Google opens Android Studio to 40-plus coding agents

By Marcus Bennett

Google is bringing more than 40 third-party coding agents into Android Studio Rabbit 2 at the Android Dev Summit, including Anthropic’s Claude Agent and OpenAI’s Codex.

The “Bring Your Own Agent” (BYOA) feature uses the open Agent Client Protocol (ACP), allowing developers to run competing agents inside the IDE (integrated development environment) rather than relying only on Google’s tools.

The market stakes are toolchain control: developers can choose Anthropic or OpenAI without leaving Google’s IDE, while Google retains the surrounding development workflow.

What changes inside Android Studio?

Google’s built-in marketplace lets developers install and run Claude Agent, Codex and Google Antigravity simultaneously. Each agent can directly access:

• Project structure
• Symbol resolution
• The emulator
• The Logcat system

This direct access gives agents a broader view of the application than a standalone coding interface. It also lets developers mix model vendors without moving their work to another tool.

How does Google test and audit coding agents?

Google released version 2.0 of the Android Bench leaderboard to measure language-model performance on complex, multi-day development projects. It evaluates three measures:

• Functionality
• Time spent
• Cost

The benchmark measures an agent’s ability to sustain work across days instead of completing one isolated coding prompt.

For code review, Google is adding Change Review because AI-generated changes often become difficult to audit. The tool analyzes pending changes, generates bug reports and explains modifications step by step.

An Android Studio Canary preview also includes a migration assistant that automatically translates older XML (Extensible Markup Language) layouts into Jetpack Compose.

Google separately is testing Lightbuild, a declarative build system based on a YAML schema. YAML is a human-readable configuration format, and Lightbuild allows both people and AI agents to read and edit the build definition.

How much faster can terminal agents work?

Google expanded the Android CLI (command-line interface) for developers who run agents in a terminal or continuous-integration (CI) pipelines. It now works as an official agent plugin for Claude Code, Codex and Antigravity.

New commands render Jetpack Compose previews directly in the terminal, letting agents inspect user-interface (UI) changes without booting an emulator. Google’s internal tests found:

• Agents completed UI tasks 77 percent faster.
• Agents reduced token consumption by as much as threefold.

The CLI can also stream physical test devices from Google labs, giving agents access to real hardware when they investigate device-specific bugs.

For memory defects, Google integrated the open-source LeakCanary memory-leak detection library into the CLI and Android Studio Profiler. It evaluates heap dumps on the computer rather than the mobile device, helping avoid app slowdown during testing.

In a Google A/B test using Anthropic’s Claude Opus 5.5, the CLI tool found and fixed memory leaks nearly twice as fast while cutting token usage by about 25 percent.

The expanded command-line toolset also lets agents:

• Capture system traces through the Android Performance Analyzer.
• Pull crash reports from Firebase Crashlytics.
• Check code optimization through the R8 compiler.

Google says language models often lack training on the newest APIs, so it provides more than 30 Android Skills. Those packages feed documentation for checking Google Play policies before submission, handling permission-security issues and adapting layouts for foldables and laptops branded as Googlebooks.

What can Gemini Nano 4 do on a phone?

Beyond developer tooling, Google is adding an alpha Agentic API to ML Kit, its machine-learning toolkit for Android. Powered by Gemini Nano 4, the interface supports local, on-device processing and can:

• Execute functions automatically.
• Save chat history.
• Manage the orchestration loop, the sequence that coordinates functions and model calls.
• Apply safety filters.

Google is also using the Model Context Protocol (MCP), a protocol for connecting models to external tools, and launching the AppFunctions API in beta as a unified tool-calling layer. Apps can expose their functions to local or system-wide agents through that layer.

An alpha Agent-to-UI (A2UI) interface for Jetpack Compose builds on MCP and AppFunctions. Instead of returning plain text blocks, it lets agents display interactive interfaces assembled from an app’s own components.

That approach gives assistants access to app services while preserving each developer’s native interface design. It also expands the role of an Android app from a destination in Search or Gemini to an action those assistants can perform.

How do apps reach Search and Gemini?

Google’s Connected Apps Platform preview links apps to Google Search and the Gemini app through Android MCP or cloud-hosted MCP servers. Developers manage those connections through a single console.

Google says more than 50 partner integrations from Adobe, Peloton and Experian are already live in the Gemini app. The company plans to roll them out to web search in the coming weeks.

The distribution stakes are measurable: Google is turning existing Android services into callable tools for two of its largest assistant surfaces, with at least 50 live partner examples before the web-search expansion begins.

What changes for XR and Android XR glasses?

Google also updated Jetpack Compose 1.13 and the AppState model so apps can adapt more easily to large screens. The extended-reality (XR) development environment has left testing, with stable releases for:

• The Jetpack XR software development kit.
• XR emulators in Android Studio.
• The Engine Hub for Unity, Godot and Unreal.

At the summit, Google and hardware maker XREAL introduced the XREAL AURA. XR chief Shahram Izadi said the glasses are “the first wired smart glasses running Android XR.”

Google lists the AURA’s core specifications as:

• A weight below 95 grams.
• Transparent, dimmable lenses.
• A 70-degree field of view.
• An external puck containing the computing unit and battery, which also serves as a trackpad.
• Qualcomm’s new Snapdragon Reality Elite chip.

Users can place up to five Android apps in the field of view at once or connect a USB-C cable to mirror a laptop screen.

Gemini is built into the glasses and can inspect both the real-world environment and virtual screens. Google says it can organize windows, offer video-game tips and visually highlight physical objects in the room.

XREAL now accepts orders in the United States, Canada and South Korea for $1,099, with shipping scheduled to begin in mid-November.

What happens next?

Together, the announcements move Android’s agent story from model selection to execution. Developers can bring competing agents into one IDE, connect app functions to Gemini and Search, and run some agent features locally on-device.

Adoption of the 40-plus-agent environment and the 50-plus connected integrations will show whether the strategy works. The next scheduled checkpoints are XREAL AURA shipments in mid-November and the Connected Apps rollout to web search in the coming weeks.