AI QA testing for Android apps: the tools, and how to run it with your own agent.
Tool descriptions come from each vendor's site, checked October 12, 2026: QA Wolf, Momentic, testRigor, Repeato, AskUI.
AI QA tools that cover Android
| Tool | What it does | Model |
|---|---|---|
| QA Wolf | AI agents and engineers build and run end-to-end tests, Android via Appium | Managed service or self-serve |
| Momentic | An AI QA agent that tests from a prompt; Android on hosted emulators | Self-serve, free to start |
| testRigor | Tests written in plain English, native Android and iOS | Self-serve |
| Repeato | No-code, computer-vision tests for Android, iOS and web | Self-serve, free version |
| AskUI | An agent runs plain-language test cases on any screen | Self-serve and enterprise |
| Your agent + Phonebox | Claude Code, Codex or Cursor builds, installs and walks the app on a cloud phone | Pay per phone-minute |
Build or buy
| AI QA product | Your agent + Phonebox | |
|---|---|---|
| Who writes the checks | The vendor's agent or your team in its editor | You, as a plain-language task in the repo |
| Who maintains them | The vendor, on managed plans | Nobody: the agent re-reads the screen each run |
| Where results go | The vendor's dashboard | Your agent's chat, PR or issue, with screen text |
| iOS | Usually yes | No, Android only |
| Cost model | Subscription or sales quote | $0.06 a phone-minute plus model tokens |
Run AI QA with your own agent
- Connect the phone to your agent through the hosted MCP server.
- Install the build: the agent uploads and installs the APK it just built.
- Give it the scope: the screens that changed 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.What a good report looks like
Ask for evidence, not opinions. Because the agent reads each screen as text, its report can quote the element that was wrong. A report in this format, with an example bug:
Bug: checkout total ignores the promo code
Build: 2.14.0-debug (versionCode 2140)
Steps:
1. Basket → tapped "Apply code", typed SPRING10, tapped "Apply"
2. Screen showed: "SPRING10 applied: −$4.00"
3. Tapped "Checkout"
Expected: total $36.00
Saw: [7] TextView "Total $40.00" #order_total
Recording: repro-checkout.mp4 (pulled with phonebox record pull)The Saw: line uses the exact format phonebox look returns, so in a real report it can be checked against the phone. Add a screen recording for bugs that need a video.
What a QA pass costs
| Pass | Phone time |
|---|---|
| One 8-minute pass on a pull request | $0.48 |
| 100 pull requests a month | $48 |
| A 30-minute pre-release pass, weekly | $7.20 a month |
Questions
What is AI QA testing?
Testing where an AI model decides the steps, from a plain-language goal, instead of following a fixed script. For mobile apps that means an agent that reads the screen, taps, types and reports what looks wrong.
Can I do AI QA testing without buying a QA tool?
Yes. A coding agent such as Claude Code, Codex or Cursor can build your APK, install it on a Phonebox cloud Android phone, walk the changed screens and report bugs with the screen text it saw. Phonebox costs $0.06 per phone-minute, billed per second, so an 8-minute pass is $0.48 of phone time.
How do I make AI QA reports trustworthy?
Ask for evidence: the build version, the exact steps, and the screen text or element the agent saw at the failing step, plus a screen recording for visual bugs. With Phonebox, an element quoted in a report uses the same format the phone returns when the screen is read.