Google opens Android Studio and CLI to third-party AI agents

Google has officially opened its Android development ecosystem to third-party artificial intelligence agents, allowing tools like Anthropic’s Claude Code and OpenAI’s Codex to interact deeply with Android Studio and the command-line interface. Announced on October 28, 2026, at the Android Dev Summit, this move enables these external agents to not only generate code but also build, test, debug, and verify fixes within the official toolchain. The update marks a significant shift from proprietary lock-in, giving developers the freedom to choose their preferred AI models while integrating them into Google’s professional-grade infrastructure.

What happened

The core of this announcement is the stabilization of the "Bring Your Own Agent" feature in the new Android Studio Rabbit 2 release. This feature supports more than 40 coding agents that comply with the open Agent Client Protocol. Rather than forcing developers to use Google’s own Gemini-based tools, the company is providing standardized access to its integrated development environment. Developers can install multiple agents, switch between them seamlessly, and allow these agents to inherit the IDE’s full context, including project structure, build configurations, and symbols. This eliminates the need for agents to reconstruct context through slow code searches or terminal guesswork.

Beyond the graphical interface, Google has released official Android CLI plugins for major agents like Claude Code and Codex. These plugins extend agent capabilities to project creation, emulator management, UI inspection, and documentation access. A key addition is the Change Review feature, which scans pending modifications for common issues and guides developers through large sets of changes before they are applied. This ensures that autonomous suggestions are vetted for quality and consistency before merging, addressing concerns about unchecked AI-generated code.

How it works

The integration relies on the Agent Client Protocol, an open standard that allows external AI models to communicate with Android tools. When an agent connects to Android Studio, it gains read access to the same contextual data a human developer sees, such as Compose Preview states and Logcat outputs. This shared context allows the agent to understand the app’s current state without re-parsing files from scratch. For terminal users, the new android preview command enables agents to render Jetpack Compose previews directly in the command line, bypassing the need for a full IDE or emulator launch. Google claims this method allows agents to complete UI tasks 77% faster while using up to three times fewer tokens, though specific measurement methodologies were not detailed.

For debugging and testing, the CLI now includes an android device remote command, granting agents access to physical devices in Google’s labs. This allows for real-world testing rather than relying solely on emulators. Additionally, the android signals command pipes production data, such as Android vitals and Firebase Crashlytics reports, into the terminal workflow. Agents can use this data to investigate memory leaks, application not responding (ANR) errors, and crashes. In a demonstrated workflow, Claude Code used the LeakCanary library to identify memory leaks across multiple devices, trace them to specific code segments, apply fixes, and rerun tests to confirm resolution.

Key details

• The "Bring Your Own Agent" feature is now stable in Android Studio Rabbit 2, supporting over 40 agents via the Agent Client Protocol.
• New CLI commands include android preview for terminal-based Compose rendering and android device remote for accessing physical test devices.
• Google reports that agents using the terminal preview command completed UI tasks 77% faster and used up to three times fewer tokens.
• The android signals command integrates Firebase Crashlytics and Android vitals data directly into the agent’s terminal workflow for production debugging.
• Android Bench 2.0 has been introduced to evaluate agent performance on complex tasks, measuring both completion time and cost.
• Matthew McCullough, VP of Product for Android, emphasized that the goal is to provide developer choice rather than steering users toward Google’s proprietary Gemini models.

Why it matters

For software engineers and technical leads, this development reduces friction in adopting AI-assisted workflows. Previously, using external AI tools often required copying code snippets between windows or relying on limited plugin integrations that lacked deep IDE context. By granting agents native access to build systems, profilers, and crash reports, Google enables a more cohesive loop where AI can verify its own work. This shifts the role of the developer from writing every line of code to reviewing and guiding autonomous agents that can handle routine debugging and UI adjustments. The ability to switch between different AI models within the same IDE also prevents vendor lock-in, allowing teams to select the best model for specific tasks based on cost or performance.

However, the speed of autonomous bug-finding introduces new challenges. As noted in the source, autonomous detection can outpace fixing, with one scanner identifying 29,000 potential vulnerabilities while only a few hundred were resolved. The introduction of tools like Change Review and Android Bench 2.0 attempts to mitigate this by providing structured evaluation and pre-merge checks. Developers must still maintain a "human in the loop" approach, as AI agents may miss subtle issues like race conditions, even if they efficiently resolve obvious memory leaks. The focus now shifts to managing and verifying AI output rather than just generating it.

What you can do

• Update to Android Studio Rabbit 2 to access the stable "Bring Your Own Agent" feature and install compatible AI agents.
• Experiment with the android preview command in your terminal to reduce token usage and speed up UI iteration tasks.
• Configure the android signals command to connect your agent to Firebase Crashlytics data for automated production issue investigation.
• Use the Change Review feature to audit large sets of AI-generated modifications before applying them to your codebase.
• Evaluate your preferred AI models using Android Bench 2.0 to understand their performance and cost implications for complex Android tasks.
• Integrate LeakCanary with your AI agent workflows to automate the detection and resolution of memory leaks during development.