Your integration_test suite is dozens of lines of fragile finders and pumpAndSettle, and it shatters the moment a layout changes. What if we moved from writing test code to defining user intent?

Agentic testing means AI agents that drive your running Flutter app, executing QA flows written in natural language, with no brittle test code to maintain. They reach the app through the Dart VM service to read and drive its widget tree, shown live on a demo app. This is not theory: I already run agents on the smoke flows that were once my most brittle scripted tests.

This is not a pitch to throw away deterministic CI. You will leave able to: draw a clear line between flows that suit agents (smoke, exploratory) and flows that must stay deterministic (precise assertions, gating, flake control); wire an agent to a running app through the Dart VM service; and handle agent non-determinism so it never guards your CI gates.