MindPattern
← The Wire
Source trail

Google Developers Blog

Public MindPattern findings, entities, and graph evidence that cite this source.

Findings
40
All-time hits
33
High value
9
Last seen
2026-07-07

Connected entities

Google Developers BlogGoogle Shuts Down Open-Source Gemini CLI on June 18, Pushes Users to Closed-Source Antigravity CLIGoogle Kills Gemini CLI: Forced Migration to Antigravity CLI by June 18 DeadlineCROSS-CATEGORY: Agent Infrastructure Wars Erupt — Google (Antigravity SDK + WebMCP), CopilotKit (AG-Google Developers Blog — Build Long-Running AI Agents with ADKGoogle ADK: Build Long-Running Agents That Pause for Days and Resume Without Losing Context — DurablGoogle Android DevelopersGoogle Launches Android CLI — 70% Less Token Usage, 3x Faster Agent DevelopmentGoogle Colab MCP Server: Official Bridge Lets Any AI Agent Control Colab Notebooks with GPU AccessGoogle ADK for TypeScript GA: Code-First Multi-Agent Framework With Native A2A and MCP — Python ADK Google Ships Official Colab MCP Server — Any AI Agent Can Now Control Notebooks Natively

Related findings

  1. 2026-07-07 / TOOLSPattern: The AI Coding-Tool Field Is Consolidating Into a Few Managed HarnessesTwo standalone tools folded into larger agent platforms within weeks: Google retired Gemini CLI (consumer access ended June 18) into Antigravity CLI, and Windsurf rebranded to Devin Desktop with Cascade retired July 1 in favor of an Agent Command Center. The market is collapsing from many independent editors/CLIs toward a handful of vendor-managed, multi-agent harnesses — Antigravity, Devin, Claude Code, Codex. For builders this raises switching-cost and lock-in considerations that didn't exist when these were thin local tools.
  2. 2026-06-23 / TOOLSGoogle Shuts Down Open-Source Gemini CLI on June 18, Pushes Users to Closed-Source Antigravity CLIAs of June 18, 2026, Gemini CLI stopped serving requests for all free, Pro, and Ultra tier users, replaced by Google's closed-source Go-based Antigravity CLI (invoked via 'agy'), with free-tier quota collapsing from ~1,000 to roughly 20 requests/day. Google had merged 6,000+ community pull requests into the 105K-star open-source project over nearly a year before restricting it to enterprise licenses, drawing a 'bait-and-switch' backlash and breaking CI/CD pipelines that depended on it. Only Gemini Code Assist Standard/Enterprise and paid API-key users retain access.
  3. 2026-06-17 / SOURCESGoogle Publishes an 'Agentic Resource Discovery' Specification to Standardize How Agents Find and Use ResourcesOn June 16–17 Google announced an Agentic Resource Discovery specification, extending its agent-interoperability push (alongside the A2A protocol) toward a common way for agents to discover the resources and tools available to them. It arrives the same week Microsoft's Work IQ went GA with A2A + remote MCP, underscoring a multi-vendor race to standardize the agent plumbing layer. Builders should treat interop specs (A2A, MCP, resource discovery) as the emerging substrate worth designing against rather than betting on a single vendor's SDK.
  4. 2026-06-01 / AGENTSGoogle Kills Gemini CLI: Forced Migration to Antigravity CLI by June 18 DeadlineGoogle is deprecating Gemini CLI and Gemini Code Assist IDE extensions for free and Pro/Ultra users on June 18, 2026, folding terminal-based AI coding into the broader Antigravity agent-first platform. Skills, hooks, subagents, and extensions carry over as Antigravity plugins. Enterprise customers with paid Gemini Code Assist Standard/Enterprise licenses retain access. The move consolidates Google's agentic developer tools into a single multi-agent backend.
  5. 2026-05-27 / MARKETSCROSS-CATEGORY: Agent Infrastructure Wars Erupt — Google (Antigravity SDK + WebMCP), CopilotKit (AG-UI), and Camunda (ProcessOS) All Ship Agent Platforms in Same 2-Week WindowThree distinct layers of agent infrastructure shipped simultaneously in May 2026: Google's Antigravity SDK + WebMCP (agent runtime + browser tool standard), CopilotKit's AG-UI (agent-to-UI protocol adopted by Google/Microsoft/Amazon/Oracle), and Camunda's ProcessOS (agent-driven business process automation). This is the infrastructure layer race for the agentic era — the equivalent of the cloud platform wars (AWS/Azure/GCP) but for AI agents. The convergence on open standards (WebMCP, AG-UI, MCP) signals that agent interoperability, not model capability, is becoming the competitive battleground.
  6. 2026-05-23 / SKILLSGoogle Ships Stable Android CLI at I/O 2026 — AI Agents Can Download Android SDK, Build Projects, and Run Apps on Physical Devices ProgrammaticallyGoogle released a stable Android CLI that enables AI agents to tap into Android Studio's capabilities without the GUI — handling tasks like downloading the Android SDK, building projects, and running apps on physical devices. This is purpose-built for AI agent integration, not just developer convenience. For builders doing mobile development with AI agents: this means your terminal-based coding agent can now build and deploy Android apps end-to-end without manual IDE interaction.
  7. 2026-05-23 / SKILLSAntigravity CLI 'agy' Replaces Gemini CLI with Multi-Model Config — Accepts Claude, GPT, and Self-Hosted Models; Gemini CLI Sunset June 18Google's Antigravity CLI ('agy') runs the same agent harness as the desktop app but in a terminal-native form, with config (~/.config/antigravity/config.toml) accepting model identifiers for Gemini, Claude Opus, GPT, and self-hosted endpoints. Gemini CLI and Gemini Code Assist IDE extensions stop serving requests on June 18, 2026. The multi-model config is the headline: you can point 'agy' at any backend, making it a vendor-neutral terminal coding agent — a potential alternative to Claude Code for teams that want model flexibility.
  8. 2026-05-22 / AGENTSGoogle Genkit Ships Middleware Layer: Composable Hooks for Agent Retry, Fallback, and Approval GatesGoogle announced Genkit Middleware on May 14, adding composable hooks that intercept generation calls including the tool execution loop. Pre-built middleware includes automatic retry with exponential backoff and jitter on transient errors (retrying only the model call, not the surrounding tool loop), model fallback that switches providers when quota is exceeded, and tool approval gates for human-in-the-loop workflows. Available in TypeScript, Go, and Dart with Python coming soon. For teams building production agents, this addresses a critical gap between prototype and reliable deployment.
  9. 2026-05-21 / PROJECTSGemini CLI Sunset Announced — Developers Must Migrate to Antigravity CLI by June 18Google announced that Gemini CLI and Gemini Code Assist IDE extensions will stop serving requests for Google AI Pro, Ultra, and free-tier users on June 18, 2026. The replacement Antigravity CLI, built in Go, shares the same agent harness as Antigravity 2.0 with cross-platform terminal sandboxing, credential masking, and hardened Git policies. Developers using Gemini CLI in their workflows have less than 4 weeks to migrate — a significant forcing function for the entire Gemini developer ecosystem.
  10. 2026-05-20 / MARKETSGoogle I/O 2026: Chrome DevTools for Agents and WebMCP Open Standard Redefine How AI Agents Interact with Web AppsGoogle announced Chrome DevTools for agents — giving coding agents like Gemini CLI, Antigravity, and Cursor direct access to web app runtimes — alongside a proposed WebMCP open web standard for browser-based AI agents. Managed Agents in the Gemini API also shipped, providing one-API-call provisioning of a fully sandboxed agent with remote execution. For SaaS companies: the browser is becoming an agent-accessible surface whether you build for it or not, and WebMCP could standardize how every web app exposes functionality to AI.
  11. 2026-05-20 / SOURCESGoogle I/O 2026: Android 17 Ships Stable Android CLI for AI Agent Workflows — Claude Code, Codex, Antigravity All SupportedAndroid 17's headline developer feature is the now-stable Android CLI, a command-line toolset designed for AI agents and LLMs including Claude Code, Codex, and Antigravity. It supports semantic code analysis, warning detection, Jetpack Compose rendering, and automated UI testing. Android 17 also unifies widget development under Compose via Jetpack Glance across mobile, Wear OS, and cars, and adds an Android Performance Analyzer with AI-assisted trace analysis. This signals Google treating agentic coding as a first-class development paradigm.
  12. 2026-05-17 / SKILLSGoogle ADK Context Engineering: Treat Context as Compiled View Over Session/Memory/Artifacts Tiers — Strict Scoped Handoffs for Multi-Agent ReliabilityGoogle's ADK architecture blog introduces tiered context engineering for multi-agent systems. Context is a compiled view over three tiers: Session (current conversation), Memory (persistent knowledge), and Artifacts (files/outputs). Two multi-agent patterns: 'Agents as Tools' (focused prompt in, result out, no history shared) vs 'Agent Transfer' (full context handoff). Explicit processors handle transformation, compaction, and caching at each tier boundary.
  13. 2026-05-17 / SKILLSGoogle ADK: Build Long-Running Agents That Pause for Days and Resume Without Losing Context — Durable State Machines + DatabaseSessionService PatternGoogle Developers Blog (May 12) published a tutorial on building agents that span days or weeks using ADK's durable state machine pattern. Key technique: replace conversation history with explicit state schemas (6 named states, linear progression), use DatabaseSessionService for checkpoint persistence across container restarts, and event-driven webhooks with state_delta for atomic state transitions before inference. Eliminates polling and prevents hallucinated progress.
  14. 2026-05-10 / SKILLSGoogle Jules Gets Gemini 3 Pro: Multi-Step Agentic Coding with Clearer Reasoning and Improved ReliabilityGoogle's always-on autonomous coding agent Jules now runs on Gemini 3 Pro, delivering improved agentic capabilities that translate to clearer reasoning, stronger intent alignment, and better reliability on multi-step tasks. Available first to Google AI Ultra subscribers. The upgrade suggests Google is positioning Jules as a direct competitor to Claude Code and Codex for autonomous background coding work.
  15. 2026-05-05 / AGENTSGoogle ADK Reaches v1.0 Stable Across Python, Go, and Java — A2A Protocol Hits v1.2 With Signed Agent CardsGoogle's Agent Development Kit achieved production-ready v1.0 across all three major languages (Python, Go, Java) with native Agent2Agent protocol support. A2A v1.2 now includes cryptographic signatures for agent card domain verification and is in production at 150+ organizations, governed by the AAIF under the Linux Foundation. The ecosystem now has 110M+ monthly MCP downloads and growing enterprise adoption of agent interoperability standards.
  16. 2026-04-26 / SKILLSGoogle ADK Tiered Context Architecture: Compile Working Context from Session + Memory + Artifacts Like a Compiler Optimizes Code — Cache-Friendliness as Hard ConstraintGoogle's Agent Development Kit treats context as a compiled system, not a string buffer. Three layers separate storage from presentation: Session (chronological interaction history), Memory (long-term cross-task knowledge), and Working Context (ephemeral optimized view). The Working Context is compiled each turn — like a compiler transforming source into machine code — by processor pipelines that compact history, retrieve relevant memory, and reference artifacts. Critically, ADK lets you order your pipeline to keep frequently reused segments stable at the front of the context window for cache hits, pushing dynamic content to the end.
  17. 2026-04-17 / HACKER NEWSGoogle Launches Android CLI — 70% Less Token Usage, 3x Faster Agent DevelopmentGoogle released Android CLI, an official command-line interface with modular markdown-based 'skills' that auto-trigger on matching prompts, plus an Android Knowledge Base searchable via 'android docs'. Reduces LLM token usage by 70% vs standard toolsets. Designed to work with Claude Code, Codex, Gemini CLI, and any agent — not just Android Studio. 298pts on HN.
  18. 2026-04-15 / SKILLSGemini CLI Ships Subagents: Google's CLI Now Orchestrates Parallel Specialized Agents with Isolated Context WindowsGoogle announced on April 15, 2026 that subagents have arrived in Gemini CLI. Each subagent operates within its own separate context window, custom system instructions, and curated tool set, keeping the main session lean while delegating sub-tasks. Parallel execution lets you dispatch multiple agents simultaneously — research five topics or refactor several components at once. Built-in subagents include Codebase Investigator (dependency mapping), CLI Help Agent, and an experimental Browser Agent. This positions Gemini CLI as a direct competitor to Claude Code's subagent architecture.
  19. 2026-04-14 / SKILLSGoogle ADK Context Engineering: Five-Layer Architecture Treats Context as First-Class System with Lifecycle and ConstraintsGoogle's open-source Agent Development Kit framework implements context engineering as a structured system with five distinct layers: system instructions, tool definitions, conversation history, retrieved knowledge, and user profiles. Each layer has its own lifecycle, optimization constraints, and budget management. The framework makes active context engineering achievable in production multi-agent systems rather than relying on ad-hoc prompt construction.
  20. 2026-04-13 / AGENTSGoogle Gemma 4 Launches: Open Model Family Enables On-Device Agents with Multi-Step Planning and Autonomous ActionGoogle released Gemma 4, an open model family enabling on-device AI agents with multi-step planning, autonomous action execution, offline code generation, and audio-visual processing. Designed for edge deployment without cloud dependency, Gemma 4 represents Google's push to make agent capabilities available on consumer hardware. For builders, this means shipping agents that work offline on phones and laptops.
  21. 2026-04-10 / TOOLSGoogle Colab MCP Server: Official Bridge Lets Any AI Agent Control Colab Notebooks with GPU AccessGoogle announced the open-source Colab MCP Server on April 7, letting MCP-compatible agents (Gemini CLI, Claude Code) create notebooks, execute code cells, manage dependencies, and rearrange outputs programmatically. Eliminates manual copy-paste between terminal and Colab. Designed to offload compute-intensive or unsafe tasks from local machines to cloud GPU runtimes.
  22. 2026-04-10 / NEWSGemini Code Assist Ships 'Finish Changes' and Code Outlines to VS Code and IntelliJ — GAGoogle's Gemini Code Assist now includes Finish Changes (Option+F), which observes in-progress edits and uses Gemini 3.0 to propagate patterns across the file without a prompt. Outlines generates concise English summaries interleaved with source code in the IDE sidebar. Both features are GA for all Gemini Code Assist users on VS Code and IntelliJ at no additional cost.
  23. 2026-04-05 / AGENTSGoogle ADK Java 1.0 Ships with Maps Grounding, Human-in-the-Loop Confirmation, and Event CompactionGoogle ADK Java 1.0.0 launched with advanced capabilities: GoogleMapsTool for location-based grounding, UrlContextTool for web content fetching, ToolConfirmation workflows that pause agents for human approval on sensitive operations (financial transactions, production DB changes), event compaction for context window management via summarization/retention strategies, a centralized plugin architecture with App-wide execution control, and Session/Memory services with Vertex AI and Firestore persistence. The Java release targets the enterprise developer community that dominates backend services.
  24. 2026-04-05 / AGENTSGoogle Launches ADK for TypeScript: Code-First Agent Framework Extends to JavaScript EcosystemGoogle released Agent Development Kit (ADK) for TypeScript, completing a four-language ecosystem (Python, Java, Go, TypeScript). The framework enables TypeScript/JavaScript developers to build AI agents using familiar tooling — npm install @google/adk, define logic/tools/orchestration in code, deploy to local/container/Cloud Run. It's model-agnostic (supports Gemini 3 Pro, Flash, and third-party models) and includes @google/adk-devtools for interactive CLI and web UI debugging. ADK for TS makes Google's agent framework accessible to the largest single-language developer community.
  25. 2026-04-05 / PROJECTSGoogle Antigravity: Full-Stack Vibe Coding Agent Integrated Into AI Studio With FirebaseGoogle's Antigravity coding agent, originally launched with Gemini 3, received a major March 2026 update integrating it directly into Google AI Studio with built-in Firebase support. The agent proactively detects when apps need databases or auth, auto-provisions Cloud Firestore and Firebase Authentication, and supports Next.js, React, and Angular. Powered by Gemini 3.1 Pro and Gemini 3 Flash, it's free in public preview and represents Google's answer to the vibe coding trend.
  26. 2026-04-04 / SKILLSGemma 4 Android AICore Developer Preview: On-Device Agentic Intelligence, 4x Faster, 60% Less BatteryGoogle announced Gemma 4 as the base for next-gen Gemini Nano 4, available via Android AICore Developer Preview. The on-device model is 4x faster and uses 60% less battery than the previous version. Android Studio now supports Gemma 4 as a local agentic coding assistant. The AICore preview will add tool calling, structured output, system prompts, and thinking mode. Developers can also use Google AI Edge for cross-platform edge deployment.
  27. 2026-04-03 / AGENTSGoogle ADK Go 1.0 Ships with Native OpenTelemetry, A2A Protocol, and YAML Agent ConfigurationGoogle released Agent Development Kit (ADK) for Go 1.0.0 on March 31, marking the Go agent ecosystem's shift from experimental scripts to production services. Key features include native OpenTelemetry integration generating structured traces and spans across model calls and tool execution, YAML-based agent configuration eliminating boilerplate Go code, a plugin system with Retry-and-Reflect self-correction, and human-in-the-loop confirmation flows for sensitive operations. The A2A protocol has been refined for seamless communication between Go, Java, and Python agents with automatic event ordering and response aggregation — completing Google's ADK language coverage across four major runtimes.
  28. 2026-04-01 / AGENTSGoogle Ships ADK for Java 1.0.0: Agent Development Kit Expands to Java with A2A Native Support and ComputerUseToolGoogle released Agent Development Kit for Java 1.0.0 on March 30, bringing the Gemini agent framework to the Java ecosystem with native Agent-to-Agent (A2A) protocol support via the official A2A Java SDK Client. Key capabilities include ComputerUseTool for browser/computer control abstraction, ContainerCodeExecutor and VertexAiCodeExecutor for sandboxed code execution, GoogleMapsTool and UrlContextTool, human-in-the-loop ToolConfirmation workflows, and plugin architecture with LoggingPlugin, ContextFilterPlugin, and GlobalInstructionPlugin. Event compaction enables context window management through summarization, and state persistence supports InMemory, VertexAi, and Firestore backends.
  29. 2026-04-01 / NEWSGoogle Announces ADK for Java 1.0.0 — AI Agent Development Kit Brings First-Class Java SupportGoogle released ADK (Agent Development Kit) for Java 1.0.0, providing native Java support for building AI agents with Spring Boot integration and support for all major LLM providers. This is the first production-ready agent framework targeting the Java ecosystem specifically, addressing the gap where most agent tooling has been Python-first. Follows JetBrains' Koog framework announcement from March 30, signaling a race to capture the enterprise Java agent developer market.
  30. 2026-03-20 / SKILLSGoogle ADK TypeScript: Typed Data Contracts Between Multi-Agent Nodes Catch Integration Bugs at Compile TimeGoogle's Agent Development Kit for TypeScript (adk-js) brings static typing to the inter-agent message boundary — the point where orchestrator agents hand off tasks to sub-agents. Defining tool input/output as TypeScript interfaces means the compiler catches schema mismatches before runtime, a class of bug that causes silent agent failures in Python-based frameworks. ADK is model-agnostic (Gemini 3 Pro/Flash, plus third-party) and deployment-agnostic (local, container, Cloud Run), with modular multi-agent hierarchies composable via the same TypeScript patterns developers already use for API design.
  31. 2026-03-20 / SKILLSGemini CLI v0.34.0 Plan Mode: Read-Only Reasoning with ask_user Clarification and MCP Context Before Any WriteShipped March 17 2026, Gemini CLI's Plan Mode (activated via /plan or Shift+Tab) puts the agent in a read-only state where it can navigate code, grep patterns, and pull read-only MCP tools (GitHub issues, Postgres schemas, Google Docs) but cannot modify files. The new ask_user tool lets it pause and prompt you with targeted questions before finalizing a plan. Only after explicit approval does it exit plan mode and begin writes, eliminating the common failure where eager agents overwrite files based on misunderstood intent.
  32. 2026-03-18 / TOOLSGoogle A2UI Spring 2026 Update: OpenClaw Integration and ADK Python Expansion at v0.8 Public PreviewGoogle's A2UI (Agent-to-User Interface) declarative standard shipped a Spring 2026 update adding OpenClaw integration and expanded ADK Python support, now at v0.8 Public Preview under Apache 2.0. A2UI lets agents generate structured UI responses (using client-trusted component catalogs, not arbitrary code) that render across Angular, Flutter, React, and native mobile from a single agent response. The OpenClaw integration is the key signal: connecting the largest agent ecosystem (210K+ stars) to Google's UI standard could establish A2UI as the cross-platform agent UI layer before competing specs gain traction.
  33. 2026-03-17 / AGENTSGoogle ADK for TypeScript GA: Code-First Multi-Agent Framework With Native A2A and MCP — Python ADK Hits v1.26.0Google launched the Agent Development Kit for TypeScript as an open-source, code-first multi-agent framework, completing a four-language SDK family alongside Python, Java, and Go. ADK Python simultaneously released v1.26.0 on March 13. Both SDKs are model-agnostic with Gemini 3 Pro/Flash optimization and built-in A2A and MCP interoperability — making ADK the first multi-language agent SDK with full protocol-native support across all variants.
  34. 2026-03-17 / TOOLSGoogle Ships Official Colab MCP Server — Any AI Agent Can Now Control Notebooks NativelyGoogle released the open-source Colab MCP Server (googlecolab/colab-mcp) on March 17, 2026, giving any MCP-compatible agent native control over Google Colab notebooks — autonomous cell creation, Python execution, visualization generation, and full notebook lifecycle management. The server bridges local agent workflows with Colab's cloud compute environment and is open to community contributions. This makes Colab a first-class execution backend for MCP-native agents without any custom integration work.
  35. 2026-03-17 / NEWSGoogle Ships Gemini Code Assist 'Finish Changes' and 'Outlines' in IntelliJ and VS Code — Closing Gap on Claude Code Agentic FeaturesGoogle launched two new Gemini Code Assist features in its IntelliJ and VS Code extensions: 'Finish Changes,' which autonomously completes partially-started code edits across a file, and 'Outlines,' which generates structural scaffolding for new files from natural language descriptions. The features ship the same week as VS Code 1.111's Autopilot mode and directly compete with Claude Code's multi-file edit capabilities. Google's IDE agent features are catching up to Anthropic and Microsoft in the agentic coding layer.
  36. 2026-03-17 / SKILLSGoogle Android Bench: Platform-Specific LLM Leaderboard for Model Selection in Mobile DevelopmentGoogle released Android Bench, the first official leaderboard evaluating LLMs specifically on Android development tasks sourced from real-world public GitHub repos, covering breaking changes across Android releases, Wear OS networking, and Jetpack Compose migration. Current top scorers are Gemini 3.1 Pro (72.4%), Claude Opus 4.6 (66.6%), and GPT-5.2-Codex (62.5%), with all models completing only 16–72% of tasks — revealing a significant capability gap that general coding benchmarks like SWE-bench don't surface. The dataset and test harness are open-source on GitHub with anti-contamination canary strings, making it the primary benchmark to run before selecting a model for any Android-focused coding agent.
  37. 2026-03-02 / NEWSGoogle Developer Knowledge API and MCP Server Public PreviewGoogle launched Developer Knowledge API with remote MCP server for machine-readable always-current developer docs across Firebase Android Cloud
  38. 2026-03-02 / NEWSGoogle Developer Knowledge API and MCP ServerGoogle launches public preview of Developer Knowledge API with remote MCP server for always-current dev docs
  39. 2026-02-27 / NEWSGoogle ADK TypeScript Open Source Agent FrameworkGoogle released Agent Development Kit for TypeScript. Open source code-first multi-agent framework. Native MCP integration. Model-agnostic. Three-way agent framework war: Google ADK vs Microsoft Agent Framework vs Anthropic Agent Skills.
  40. 2026-02-13 / TOOLSGoogle Antigravity: Agent-First IDE Free in Public PreviewGoogle's .4B Windsurf acquisition becomes free agent-first IDE with Manager View
Open latest cited source