Android Studio integrates external AI agents and CLI tools for autonomous development

At the 2026 Android Dev Summit, Google unveiled significant updates to its developer tooling ecosystem, focusing on deep integration with artificial intelligence agents. The announcements, released on October 28, 2026, center on enabling developers to use their preferred coding agents and large language models directly within Android Studio and via the command line. These changes aim to streamline workflows for building, testing, and optimizing Android applications by granting AI agents first-class access to the Android toolchain.

What happened

The core update is the stabilization of Bring Your Own Agent (BYOA) in Android Studio Rabbit 2. This feature allows developers to connect over 40 different ACP-compatible coding agents to the IDE through the open Agent Client Protocol. Instead of being locked into a single provider, teams can now install multiple agents from a marketplace and switch between them as needed. This integration gives external agents access to Android Studio’s internal project understanding, including symbol resolution, build systems, and Compose Preview rendering.

Simultaneously, Google expanded Android CLI to support autonomous operations in terminal environments and continuous integration pipelines. New capabilities include streaming connections to real physical devices from remote labs, rendering Compose Previews without an emulator, and installing official plugins for popular agents like Claude Code and Codex. These tools allow agents to build, run, and debug apps with greater independence, reducing the need for manual intervention during repetitive tasks.

Performance tooling also received major agentic upgrades. LeakCanary is now integrated into Android CLI, enabling agents to analyze heap dumps locally rather than on-device, which significantly reduces token usage and execution time. Additionally, the Android Performance Analyzer now supports command-line tracing and analysis, allowing agents to capture system traces, propose fixes, and verify improvements automatically. These features work in tandem with new production signal tools that bring Firebase Crashlytics and Android vitals data directly into the terminal.

How it works

The Bring Your Own Agent framework relies on the Agent Client Protocol (ACP), an open standard that facilitates communication between the IDE and external AI services. When a developer selects an agent in Android Studio, the IDE exposes its internal state—such as project structure and UI previews—to the agent. This context allows the AI to generate code that aligns with existing patterns and immediately visualize changes through Compose Previews. For code reviews, a new Change Review feature lets agents analyze pending Git changes, identify potential issues, and suggest fixes based on a structured report.

On the command line, Android CLI acts as a bridge between AI agents and the Android platform. Commands like android device remote provision real hardware from Google’s secure device labs, while android preview renders UI components headlessly. This headless rendering enables agents to verify visual correctness across different screen sizes without launching a full emulator. For performance debugging, the android profiler leakcanary subcommand processes heap dumps on the local machine, offloading heavy computation from the test device and providing structured data that agents can interpret efficiently.

Key details

• BYOA Stability: Bring Your Own Agent is stable in Android Studio Rabbit 2, supporting over 40 ACP-compatible agents.
• Android Bench 2.0: A new leaderboard evaluates LLMs on real-world Android tasks, grading solutions on functionality, cost, and time.
• Physical Device Streaming: Android CLI now supports connecting to real remote devices via android device remote for authentic testing.
• LeakCanary Integration: The android profiler leakcanary command analyzes heap dumps locally, reducing token usage by up to 25% in internal tests.
• XR Emulators: Emulators for XR glasses are now stable, featuring customizable 360-degree environments and hardware LED indicators.
• R8 Analysis: AGP 9.3 includes the R8 Configuration Analyzer, which scores keep rules and helps optimize app size.

Why it matters

For software engineers and technical leads, these updates reduce the friction associated with integrating AI into established workflows. By supporting multiple agents through BYOA, organizations can avoid vendor lock-in and choose models that best fit specific tasks, such as migration or debugging. The ability to run agents in CI pipelines via Android CLI means that automated testing and code generation can happen at scale, without requiring expensive IDE licenses for every build step. This shift enables more consistent and faster iteration cycles, particularly for teams managing large codebases.

The focus on agentic performance analysis addresses a common bottleneck in mobile development: identifying and fixing subtle memory leaks and performance regressions. Traditionally, these tasks require manual profiling and deep expertise. By automating the capture and analysis of heap dumps and system traces, developers can resolve issues earlier in the cycle. The integration of production signals into the terminal further shortens the feedback loop, allowing engineers to investigate live crashes and ANRs without switching contexts between monitoring dashboards and code editors.

What you can do

• Update to Android Studio Rabbit 2 to enable Bring Your Own Agent and explore the agent marketplace.
• Install the Android CLI plugin for your preferred coding agent, such as Claude Code or Codex, to enable terminal-based automation.
• Use the android preview command in your CI pipelines to verify Compose UI changes without launching emulators.
• Run android profiler leakcanary to automate memory leak detection during local development or testing.
• Try the Migration Assistant in Android Studio Canary to convert XML layouts to Jetpack Compose using AI guidance.
• Evaluate different models using Android Bench 2.0 results to select the most cost-effective agent for your specific development tasks.