What We Handed Over to AI Agents — and How to Do the Same

I'll tell you how AI agents deliver real value in a large mobile project, using the Cian Android app as an example. AI knows the technologies, but it doesn't know your project — so the value comes not from the model itself, but from everything built around it: skills, MCP connectors, and framing tasks correctly for a specific codebase.

Through that lens, we'll walk through applied cases that made it all the way into practice. We'll see how an agent builds UI without our involvement, with validation right on the device.

We'll look at automated test coverage and how simplifying the pipeline made writing tests 4× faster. We'll touch on migrating legacy modules to a new architecture with no developer involvement, and on finding duplicate resources using ML algorithms.

The talk will be useful for anyone who wants to move from experimenting with AI to systematically adopting it in production development.