I just couldn't love it, and frankly I don't get the hype around it. I recently found that all my use cases can be served by either:

1. A general purpose LLM chat interface with high reasoning capacity (GPT-5 thinking on web is my go to for now)

2. An agent that has unrestricted token consumption running on my machine (Claude Code with Opus and Amp are my go to for now).

3. A fine-tuned, single purpose LLM like v0 that is really good at one thing, in this case at generating a specific UI component with good design aesthetics from a wireframe in a sandbox.

Everything else seems like getting the worst of all worlds.