Google used this year's Android Dev Summit to make its developer tools agent-agnostic, and to push Gemini deeper into apps, search and a new pair of smart glasses.
The headline change is in Android Studio Rabbit 2, which adds a "Bring Your Own Agent" feature. Developers are no longer stuck with Google's own AI tools. Using the open Agent Client Protocol, they can now run more than 40 coding agents directly inside the IDE, including Anthropic's Claude Agent and OpenAI's Codex alongside Google's own Antigravity, all through a built-in marketplace. Those agents get direct access to project structure, symbol resolution, the emulator and Logcat, the tools a human developer would normally reach for.
Checking the agents' work
Google also shipped version 2.0 of its Android Bench leaderboard, which scores AI models on multi-day development projects by functionality, time and cost. Because AI-written code tends to get messy fast, a new Change Review tool analyzes pending changes, generates bug reports and explains edits step by step. A migration assistant in the Android Studio Canary preview will translate older XML layouts into Jetpack Compose automatically, and Google is testing a YAML-based build system called Lightbuild that both people and agents can read.
For developers who prefer the terminal, the Android CLI now ships as an official plugin for Claude Code, Codex and Antigravity. New commands let agents render Compose previews in the terminal without booting an emulator. Google says internal testing found agents completed UI tasks 77 percent faster and used up to three times fewer tokens with that feature, though those figures come from Google's own benchmarks rather than independent testing.
The open-source LeakCanary memory-leak detector is now built into both the CLI and the Studio Profiler, evaluating heap dumps on a desktop machine instead of the phone so the app under test doesn't slow down. In an A/B test Google ran using Anthropic's Claude Opus 5.5 model, the CLI reportedly found and fixed memory leaks nearly twice as fast while cutting token use by about 25 percent. More than 30 "Android Skills" now feed agents documentation on things like Play policy checks and foldable layouts, covering gaps where models haven't been trained on newer APIs.
Gemini moves into apps and search
Outside the IDE, Google is adding an alpha Agentic API to ML Kit, powered by Gemini Nano 4, for on-device agents that can execute functions and manage their own orchestration loop locally. A beta AppFunctions API, built on the Model Context Protocol, lets apps expose their functions to local or system-wide agents, and a new Agent-to-UI interface lets agents respond with interactive app components instead of plain text.
Google is also piloting a Connected Apps Platform that links apps into Search and the Gemini app via MCP. The company says more than 50 partner integrations, including Adobe, Peloton and Experian, are already live in the Gemini app, with a web search rollout coming in the following weeks.
A new pair of glasses
Google paired the software push with hardware: XREAL's AURA, built on Android XR. Google's XR chief Shahram Izadi called it the first wired smart glasses running the platform. The glasses weigh under 95 grams, have transparent dimmable lenses with a 70-degree field of view, and offload computing and battery to an external puck built on Qualcomm's new Snapdragon Reality Elite chip. Wearers can pin up to five Android apps in view or plug in over USB-C to mirror a laptop screen, with Gemini able to see both the real world and the virtual windows to organize them or flag objects in the room.
The XREAL AURA is available to order now in the US, Canada and South Korea for $1,099, with shipping due to start in mid-November.
ADFiled by The AI Desk
Models, assistants and the companies and chips behind them, reported from what was released and what was claimed, with the difference kept clear.
More from this desk →
Be the first to comment
Join the argument. No password, just your email or a passkey.