My friend Pacu

PACU is an acronym I’ve started using with LLMs, short for “Pause and Check Understanding.” The key is it forces your agent to regurgitate its interpretation of your prompt back to you. If you spend a lot of time using AI coding tools you probably have your own variant in your AGENTS.md:

If the user says pause and check understanding or any case variant/abbreviation (for example: PACU, pacu, PaCu): perform no work, paraphrase your understanding of the mission, and wait for explicit confirmation before proceeding.

I find myself typing PACU multiple times a day, because using it makes the agent actually explain what it thinks I said. To the point where one PACU is often not enough to get in sync: saying things like “rePACU” or “rePACU everything from beginning to end” leads to progressively increasing clarity for both me and (I think/hope) the AI.

I don’t think the prompt itself is the important bit (though having a nice easy to type acronym certainly helps) what matters here is the workflow. Having a quick and easy way to force the agent to re-express itself and enumerate what it (mis)understood before letting your agent go off on an hour long coding tear has greatly reduced my odds of ending up with a pile of vibe spaghetti. Maybe you’d like to teach your agents how to PACU too?