Android Studio Just Opened to 40+ AI Agents — Claude, Codex and Antigravity in One IDE, Explained Honestly

Nothing here is sponsored — no affiliate links on this page. Every fact below comes from Google's Android Dev Summit 2026 announcements, reported October 28–29, 2026. We have not run the new Android Studio agent features hands-on, and this is researched analysis, not hands-on testing.

Google has stopped treating Android Studio as a one-vendor IDE. At Android Dev Summit 2026, Android Studio Rabbit 2 went stable with "Bring Your Own Agent": developers can now run more than 40 coding agents directly inside the IDE — including Anthropic's Claude Agent, OpenAI's Codex and Google's own Antigravity — over the open Agent Client Protocol. If you build Android apps, this is the announcement that changes your daily workflow, not the hardware on the side of the stage.

What actually shipped for developers

Bring Your Own Agent, in the IDE. With Android Studio Rabbit 2, developers are no longer limited to Google's own AI tools. They can install and run Claude Agent, Codex and Antigravity through the built-in marketplace, alongside more than 40 agents compatible with the Agent Client Protocol. These agents get direct access to IDE tools: project structure, symbol resolution, the emulator and Logcat. That access is the whole point — an agent that can only read your open file is a chatbot; an agent that can see the project, run the emulator and read the logs can actually debug.

Compose previews without the emulator. Agents can render Jetpack Compose previews without running an Android Emulator or deploying an APK, and Firebase-powered device streaming lets developers test on real devices remotely. In Google's internal tests, rendering Compose previews in the terminal made agents finish UI tasks 77 percent faster with up to three times fewer tokens. Treat that as Google's number, not an independent benchmark — but the mechanism is easy to believe: booting an emulator for every UI check is exactly the kind of slow loop that eats agent time and tokens.

Android CLI becomes an official agent plugin. The Android CLI is now an official agent plugin for Claude Code, Codex and Google Antigravity, with new capabilities for device testing, UI validation, performance profiling and app quality analysis. It also pulls in LeakCanary for memory-leak detection: in a Google A/B test with Claude Opus 5.5, the CLI tool found and fixed memory leaks nearly twice as fast with about 25 percent fewer tokens, because heap dumps are evaluated on the computer rather than slowing down the device. Perfetto system traces help investigate cold startup, UI jank and background wakeups, and the Analyze feature checks app bundles and libraries against the same logic Google Play uses, before upload.

30+ Android Skills feed agents the docs they lack. Language models often have no training on the newest APIs, so Google now provides more than 30 Android Skills across more than 20 areas. They feed documentation to agents for tasks like checking Google Play policies before submission, handling permission security issues, optimising Compose performance, laying out Glance widgets, and adapting layouts for foldables and laptops. This is the unglamorous part of the announcement and possibly the most useful: most agent failures on a real codebase are not reasoning failures, they are "this API changed six months ago" failures.

New APIs turn apps into agent tools. Beyond the IDE, Google is shipping the APIs that let developers build agent features into apps. The ML Kit Agentic API, in alpha, extends the ML Kit Prompt API with Gemini Nano 4 capabilities for on-device agents: it automatically executes functions, saves conversation history, manages the orchestration loop and includes automatic guardrails. AppFunctions, now in beta, is Android's unified tool-calling layer, letting apps expose functionality to on-device or system agents through Android MCP. A2UI, in alpha for Jetpack Compose, lets agents return interactive interfaces built from an app's own components instead of plain text blocks. The Connected Apps Platform preview — covered separately on ToolsDig on October 28 — is the distribution side of the same idea, linking apps into Gemini and Search; this article is about the building side.

Verification and migration tools, still early. Google also released Android Bench 2.0 to measure models on multi-day development projects by functionality, time and cost, and a code verification tool called Change Review that analyses pending changes, generates bug reports and explains modifications step by step. The Android Studio Canary preview adds a migration assistant that translates older XML layouts into Jetpack Compose, and Google is testing Lightbuild, a declarative YAML-based build system designed to be readable by both humans and agents. Note the stages: Bench and Change Review are new tools to evaluate, not proven workflows, and Lightbuild is a test, not a replacement for your build today.

What this does not mean

First, "40+ agents" does not mean 40 good agents for your project. Protocol compatibility is not quality. An agent still has to understand your architecture, your tests and your release process, and the only honest way to find out is to run the same real task — a bug fix with a failing test, a Compose migration, a permission audit — through two agents and compare the diff, the test result and the token bill.

Second, the speed and token numbers are Google's internal tests. "77 percent faster" and "three times fewer tokens" for terminal Compose previews, and "nearly twice as fast" for LeakCanary fixes, all come from Google's own testing, including one A/B test with a single model, Claude Opus 5.5. They are a reason to try the tools, not a promise about your codebase.

Third, most of the app-facing APIs are not stable yet. The Agentic API is alpha, A2UI is alpha, AppFunctions is beta, and the Connected Apps Platform is a preview. Building a production feature on an alpha API is a bet on Google's roadmap, not a finished platform — prototype on it, and keep the fallback path that works without an agent.

What to do this week

• If you build Android apps: install Rabbit 2 and run your existing agent — or two rival agents — on one real task you already know the answer to. Score the diff, the tests and the time, not the demo.
• If you run agents in CI or the terminal: try the Android CLI as an official plugin, and turn on terminal Compose previews for UI work before you change anything else — that is where Google's biggest claimed saving lives.
• If you ship to Play: run the Analyze check and the relevant Android Skills (Play policy, permission security, foldable adaptation) before your next submission, not after a rejection.
• If you want your app to be agent-ready: prototype with AppFunctions in beta and A2UI in alpha, but ship the non-agent path first — the alpha label means the interface can still move under you.
• Do not judge this by the agent count. The test that matters is mundane: give an agent a real bug, a failing test and a device log, and count how often it lands a clean, reviewed fix without inventing an API that does not exist. That is the whole product.

What we don't know yet

• Independent results: no independent benchmark yet confirms Google's speed, token or leak-fix claims on real-world Android codebases.
• Agent pricing: the announcement does not state pricing changes for the agents themselves — each agent still runs on its own provider's terms, which Google did not spell out here.
• How agents are ranked or surfaced: whether the IDE marketplace will rank, recommend or restrict agents, and on what basis, is not stated in the announcement.
• Stability timelines: no general-availability dates are given for the alpha and beta APIs in this announcement.

None of that makes the announcement small. Android's main IDE just stopped being tied to one vendor's AI tooling, and the same summit pushed agents into the apps themselves — on-device with Gemini Nano 4, through MCP tool-calling, and into Search and Gemini. The IDE war is now an agent-quality war, fought on your codebase. Pick your agent the way you would pick a hire: on a real task, with a real test suite, and a bill you can read.

Sources: Google, Android Dev Summit 2026 announcements, October 28–29, 2026, as reported by FoneArena (published October 29, 2026) and The Decoder (published October 28, 2026); Google's Connected Apps Platform coverage on ToolsDig, October 28, 2026.