Agent Device: AI-Native Mobile Automation for iOS and Android
Article Summary
Callstack just released a CLI tool that finally lets AI agents actually test mobile apps. No more agents that can write code but can't click a button.
Agent Device is an open source CLI tool that bridges the gap between AI coding agents (like Claude Code) and mobile UI automation. While AI agents excel at manipulating code, they've historically struggled with reliable mobile app interactions on iOS and Android.
Key Takeaways
- Provides lifecycle, snapshot, click, fill, scroll, and find commands for mobile automation
- Generates token-efficient accessibility trees that LLMs can process effectively
- Replay functionality enables deterministic reruns using recorded .ad scripts
- Built-in skills support gives agents immediate API proficiency without trial-and-error
AI agents can now handle complete mobile UI workflows autonomously, reducing manual testing intervention and accelerating delivery cycles.
About This Article
AI agents like Codex and Claude Code can't reliably interact with mobile UIs, so developers end up manually stepping in to handle iOS and Android workflows that should run on their own.
Callstack built agent-device with snapshot-ref notation (@e1, @e2) that creates compact accessibility trees small enough for LLM context windows, along with replay scripts (.ad files) that let you run the same steps again without guessing what to do next.
Teams get more out of their agents and switch contexts less often. The unified CLI removes the friction of dealing with different platforms and cuts down on the debugging headaches that come with flaky mobile automation.