
How AI Agents Struggle with Flutter: 7 Gaps Experience Still Closes
We let agents write a large share of our Flutter code, and we review a lot of Flutter written by agents nobody supervised. The difference is not typing speed — it is seven decisions an unsupervised agent gets wrong every single time: it recomputes derived state instead of keeping one source of truth, ships no tests or benchmarks, wires complex async into if-pyramids instead of streams, treats design tokens as constants instead of a Theme, invents a new UX vocabulary on every screen, concatenates strings where ICU is required, and simplifies uniformly instead of knowing where complexity has to stay. Here is what each one looks like in Dart, why the agent lands there, and what we do instead.













