Google Engineering Blog & Tech Articles

The MaxText team successfully reproduced AI2’s OLMo 3 7B language model from scratch on Google Cloud TPUs using JAX/XLA, precisely matching the original PyTorch-on-GPU reference across pre-training and mid-training stages on all held-out evaluations. The implementation achieved up to 57.4% Model Flops Utilization (MFU) and demonstrated robust infrastructure portability by surviving mid-run cluster resizes and cross-generation TPU shifts without requiring recipe alterations. Crucially, the exercise proved the necessity of comprehensive held-out validation by catching a silent data-loader memorization bug that artificially depressed training loss and would have otherwise faked a performance win.

Google Cloud API Gateway now acts as a native remote Model Context Protocol (MCP) server, eliminating the need to build and maintain custom middleware to expose REST APIs to AI agents. By simply adding specific annotations (like x-google-api-management.mcp) to existing OpenAPI 3.x specifications, developers can instantly convert standard REST operations into discoverable, agent-ready tools. The gateway automatically transcodes incoming MCP JSON-RPC requests into REST calls, ensuring that your existing authentication, quotas, and logging policies apply seamlessly to agent traffic without requiring new infrastructure.

The Google Antigravity SDK now empowers developers to execute offline, agentic workflows locally using models like Gemma 4 26B A4B via LiteRT. This update facilitates powerful hybrid orchestration architectures, allowing a cloud model to act as a lightweight planner while local models securely handle token-intensive tasks—like code auditing and patching—directly on-device. Furthermore, the SDK provides drop-in support for OpenAI-compatible inference servers like Ollama and vLLM, enabling the seamless creation of privacy-first, autonomous local utilities.

Unlock premium Google Colab compute with Google AI. Subscribers now get priority accelerators, Premium GPUs, and background execution for long training runs.

Google has partnered with Speakeasy to open-source their OpenAPI code generation suite under the AGPLv3 license, a strategic move prompted by the sudden shutdown of Google's previous proprietary SDK provider. The newly open-sourced suite equips developers with deterministic, multi-language SDK generators that natively support strict typing and SSE streaming, alongside tools for compiling agent-native CLIs and documentation MCP servers. Engineering teams can now safely integrate this robust tooling directly into their CI pipelines to automatically generate reliable client libraries for their own APIs, all while retaining complete licensing control over the output code.

Agent Anomaly Detection is a new, out-of-band oversight layer for the Gemini Enterprise Agent Platform that analyzes OpenTelemetry traces and tool calls to catch behavioral risks without adding runtime latency to live requests. It utilizes a multi-tiered detection pipeline—combining lightweight statistical scanning with deep LLM-based reasoning—to identify logical anomalies and policy violations grounded in the OWASP Agentic Top 10. Developers can triage these automated findings within Security Command Center or leverage the exposed API to programmatically block subsequent tool calls when an agent breaches defined risk thresholds.

This blog post explores how to transition AI agents from static, build-time security controls to dynamic runtime governance using the Gemini Enterprise Agent Platform. It highlights three primary managed defenses: Model Armor for screening edge prompts, Semantic Governance Policies for evaluating tool intent against business rules, and Agent Anomaly Detection for catching multi-turn exploits. By shifting these capabilities to the platform level, security administrators can dynamically enforce policies and neutralize complex attacks without needing to modify or redeploy the agent's underlying code.

Imagine going to sleep after writing a single Markdown specification and waking up to find that an A...

While end-to-end benchmarks like SWE-bench provide broad performance scores for AI agents, they are often expensive, slow, and lack the root-cause diagnostics needed to explain exactly where an agent's logic broke down. To solve this, developers should adopt behavioral evaluations—fast, local, unit-style tests that assert on discrete intermediate actions, such as verifying specific tool calls or file modifications rather than final string equality. By building these inexpensive micro-checks alongside macro benchmarks, engineering teams can confidently iterate on system prompts and upgrade models without the risk of regressions.

Google has officially released version 1.0 of the Agent Development Kit (ADK) for Kotlin, achieving full feature parity with the Python and Java ADK cores to enable idiomatic, multi-agent AI development. Built on Kotlin Multiplatform (KMP), the framework leverages Kotlin Symbol Processing (KSP) for zero-reflection, type-safe function calling, alongside advanced orchestration capabilities like human-in-the-loop workflows and context compaction. Additionally, the release introduces a robust suite of Android-first extensions, allowing mobile developers to integrate local models via LiteRT-LM, cloud reasoning through Firebase AI, session persistence using Room, and semantic memory powered by AppSearch.

The Gemini Enterprise Developer Experience (DevEx) program conducts ongoing sprint testing of end-to-end developer workflows to identify and rapidly resolve friction points without relying on internal shortcuts. This recent sprint focused on optimizing enterprise AI governance, including refining setup prerequisites, securing extension configurations, and clarifying policy enforcement mechanics to ensure a smoother, more reliable deployment. Developers can now leverage updated documentation and standardized code samples to improve their experience with Agent Gateway and Semantic Governance configurations.

The recent Google for Startups AI Agents Challenge revealed that the most successful multi-agent systems rely on foundational software engineering patterns rather than just raw model power. Winning architectures consistently implemented bidirectional MCP for seamless inter-agent communication, async event buses for parallel execution, strict unified validation for model fallbacks, and tiered routing to minimize expensive inference calls. By prioritizing these structural practices over simple linear prompt chains, developers can build more resilient, low-latency, and cost-effective agentic workflows.

Astroparticle physics sits at the exciting intersection of astrophysics and particle physics and stu...

Google Cloud has natively integrated TPU support into the vLLM serving engine, allowing developers to elastically scale high-demand embedding pipelines using Google Kubernetes Engine (GKE). To handle massive 15K+ token contexts for models like Qwen3-Embedding-8B, the engineering team implemented TPU-specific optimizations such as hardware-safe tensor alignment, JAX/XLA compilation pre-warming, and a hybrid StepPool architecture for chunked prefill management. These enhancements achieve near-perfect numerical parity with reference GPU baselines, and developers can immediately leverage the open-sourced setup recipes on the AI-Hypercomputer GitHub to build their own high-throughput semantic retrieval applications.

Moving live voice agents from demo to production requires rigorous, automated testing to handle the unpredictability of real multi-turn conversations. ADK now provides native live evaluation, allowing developers to test graph-based agent workflows against LLM-driven simulated users that generate actual audio via Gemini TTS. By defining evaluation scenarios and natural-language rubrics, you can automatically score audio responses and tool executions, inspect the resulting transcripts in ADK Web, or run the CLI directly in your CI/CD pipeline.

Building autonomous AI agents that mutate production state requires moving beyond soft system prompts to a robust zero-trust architecture. To secure Google Agent Development Kit (ADK) workflows against prompt injections and malicious execution, developers must implement hardware-backed cryptographic signatures for database writes, kernel-level sandboxing with gVisor for dynamic code, and deterministic semantic gateways for I/O validation. By enforcing these hard security boundaries at the infrastructure level, you can safely deploy multi-tool AI agents without risking unauthorized data manipulation or server compromise.

HeyGen ported their 18B+ parameter Avatar IV video generation model to Google Cloud's Trillium (v6e) TPUs via torchax and XLA, utilizing FSDP and Ulysses sequence parallelism across an eight-chip mesh. To achieve a 1.86x speedup for real-time streaming, the engineering team pipelined exposed all-to-all collectives, aligned sparse attention block sizes to eliminate mask padding, and bypassed softmax serial dependencies using a precomputed Cauchy-Schwarz upper bound. These custom Pallas kernel and compiler optimizations were deployed only after passing rigorous two-tier quality gates to guarantee byte-identical or mathematically equivalent pixel outputs.

Credentio is a newly released, open-source C++ library from Google that allows developers to integrate high-performance, local-first validation of C2PA Content Credentials into their client and server applications. By processing assets entirely locally with a highly optimized memory footprint, the library delivers instant validation verdicts for multi-gigabyte media files without incurring cloud latency, bandwidth costs, or data privacy risks. The library currently features deep manifest parsing alongside configurable trust list integration, and is available now on Google Source with future plans to support full credential generation and embedding.

Deploying secure, real-time Edge AI on Raspberry Pi is now simplified using LiteRT and lightweight Gemma open models. LiteRT optimizes CPU and GPU performance, delivering fast token speeds for models like Gemma4, enabling real-time local reasoning for robotics. Developers can quickly convert, quantize, and run these models using the lightweight LiteRT CLI tool. Support for Hailo AI accelerators is also coming very soon.

As AI coding assistants shift the developer's primary role from writing boilerplate to reviewing and maintaining systems, language choice becomes critical for long-term architectural integrity. Go directly addresses this new paradigm by utilizing its strict compiler, integrated toolchain, and uncompromising readability to provide deterministic guardrails that help AI models self-correct and generate highly standardized code. By enforcing ecosystem-wide consistency and strict backward compatibility, the Go platform empowers engineering teams to efficiently verify, optimize, and maintain high-velocity, AI-generated output in production environments.

Agent Plugins 1.0.0 is a new, vendor-neutral directory specification—backed by Google, Amazon, Microsoft, and others—for packaging Agent Skills and MCP servers into a single portable unit. By standardizing the manifest (plugin.json) and utilizing a fixed directory layout, it eliminates the need for developers to maintain separate wrappers or configurations to support different AI coding agents and IDEs. Google has officially joined as a Core Maintainer and already rolled out support in the Agents CLI and Data Agent Kit, allowing developers to start building and distributing interoperable plugins today.

The 2026-07-28 Model Context Protocol (MCP) specification replaces legacy stateful constraints with a fully stateless core, enabling cloud-native horizontal scaling, serverless deployments, and standard round-robin load balancing. This architectural shift introduces standardized HTTP headers for efficient routing without deep packet inspection, caching controls, and Multi Round-Trip Requests (MRTR) to handle interactive and long-running tasks without blocking connections. Developers can immediately begin migrating their agentic applications to this highly scalable infrastructure using the newly available beta SDKs for Python, TypeScript, Go, and C#.

Google Cloud API Gateway now offers a model routing feature in Public Preview, allowing developers to dynamically route traffic to models like Gemini, Claude, or OpenAI OSS-GPT without hardcoding endpoints or managing open-source proxies. Developers can easily configure these routing rules directly within their OpenAPI 3.x specifications by mapping virtual model names to specific backend targets on a shared host. Once deployed, the Gateway acts as a serverless ingress layer that accepts standard OpenAI-compatible requests, automatically transcodes the payload to the native schema of the target model, and routes the traffic on the fly.

Real-time AI agents break traditional request-response load balancing paradigms because they rely on long-lived, stateful bidirectional streams that obscure true server capacity. To solve this, developers must implement application-level session tracking directly within the runtime to accurately measure the committed concurrent workload of active conversations. By feeding these precise session counts alongside standard CPU utilization metrics into a hybrid routing algorithm, infrastructure can effectively distribute stateful AI traffic and prevent individual backend bottlenecks.

Agent Platform's evaluation service is now generally available, providing developers with a unified engine to measure agent quality consistently across local development experiments and live production traffic. You can evaluate agents using over 20 pre-built metrics, DeepMind-backed adaptive rubrics, or custom code-based and LLM-as-a-judge metrics stored in a centralized, versioned registry. The service integrates directly into existing workflows via the Agent Platform SDK, agents-cli, and ADK, offering built-in user and environment simulators to automate complex multi-turn testing and streamline CI pipelines.

To prevent context window bloat and reduce token consumption, Genkit Go introduces Agent Skills based on a progressive disclosure architecture. Developers can package specialized instructions, scripts, and references into modular SKILL.md bundles where only the frontmatter metadata is initially exposed to the agent's system prompt. When a task matches the skill's description, Genkit's middleware dynamically loads the full instruction body and associated assets, ensuring the model accesses precise workflows exactly when needed.

Google's open-source TPU microbenchmark suite provides developers with granular performance metrics across Network, Compute, HBM, Host Transfer, and Attention components to validate real-world hardware capabilities. By leveraging these benchmarks to establish a Roofline model, engineers can accurately diagnose whether their machine learning workloads are compute-, memory-, or network-bound. This empirical baseline directly guides targeted software optimizations—such as kernel tuning, mesh sharding, and rematerialization—to maximize hardware utilization for large-scale model deployments.

This second installment explores how Ray’s higher-level libraries—Serve, Data, and Train—abstract the complexities of running AI workloads on Google's TPU slices. Ray Serve uses a simple topology configuration to correctly gang-schedule large multi-host models, while Ray Data eliminates data-loading bottlenecks by feeding accelerators directly with native JAX batches. Finally, JaxTrainer streamlines distributed training across TPUs by automatically handling cross-slice coordination, checkpointing, and fault tolerance.

Tunix is Google’s new JAX-native post-training library designed to eliminate TPU idling bottlenecks when training multi-turn, tool-using LLM reasoning agents. It maximizes hardware throughput by combining highly concurrent, asynchronous rollouts with a decoupled producer-consumer pipeline, ensuring the trainer is constantly fed even while agents wait on network I/O or environment steps. Additionally, Tunix provides plug-and-play abstractions and continuous macro-level profiling, allowing developers to easily integrate custom open-source environments and optimize complex distributed workflows without massive code rewrites.

Ray 2.55 introduces official, first-class support for Google Cloud TPUs, enabling developers to run distributed Python workloads on Google's accelerators using the familiar Ray task-and-actor APIs. To handle the strict networking requirement of keeping multi-host TPU "slices" together over their Inter-Chip Interconnect (ICI), the KubeRay Operator on GKE automatically provisions and labels the underlying hardware layout. Ray Core utilizes these labels via its slice_placement_group() primitive to atomically reserve complete slices, allowing developers to deploy jobs through KubeRay, Ray Train, or Ray Serve simply by declaring a hardware topology (like "4x4") without writing custom placement code.

Conductor has evolved from a Gemini CLI extension into a portable plugin, bringing conversational Spec-Driven Development (SDD) to ecosystems like Antigravity CLI and Claude. Rather than relying on strict command sequences, developers can now chat naturally with their AI assistant while it dynamically manages persistent markdown artifacts (like spec.md and plan.md) in the background. This update eliminates workflow friction while ensuring your repository remains a version-controlled, single source of truth for your project's architecture and state across different AI tools.

Google Cloud has partnered with Parallel Web Systems to natively integrate Parallel's search infrastructure as a web grounding provider on the Gemini Enterprise Agent Platform. This integration enables developers to anchor their AI agents in verifiable, real-time web results, significantly improving factual accuracy for complex enterprise workflows. Additionally, the partnership offers expanded architectural flexibility, allowing users to programmatically extract, permanently cache, and process web data alongside other large language models.

To resolve the scaling bottlenecks and runtime errors caused by monolithic system prompts, engineering teams should treat prompts as build artifacts by modularizing instructions into reusable templates. By running these modular "skill files" through a transpiler, developers can enforce static validation, catch missing dependencies at build time, and integrate prompt generation directly into their CI/CD pipelines. This deterministic approach prevents code drift and ultimately establishes a safe framework where agents can propose updates to their own logic via standard pull requests.

We're excited to introduce LiteRT.js, the newest member of the LiteRT family! LiteRT.js is our powerful solution for running machine learning models directly in the browser, extending Google's cross-platform edge AI runtime to the web. Built for JavaScript developers, LiteRT.js delivers state-of-the-art ML model inference performance on WebGPU and upcoming WebNN, with a fallback to WebAssembly for CPU. This post provides a quick tour of LiteRT.js and gives web developers everything they need to get started.

On May 23, 2026, fresh off the stage at Google I/O, our Google Developer Experts (GDEs) converged on...

Distributed AI training is notoriously fragile because losing a single machine typically crashes the entire multi-node job, forcing a time-consuming, full-workload infrastructure restart. To address this, Google’s JAX ecosystem utilizes elastic training via Pathways, which converts a hardware failure into a catchable Python exception so the running process can survive. When an unplanned failure occurs, the system automatically replaces only the broken worker, restores the last viable checkpoint from Cloud Storage, and resumes training in place—minimizing total downtime to under two minutes without ever restarting the main controller process.

Answering the questions of "why we built ADK 2.0". This explains the rationale, some of the features, and why a developer should consider upgrading. This will be published the day after ADK go 2.0 launches.

The open-source Genkit framework has introduced the Agents API, a full-stack tool designed to simplify the complex plumbing of conversational AI by packaging message history, tool loops, and streaming into a single interface. The API supports flexible, server- or client-managed state persistence—allowing for advanced workflows like history branching, long-running detached tasks, and multi-agent coordination—while seamlessly connecting backends to frontends via a unified wire protocol. Currently available in preview for TypeScript and Go, it also integrates with the Genkit Developer UI to allow developers to easily test, debug, and inspect agent snapshots without writing client code.

The Google Cloud Workbench Notebooks extension for VS Code has officially launched, allowing developers to connect their local IDE to scalable, cloud-based Jupyter environments. This integration streamlines the machine learning lifecycle by eliminating context switching and providing direct access to high-performance Google Cloud infrastructure. To support transparency and community-driven innovation, the newly released extension is fully open-sourced and available on GitHub and the VS Code Marketplace.

The Agent Development Kit (ADK) for Go 2.0 has been released, introducing a first-class, graph-based workflow engine to help developers compose complex, multi-agent applications. This update adds built-in primitives for human-in-the-loop (HITL) orchestration, dynamic execution using plain Go code, and automated resilience features like exponential backoff retries. By unifying the execution model, both single-agent applications and intricate graphs now run on the same runtime, simplifying telemetry and state persistence.

Building AI agents often leaves developers uncertain if prompt tweaks to fix single errors will accidentally cause widespread regressions in production. To bridge this gap, Google has introduced a new developer skill for coding agents that automates a five-stage evaluation flywheel: preparing data, running inference, grading with adaptive AutoRaters, analyzing failure clusters, and executing targeted optimizations. Running continuously against production traffic or on-demand via synthetic scenarios, this tool allows developers to describe testing goals in plain language while an independent evaluation service safely validates and counts actual performance improvements.

AI coding agents are rapidly shifting from reactive assistants that complete tasks when prompted to ...

How a Python agent and a Go agent collaborate on contract compliance using the Agent2Agent protocolY...

Celebrating the first anniversary of the Agent-to-Agent (A2A) protocol, this blog post highlights how the framework enables autonomous AI agents to securely collaborate and hand off tasks without the rigidity of traditional APIs. By delegating complex workflows to specialized peer agents, A2A prevents context pollution, ensures data privacy, and simplifies application design through modularity. To demonstrate this ecosystem in action, the post spotlights FoldRun—an agentic interface for life sciences that orchestrates complex protein structure predictions—alongside diverse A2A use cases spanning commerce, data streaming, DevOps, and telecommunications.

An open specification for finding and verifying tools, skills, and agents across the web.Agents are ...

This post introduces three architectural patterns designed to integrate Model Context Protocol (MCP) Apps and Agent-to-User Interface (A2UI) to solve the tradeoff between highly custom iframe environments and native, declarative rendering. By combining these approaches, developers can serve native-feeling UIs directly over MCP servers, embed complex and stateful iframe apps securely inside declarative views, or inject generative UI components into legacy systems. Ultimately, these hybrid frameworks empower engineering teams to deliver secure, performant, and brand-consistent agentic user experiences tailored to their specific project constraints.

Google has officially launched the TPU Developer Hub, a centralized educational resource designed to help model builders and developers maximize the performance of Google Cloud TPUs. The hub offers code-first resources, open-source recipes, and deep-dive documentation covering hardware architecture, software optimization, debugging, parallelism, and networking. These materials are tailored for both human developers and AI-assisted tools to streamline everything from large-scale training to low-latency inference workloads.

Google is enhancing Sign in with Google by introducing new OIDC standard claims—specifically auth_time and amr (Authentication Methods Reference) to provide developers with deeper session metadata. These updates allow verified apps to verify the "freshness" of a user's login and the specific authentication methods used (such as MFA or hardware keys), enabling more dynamic, risk-based access controls. By leveraging these federated identity signals, platforms can better prevent account takeover and fraud while implementing granular security policies like step-up authentication for sensitive actions.

DiffusionGemma is an experimental text-generation model built on the Gemma 4 architecture that uses diffusion-based parallel generation instead of token-by-token autoregression, enabling much faster inference, bidirectional context awareness, and real-time self-correction while remaining deployable on consumer GPUs. Its architecture generates and refines 256-token blocks in parallel through iterative denoising, allowing it to handle complex constraint-based tasks such as Sudoku more effectively than traditional language models and demonstrating strong gains from fine-tuning. The model integrates with vLLM and other popular inference frameworks, giving developers access to a new non-autoregressive approach that combines high performance, efficient long-context scaling, and straightforward customization and deployment.

Google has announced the Google Colab Command-Line Interface (CLI), a new tool that allows developers and AI agents to connect local terminals to remote Colab runtimes for frictionless execution. The lightweight CLI enables users to easily request high-powered GPUs, run local Python scripts remotely, and seamlessly retrieve artifact logs or models like fine-tuned Gemma 3 adapters. By integrating directly into standard terminal environments, the tool is highly programmable and ready to be used by AI agents such as Antigravity or Claude Code to manage complex machine learning pipelines.