Why AI loops stall: the hidden cost of chasing a target that can't be reached
Knowing When to Stop: The Art of Making a Loop Converge
Loop engineering promises to let AI agents work autonomously, but a loop is only as good as its verifier. Drawing on interviews and experiments, this piece breaks down the four conditions for convergence—target state, observable state, precise edits, and a stopping rule—and shows why code loops thrive while open-ended visual tasks stall. The economics are brutal: returns are logarithmic, and past a plateau, extra iterations can actively hurt. A loop is tuned to its stack, and bespoke loops don't generalize for free.
Loop engineering is not the practice of making an agent retry. It is the practice of making each cycle reduce the distance between the current state and a desired state. A loop is not yet a direction.