boyds-law-of-iteration ADG-0004

Boyd's law of iteration

“Speed of iteration beats quality of iteration.”

Origin

John Boyd's OODA-loop work on air combat, as applied to software by Jeff Atwood (2007). About learning faster than the opponent or the problem changes, on the assumption that each iteration teaches something.

Why it should, or should not, apply to models

An agent iterates for nearly free, so if the law transfers it transfers overwhelmingly — except that the law's silent premise is that each iteration carries real feedback. A model iterating on its own judgement gets no new information per loop and can iterate quickly toward a worse answer. So the prediction is sharp: fast iteration wins when a grounded signal closes each loop, and is worthless or harmful when the loop is closed by the model's own opinion.

Standing: Holds, narrowed

Standing counts claims a person has reviewed. Claims reviewed by AI alone are listed and marked, and do not move it.

Where it holds only under a condition

Claims that keep the heuristic but bound it.

When a model gets a real environment signal after acting — a test result, a tool error, a task-success indicator — reflecting on that signal in words and retrying substantially improves success rates on multi-step coding and decision-making tasks over a single attempt.
Fixing its own mistakesReviewed by AI and Russ Miller
Iterating helps substantially when each loop is closed by a real environment signal — a test result, a tool error.
Without an external, ground-truth signal — a failing test, a compiler error, a verifier's output — a model's own critique of its reasoning is not a reliable improvement signal, and asking it to review and revise a correct answer often turns it into a wrong one.
Fixing its own mistakesReviewed by AI and Russ Miller
Iterating on the model's own critique does not help and can degrade a correct answer. Speed without a signal is not learning.
With weights frozen and no stronger model in the loop, an agent can raise its own pass rate by rewriting the scaffolding it runs inside — clustering its failures, proposing minimal edits to prompts, tools and policy, and keeping only edits that improve one split without degrading another: all 9 of 9 model-by-benchmark pairs improved, by up to 40.6 percentage points (GLM-5 on AppWorld, 44.4% to 85.0%), with weak models gaining most; but the promotion gate reads the held-out split, so the reported held-out gain is not clean generalisation, and there is no matched-compute baseline.
Fixing its own mistakesReviewed by AI
Many fast iterations on the harness raised pass rate, but only because a regression gate rejected the ones that did not.

Add a case where it holds or breaks

A break is worth more than a hold. If you have seen this fail for a model, with the setup written down, that is the most useful thing you can add here.