I’ve found that capturing a screenshot of localhost:3000 and attaching it when re-prompting the agent lets the AI review its own work and suggest improvements. Adding native support for this feedback loop would make frontend development far easier.
Yes, I dont like front end.