It tackles one of the annoying parts of mobile QA, turning test cases into something that runs reliably on devices.
With CoWork, I can upload the existing test cases then it plans the steps and executes them on a real device and the part that really caught my attention is how it handles changes during execution. If the app has a different label, an unexpected popup, or the flow changes, it doesn't just blindly continue. It can replan and bring a human into the loop when a decision actually matters. That feels much more practical than treating AI as something that should always make the final call.
I also like that it's focused heavily on mobile testing, including Android, iOS and Flutter, rather than trying to be an AI tool for every type of testing at once.
The caching is the bit I'd worry about. If it remembers which elements it poked last run it stops looking, and a screen that quietly moved a button still passes because the cached path resolves fine. That's the failure I'd rather have loud and slow than fast and confident, and it's the same shape as every other agent problem, plausible output that nobody complains about. The Gherkin file falling out of each pass is the real feature, because it's the one artifact a human can read and disagree with.
QApilot
@asadmalik901 - Fair concern. The cache is a hint rather than a bypass, resolution still runs and the assertion still has to pass for the test, so a moved button only slips through if the cached path also satisfies the check.
Agreed on the Gherkin. It's the only artifact a human can argue with, which is why we write it on pass only.