> You often need a real environment to iterate, even for the agent: debug, inspect logs, run the dev server, and use the frontend to actually verify behavior. Without that shared runtime, the agent’s feedback loop isn't tight enough, and you can't verify changes as well.

Cloud environments can do that (e.g., Github Copilot). They can even spin up playwright and take screenshots of an app and do frontend tests.