How do executives use AI without losing their gut feel?

How execs use AI without dulling their gut feel

Executives keep their gut feel by deciding first, then asking AI. Write your instinct in one sentence before opening the model. Use AI as a red team against your call, not as the call. Kahneman's System 1 is built by reps; if you stop forming a view before consulting the tool, the muscle atrophies and you stop noticing.

An executive's gut is not a mystical thing. It is a System 1 pattern-recognition layer, built over thousands of decisions, that fires fast and feels like knowing before you can explain. Daniel Kahneman spent half his career documenting exactly how this works in Thinking, Fast and Slow — System 1 generates a fast read; System 2 evaluates it. The problem AI introduces is subtle and was not on Kahneman's radar in 2011: it offers a slow, deliberate-sounding System 2 process you can borrow on demand, and if you borrow it too eagerly, you stop ever firing your own System 1. The muscle does not announce its atrophy. It just shows up one day as the realization that you no longer have a view on anything until you have asked the model.

I work with founders and a couple of C-level operators who are now consciously trying to use AI without losing their edge, and the practitioner protocol I have converged on is small, almost embarrassingly simple, and it works. It starts before you open the model. Whatever the decision is — should we extend this offer, do we ship Thursday or pull the release, is this candidate the one — you write your gut answer in one sentence first. Just one. The sentence has to include a verb and an actual position. "Ship Thursday because the regression risk is smaller than the morale cost of slipping again." Not "I lean toward shipping but want to think about it." A position, written, before AI gets a vote. This is the part founders skip, and it is the part that protects the muscle.

Then — and only then — I bring in the model. The prompt I use is some version of: "Here is the decision I am about to make and my one-sentence reason. Red-team it. Tell me the strongest version of the opposite case, the three failure modes I am not seeing, and the cheapest test I could run before committing." The model is not the decider. It is a sparring partner whose only job is to make my gut answer either survive its first real challenge or get killed cleanly. Ethan Mollick, in Co-Intelligence, calls this "always invite AI to the table," and the framing matters — invited guests do not vote.

The cautionary tale that drove this home for me this month was the DN42 incident, which hit number one on Hacker News with 1,461 points. An operator gave an AI agent a credit card and the autonomy to research a small hobbyist network called DN42. The agent — operating exactly as instructed, with no malice — provisioned five m8g.12xlarge AWS instances and burned $6,531 in 24 hours scanning the network at hourly intervals. The agent was not wrong, in a System 2 sense. It was correctly executing its mandate. What was missing was the human gut feel that says this is wildly disproportionate to the size of the problem. That is exactly the read a senior infrastructure executive's System 1 would have fired in under a second, and exactly the read the operator never got to form because they had delegated the entire loop. The bill was eventually negotiated down to $1,894, but the lesson is bigger than the dollars.

Two more habits that compound the gut-first protocol. The first is timing. I do not consult AI on decisions where I have already made the call internally — if my one-sentence answer is "we are not extending this offer," I act, I do not ask the model to give me a second opinion I will only use to second-guess myself. This is a Kahneman finding: the more System 2 cycles you spend on a System 1 verdict you have already reached, the more you erode confidence without improving accuracy. The second is the after-action review. Once a week, I write down three decisions I made that week with the AI involved, and what my one-sentence gut answer was versus what the model contributed. Over six months you can see whether your instinct is getting sharper or duller. Mine has stayed roughly the same on people decisions, gotten sharper on technical scope decisions, and gotten measurably worse on pricing — which told me where to deliberately stop using AI and go back to forming the call alone.

The honest punchline is that AI does not have to dull your judgment, but it will if you let it. Used as a red team after you have formed a view, it sharpens you. Used as the first move before you have formed one, it slowly replaces you in the loop you used to run. The DN42 operator did not lose money because the agent was bad; they lost money because no human gut was watching the meter. Make sure yours still is.