The Flutter team published its 2026 roadmap on February 24. I read it in September, which is late, but I read it right after moving a game's web build from Flutter to React because hot reload kept wiping our state. That gives you a specific lens. I was not reading for what is exciting. I was reading for whether the thing that pushed me off Flutter Web is on the list.
The six things they say they will do
The roadmap is organized around six themes. Stripped of the adjectives:
Finish Impeller on Android. The legacy Skia backend is being removed on Android 10 and above, so Impeller becomes the only renderer there. This is the shader-jank fix finally being the default instead of a flag. If you ever shipped a Flutter animation that stuttered the first time it ran on a mid-range Android phone, this is the line for you.
WebAssembly as the web default. The stated intent is that Wasm becomes how Flutter Web ships, for native-quality performance in the browser. This is about runtime speed of the deployed app, which was never my problem.
GenUI and agentic apps. A Flutter GenUI SDK and an A2UI protocol so that a model can generate UI at runtime instead of picking from screens you already built. This is the headline item and the one I trust least, in the sense that I do not yet know what a good version of it looks like. An LLM that emits a widget tree is easy to demo. An LLM that emits a widget tree a user can rely on is a different product.
Full-stack Dart. Dart on the server, Dart for the whole app.
AI-enhanced developer experience, including MCP servers for Dart tooling so that an agent can do large refactors against a real analyzer instead of guessing at text. Of the AI items, this one I want tomorrow. I already lean on an agent to write Dart. Having it able to ask the analyzer what a symbol resolves to, instead of pattern-matching, is the difference between a helpful pair and a confident one.
Decoupling Material and Cupertino into standalone packages so each can move faster than the framework release cycle.
The thing I looked for and did not find
Web hot reload is not mentioned. Neither is anything about state preservation across reloads on web.
I want to be careful here. A roadmap post is not a changelog, and the absence of a line does not mean nobody is working on it. But it does mean it is not one of the six things the team chose to promise for the year. The web items that made the cut are about Wasm and runtime performance. My problem was the inner loop: edit, reload, and the game forgets where it was. That is a developer experience item, and the developer experience theme in this roadmap is about AI tooling, not the reload cycle.
For the game, the decision was right and this roadmap does not make me regret it. For mobile, where Impeller finishing is real, measurable, and shipping, I am staying exactly where I am.
What I am actually going to try
The MCP server, as soon as it is more than experimental. If it works the way the roadmap describes, the first thing I would point it at is the sequential Dio pipeline that drifted from the React app, because that was a refactor where the hard part was knowing every call site, and that is precisely what an analyzer-backed agent should be good at.
GenUI I will watch. I do not have a screen in any of my apps that I want a model to invent at runtime. When I do, I will know it, and I will read the SDK then.
Source: The Flutter Blog, “Flutter & Dart's 2026 roadmap,” February 24, 2026.