The advice to stop prompting and start running agents in loops usually arrives with no mention of what it costs to run them. It is good advice when token usage is not a concern. For everyone else, it assumes a budget you might not have.
Two things the slogan skips.
First, a loop needs a finish line it can check, something a machine can verify as done. That is clean for a backend that returns JSON. It is still an open question for work with no clean answer, like whether a UI looks and feels right. People building this daily have not settled it, so the advice to just loop it hands you the part no one has solved yet.
Second, learning what a good loop looks like for your own project costs real money. You find out by getting it wrong, and each wrong run bills you. Experimenting your way to a good loop is not free, most practitioners cannot run unlimited experiments, and the advice glosses over that.
So someone takes the slogan at face value, points a loop at a fuzzy finish line, and pays to discover the output was close but not what they meant. The bill and the gap both fall on the person who could least afford the lesson.
How are you writing a finish line for work that has no clean pass or fail?
Written by Duane Grey
AI Strategy & Implementation
Independent AI consultant helping companies cut through hype and deploy systems that produce real results.