Google previews Bring Your Own Agent support in Android Studio Rabbit 2 Canary
Agent choice
Android Studio Rabbit 2 Canary previews Bring Your Own Agent support for ACP-compliant agents including Claude Agent, Codex and Google Antigravity.
IDE control
Android Studio supplies project graphs, build setup, diagnostics, SDK tooling, Compose Preview support and emulator controls to connected agents.
Preview status
The feature is currently a Canary-channel preview, with no stable-channel release date or uniform support guarantee across all agents.
Google is previewing Bring Your Own Agent support in Android Studio Rabbit 2 Canary, allowing Android developers to connect ACP-compliant coding agents such as Anthropic’s Claude Agent, OpenAI’s Codex and Google Antigravity directly inside the IDE.1
The move broadens Android Studio’s agent strategy beyond a single first-party assistant. Rather than requiring teams to standardize on one model or provider, Google is positioning the IDE as the control plane: Android Studio supplies Android-specific context, tools and guardrails, while developers choose the coding agent that fits their workflow, account structure or enterprise policy.1
The preview is available in the Canary channel and is aimed at teams already experimenting with AI coding agents, enterprise harnesses or multiple AI plans. AndDev reported that the change affects the agent integration layer, not the Android build system itself. Bring Your Own Agent uses the Agent Client Protocol, or ACP, to pass project information and IDE tools to a connected agent.1
Android Studio remains the source of Android project awareness. Through ACP, the IDE can give a connected agent information such as the project graph, build setup and platform details, helping the agent focus on the modules, files and configuration relevant to the task.1
That model is significant for mobile teams because Android development often depends on context that generic coding clients do not automatically understand: Gradle configuration, build variants, SDK tooling, Compose previews, emulator state and platform-specific diagnostics. By exposing those capabilities from the IDE, Google is attempting to let external agents operate inside the existing Android workflow rather than alongside it.1
For engineering managers, the practical implication is that agent choice can become more decoupled from IDE standardization. A team could continue using Android Studio as its primary development environment while evaluating providers based on model quality, security posture, licensing, quota, authentication method or enterprise procurement requirements.1
According to AndDev, Android Studio’s Bring Your Own Agent preview can connect agents to Android Studio project-aware infrastructure, build diagnostics, Android SDK tools, Jetpack Compose Preview support and emulator controls.1
The source also says integrated agents may be able to read, write and edit files, execute shell commands, run tests and search the web when those capabilities are available in the agent workflow.1 That makes the integration broader than chat-based code suggestion. The agent can potentially participate in build-debug-fix loops, UI iteration and runtime validation from inside the IDE.
A typical workflow could involve a developer asking an agent to diagnose a failing build, inspect the relevant module, edit source files, run tests and validate behavior against an emulator. The value of the integration is not just that the agent can generate code, but that Android Studio can provide the operational surface around it: build output, project topology and device or emulator interaction.1
The preview also puts permission review at the center of the agent workflow. AndDev notes that the integration includes granular permissions and approval pauses for riskier actions.1
That control layer matters because the same capabilities that make an agent useful inside an IDE also increase risk. File edits, shell commands, test execution, web access and emulator control can affect source trees, local environments and potentially sensitive project information. Permission gates give teams a way to distinguish low-risk assistance from actions that require explicit developer approval.1
For organizations adopting the preview, evaluation should include more than answer quality. Teams should test what the agent can see, which tools it can invoke, when approval prompts appear and whether the workflow matches internal rules for source access, generated code review and build execution.
The feature remains a preview in Android Studio Rabbit 2 Canary. AndDev reports that Google has not established a stable-channel release date, a minimum stable Android Studio version, a required Android Gradle Plugin version or uniform support across all ACP-compliant agents.1
That makes Rabbit 2 Canary a testing checkpoint rather than a production migration target. Teams interested in the feature should evaluate it in an isolated environment, connect one provider at a time, verify Android-specific tool access and keep their existing workflow available as a fallback.
The initial named agents are Claude Agent, Codex and Google Antigravity. Additional options depend on ACP compatibility and the agents available through Android Studio’s settings registry.1 Provider-specific details, including login methods, quotas, model access and plan support, remain subject to each vendor rather than Android Studio itself.1
The preview does not eliminate Google’s built-in Gemini experience in Android Studio. AndDev reported that the built-in Gemini route will continue to be offered, while Google recommends Antigravity for what it describes as the best Gemini experience in this workflow.1
That creates two Google-aligned paths: the built-in Gemini agent and an ACP-connected Antigravity agent. For teams already standardized on Gemini, the choice may come down to whether the existing built-in experience is sufficient or whether Antigravity’s agent workflow, model access and quota structure better fit the organization’s needs.1
The broader signal is that Google is not treating Android Studio solely as a container for one assistant. With Bring Your Own Agent, the company is positioning the IDE as the orchestration layer for coding agents in Android development — one that can provide project context, Android tooling and human approval gates while leaving room for multiple model providers.
Darktrace’s new Signal Labs framed agentic AI risk as a monitoring problem, but the technical finding is narrower and more urgent: conversation histories and evaluation sandboxes can become inputs attackers or agents may try to manipulate. For platform teams, prompts, local histories, tool permissions and test harnesses now require controls similar to other security-sensitive state.
OpenAI’s GPT-6 Sol and Luna are positioned less as headline-grabbing frontier models than as lower-cost workhorses for recurring coding, support, and clerical agents. For engineering teams, the practical question is not whether they top every benchmark, but whether they reduce the total cost of accepted work after retries, caching, latency, and human repair are counted.
Kiteworks’ precautionary request that customers take file-transfer systems offline shows how vendors may respond when threat intelligence indicates imminent exploitation but no patch, CVE or confirmed compromise is available. For enterprise security teams, the episode underscores that uptime decisions for exposed file-sharing platforms may need to account for threat timing, not just patch status.
OpenAI’s disclosure that agents unexpectedly interacted with SEC and Census data sources highlights a public-sector blind spot: autonomous AI clients can look like ordinary web traffic while behaving unlike ordinary users. Security leaders now face questions over authentication, rate limits, audit trails and disclosure rules when no conventional breach is found.
Bring Your Own Agent
A model that lets developers connect an external coding agent to the IDE instead of using only the IDE vendor’s built-in assistant.
Agent Client Protocol
The protocol boundary Android Studio uses to pass project context and tools to compatible coding agents.
Canary release
An early preview channel for Android Studio features that is intended for testing, not production-standard stability.
Project graph
A structured view of a project’s modules, dependencies and build configuration that helps tools understand where code and build tasks fit.
Comments