Google opens Android Studio to third-party AI coding agents

Opening the dev environment

Google is significantly altering how developers build Android applications by opening the Android Studio environment to external AI tools. Through a new 'Bring Your Own Agent' feature, developers are no longer restricted to using only Google's native AI offerings. They can now integrate more than 40 different coding agents into their workflow using the open Agent Client Protocol.

The update allows for the simultaneous use of popular models like Anthropic's Claude Agent and OpenAI's Codex alongside Google's own Antigravity. These agents gain deep access to the internal tools of the IDE, including the project structure, symbol resolution, and the emulator. This move aims to give developers more flexibility in choosing the best AI partner for their specific coding needs.

To help measure the effectiveness of these various models, Google also released version 2.0 of the Android Bench leaderboard. This system evaluates how different language models perform on complex, multi-day development projects. It specifically looks at functionality, the total time spent on tasks, and the overall cost of the AI's output.

Optimizing agent workflows

To make AI coding more efficient, Google introduced new terminal tools and specialized skills designed to reduce token usage. Tokens are the basic units of text that AI models process, and reducing them helps lower the cost and speed of development. Internal tests showed that these new CLI tools helped agents finish UI tasks 77 percent faster while using up to three times fewer tokens.

The expanded Android CLI command-line tool now supports official agent plugins for Claude Code and OpenAI's Codex. These tools allow agents to render Compose previews directly in the terminal, which lets them visually check user interface changes without needing to boot up a full emulator. Additionally, the CLI can stream physical test devices from Google labs so agents can investigate bugs specific to certain hardware.

Debugging has also received significant upgrades with the integration of the open-source LeakCanary library. This library helps detect memory leaks by evaluating heap dumps on a computer instead of the mobile device itself. This approach prevents the app from slowing down during the testing phase. In one test using Anthropic's Claude Opus 5.5, this tool helped find and fix memory leaks nearly twice as fast while cutting token usage by about 25 percent.

Local and connected agents

Google is also shipping new APIs that allow developers to build agent features directly into their own apps. For tasks that need to happen on the device without internet access, Google added an alpha Agentic API to its ML Kit. This API is powered by Gemini Nano 4 and handles the orchestration loop, chat history, and safety filters automatically for local processing.

To connect apps to broader system-wide agents, the company is launching the AppFunctions API in beta. This serves as a unified tool-calling layer, allowing apps to share specific functions with both local and system agents. Furthermore, the new Agent-to-UI interface for Jetpack Compose allows agents to display interactive components from an app instead of just providing plain text responses.

For wider reach, the Connected Apps Platform will allow developers to link their services to Google Search and the Gemini app. Over 50 partner integrations from companies like Adobe and Peloton are already live in the Gemini app. These integrations will begin rolling out to web search in the coming weeks, allowing users to interact with third-party services through Google's primary AI interfaces.

Hardware and XR updates

The updates extend beyond software to include new hardware capabilities for extended reality. Google partnered with XREAL to unveil the XREAL AURA, which are the first wired smart glasses running Android XR. These glasses weigh under 95 grams and feature transparent, dimmable lenses that provide a 70-degree field of view while the battery and computing unit sit in an external puck.

The XREAL AURA uses the Snapdragon Reality Elite chip to display virtual workspaces and can mirror a laptop screen via a USB-C cable. Because Gemini is integrated into the glasses, the AI can see both the real-world environment and the virtual screens simultaneously. This allows the AI to perform tasks like organizing windows, offering video game tips, or highlighting physical objects in the room.

Software support for these types of devices is also maturing as the Extended Reality development environment moves out of the testing phase. Google has released the Jetpack XR SDK and stable emulators for Android Studio. These tools, along with the Engine Hub for popular game engines like Unity and Unreal, are now available to help developers build for large screens and XR hardware.

New developer resources

To bridge the gap between AI capabilities and the latest software updates, Google is providing over 30 new Android Skills. These skills feed specific documentation to AI agents to help them handle complex tasks like checking Google Play policies. They also assist with handling permission security issues and adapting app layouts for foldables and laptops.

For those moving away from older development methods, the Android Studio Canary preview now includes a migration assistant. This tool automatically translates older XML layouts into Jetpack Compose, which is Google's modern toolkit for building native UI. This helps developers modernize their codebases more quickly as they transition to AI-assisted workflows.

Finally, Google is testing Lightbuild, a new declarative build system based on a YAML schema. This system is designed to be readable and editable by both humans and AI agents. By using a standardized format, Google hopes to make it easier for agents to understand and modify the build process of Android applications.