OODA loop
Observe, Orient, Decide, Act. The four-step decision cycle from John Boyd. Applied to AI output, it's the verify-then-send discipline that catches hallucinations before they ship.
What it does (the operator translation)
The model drafts a follow-up text in 3 seconds. You read it in 8. The 8 seconds is the OODA loop.
Observe. Read the draft. The whole thing. Names spelled right? Address correct? Price match the listing?
Orient. Compare against what you know. Does this sound like you, or like a Zillow bot? Does the comp the model cited actually exist in Realtracs?
Decide. Edit, regenerate, or send. One of three. Don't paralyze on the fourth option.
Act. Hit send. Or don't. The point is the cycle ran before the message left your phone.
Simon Willison's 2025 weblog made the same point a different way: verification is now the expensive part. The model drafts fast. What costs you is making sure the draft is right. OODA names the discipline. Yan/Husain Part II on operations walks the same logic for production systems.
The contrast with vendor-pitched "set it and forget it" automation: a Zap that runs unsupervised has no OODA cycle. No observe step. No verify. That's not automation — that's a quiet bet that nothing changed since you wired it.
Why a working REALTOR cares (the breakpoint)
For every working agent. The OODA loop is the operator move that catches hallucinated comps, wrong square footage, fabricated school ratings, and "stunning" sneaking back into your follow-up text. 8 seconds per draft is the cheapest insurance in the stack.
What this is NOT (the category-flip)
OODA is NOT a slowdown. The verification step is what makes the 80/20 rule actually work — model does 80% of the draft, you spend the 20% on the OODA pass. Skip the loop and the speed gain disappears the first time the model fabricates a school district.
Related terms
80/20 rule · Hallucination · CRAFT framework · Context card
Where this comes up in The Listing Machine
OODA is the verification discipline taught in week three of The Listing Machine. Students run real outputs against real listings — and learn that the 8-second check is non-negotiable before any AI draft leaves their phone.