Google just handed its rivals the keys to Android

Google is granting coding agents like Anthropic's Claude Code and OpenAI's Codex expanded access to its Android development tools, including devices, performance profilers and production crash reports. This enables agents to pinpoint bugs, test alterations, and validate their own fixes within Android Studio and via the command-line Android CLI.

The company has opened both platforms to third-party agents, rather than pushing developers towards Google's Gemini model. Matthew McCullough, vice president of product for Android, emphasized granting developers and their agents the freedom to select their preferred models and coding agents. Senior product manager Adarsh Fernando stated that the objective is to equip developers and their agents with the tools and knowledge necessary to construct superior Android applications more swiftly.

The Bring Your Own Agent feature now accommodates over 40 coding agents that adhere to the open Agent Client Protocol, including Anthropic's Claude Agent, OpenAI's Codex, and Google's Antigravity. Developers can now install multiple agents and switch between them without leaving the IDE, and connected agents inherit the IDE's understanding of a project, including its structure, build configuration, symbols, and tools like Compose Preview, the emulator, and Logcat.

Google also unveiled a Change Review feature that scrutinizes an agent's proposed changes for prevalent issues and can guide developers through extensive modifications before any implementation. This initiative is extended beyond the IDE with official Android CLI plugins for Claude Code and Codex, covering project creation, emulator management, UI inspection, and Android documentation.

An updated command, android preview, permits coding agents to create Jetpack Compose previews within the terminal, enabling agents to accomplish UI tasks 77% faster while utilizing up to three times fewer tokens. Google's tools also enable agents to detect memory leaks, performance issues, and production crashes post-app development.

For instance, Claude Code collaborated with LeakCanary, an open-source leak-detection library, to examine critical user journeys for memory leaks using a single developer instruction. Claude Code executed these journeys across multiple devices concurrently, analyzed heap dumps via the CLI’s LeakCanary integration, traced the leaks back to the culpable code, applied the fixes, and confirmed their efficacy.

This process enabled capable models to accomplish these investigations nearly twice as quickly while utilizing up to 25% fewer tokens. Nevertheless, the development of autonomous bug-finding surpasses that of autonomous bug-fixing, as demonstrated by an Anthropic scanner that identified nearly 29,000 possible vulnerabilities in open-source code while only a few hundred were rectified.

A companion android signals command transmits Android metrics and Firebase Crashlytics data directly into the same terminal workflow, offering agents a direct connection to production crashes, ANRs, and memory concerns. However, this does not render the human element obsolete, as developers who witnessed Claude Code remedy a Linux kernel issue only to have reviewers identify a race condition can attest.

Google is also releasing Android Bench 2.0 to assess how models and agent harnesses execute on more intricate Android development tasks. This benchmark evaluates whether completed apps function correctly, measuring completion time and cost. The post "Google just handed its rivals the keys to Android" was originally published on The New Stack.