Most Decisions Are Made Before We Think We're Making Them

By Momar Lissa Ndiaye ("MLN"), Founder & CEO, weyoga Inc.

The Thesis movement of this series argued that the scarce layer in AI is recognition, not intelligence. The Mechanics movement, which begins here, asks the prior question: recognition of what, exactly? What is actually happening inside human decisions that makes them so resistant to information — and so responsive to being seen?

Start with the claim in the title, stated at full strength: by the time a decision feels like a decision, most of it has already happened.

This claim needs two terms, and the reader should slow down here, because everything in this essay — and much of this series — depends on the distinction between them. Decision formation is the upstream process by which a choice takes shape: which options exist, which evidence counts, which outcome the whole apparatus is leaning toward. Decision ratification is the downstream, felt experience of choosing — the deliberation, the pro-con list, the conversations with friends, the late-night weighing. Our entire folk model of choice assumes these are the same event. The claim of this essay is that they are different events, separated in time, and that by the point of ratification, formation has usually converged. What we experience as choosing is, more often than we would like, confirming.

Formation is not mysterious; it runs on three mechanics, and they are worth stating as first principles rather than examples, because later essays will build on them as a model. Framing determines which options ever made it onto the list — the person "deciding between two job offers" already made the larger decision, to leave, weeks earlier, without ceremony and usually without noticing. Evidence selection determines what the deliberation will weigh — by the time we deliberate, we have spent days gathering reasons, and we did not gather them neutrally; we collected what the forming conclusion needed. Precedent supplies the direction for both: the strongest single predictor of what a person will choose in a situation is what that person chose in structurally similar situations before — nearly invisible from inside any one decision, nearly decisive across a sequence of them.

The evidence for precedent's weight is one honest look at anyone's history, including your own. The founder who has "carefully decided" to take on a demanding new commitment for the fourth consecutive spring. The investor whose "case-by-case judgments" somehow always overweight the same kind of story. The person who has "thought hard" about each of three relationships that ended for the same reason. In every case the deliberation was real, effortful, and sincere. And in every case, the conclusion was largely settled before deliberation opened. This is the sentence to sit with, because it inverts where we assign fault: the reasoning was sound; the inputs were rigged, by history, upstream. Most accounts of poor decisions blame faulty reasoning at the moment of choice. The record suggests something stranger — people reason well, repeatedly, about option sets and evidence pools that precedent assembled for them, and then sign the result as fresh judgment.

Now, the steelman, because this claim has distinguished opposition, and because the deterministic reading of it must be explicitly refused. Humans do deliberate to real effect; people change careers against type, leave patterns, surprise their own histories. Deliberation is not theater, and nothing here says people are predetermined. The claim is about where agency operates, not whether it exists: deliberation works on what it can see. When the pattern driving formation is visible to the person, deliberation can engage it and genuine departure becomes possible — this is precisely what makes good therapy, great mentorship, and long marriages so consequential. When the pattern is invisible, deliberation runs anyway, feels identical from inside, and lands where formation was always going to land. The difference between the two cases is not willpower, intelligence, or information. It is recognition.

One distinction should be installed here before it can be misread later: patterns make behavior predictable; visibility makes it changeable. Prediction and intervention are different operations. The existence of precedent means an outside observer could forecast the ratification; it does not mean the ratification was inevitable — only that it was inevitable unseen. That asymmetry between the seen and unseen pattern is where the next essay lives, so I flag it now and leave it.

And here, without ceremony, is the definition this series has been circling since its first essay. Behavioral intelligence is not better reasoning, more knowledge, or more memory. Behavioral intelligence is the capacity to recognize upstream behavioral signatures — the early marks of framing, evidence selection, and precedent — before formation becomes ratification. In humans, it is the rarest form of self-knowledge. In machines, it has never existed at all.

Why not? Notice what this implies about timing, because it sets up everything the Infrastructure movement will say. If formation precedes the felt decision moment, then the moment of maximal leverage is earlier than any moment at which a person would think to ask for help. Pull interfaces — which wait for a formulated question — arrive after framing, after evidence selection, after precedent has voted. They arrive for the ratification. And the limitation of even the most capable model here is not intelligence but observability: ask the frontier model whether to take the job, and it will reason impressively about the job. What it cannot say — because the information was never allowed to be observed — is the sentence that changes the decision's trajectory: the last three times you faced this shape of choice, you chose the same way, for reasons you later regretted in the same words. That sentence requires no additional intelligence whatsoever. It requires yesterday. Essay 3's architectural argument, again, now with its mechanism exposed.

None of this diminishes human agency; it locates it. Agency does not live primarily in the deliberation moment, where we have romantically placed it. It lives upstream — in whether formation is seen before it finishes. A person who can see their precedent is, for the first time, participating in the actual decision. Everyone else is participating too, in a sense. Just not at the moment they think they are.

The next essay takes this to its uncomfortable statistical conclusion: just how predictable a life actually is under recurring conditions — and why that predictability, correctly understood, is the best news in this entire series.

Part of The Recognition Layer →

Momar Lissa Ndiaye ("MLN") is the Founder & CEO of weyoga Inc., a Delaware company. — weyoga.ai · mln@weyoga.ai