AI-Assisted Mobile & Web Testing
One person, one day, and an agent that doesn't just run the test script — it notices when the app changes underneath it and rewrites the script to match.
Tester
Maestro CLI
Claude Code
running the loop
step 3 — element moved, flow rewritten mid-run, no one paged
Replaced
Manual testing on attached devices & emulators
Built with
Claude Code orchestrating Maestro flows
Built by / in
One person · 1 day
Result
~70% less tester time
Before
The challenge
Mobile and web QA was done by hand, on whatever devices and emulators a tester happened to have attached to their own machine. Every release meant a person manually walking through the same flows, start to finish, with no way to move faster than a person physically could.
That effort repeated across three separate surfaces, each release, every time:
FOMO B2B
Manually walked, release after release
FOMO B2C
Manually walked, release after release
FOMO Admin
Web application, manually walked each cycle
Running
The approach
Built in a single day, by one person. Claude Code was given detailed business context on how each app is meant to work, then instructed to generate Maestro test flows in detail across all three surfaces.
- 01
Context in
Claude Code is briefed on the intended behavior of FOMO B2B, FOMO B2C, and FOMO Admin — not just the UI, but how each flow is supposed to work.
- 02
Flows generated
Claude Code writes detailed Maestro test flows covering all three applications, ready to run through the standard Maestro CLI.
- 03
Two lanes, run in parallel
A tester runs the flows directly through the Maestro CLI. At the same time, Claude Code runs its own copy independently, in a loop.
- 04
Self-adjusting, not self-reporting
When Claude Code's loop hits something broken, it identifies the issue and adjusts the test flow itself — instead of waiting for a person to notice the failure and go fix the script by hand.
Pass
The results
Not by replacing the tester — by handing off the repetitive, mechanical part of test execution to an agent that also notices when something breaks and adapts the flow itself. That leaves the tester's time for the judgment calls a script can't make, instead of repeating the same manual walkthrough release after release.
70%
Tester time returned
Honest scope note
This is an internal tool, not a product — built to test Zopflix's own apps, not sold or offered to clients.
- Runs on the devices and emulators already on hand — not a dedicated device lab or farm.
- Scoped the same way every other proof point on this site is scoped: precisely, and without overstating it.