’t want to manage GPU clusters.
Enterprise-ready safety: Gemma undergoes rigorous pre-release safety evaluations, data filtering, and red-teaming, and pairs with ShieldGemma 2 to help you meet enterprise compliance requirements.

Build with Gemma: What to do this week

Great technical architecture isn't about finding one model to do everything. It’s about assembling the right tool for each job so you can move faster, protect your runway, and ship a superior product.

Here’s my challenge to your engineering team this week:

Audit your model calls: Look at your logging dashboard and identify three high-volume, deterministic tasks (such as intent classification, JSON validation, or summarization) currently running on your most expensive models.
Benchmark Gemma: Run a quick test with a compact Gemma model locally or on a single endpoint. Measure the latency and calculate what happens to your gross margins when that workload runs with lower inference cost.
Redirect your runway: Take the capital and engineering hours you save on compute and invest them back into your core differentiators.

You can download the Gemma weights directly or deploy them through Model Garden. If you need compute credits and technical architecture reviews to get up and running, the Google for Startups team is ready to help you build - learn more.

Best practices guide for customizing Gemini models via Reinforcement Learning (RL)

Fri, 25 Sep 2026 16:00:00 +0000

Reinforcement learning (RL) has been a keystone of modern LLM post-training, but it demands large training clusters and access to model internals that external customers can't have with proprietary models like Gemini. So here at Google Cloud, we packaged it into a managed RL fine-tuning service (RLFT service) — you bring prompts and a reward function; we handle the infrastructure and the proprietary model internals.

Now, you can adapt Gemini with the service — teaching the model from a reward signal you define, rather than from a fixed set of labeled answers. This unlocks a class of problems that supervised fine-tuning (SFT) struggles with: tasks that are hard to demonstrate but easy to score.

In this guide, we will walk through practical best practices for using RL fine-tuning service. We'll start with a short tour of the RL training loop, how to decide if and when to use RL, and introduce how to get the most value from this approach.

What is RLFT?

RLFT adapts Gemini from a reward signal you define rather than labeled answers. Instead of authoring a large set of gold examples, you write one program that scores a response and the service improves the model against it — unlocking tasks that are hard to demonstrate but easy to verify: you can't hand-write the ideal SQL for every schema, but you can run the query and check the result.

At each training step the service generates multiple candidate responses to your prompts, scores them with your reward, and improves the model so that higher-scoring responses become more likely while it stays close to the original Gemini. The reinforcement learning that makes this work is fully managed — you never configure it. The one thing you own, and the thing that most determines your results, is the reward.

Three properties define what RLFT can and can't do:

It learns from the model's own outputs: It refines what the model already produces rather than copying an external target, so it tends to disturb unrelated capabilities less than SFT.
It rewards outcomes, not paths: Any response that reaches a good result earns reward, which fits open-ended tasks with many valid solutions.
It amplifies existing competence: It makes occasional success reliable, but it can't teach a skill the model never demonstrates.

When to use RLFT

Prompting and SFT handle most adaptation; exhaust them first. RLFT earns its keep when you can grade a response but can't cheaply author it, when SFT has plateaued on the metric that matters (faithfulness, schema validity, tone), or when the task has many equally valid answers a single reference target would wrongly penalize.

SFT and RLFT are complementary, not competing:

Direct RLFT when the base model already succeeds part of the time — enough for the reward to tell better answers from worse ones.
Two-stage SFT → RLFT when you have SFT data or the base success rate is too low for RL to gain traction. Use SFT as a short, cheap warm start — kept light, since over-fitting the demonstrations leaves less room for RL to improve — then continue into RL via Continuous Tuning, which initializes RL from the SFT checkpoint.

Across early adopters, these patterns show where RLFT delivers the most value — each scoring an outcome the business cares about but could never cheaply demonstrate.

Use cases for RLFT

AI-powered NPCs in games

What: In-character, on-brand dialogue held across long, multilingual, multi-turn conversations.
Problem: Off-the-shelf models break immersion — wrong language, hallucinated items, ignored players, repetitive loops.
Objective and reward: A Gemini autorater (LLM-as-a-judge) scores each turn on persona, flow, and game-state syntax, penalizing format and language errors.
Results: Loops and language drift disappeared and state syntax held, making shippable in-game characters viable at scale.

Structured entity extraction

What: Pulling a set of items from unstructured documents, such as supplier invoices and shipping manifests, into structured records automatically.
Problem: The long tail where SFT plateaus — missing required fields (recall) or inventing ones that aren't there (precision).
Objective and reward: A rule-based precision/recall reward forces every field to be grounded in the source text, not imitated from one gold answer.
Results: Field-level accuracy rose on noisy real-world documents where tuning had stalled, turning a manual review step into an automated one.

Content moderation

What: Applying intricate policies and decision trees at scale.
Problem: Models hallucinate false positives or reward-hack with invalid formats to dodge evaluation.
Objective and reward: A Cloud Run reward pairs format validation with a deterministic grader to enforce multi-step policy adherence.
Results: The model handled complex exemption carve-outs, sharply cut false positives, and stopped reward hacking — reducing the human-escalation volume that makes moderation expensive.

Code measured by execution

What: SQL or API calls graded on whether they actually run against customer data.
Problem: SFT mimics one reference query and breaks on unseen proprietary schemas.
Objective & Reward: A code-execution reward runs the code in a secure sandbox and pays out only if it compiles, executes, and returns the correct result.
Results: The model produced first-attempt executable queries at closed-frontier quality and lower inference cost, letting non-technical users query proprietary data in natural language.

Presentation slide generation via HTML

What: Multi-slide decks authored as HTML/CSS.
Problem: Training on text alone is blind to visual quality — overflows, clipped elements, and inconsistent styling slip through unnoticed.
Objective & Reward: A code-execution reward renders the slides and scores visual design, layout integrity, structural completeness, and rubric adherence.
Results: The model emitted modular, well-styled decks with cohesive themes and no layout overflow.

Where to start?

A dataset. A diverse set of prompts with a held-out validation split is enough for a first run — confirm the loop converges and reward moves the right way, then scale. Keep train and eval strictly separated; a contaminated eval hides overfitting.
A reward function. Your task specification as code or configs, and the dominant driver of quality. A good reward correlates with human preference, is robust to malformed output (catch the failed parse and return a clearly negative score rather than crashing), and resists reward hacking — ensemble judges, penalize length, floor degenerate outputs, and prefer a verifiable check over a model's opinion. Validate it offline before launch.

The service handles the rest; start from the defaults, watch reward and eval curves in the console, and take the checkpoint where validation reward saturates rather than the last step.

Get started today

What will you build? The tools are ready and waiting.

Documentation: Reinforcement Learning Fine-Tuning

Power your agents: Gemini 3.8 Live with Live Avatar is now generally available

Thu, 24 Sep 2026 15:00:00 +0000

Following our announcement of Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking last week, we are thrilled to share that Gemini 3.8 Live with Live Avatar is now generally available in Gemini Enterprise. First previewed at Google Cloud Next 2026, the technology is now officially ready for enterprise production.

As enterprise voice AI evolves beyond basic speed and cost metrics, our priority has shifted to making each interaction even higher quality. Gemini 3.8 Live already delivers a native speech-to-speech foundation for fluid, responsive dialogue. The Live Avatar feature brings an interactive visual presence to conversational video agents across web, mobile, and interactive kiosks. Together, you’ll have access to:

Video avatars: Conversational video with the Live Avatar feature can generate video avatars with synchronized lip-syncing. Note: Custom avatar feature is available via allowlist only.
Fluid dialogue: Native speech-to-speech means more natural interruption recovery without dropping conversation context or backend transactions.
Tool calling: It executes tools and API calls in the background while continuing the conversation, so the model can acknowledge requests and keep chatting while tasks finish in the background.
Breaks language barriers: Gemini 3.8 Live understands and speaks 97 languages, with automatic language detection.
Make it easy for agents to see what your user sees: Live visual understanding can process live camera feeds and screen shares alongside audio, all at the same time.

Though Gemini 3.8 Live Extended Thinking remains in private preview, Gemini 3.8 Live with Live Avatar is now available with US and EU endpoints, with provisioned throughput, enterprise compliance, and strict data governance. Try the model and its features in Gemini Enterprise, and start building now with API.

Trust and transparency at its core

To safeguard identity and prevent misuse, customers can deploy from a library of curated, pre-built avatars, while custom avatar creation is gated behind a strict enterprise allowlisting and verification process. Furthermore, all generated audio and video streams carry imperceptible SynthID watermarks, ensuring AI-generated content remains transparent and verifiable.

Three demos of Gemini 3.8 Live with Live Avatar in action

#1: Interactive custom avatar

^{Create an interactive custom avatar in a step-by-step process}

Watch how Gemini 3.8 Live makes it possible to build a custom avatar by adding system instructions, uploading a single reference photo and audio file sample.

Demo #2: Live video understanding in a voice-first claims intake

^{Streamline claims intake with live video understanding using Gemini 3.8 Live}

In this demo, you’ll see a common use case come to life using Gemini 3.8 Live: an intake agent. In this scenario, we chose an insurance claims agent. You talk and show the damage on camera, and the claim notebook fills itself in as you go. In the background, an ADK agent team checks the policy, applies the intake rules, and builds the adjuster packet. To dive deeper, check out the open-source code.

#3: A real-time voice AI agent with Google ADK and Gemini Live API

^{Build a live voice agent using Google ADK and Gemini Live API}

See how developers can use Agent Development Kit (ADK) to define agents, manage runners and session memory, and stream real-time audio directly to the Gemini Live API without a traditional speech-to-text pipeline.

How our customers are innovating with Gemini 3.8 Live with Live Avatar

Cox Automotive built an AI-powered shopping assistant for Autotrader that uses live screen-highlighting and tool-calling capabilities to guide car shoppers through vehicle search, comparison, and financing — in real time, through natural conversation.

“Shoppers increasingly expect to describe what they need in their own words rather than work through filters and menus. Autotrader’s new conversational AI Avatar brings that experience to vehicle discovery by matching natural conversation to the right inventory. It is another step toward our vision of connected intelligence, where every consumer interaction draws on the full depth of Cox Automotive data.” — Marianne Johnson, EVP and Chief Product Officer, Cox Automotive.

“We're building a personal AI that knows you, speaks your language, and is always on your side. Today, it handles over a million live calls daily across nine Indian languages. Gemini 3.8 Live improved interruption handling, multilingual conversations, and tool-call reliability. This AI doesn't just answer calls; it gets things done for you.” — Akhilesh Damaraju, CEO, Equal AI.

“We're excited that Gemini 3.8 Live and Agentforce are coming together to reimagine what's possible in intelligent service. This collaboration between Salesforce AI Research and Google combines real-time, multimodal capabilities with agentic AI to explore new ways to create richer, more intuitive customer experiences from first contact to resolution." — Bob Van Osten, VP of Product for Agentforce, Salesforce.

“Our team has been very impressed with Gemini 3.8 Live throughout testing and benchmarking! The updates made to Voice Activity Detection and the improvements to overall latency are huge steps forward and further our ability to deliver the highest quality AI Assistant on the SPECS platform.” — Eric Walsh, Software Engineer, Specs.

Start building today

Gemini 3.8 Live with Live Avatar are available now for enterprise customers:

Try the model in Gemini Enterprise
Access the Gemini Live API documentation for integration guides
Review pricing on our pricing page
Skill for building real-time, bidirectional streaming apps with the Gemini Live

If your application requires specialized, modular audio capabilities, explore our other audio models:

Reach out to your Google Cloud sales representative to activate provisioned throughput, discuss customized deployment architectures and allowlisting for custom avatar.

How growing Latin American midsize businesses are building in the AI era

Thu, 24 Sep 2026 14:30:00 +0000

Latin America’s small and medium-sized businesses are the heartbeat of the region's economy — accounting for more than 60% of total employment in the region, according to United Nations estimates. And just like their enterprise peers, everywhere you look, ambitious teams are moving fast to embrace AI.

Many have already transitioned from experimenting with generative tools and agentic workflows to using them every day to work smarter, save time, and deliver exceptional customer experiences. These growing businesses are particularly focused on maximizing the benefit they get from their investments in AI, whether that’s using a fast, low-cost model to summarize daily emails or deploying an advanced model for complex data analysis, teams can match the right AI capability to their exact task and budget.

It’s this range of options, and a familiarity with the broader suite of Google business, media, and advertising tools that has led many SMBs to choose Google Cloud, and Gemini Enterprise in particular, as their AI platform of choice. By doing so, they’re able to build custom AI agents, streamline daily tasks and paperwork, and offer customers instant support with the speed and reach needed to compete on a global scale.

With our unique front row seat, we’ve seen the benefit SMBs are getting from leveraging Gemini Enterprise, not only for generative AI, but as a catalyst for adopting other essential cloud tools like Google Kubernetes Engine and BigQuery for complete end-to-end modernization. The number of Latin American-based small and medium businesses using Google Cloud AI tools has grown 8x year-over-year and the number of Brazil based small and medium businesses using Google Cloud AI tools has grown 9x year-over-year. This rapid adoption spans our Gemini models, Gemini Enterprise, and core Cloud infrastructure, and are helping businesses to:

Roll out better customer support systems to help escalate and resolve customer support calls more quickly.
Automate repetitive actions in areas like payroll and accounting.
Help more employees understand and leverage data at work — even those not trained as data analysts.
Rapidly create and implement new designs for marketing collateral.
Help more people build their own AI agents to help them in their everyday jobs.

As we head into today’s Google Cloud Summit in Brazil, we were proud to showcase nearly 20 of our newest Latin American SMB customers using Google AI to reduce busywork, serve their customers faster, and grow their businesses.

Announcing new Latin American customers putting Google AI to work

AdGoat, an Argentina-based adtech company processing more than 10 billion annual ad requests across more than 100 global websites. It uses Cloud Run, the Gemini API, and Gemini Enterprise to automate content analysis, ad bidding, and audience targeting to help e-commerce brands drive higher campaign returns.
Angelus, a Brazilian dental and healthcare manufacturing company, uses Gemini Enterprise to streamline project management across its research and development department. This enables its teams to automatically pull technical project data into pre-approved templates aligned with the company’s brand identity and regulatory requirements.
BunkerDB, a marketing science company operating across Latin America, uses Gemini Enterprise, Cloud Run, and Cloud Storage to power an AI platform that organizes marketing assets, checks brand compliance, generates or adapts multimodal content, and predicts ad performance before launch. All of this helps it reduce creative turnaround times from weeks to hours and cut cost per lead by up to 25%.
Caffeine Army, a Brazilian wellness and high-performance company that connects people with solutions in nutrition, sports, and well-being, deployed BigQuery and Gemini Enterprise on Google Cloud to unify customer purchase insights, enabling faster creative campaign turnarounds and boosting team productivity across the organization.
Convert, a Brazilian marketing and analytics provider, uses Looker, BigQuery, and Cloud Run to power five specialized AI agents that answer complex business questions in natural language, speeding up report deliveries by 65% and reducing operational costs by 32%.
Growth Digital, a Google Ad sales rep operating across 13 Latin American countries, used BigQuery and Gemini Enterprise to build over 113 AI agents, enabling teams to build proposals 5x faster, cut campaign reporting time by 80%, and reduce financial error rates to under 0.01%.
GrupoTusMaquinas.com, an equipment management platform based in Chile, deployed Google Cloud AI tools and Gemini models to create digital tracking profiles for trucks and machinery, allowing businesses to query fleet status in plain language and manage vehicles regardless of brand or location.
HealthAtom, a healthcare technology company, uses the Gemini API, Firestore, and Cloud Functions to power AI assistants across its clinical platforms, automating appointment scheduling and medical record reviews while supporting 80 million annual patient interactions.
KLog.co, a Chilean logistics technology company digitizing freight forwarding across Latin America, uses Gemini Enterprise, BigQuery, and Google Workspace to automate cargo tracking and shipping paperwork, cutting manual data entry errors by over 90% and increasing document processing capacity tenfold.
NEEOH, a leading Brazilian out-of-home advertising communication platform, uses Gemini Enterprise to standardize secure AI usage across its organization, enabling teams to generate campaign copy and build pitch proposals faster while keeping corporate client data secure.
Luxia Agro, an Argentinian foreign trade supplier of crop protection products, uses Gemini 3.5 Flash and Gemini Enterprise to automatically pull key details from complicated shipping emails and update their central business systems. This allows it to automate 80% of foreign trade operations and cut manual processing errors in half.
Macal, a Chilean auction company, uses the Gemini Enterprise, Cloud Run, BigQuery, and Security Command Center to automatically verify property records and modernize its technology systems, cutting software development times from weeks to days and lowering infrastructure costs by up to 30%.
Ninecon, a Brazilian tech consulting firm, deployed Gemini Enterprise to integrate AI directly into employee workflows, allowing managers to track usage patterns and optimize project turnaround times with real-time insights.
Nova Gestões, a customer service and operations provider in Brazil, uses Cloud Speech-to-Text and Gemini Enterprise to translate and analyze 100% of customer calls in real time, reducing post-call manual data entry and boosting team productivity by 30%.
Romi, a Brazilian industrial machinery manufacturer, uses the Gemini API and Gemini Enterprise to power an interactive chat assistant directly on CNC machine HMI (human machine iInterface), giving factory operators instant answers grounded in official manuals and generating QR codes for step-by-step instructional videos.
Supermercados El Dorado, a leading supermarket chain in Uruguay, leverages Google Compute Engine and Gemini Enterprise to modernize legacy testing infrastructure and connect custom AI agents within their daily workflows, boosting team productivity across departments.
Tryvia, a Brazilian IT and business solutions provider, uses Google Cloud, Looker, and Gemini Enterprise to move off legacy physical servers, giving teams real-time reporting dashboards and AI tools that speed up software development and daily tasks.
Via Cristais, a major highway operator in Brazil, leverages Google Contact Center as a Service to speed up emergency routing for highway accidents, reducing caller wait times, improving driver satisfaction, and mitigating the impact of call center staff turnover.
WeSpeak, an AI conversational platform for the hospitality industry in Latin America, uses Cloud Run, Gemini Pro, and Gemini Flash to automate end-to-end guest interactions across messaging channels like WhatsApp and Instagram. This has helped it achieve an 85% resolution rate and a 2x increase in overall sales volume for hotel clients.

Helping your team build AI skills

To help growing teams get the absolute most out of AI, we’ve created easy, no-cost learning programs that anyone can use:

Programs for small and medium businesses: Explore beginner-friendly training paths or join specialized programs to learn how to build custom AI assistants for your day-to-day work.
Google skills for organizations: Access thousands of free, on-demand AI courses and hands-on practice labs designed by experts at Google Cloud and Google DeepMind.
Get certified: Help your staff gain industry-recognized AI certificates through guided courses, expert mentoring, and skill badges.

By offering easy-to-use tools and free training — from everyday office apps in Workspace to advanced AI on Google Cloud — Google is here to help Latin American businesses thrive today and in the future.

The three things today's hottest startups are looking for in their AI stack

Thu, 24 Sep 2026 13:00:00 +0000

Google Cloud has become the platform of choice for startups building AI.

Our uniquely complete stack — including a choice of first- and third-party compute and models; our platform for building and managing agents; and our products for securing AI workloads — has emerged as the single most important driver of this growth and it is powering AI development for many of the most exciting and innovative startups in the world.

As a result, startups are choosing to build and run on Google Cloud at a higher rate than they were three years ago, at the start of the AI era.

Given how quickly the technology industry moves in the AI era, the choices startups make can be notable. As we’ve worked together and watch many of these leaders scale, we’ve observed a few important trends emerging over the past several months. We expect these decisions will continue to shape the choices startups make about the platforms and technology they use:

Gemini Enterprise, which includes our tools for managing TPU and GPU clusters, services for building and managing agents, and APIs to access both first- and third-party models, is growing significantly with startups. And when startups use Gemini Enterprise, they also tend to use our “core cloud” services like Storage, BigQuery, or GKE.
Gemini models — as well as several of the third-party models available through Gemini Enterprise — are providing very strong price-performance for startups. These customers are increasingly deploying both our frontier models and “workhorse” models as their AI to power workloads as diverse as scientific research, generative media creation, and financial analysis.
Access to compute on GPUs and TPUs is critical for AI and the ability to choose one — or both — is unique to Google Cloud. But importantly, startups almost always use additional products from our stack alongside these chips, like models, tools for building agents, or services like BigQuery or GKE. These additional technologies illustrate how the needs of startups are rarely singular, and just how much value they find in having ready access to a strong suite of second-, third-, and fourth-level technologies beyond just compute.

We can see the demand for these technologies first-hand in some of the recent deals we have struck in the past 60 days with a number of leading startups across sectors:

Artificial Agency, a startup focused on generative behavior in games, is running critical AI workloads and research on Google Cloud, where it is using NVIDIA GPUs for model training and inference, as well as Gemini models and Cloud Storage.
Arya Health is building the AI workforce for healthcare, deploying agentic AI to perform the non-clinical administrative work that limits providers’ ability to deliver and expand care. Arya’s AI agents work across scheduling, intake, recruiting, onboarding, compliance, after-hours operations, and other critical workflows, interacting through voice, text, email, and providers’ existing systems. Arya uses a range of Gemini models across its agentic infrastructure, including Gemini 2.5 Pro, 3.1 Flash, and 3.5 Flash Lite, selecting models based on the reasoning, speed, and cost requirements of each workflow.
Casco is a cybersecurity startup whose autonomous agent swarms execute sophisticated, multi-step attacks to uncover vulnerabilities across enterprise applications, cloud environments, and infrastructure. Its architecture combines advanced reasoning models for complex, long-running tasks with fast models such as Gemini 3.5 Flash for focused subagent work. Google Cloud’s model portfolio and infrastructure, including Provisioned Throughput, help Casco match each workload with the right combination of intelligence, speed, and capacity.
CodeRabbit has been a pioneer in independent AI code review and has expanded that layer into Agentic Change Management, the control plane for agentic software development. They use our Cloud Run and Storage products to underpin their application, and are now beginning to leverage Gemini 3.1 Pro and other Gemini models to power use cases like analyzing how a single code change impacts an entire project, writing clear and contextual review comments to explain logic bugs, and instantly generating one-click fixes for developers.
Comfy offers a platform for creatives to build brand-consistent content and media with generative AI. They are utilizing a mix of NVIDIA systems and Google media generation models, like Veo and Nano Banana, in their platform.
MicroAGI, a German AI robotics startup, recently announced they would access NVIDIA Blackwell systems for model training through Google Cloud. They will also use Gemini Enterprise Agent Platform and AI models on Google Cloud to help robotics process multimodal information like video.
Ineffable Intelligence, the London-based superintelligence startup, recently announced a partnership with Google Cloud to access NVIDIA Vera Rubin systems as well as high-efficiency AI networking and storage.
PEAR Health Labs built and runs its AI health and fitness companion and agentic health platform entirely on Google Cloud, using our Gemini 3.5 Flash model as well as our data cloud products.
Roboforce is a physical AI company building scalable robots for industrial environments. They recently signed a new agreement with Google Cloud to utilize our GPU-powered VMs, which will be used to train and serve their custom and post-trained physical AI models.
xFigura offers a platform that gives architecture teams a single, secure canvas that brings generative models into one place for AI-driven design. Built on Google Cloud, xFigura uses models like Nano Banana and Omni to help designers generate and refine concepts in plain language, with authorship staying with the architect.
SciFin is a startup that helps revenue teams uncover "revenue reality" by converging fragmented sales context across CRM systems, documents, emails, and other sources. They are utilizing Gemini models, infrastructure, and Gemini Enterprise to build and run their platform.

These new and expanding customers represent some of the most exciting names in AI and are demonstrative of the type of successes that startups are having with Google Cloud’s uniquely complete stack for building AI.

To learn more about Google Cloud’s work with leading AI startups, or to get started building with us, visit here.

Changing the game: Using agentic AI to secure infrastructure code

Sat, 19 Sep 2026 01:00:00 +0000

Blog authors: Andrés Lagar-Cavilla, Distinguished Engineer; Stella Voutsina, Sr. Technical Program Manager; Parthasarathy Ranganathan, VP, Engineering Fellow

AI is accelerating software development at an unprecedented pace. But as code generation scales, so do the challenges of securing the code, especially emerging AI-based vulnerability exploitations. To meet these challenges, the Google AI and Infrastructure team is transforming how we approach security. In this article, we discuss new AI-native agentic methods that we’ve developed that systematically embed high-precision, pervasive vulnerability scanning and patching directly into Google’s software development lifecycle. By continuously scanning every code change across hundreds of millions of lines of code that we deploy onto our infrastructure, we are preventing hundreds of vulnerabilities per month from ever reaching our code base or production, defending our global network, AI infrastructure and our users.

Solution architecture and implementation

Pervasive pre-submit agentic scanning: security as part of ongoing software development

Traditionally, the technology industry relies on large one-off security scans that are slow and lack sufficient context. As a result, they often find vulnerabilities too late. Our approach instead focuses on pre-submit scanning, where we evaluate each code check-in (across every layer of the stack) in real-time using AI agents. By integrating the pre-submit scan into the tools developers already use, security becomes a continuous routine process, similar to rule checkers, readability reviews or other software development tools. Also, from an AI perspective, scanning each individual code change requires much less context than performing a large one-off scan, significantly improving the scan’s effectiveness.

The importance of localized threat models

For this initiative, we evolved Mantis, our open-source multi-agent review harness, to increase the precision of our security agents by matching them with a cohort of robust localized threat models. Rather than relying on static decoupled documents, the threat models use live codebase metadata. The scanning agent improves its accuracy further using a dependence call graph across packages and libraries to expand and refine its threat model context. Making threat models part of our ongoing vulnerability scanning encourages developers to continuously update threats and dependencies, keeping the models up-to-date. Using localized and precise threat model data translates to dramatic accuracy improvements, bringing our false-positive rates down to 3% in some cases.

Specialized triage agents speed up development

Vulnerability scanning as part of code check-in requires it to respond quickly to the developer or agents generating the code, so as not to impede engineering productivity. To get responses with low latency, we run a two-step validation process. First, we run a quick lightweight scan that validates its findings against a specialized triage agent. This agent programmatically checks the actual structure of the code (using abstract syntax tree parsing, call-graph traversal, and pre-indexed domain safety rules) to prove that the vulnerable path is actually reachable by an attacker. This agent gets over 92% precision and completes its work in less than a minute. Then, a post-submit scan as part of nightly integration testing serves as a second layer of defense, using off-peak cycles to test for vulnerabilities that may have been introduced across multiple changes.

Bug fix agents close the loop

Finding vulnerabilities is only half the battle. The last component of our solution is an automated bug-fix agent that uses the scan results and generated proofs (snippet of code that demonstrates how the vulnerability is exercised) to autonomously construct precise fixes that are consistent with our internal coding standards. The agent submits the fixes for human review as part of the original change request’s review, further reducing the time between detection and resolution.

Learnings and call to action

Embedding continuous scanning directly into the software development lifecycle has been a game changer at Google; its suggestions are widely adopted, and it’s prevented a multitude of vulnerabilities from being introduced into the codebase. But any organization wishing to improve security can adopt a similar AI-native approach, following these principles:

Keep systems separate: To prevent bias, keep the harnesses, rules, and context for each of your development, scanning, triage agents separate. Pair lightweight AI scans with deterministic, structural validation to drive down latency and improve accuracy.
Use context wisely: Feed your agents your existing threat models. Precise context is the answer to reducing false positives, and up-to-date threat models set a high floor on a team's security posture by improving the rate of true positives in presubmit scanning.
Build a good harness: While the choice of the underlying model is important, using a multi-agent harness can have substantial impact, by helping compensate for variability in model choice.
Automate the fix: Use agents to also propose human-in-the-loop fixes, to further reduce time-to-resolution.

If you want to get started on your own AI-native security transformation, Mantis is now available as open source for you to use and benefit from. You can also learn more about the fundamentals of cybersecurity and the other platforms that power this agentic pipeline: Google Cloud, Gemini Enterprise and Gemini models running on Trillium and Ironwood TPUs. And you can get inspiration from how agentic vulnerability scanning and remediation defends Google Cloud customers as an integral part of Google Cloud’s secure software development lifecycle (SDLC) effort.

^{With special recognition to critical team members who made this delivery possible: Stella Voutsina (Lead Program Manager), Yulong Zhang (Senior Staff Security Engineer, Mantis), and Nick Galloway (Staff Security Engineer, Mantis).}

How Orange built FinOps accountability, and why agents are next

Wed, 16 Sep 2026 16:00:00 +0000

At Orange, the leading France-based multinational telecom provider, there are days when engineering teams set aside their delivery backlogs and spend the day cleaning up cloud spend together. There's a leaderboard. There are goodies on the line. Experienced practitioners guide the newcomers, so people learn the work while doing it. By the end of the day, sponsors can see the results.

Orange calls these FinOps Clean Days. Together with gamified hackathons, they've earned the company's 100-plus person FinOps community a Net Promoter Score within the organization that’s above 70.

Those numbers point at something the wider industry is wrestling with. Recent State of FinOps reports identify getting engineers to take action as one of the top challenges organizations face. Moving from awareness to action means finding ways to build FinOps accountability, and to get teams to genuinely care.

That makes FinOps a business change problem. And business change problems have known solutions. We spoke with Camille Marini, the FinOps lead at Orange, to get a deeper understanding of how the company overcame these hurdles to accelerate AI adoption and ROI, and how your organization might follow the same course.

Why the Clean Days work

Orange has held two principles since it set up its FinOps team. First, Cloud FinOps is a shared responsibility, with every stakeholder in a project involved in their own way. And the only path to that shared responsibility runs through communication and a deliberate change effort.

“We insisted on the concept of shared responsibility across the organization for our FinOps practices,” Marini told us. “It’s very similar to how we approach cloud security. We needed to make teams understand that every single stakeholder in a project is involved in FinOps, each in their own way, if we are going to achieve responsible and impactful AI spending and usage.”

Those principles led Orange to create a FinOps Community of Practice, with support from Google Cloud Consulting. The team ran it on standardized communication channels so the methodology reached well beyond the central group, and kept the meetings actionable, sharing optimizations and billing updates so every session provided value.

The Clean Days came from a clear-eyed reading of how agile teams actually operate. In agile environments with deployment running constantly, optimization work rarely wins against the sprint. Delivery priorities, backlogs, and daily operations take the available time first. So Orange created protected time, made it collaborative, and made it fun.

McKinsey's four building blocks of change explain why this approach lands. Any large organizational change, the framework holds, requires action across four areas:

Conviction and understanding: "I know what is expected of me and I agree with it."
Formal mechanisms: "The structures, processes, and systems reinforce the change."
Role modeling: "I see my leaders and colleagues behaving differently."
Talent and skills: "I have the skills and opportunities to behave in a new way."

Map Orange's practice onto those blocks and the pattern is visible. Gamification and rewards give engineers colleagues to emulate: The leaderboard makes different behavior visible, and sponsors see the quick wins for themselves. Experienced practitioners guiding novices builds talent and skills through the community itself. The regular sessions, sharing optimizations and billing updates, build the conviction that comes from knowing where the money goes.

FinOps activities mapped to the four building blocks of change, with the points where AI agents can reinforce them.

What happens beyond 100 people

A community of 100 engaged people is an achievement. But in an organization with thousands of engineers, no central FinOps team can reach everyone directly. The question for leaders is how to extend what a community like Orange's creates — the awareness, the shared ownership, the habit of acting — to people the FinOps team will never meet.

This is where AI agents extend the capabilities of a FinOps team with two core benefits. They take on complex, time-intensive activities that previously needed a human, and they reduce friction around FinOps for individuals across the business.

Getting teams to adopt them takes a strategy aimed at your own organization's pain points, which often come from high cognitive load, unclear accountability, or competing priorities. Start by finding where engagement drops off in your FinOps lifecycle:

An awareness gap: If teams are unsure of their spend impact, an insight agent can push real-time cost data into their daily tools.
A bandwidth gap: If engineers are too busy with backlogs, a remediation agent can identify quick wins and present them as ready-to-merge code changes.
A complexity gap: If reporting feels like a manual chore, an orchestration agent can gather the data and simplify the process.

Start with trust, then add autonomy

The sensible path runs in sequence. Establish the community practice, the way Orange did. Then introduce read-only agents that inform and suggest. Only once those are established across the community should you build agents that execute changes. Direct action carries operational risk, so manage it carefully. It's also where significant wins often sit.

How you build depends on who's building. For teams that want to deploy quickly with minimal code, the Gemini Enterprise App provides a no-code environment for creating agents. For developers who need granular control, the Gemini Enterprise Agent Platform (formerly Vertex AI) offers advanced tools for launching and governing agents built with frameworks like the Agent Development Kit (ADK).

Cloud FinOps is moving beyond centralized reporting toward action that happens where the work does. The organizations getting there start with the culture, then use agents to carry it further than any one team could reach.

Orange's numbers came out of the community work. Building that foundation is the part worth copying first. When you're ready to extend it, Google Cloud Consulting can help you shape the community practice, and the Gemini Enterprise App is a low-lift way to put your first read-only agent in front of your teams.

Cloud CISO Perspectives: How Google monitors AI threats and advances AI defenses

Wed, 16 Sep 2026 16:00:00 +0000

Welcome to the first Cloud CISO Perspectives for September 2026. Today, Sandra Joyce shares the latest details on Google’s visibility into how attackers are using AI, and how we’re using AI to stop them.

As with all Cloud CISO Perspectives, the contents of this newsletter are posted to the Google Cloud blog. If you’re reading this on the website and you’d like to receive the email version, you can subscribe here.

aside_block: <ListValue: [StructValue([('title', 'Get vital board insights with Google Cloud'), ('body', <wagtail.rich_text.RichText object at 0x7fd621542b50>), ('btn_text', 'Visit the hub'), ('href', 'https://cloud.google.com/solutions/security/board-of-directors?utm_source=cgc-site&utm_medium=et&utm_campaign=FY26-Q2-GLOBAL-GCP39634-email-dl-dgcsm-CISOP-NL-177159&utm_content=-&utm_term=-'), ('image', <GAEImage: GCAT-replacement-logo-A>)])]>

‘Spellcheck for cybersecurity’ and beyond: How Google monitors AI threats and advances AI defenses

By Sandra Joyce, VP, Google Threat Intelligence

Sandra Joyce, VP, Google Threat Intelligence

Anyone operating in security knows that speculation is a major liability during periods of technological disruption. While there is plenty of hype and understandable concern around how threats might use and target AI, a CISO’s AI security strategy has to be anchored in ground truth.

Google operates at a rare intersection as both a frontier AI lab and a security company with a frontline view of global incidents. This dual vantage point allows us to understand how AI is built, and exactly how AI is being targeted in the wild. To provide the operational realities that security and business leaders need in the AI era, Google Threat Intelligence Group (GTIG) recently released our latest AI Threat Tracker.

When we strip away the noise and look at the telemetry, the real threat landscape boils down to three structural shifts that CISOs must address:

AI is reshaping how software is built.
AI is expanding the attack surface.
AI is enhancing threat capabilities.

Today, we’re sharing details on Google’s visibility into these three challenges, and our approach for solving them.

Building securely in the AI era

AI has fundamentally altered software development velocity. Across the industry, autonomous agents and AI workflows now push code into production at unprecedented speed. This creates exciting opportunities for innovation, yet CISOs are faced with the difficult task of mitigating enterprise risk while maintaining business momentum.

We’re seeing threat actors turn our greatest engineering shortcut against us by contaminating upstream packages that AI assistants are trained to suggest and trust. GTIG believes that malicious contamination of AI-assisted coding practices has been contributing to the significant growth in large-scale, open-source software supply chain compromises we observed in 2025 and early 2026.

The solution to a machine-speed threat landscape isn't slowing developers down — it’s building security natively into the AI pipeline. Part of this process involves in-editor guardrails for developers that create a real-time 'spellcheck for cybersecurity.'

We’re also monitoring adversaries targeting agents. The financially-motivated threat actor TeamPCP (UNC6780) has implemented more than half a dozen methods to exploit AI tools and open-source software development practices, including hijacking AI toolkits, prompt injection, and blinding AI scanners with toxic prompts to obfuscate malicious payloads.

Just as word processors underline typos without forcing the writer to stop, security controls must sit natively inside the developer’s editor and agentic workflows, instantly flagging poisoned packages, toxic prompts, and misconfigured toolkits.

Crucially, this can’t stop at the editor. Traditional security suffers from context blindness: Code editors can’t see cloud configurations, delivery pipelines miss runtime exposure, and production teams can’t easily patch root-cause blueprints.

Bridging this gap requires an integrated code-to-cloud approach — the exact design principle behind platforms like Wiz Code. The underlying approach is to ensure code is continuously verified against live cloud realities before it ships.

When organizations think about AI-driven code analysis, the default assumption is to pick one frontier model and point it at their repository. However, our research and telemetry show that single-model security creates a dangerous monoculture: No single AI model can discover every vulnerability, and threat actors are already testing inputs that can blind specific LLM safety filters and scanners.

To secure this expanding attack surface, CISOs should avoid the trap of managing AI through disconnected silos... The future of cloud and AI defense needs to be built on a unified and dynamic graph that connects your code, your models, your data lineage, and your runtime identities into a single living map.

To solve this, Google takes a deliberate multi-model approach. By orchestrating several foundation models — including Gemini, commercial, and open-source — we cross-validate findings, strip out false positives, remediate code, and identify complex logic flaws that a single model misses. We’re smarter with more than one “brain.”

Securing AI

Securing the development lifecycle is only half the battle. We also need to prevent adversaries from exploiting AI attack surfaces and weaponizing over-privileged agents. Threat actors are targeting AI workloads with techniques that include:

LLMJacking: Cybercriminals and state-sponsored groups target GPU access to support running their AI models and agentic workflows. In one notable intrusion Mandiant investigated in April, a threat actor gained initial access to a victim’s cloud environment from an exposed personal access token, and used it to deploy unauthorized AI infrastructure and scale high-performance compute resources, leaving the victim to absorb the hardware and platform costs.
Targeting of AI data and access: Cybercriminals now recognize that your custom prompts, agent instructions, and fine-tuned models represent high-value crown jewels. In Q2 2026, Mandiant investigated multiple data theft extortion operations where threat actors stole proprietary AI data, including models, skills, prompts, source code, and related research. Demand is also surging for AI account credentials in underground marketplace forums, with some sellers offering steep discounts for consumer accounts at up to 99% off retail prices.

To secure this expanding attack surface, CISOs should avoid the trap of managing AI through disconnected silos. Don’t treat agent access policies, model inventories (AI-BOMs) and shadow AI as separate challenges because these risks are deeply connected. The future of cloud and AI defense needs to be built on a unified and dynamic graph that connects your code, your models, your data lineage, and your runtime identities into a single living map.

Pioneered by the Wiz Security Graph, this approach serves as the contextual engine for Google AI Threat Defense (AITD) — our broader autonomous security framework that fuses the reasoning power of Gemini and other frontier models, the contextual risk prioritization of Wiz, the code remediation capabilities of CodeMender, and the frontline expertise of Mandiant to stay ahead of AI-driven attacks. Crucially, this context is not siloed; it directly feeds Google Security Operations, ensuring that security operations teams can continuously identify, prioritize, and sever toxic attack paths at machine speed.

Defending against AI threats

Threat actors are rapidly moving beyond simple prompt generation toward fully-automated, multi-agent attack pipelines.

Cyber Defense Summit 2026: Security in the AI era

In one notable intrusion investigated by Mandiant, a financially-motivated actor compromised an organization's cloud infrastructure and deployed an autonomous agent framework. The threat actor used an AI coding chatbot, a prompt, and a set of agent instructions to plan, build, and execute a mass credential harvesting campaign in less than six hours.

We’re also tracking adversaries using AI as an intelligent orchestrator across the entire attack lifecycle. GTIG recently observed a PRC-nexus espionage group experimenting with a tool called CC Switch to cycle across multiple accounts and swap AI models — like Claude, Codex, and Gemini — picking the best model for specific tasks, such as writing exploit scripts and drafting lures. While the underlying hacking tools aren’t new, AI turned what had been a disjointed manual process into a smooth and automated workflow.

To take advantage of your deep context, it’s imperative to shift from manual, human-scale incident response to machine-speed security operations. We can no longer rely on human analysts manually triaging endless backlogs of static alerts.

While these machine-speed attacks sound daunting, defenders actually hold an asymmetric advantage. Even when armed with autonomous AI, an attacker operates from the outside with limited context — probing in the dark, guessing connections, and hoping a compromised credential leads to a useful asset.

Defenders, on the other hand, possess deep context that attackers don’t have. You know your code, cloud configurations, user identities, deployment realities, and internal architecture better than anyone. When you feed this rich, multi-dimensional internal observability into security models, AI defense becomes inherently faster and more accurate than AI offense.

By codifying our frontline threat intelligence directly into these AI models, these autonomous agents can continuously monitor for, investigate, prioritize, and remediate attacks.

How Google is helping defend the ecosystem

As adversaries adopt AI, we have a unique opportunity to disrupt them at the source. As a major security and AI provider, we take this responsibility seriously, using multiple levers to stay ahead.

Disabling malicious infrastructure. If you use Google tools to facilitate an attack, you lose access to those tools. We proactively disable the projects, accounts, and assets of known bad actors.
Hardening our AI models and classifiers. We operate a continuous feedback loop for our AI models. By feeding threat intelligence directly back into product development, our models learn to recognize and refuse malicious requests before an attack can even be generated.
Automating vulnerability hunting and patching also disrupt adversaries. We are moving from manual patching to AI-driven hunting. Tools like CodeMender automatically fix critical vulnerabilities in the code itself.
Developing advanced defenses and threat models. Our teams at Google DeepMind are building specialized defenses for generative AI — deploying active monitoring across our entire ecosystem to identify misuse in real-time.

Securing the AI era can’t be achieved with the disconnected, manual tools of the past, and you can only defend against an AI-powered threat with an AI-powered defense. To tip the scales back in favor of defenders, we must transition to a continuous, machine-speed model of protection — and at Google, we are committed to building that secure future alongside you.

To learn more about our approach to securing the AI era, please check out our new Mandiant AI Risk and Resilience report.

aside_block: <ListValue: [StructValue([('title', 'Learn something new'), ('body', <wagtail.rich_text.RichText object at 0x7fd61b4baf10>), ('btn_text', 'Watch now'), ('href', 'https://x.com/googlecloud/status/2090213589558698309?s=20'), ('image', <GAEImage: Cloud-CISO-Perspectives-logo-A>)])]>

In case you missed it

Here are the latest updates, products, services, and resources from our security teams so far this month:

A manufacturing blueprint for secure agentic AI: AI and agents have arrived on the factory floor. Today’s CISOs and business leaders must balance innovation with precision, physical safety, and operational resilience. Read more.
Proactive cyber defense for governments and enterprises: Our new Fairwind Program is a limited access program for governments and trusted partners to use our most advanced cyber defense capabilities. Read more.
Getting started with the Mantis harness to find and fix bugs: Mantis is part of how Google finds and fixes vulnerabilities at machine-speed. The open-source AI harness creates a more effective repository analysis. Read more.
Breaking into Google's GFile for $100,000: Learn about how a vulnerability — that was not exploited and has now been patched — could have allowed attackers to chain unauthenticated, undocumented internal APIs with overly-permissive shared file libraries to achieve unrestricted data access across core infrastructure. Read more.
Introducing new session management tools with native, granular controls: New Google Cloud session controls are deeply integrated and a granular feature of Context-Aware Access. Here’s what you need to know. Read more.
How Blackline prevents data exfiltration with VPC Service Controls: We’re excited to share new policy intelligence capabilities in VPC-SC that help drive operational simplicity: Violation analyzer and violation dashboard. Read more.
Introducing Continuous Vulnerability Assessment: You can detect exposure to new vulnerabilities the moment they’re published with Wiz CVA. Read more.
How developers prevent production risk at the source: Fixing security vulnerabilities in code takes seconds, while patching in production creates high operational costs and risk. Discover how empowering developers as your first line of defense eliminates exposure across every phase of your software pipeline. Read more.
Wiz achieves GovRAMP High authorization: Delivering unified cloud security and accelerating secure modernization to protect citizen data and critical infrastructure. Read more.

Please visit the Google Cloud blog for more security stories published this month.

aside_block: <ListValue: [StructValue([('title', 'Join the Google Cloud CISO Community'), ('body', <wagtail.rich_text.RichText object at 0x7fd61b4bb910>), ('btn_text', 'Learn more'), ('href', 'https://rsvp.withgoogle.com/events/google-cloud-ciso-community-interest-form-2026?utm_source=cgc-blog&utm_medium=blog&utm_campaign=FY25-Q1-global-GCP30328-physicalevent-er-dgcsm-parent-CISO-community-2025&utm_content=cisop_&utm_term=-'), ('image', <GAEImage: GCAT-replacement-logo-A>)])]>

Threat Intelligence news

AI Threat Tracker: From prompting to autonomy: In the newest Google Threat Intelligence Group (GTIG) report on the adversarial misuse of AI, we’ve observed adversaries transition from basic prompting to agentic AI workflows and AI-enabled automation, including threat actors compromise a cloud resource, then plan, build, and execute an agent-enabled mass credential harvesting campaign in under six hours. Read more.
Financially-motivated threat actor BREEZE COMET targets Brazil: Learn about BREEZE COMET’s tactics and toolkit, and our mitigation recommendations and detections to support organizations in defending against this active and developing threat. Read more.
JFrog Artifactory under attack: Wiz Research has identified active, in-the-wild exploitation of three critical and high-severity vulnerabilities impacting JFrog Artifactory. Attackers are chaining these vulnerabilities to bypass authentication and gain administrative control. Read more.

Please visit the Google Cloud blog for more threat intelligence stories published this month.

Now hear this: Podcasts from Google Cloud

Cloud Security Podcast: Patching browsers with AI, agents, Rust, and your tabs: Jasika Bawa and Doug Turner of Chrome Security explore how Google Chrome now uses AI agents to autonomously identify and patch security vulnerabilities at an unprecedented scale, significantly accelerating the browser's update cadence. Listen here.
Cloud Security Podcast: All about Project Atlas, Wiz's AI vulnerability research: Nir Orfeld, head of vulnerability research, Wiz, discusses how his team uses multi-agent AI systems for discovering high-impact zero-day vulnerabilities in cloud infrastructure. Listen here.
Cloud Security Podcast: How Google eliminates classes of vulnerabilities at scale: How do you build the foundations for a secure Google-scale enterprise that stays secure even if an AI is writing the code and nobody has time to review it? Christoph Kern, principal security engineer, Google, explores what secure-by-design really means in the AI era. Listen here.

To have our Cloud CISO Perspectives post delivered twice a month to your inbox, sign up for our newsletter. We’ll be back in a few weeks with more security-related updates from Google Cloud.

Google is a Leader in the 2026 Gartner® Magic Quadrant™ for Enterprise AI Assistants

Wed, 09 Sep 2026 18:00:00 +0000

We are excited to share that Gartner has named Google a Leader in its inaugural 2026 Magic Quadrant for Enterprise AI Assistants. In this comprehensive evaluation of top enterprise AI assistant vendors, Gartner placed Google in the Leaders quadrant for its evaluation across both Completeness of Vision and Ability to Execute.

Gemini Enterprise helps organizations bring helpful, secure AI directly into the daily work of their employees. It moves teams past basic chat interactions to automating multi-step, end-to-end workflows with AI agents. It connects with the tools and infrastructure companies already use and scales easily and cost-effectively, all with security and governance in place.

We see this recognition from Gartner as validation of our goal: creating a unified platform where everyone — from business users to developers — can work alongside AI agents to accomplish more together.

Our take on Google as a Leader

Amid a complex landscape of standalone AI tools and emerging platforms, Gemini Enterprise emerges as a unified, open agentic platform backed by Google’s full AI stack, with strengths mentioned in the report such as:

Unified “AI front door”: Gemini Enterprise unifies enterprise chat and search, first and third-party agents, a no-code agent designer, Google Workspace integration, and third-party connectors all in one platform, eliminating the need for organizations to piece together disparate AI tools.
Open connectivity: Gemini Enterprise offers extensive connectivity beyond Google's ecosystem — extending to Microsoft 365, other third-party software, and internal enterprise data sources. This open connectivity lets organizations adopt Gemini Enterprise alongside their existing infrastructure without costly system overhauls.
Simple economics: Gemini Enterprise offers a straightforward pricing model, with actions like chat and search included in the base SKU. Organizations can select per-user seat subscriptions, as well as a pay-as-you-go option that lets users run agent workloads without hitting quota limits mid-task.
Built-in governance: Gemini Enterprise provides robust agent governance out-of-the- box at no extra cost, enabling enterprises to seamlessly manage users, agents, and data permissions while curbing security risks and agent sprawl.
Full-stack depth and scale: Google’s vertically-integrated stack provides global infrastructure, custom silicon, world-class models, and a secure, enterprise-ready foundation — all optimized for security, interoperability, and cost.

To read the full report, download it here.

Accelerating our vision with the latest Gemini Enterprise advancements

Over the past months, we have accelerated Gemini Enterprise with significant product advancements:

Tailored industry solutions: We introduced specialized solutions for legal and financial services last month. These tailored solutions bring pre-built agents, domain-specific skills, secure data connectors, and an open partner ecosystem, allowing for rapid deployment. They are also built on top of Gemini Enterprise’s governed control plane so you can create and manage workflows in these highly regulated industries.
Google Antigravity in Gemini Enterprise: With the introduction of AI developer tools in Gemini Enterprise, enterprises can now easily enable agentic dev tools like Antigravity and Android Studio for their developer teams with eligible Gemini Enterprise licenses, and maintain full governance and observability inside the admin console.
FinOps and cost controls: New FinOps and cost-control capabilities were introduced last month, allowing organizations to optimize AI spend through more pricing options, Flexible Savings Plans, and granular spend management.

Gemini Enterprise customers are also seeing the value

Hearing this recognition from Gartner is great, but it's not just them—our customers are seeing this real-world value too, and it's driving a positive impact across their organizations every day.

Check out what our customers are saying about the recent product advancements.

“Deploying Antigravity in Gemini Enterprise allows Accenture to arm our engineers with Google DeepMind’s premier technology on the secure, trusted foundation of Google Cloud. Abstracting away operational complexity ensures our teams don't have to choose between developer speed and enterprise-grade governance — freeing them to deliver high-velocity engineering and transformative value for our clients.” — Chetna Sehgal, Global Practice Lead, Accenture Google Business Group. Learn more here.

“Cleary is committed to embedding AI into our workflows in strategic and competitive ways. Using Google’s Gemini Enterprise, which can slot in seamlessly with other daily work tools, we can unlock greater efficiencies for our teams and help them deliver even higher quality work for our clients.” — Jeff Karpf, Managing Partner, Cleary Gottlieb. Learn more here.

“As a design partner for the Financial Research agent, Deutsche Bank has helped shape this capability in view of the realities of a highly regulated industry – from data protection and governance to the workflows our teams use every day,” – Marie-Jeanne Deverdun, Chief Technology, Data and Innovation Officer, and Member of the Deutsche Bank Management Board. Learn more here.

“We’re thrilled to partner with Google Cloud in the early adoption of Gemini Enterprise for Legal. We look forward to integrating Google’s technology to streamline workflow and further support our litigators in shaping outcomes critical to our clients’ futures.” — Joe Petrosinelli, Chairman, Williams & Connolly. Learn more here.

Get started today

Read the report: Download your complimentary copy of the 2026 Gartner Magic Quadrant for Enterprise AI Assistants to explore the full analysis
Explore Gemini Enterprise: Discover how your organization can deploy governed, connected agents across every team at cloud.google.com/gemini-enterprise

How KDDI built Buffmee, a faster, reliable consumer RAG app

Tue, 08 Sep 2026 16:00:00 +0000

When building consumer-facing generative AI applications, balancing high generation quality with fast response times across diverse media types, can be challenging. KDDI, a major telecommunications carrier in Japan, tackled this challenge head-on when they developed Buffmee, their consumer Retrieval-Augmented Generation (RAG) app.

Buffmee is an interactive AI service built on the concept of 'AI that helps you grow.' By grounding responses in over 100 sources — including books, magazines, and web media — it helps users search for information, summarize key points, and explore personalized learning and hobby interests. By citing sources, Buffmee alleviates concerns about information reliability, allowing users to safely deepen their knowledge. To achieve this, KDDI collaborated closely with their development partner KDDI iret, Google Cloud Consulting and our specialized AI engineers.

As part of their app launch, the engineer team needed to ground a massive variety of proprietary content, including books and magazines. However, they struggled with latency issues that prevented them from meeting their target response times, and they needed a reliable way to ensure hallucination-free results.

Buffmee App Description and Images

To meet these performance targets, organizations need a systematic approach to AI evaluation and real-time bottleneck identification. That is why we are sharing the automated evaluation framework and performance optimization techniques that helped KDDI successfully launch their application.

The results were inspiring: KDDI reduced total application response latency by 38%, successfully hitting their target response performance. They also achieved a nearly 18% improvement in TTFT.

"Our vision hinged on a platform where content, once ingested, would instantly function as a working RAG system. Google's careful, hands-on guidance made that a reality — we're sincerely grateful for their support." — Shunya Onoda, AI Product Department, KDDI.

With these performance and accuracy improvements, Buffmee now empowers users to safely explore their favorite media through interactive Q&A and deep-dive analysis, delivering a highly personalized experience while maintaining strict trust and compliance for content providers.

Let’s deep dive into how they achieved these results.

Establish automated evaluation for diverse content

Traditional manual testing requires immense effort and cannot scale to accommodate a large content library. To solve this, the development team designed a systematic AI evaluation process using Gemini Enterprise Agent Platform Evaluation Service.

By implementing automated evaluation frameworks like LLM-as-a-Judge and the Rule of Hundreds, the team replaced labor-intensive manual testing with a data-driven process. They ingested their extensive document corpus, constructed hundreds of automated evaluation tests, and built a comprehensive benchmark dataset to measure the reliability of answers for each use case. As a result, the team improved their groundedness scores by 25%, helping deliver highly accurate and reliable outputs.

KDDI's automated evaluation loop: AI generates questions and scores answers, while humans calibrate thresholds and analyze edge-case failures.

Identify bottlenecks and optimize performance with an agentic loop

To improve response speeds, the team implemented BigQuery Agent Analytics and the Agent Development Kit (ADK) log analysis agent. By analyzing actual production logs, they visualized how skill division and prompt bloat—especially with highly complex, multi-page system prompts — impacted the Time To First Token (TTFT).

The team optimized the system prompt, including the inline integration of skills, and reviewed the sub-agent routing. This allowed them to identify and resolve deep-stack bottlenecks in real time without sacrificing response accuracy.

Four core principles for reliable evaluation

To achieve these results, the team implemented four core technical practices:

Transitioning to binary evaluation: By selectively moving away from ambiguous 1–5 ratings to a binary "pass (1) / fail (0)" system for critical metrics, the team minimized variance and noise, helping improve automation accuracy.
Strategic content sampling: Rather than attempting to evaluate every single document, the team classified their entire corpus along a two-dimensional grid: File Format (Web articles, EPUBs, PDFs, structured data) and Media Composition (Text-heavy, image-heavy, or mixed). By selecting representative samples from each cell of this difficulty grid, they reduced the evaluation workload by 75% while maintaining comprehensive test coverage.
Thresholds grounded in product judgment: Instead of relying solely on default tool parameters, the product owner reviewed randomly sampled answers alongside their automated scores to calibrate and establish what "good enough to ship" actually meant for the user experience.
Modular splitting of massive prompts into ADK Skills: Because massive system prompts exceeding 800 lines can cause LLM attention drift and latency degradation, the team split prompts by function into Agent Development Kit (ADK) Skills, dynamically loading only the required logic to optimize response times.

Get started

Building scalable, reliable generative AI applications requires both automated evaluation and deep performance analytics. To apply these techniques to your own applications:

Measure quality systematically with the Gen AI evaluation service
Structure your agents with Agent Development Kit and apply progressive disclosure deliberately
Ground your agents with Agent Search and inspect your retrieval queries

AI & Machine Learning

Architecting autonomy: A design pattern for high-compute, governed agents in biotech

A three-layer pattern

The execution layer: scaling the muscle - high-compute execution

The cognitive layer: Multi-agent reasoning

Performance best practices for high-compute pipelines

The enterprise governance and asymmetric autonomy

The pattern, and the path forward

Four evaluation engineering lessons learned from building agent plugins

1. Grade the execution steps instead of the summary

2. Design evals for asynchronous operations

3. Test live runs and "dry runs" side-by-side

4. Use “golden sets” as durable baselines

Where to go from here

What the Gemini agent means for startups

What is the Gemini agent?

Five takeaways for startups

Founders already building

Build with Google for Startups

How Visko streams real-time world models on Google Cloud TPUs

Redefining interactive media with Live World Models

Live video is a hard-deadline problem

Architecting the multimodal pipeline on a single TPU slice

Real-time performance breakdown: Hitting the clock

Inside the stack: Three optimizations that unlocked real-time streaming

1. Ulysses context parallelism across all TPU chips

2. Custom Pallas splash attention from MaxDiffusion

3. A 5.8x audio speedup by moving from torchax to pure JAX

Field notes for AI builders: Cloud TPU lessons

Building real-time generative models on Cloud TPU? Start here

Predictable AI spend in BigQuery: Introducing cost management for generative AI functions

Quotas for managing costs

Which functions are covered?

Managing your quotas

Monitoring usage

Token quotas vs. billing spend caps

How enforcement works

Sizing your input data: AI.COUNT_TOKENS

From tokens to budget: a worked example

Best practices for managing AI cost

Get started

Gemini 4 Argon available on Google Cloud, built to help solve your business’ hardest problems

Four ways Argon changes how your business works

1. Hand off your hardest work

2. Precise, verifiable results on the work that drives your business

3. Build and ship products faster

4. Stronger defense for your code and business

Enterprise customers and partners in our early access program are already seeing strong wins with Argon.

Get started

Welcome to Gemini at Work 2026: Introducing the Gemini agent

Introducing the Gemini agent

Powering Gemini: Skills, tools, and context

Gemini in Google Workspace

Data analysis, data engineering and machine learning

Industry-specific specializations

Securing and governing agents

Leading cost controls and spending options

Infrastructure: 80% better price-performance

Supercharged by our ecosystem

Gemini Enterprise in action: Global momentum

Europe and Middle East

Asia Pacific

Latin America

North America

Our commitments to you

Innovation in Ireland: How Irish brands scale with Gemini Enterprise

Ryanair supports millions of passengers through AI-backed operations

Smyths Toys rethinks the customer experience with agentic support

The Irish Revenue Commissioners: Advancing public sector data

Virgin Media Ireland: Accelerating workflows

Irish startups scaling with enterprise tools

Hexis: Scaling personalized sports nutrition

Kitman Labs: Unifying performance analytics

Spryt: Automating healthcare coordination

IMPT: Automating carbon offsetting

Transform ideas from planning to production

What’s new in AI infrastructure and orchestration in September

Product, technology, and tools updates

Practitioner guides and how-tos

Customer and partner updates

Sizing your input data: `AI.COUNT_TOKENS`