Play Console tells you when a vital gets worse. It doesn't tell you which change caused it. Finding the root cause still means capturing traces, reading flame graphs and searching release diffs by hand. That's slow, it's hard for most Android developers, and it's easy to get wrong. AI agents can make this loop much faster, but only if they work with the right tools and are held to evidence. In this talk, we share how we built that setup for a large Android app, and what we learned from using it on real startup, jank, crash and ANR regressions. You will learn: • How to go from a Play Console vital to a root cause: metrics, traces, code changes and validation. • How to give an agent skills and MCP tools to read Perfetto traces, production metrics and crash reports. • Why the agent is often wrong, and the simple checks that catch it before it costs you a day. • How to prove a fix with Macrobenchmark and production experiments. No performance expertise needed. You'll leave with a practical blueprint for building your own AI-assisted performance workflow: start with one skill, one tool and one evidence rule.