I trust my first instinct. Twenty years of early calls taught me that the gut is often the part of you that finished the sentence before the slide deck caught up. Instinct is how you raise a hand in a room that isn't ready. Instinct is how you bet the company when consensus is still laughing.
And instinct without a check is how you confuse motion with alignment. So after the first instinct, I ask a question I want every operator — and every AI system we build — to ask out loud:
Was my confidence well placed in the plan?
Not "do I feel sure?" Feeling sure is the model's default state and the founder's favorite costume. Well placed means something stricter: does the work I'm doing still match the plan I said I was executing — and if my certainty is thin, did anyone get a second opinion before we shipped?
The flat loop feels like progress
Watch what happens when a complex, multi-part request hits a system that only knows how to execute. No decomposition. No written plan. No revise step. Just a flat tool loop: try, try, try again — confident the whole way. It demos beautifully. It produces words, tickets, and a sense of speed. What it does not produce is a place to stand when part three contradicts part one, or when the answer comes back at half confidence and nobody asks again.
That is the failure mode I care about: complex work treated like a single swing. You skip the plan because planning feels slow. You skip the revise because revising feels like doubt. You skip the second opinion because the first answer arrived with good grammar. Grammar is not evidence. Confidence is not a finish line.
I wrote about why the demo is not the product — the invisible 80% still costs what it always cost. Same pattern here. The flat loop is the demo of thinking. The product of thinking is a loop you can audit: break it down, do the work, check the work, fix what drifted.
Decompose → execute → revise
The discipline is old. AI just made the counterfeit version cheaper. When the request has more than one job inside it, the honest sequence is:
- Decompose. Name the parts. Write the plan in language a skeptic could follow. If you cannot name the parts, you do not have a plan — you have a vibe with a deadline.
- Execute. Do the work against those parts, not against the loudest next impulse. First instinct chooses the direction; the plan chooses the next step.
- Revise. Compare the result to the plan. Branch when a path fails. Revisit a thought you already had if the evidence moved. This is not indecision. This is how intelligence stays honest when the ground moves weekly — the same reason I argue for intelligence over laminated playbooks.
Thoughts that can branch and be revisited are not weakness. They are how you refuse to pretend the first draft of the plan was scripture.
Low confidence without a second opinion
Here is the tell I watch for in people and in systems: an answer arrives with thin certainty — hedged, partial, "probably," "I think" — and the organization treats it like a green light because something came back. No second pass. No human in the loop. No "show me where this still matches the plan."
That is how you get polite fiction in a customer email, a price nobody authorized, a roadmap item marked done that is still a costume. At CI Web Group we keep a human where errors are expensive for exactly this reason. Confidence is the product's default state; correctness is something you engineer — and engineering includes the boring second look when the first look was only forty percent sure.
If you are scared and all in, good — fear keeps you from worshiping the first answer. All-in keeps you from freezing. The pair only works if you still check alignment.
Check everything against the plan
Before you call it done, run the alignment pass — on yourself, your team, your agents:
- What did we say the outcome was, in one sentence?
- Which parts of the work map to which parts of that sentence?
- Where did we improvise, and did improvisation still serve the plan — or replace it?
- Where was confidence thin, and who gave the second opinion?
- If we ask the same question tomorrow, would we still defend this path?
This is not bureaucracy. This is how you keep first instinct from becoming first-and-only instinct. The operators who will own this era are not the ones who never trust their gut. They are the ones who trust it, write the plan, execute the plan, and then have the spine to ask — out loud — whether the confidence was well placed.
The rule, stated plain
Trust your first instinct. Write the plan it implies. Decompose the work. Execute against the parts. Revise when reality talks back. When confidence is thin, get a second opinion before the customer does. Check everything — not to kill speed, but to keep speed pointed at the thing you said you were building.
Was your confidence well placed in the plan? If you cannot answer that without flinching, you are still in the flat loop. Get out. The future does not belong to the first answer. It belongs to the aligned one.