WTF Is a Loop?

Matt Van Horn, on the phrase that ate AI-coding Twitter this week:

stop being the thing in the loop. Write the loop once, give it skills worth calling and feedback so it can check itself, cap it so it halts, and let it run on cron while you go decide what to build next.

Good piece, and unusually honest for the genre: it concedes the cron-job jibe is half right and quotes the engineer who noted the only agentic thing about his agents is the Anthropic bill. The headline number is Boris Cherny landing 259 PRs with 100% of his recent Claude Code contributions written by Claude Code.

But look at whose loop that is. Cherny builds Claude Code: a mature repo, a strong test suite, an automated definition of "done." That's the best possible case for a self-checking loop, not the median one. The advice generalizes; the receipts don't.

Which is the part the discourse skips. A loop compounds exactly where verification is cheap and reliable. Everywhere else, in most knowledge work and plenty of ordinary software, it's a faster machine for confident mistakes. So the first question isn't what to automate. It's what you can actually check.