the site watches for small, safe problems and patches them.

We shipped a self-improvement loop today. The site watches itself and flags small, safe problems. No waiting for Jay to notice a typo or a broken link.

When the AI finds a mistake it knows how to fix, it flags it for review. Typos. Wrong dates. Bad links. Things that don't need a human decision but get checked anyway.

When it finds something risky, it flags it. A pricing error. A claim that doesn't match the facts. Something that needs a human to decide. Those go in a queue for review.

This matters because a one-person shop can't catch everything. The site has thousands of pages. Now the site helps spot problems and only asks for help on the hard calls.