Steve Ruiz's prompt hunts implicit state machines

Steve Ruiz posted a prompt he likes, and it is one line. "find all of the implicit state machines in my app". That is the whole thing.

Jamon ran it against his game codebase and reported back that he was up to 200 implicit state machines, followed by an ellipsis and a wide-eyed emoji. David K, replying to Jamon, offered the obvious next step, make them explicit and he will give you a nice way to visualize them.

Jamon

@jamonholmgren

X

I'm up to 200 implicit state machines in my game code base.

Why this one travels

An implicit state machine is the thing you get when a feature has modes but nobody ever named them. A few booleans, a nullable field, a flag that only means something while another flag is set, and the transitions live in whichever conditionals happen to run. Nobody wrote the diagram because nobody decided there was one.

That makes it a good target for a model. The prompt does not ask for a refactor, a migration or a new abstraction. It asks for a list. The agent reads the code, names the states it thinks exist, and hands you back a set of design decisions that were never written down anywhere. You can scan that list and disagree with it in a few minutes, which is not true of most things an agent produces.

For working engineers, note what is not being claimed here. Nobody in this exchange validated the 200. Jamon's number is a count the model produced on one codebase, and it is exactly the kind of number that inflates when a model is rewarded for finding things. Some fraction of those are two booleans that were always going to be two booleans. The useful move is treating the output as a list of candidates to argue with, not an inventory.

Audit prompts like this are cheap to run and cheap to throw away, which is a better ratio than most agent workflows manage. If the list is garbage you lost a minute. If three entries on it are real, you found bugs you were not looking for.