Also +1 on not full-sending AI. I use it for drafts, boilerplate, and thinking partners — not trust-on-first-output.

Totally fair take — and I think we’re actually closer than it looks.
Those 3 sites aren’t “unique projects,” they’re the same scaffold on purpose: a tiny experiment in pattern + variation. I’m testing a repeatable pipeline (structure, deployment, naming, constraints), then iterating the copy

This is an experiment of having AI do the whole thing. Not touching the code, just prompting it done - it's low stakes not on a customer or employer site.

Right there with you - just had Claude create three new sites last night.