AI coding tools are very good at producing a page that looks finished. They are much less good at noticing the things that only show up when someone else actually uses the page: a call-to-action that points nowhere, an Open Graph image that never loads, a form input with no label.
None of these are hard to fix. Almost all of them are invisible until a real user, or a QA pass, goes looking.
That gap is what "finish what your AI started" means in practice. The build gets you 90% of the way. The last 10% is the pass that catches what speed skipped, and it is exactly the kind of checklist work that is easy to automate and easy to skip when you are moving fast.
A second pass does not need to be a person. It needs to be systematic: run the checks, get the flags, fix what matters, ship. The habit is the re-check, not the first check.