Your designer wants to review the latest app changes, but they cannot simply run flutter run. Your product manager wants to tap through a feature before it is merged, and QA wants to validate a flow on a real device without waiting for a full release pipeline. Flutter Go was built to solve this exact collaboration problem: build the app, share a QR code, and let teammates instantly open and test it on their own devices.

In this talk, I will show what it took to make that workflow possible. We will look at why standard Flutter tooling was not enough, what assumptions the Flutter engine makes about building and running apps, and how we modified parts of the engine and worked with the Dart VM to create a faster iteration loop for the whole team.

What if your AI didn't just write your code, but actually used your Flutter app?

In this talk, I’ll show Marionette MCP, an open-source MCP server that lets AI agents connect to a running Flutter app and actually use it. The agent can understand the current UI, decide what to do next, and execute complete user flows, adjusting its actions as the app responds, rather than following predefined scripts.

This will be a demo-first session: less theory, more real world uses. We’ll start with simple interactions, move to a realistic user journey, and then show how Marionette can be extended to expose hidden runtime capabilities. By the end, you’ll have a clear mental model for when Marionette is useful, how it differs from code-focused tooling, and where it fits in a Flutter workflow.