Test your Android app with an AI agent.
Close the loop on mobile changes
On the web, an agent can open a browser and check its own work. On Android it usually stops at “the build passes”, and you find the broken onboarding screen yourself. A cloud phone removes that gap: the agent builds, installs, looks, taps and reports, before you ever open the emulator.
| Scripted UI tests | AI agent on a cloud phone | |
|---|---|---|
| Best for | Known regressions on every commit | New flows, exploratory checks, bug reproduction |
| Written as | Espresso or Appium code | A plain-language task |
| Breaks when | IDs and layouts change | The app genuinely behaves differently |
| Output | Pass or fail | What it saw, step by step, with screen text |
Set it up once
Connect your agent with the setup prompt. It installs the Phonebox CLI and signs you in; you never paste a key into the chat.
Read https://phonebox.dev/setup.md and set me up with Phonebox from here: get me signed in and ready, then find out what I want to use phones for and help me with that.Create one phone for testing and keep reusing it. A parked phone keeps your app, its data and its signed-in test accounts, and costs nothing between builds.
Install every build in one command
./gradlew assembleDebug
phonebox apps "$PHONEBOX_PHONE" install app/build/outputs/apk/debug/app-debug.apkThe CLI uploads the APK, reads its manifest, installs it and waits until the phone lists the app at the new versionCode. One upload can install on many phones, and reinstalling the same file reuses the upload.
Ask for the walkthrough
Give the agent the scope of the change and the standard you expect:
Build the debug APK and install it on my Phonebox phone.
Walk through the onboarding and settings screens I changed in this branch.
Read the screen before every action. Report anything that looks broken,
with the screen text you saw. Park the phone when you finish.The agent reads the screen as a numbered list of elements, taps by visible text, and reads the screen again after every action. When something is wrong, it can quote the exact text on screen, which makes its bug reports concrete.
For sign-in steps that need a real person, such as a code sent to your own phone, the agent can send you a live view link, wait, and continue.
Go further
- Test your Android build with a coding agent: the full CLI walkthrough, version checks and reinstall rules.
- Android MCP server: the same workflow through MCP tools.
- Cloud phones for AI agents: how to choose a device platform.
Questions
Can an AI agent test my Android app?
Yes. A coding agent such as Claude Code, Codex or Cursor can build your APK, install it on a Phonebox cloud Android phone with one command, walk through the screens it changed and report what it saw. It reads the screen as text, so it can check labels, error messages and navigation without brittle selectors.
Does AI testing replace Espresso or Appium tests?
No. Scripted tests are still the right tool for regressions that must run on every commit. An agent is best at exploratory checks: walking a new flow, trying edge cases a person would try, and describing what looks wrong. Many teams use both.
Which APKs can I install?
Any signed APK up to 500 MiB, debug builds included. Android App Bundles (.aab) and split APKs must be built into an APK first.
How much does it cost to test on a cloud phone?
$0.06 per phone-minute, billed per second with a one-minute minimum. A ten-minute exploratory pass costs $0.60. Parked phones cost nothing and keep the installed app and its sign-ins for the next build.