AI agents got good at the web quickly. They can read a page, fill in a form, click through a checkout. But most of what people actually do with a computer now happens on a phone, inside apps — and a lot of those apps have no website and no public API. An agent working through a browser simply can't reach them.
The same gap shows up from the other side. If you ship an Android app, you've probably wanted to know how a real person gets through it on a real phone — not a script that breaks the moment a button moves, but something that looks at the screen and figures it out. Testing that by hand on every release is slow, so the bugs that slip through are the ones your users find first.
Both problems have the same answer: give the agent a phone. Not a simulator or a screenshot mock-up — a full Android device, with real apps, a real store and a real browser, that the agent uses through the same surface you do: the touchscreen. It looks, it decides, it taps. And it reports back with what it saw.