Android automation with AI: let an agent use the phone for you.
Four ways to automate Android
| Best at | Breaks when | Runs on | |
|---|---|---|---|
| ADB shell scripts | Fixed sequences you control end to end | Coordinates or timing change | A USB phone or emulator next to the script |
| Tasker, MacroDroid | Personal triggers on your own phone | An app updates its screens | Your own phone |
| Appium, Espresso, UIAutomator | Regression tests on every commit | IDs and layouts change | Emulators or a device farm |
| AI agent on a cloud phone | New flows, apps without APIs, judgement calls | The task is genuinely ambiguous | Phonebox's cloud, driven over MCP or API |
They combine well. Keep scripts for flows that never change, and give an agent the work that needs reading and deciding.
How an AI agent drives the phone
- Observe. The agent reads the screen as a numbered list of elements with their labels, plus a screenshot.
- Act. It taps by visible text, types, swipes or presses keys, up to 20 actions in one call.
- Check. It reads the new screen and confirms the result before the next step.
snapshot snp_4f2k7m3q6z3a
app com.android.launcher3
keyboard hidden
screen 1080x2400
[1] EditText "Search apps" #search clickable editable (540,250)
[2] TextView "Chrome" #icon clickable (200,1900)
[3] TextView "Settings" #icon clickable (500,1900)
[4] TextView "Play Store" #icon clickable (800,1900)Reading text rather than pixels is what makes it reliable: “tap Sign in” still works after a redesign moves the button. Phonebox never retries a write action on its own, so a slow response can't become a double purchase.
Try it
Paste the setup prompt into your agent, then give it a small task you can verify:
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.Use Phonebox to open the Settings app on my phone,
turn on dark mode, and confirm it is on by reading the screen.
Read the screen before every action. Park the phone when you finish.For a product that automates phones for your own users, use the TypeScript SDK or REST API: give each user a persistent phone, and park it between tasks so it costs nothing.
Related
- Cloud phones for AI agents: what they are and how to choose one.
- Android MCP server for Claude Code, Codex and Cursor.
- Test your Android app with an AI agent.
Questions
Can AI automate an Android phone?
Yes. An AI agent such as Claude Code, Codex, Cursor or your own model can read an Android screen as text, decide the next step, tap and type, then check the result. With Phonebox the phone runs in the cloud and the agent connects through MCP, a CLI, a TypeScript SDK or REST.
How is AI automation different from Tasker, MacroDroid or Appium?
Macro apps and Appium follow steps you wrote in advance, so they break when a screen changes or something unexpected appears. An AI agent decides each step from what is on the screen now, so it can handle new layouts, pop-ups and apps you never scripted. Scripts stay cheaper and faster for fixed, repeated flows.
Can an AI agent use Android apps that have no API?
Yes. That is the main reason to give an agent a phone: it uses the app's own screens the way a person does. It reads labels, taps buttons and types into fields, and it can hand the phone to a person for a one-time code or a payment.
Do I need a rooted phone or ADB for AI Android automation?
Not with a cloud phone. Phonebox phones are ready to control through its API and MCP server, so nothing runs on your own phone and you don't need USB debugging, root or an emulator.
What does AI Android automation cost?
With Phonebox, $0.06 per phone-minute while the phone runs, billed per second, plus whatever your model costs. Parked phones are free and keep their apps and sign-ins. New accounts get $2 of credit with no card.