How Android Studio’s New AI Agent Supercharges Mobile Development

Android Studio Otter 2 (Canary) introduces an AI Agent built on three core concepts: Tools, Context, and the Model Context Protocol (MCP). These enable the IDE to execute development tasks rather than only returning chat text.

Tools

The Agent can invoke built‑in tools such as Find Files, read_file, version_lookup, and gradle_sync. For example, when a developer asks “Where is the theme defined?” the Agent runs Find Files theme.kt and returns the file location without scanning the entire codebase, conserving tokens.

Developer: “Where is the theme defined?” Agent runs Find Files theme.kt and returns the file location.

Typing /tools in the chat lists all currently available tools.

Knowledge Base

To mitigate the lag of large‑model training data, Android Studio bundles a search_android_docs tool that queries the latest Android, Firebase, and Kotlin documentation, allowing the Agent to retrieve up‑to‑date API information such as the newest Navigation 3 APIs.

Model Context Protocol (MCP)

MCP extends the Agent’s capabilities by allowing it to connect to external services. An example integration is a locally running Figma MCP server, enabling commands like “capture the selected Figma content and convert the current screen,” thereby bridging design and development workflows.

Context

The Agent automatically gathers project metadata (project name, open file, etc.) as context. Developers can define custom rules to keep the Agent focused, for example by adding an agents.md file that stores shared conventions (e.g., Hilt setup, code style) and by placing hierarchical agents.md files in sub‑modules.

agents.md : shared configuration file for project conventions.

Hierarchy : the Agent searches upward for agents.md to apply module‑specific context.

End‑to‑End UI Development Demo

The Canary demo walks through a complete UI creation cycle using natural‑language prompts:

Prompt generates a development plan (libraries, tools). After user confirmation, the Agent produces ViewModels, Jetpack Compose UI code, and unit tests.

Developers can drag a UI screenshot into the chat; the Agent analyses the image and emits corresponding Compose code, reusing existing components and generating new Vector Drawables when needed.

Static analysis and Lint are leveraged for automatic code fixes (e.g., removing unused imports, updating deprecated APIs), reducing costly Gradle builds.

When a design changes, uploading the new mockup lets the Agent detect differences and adjust the existing code to match the updated UI.

The Render Compose Preview tool renders the updated UI directly in the chat window for instant verification.

Natural‑language UI tweaks such as “left‑align, increase font size, add a soft gradient background” are translated into code modifications.

UI Check mode evaluates accessibility and screen‑adaptation issues; if a contrast problem is found, clicking “Fix with AI” makes the Agent propose and apply a suitable color fix, then re‑runs the check.

The demonstration highlights the Agent’s multimodal abilities—understanding design assets, performing image‑based diffing, generating previews, and conducting automated UI checks—suggesting that future Android development may become largely voice‑driven.

Reference

Video: https://www.youtube.com/watch?v=jTlW8JeCClA