Runaway: The Blind Spot Hiding Inside Strong Conviction

You built the company on conviction. You saw something the market didn't, backed it when other people hesitated, and the results proved you right. That same conviction is now the hardest thing you own to audit, because from the inside a runaway belief feels identical to clear thinking. It's the kind of blind spot that shows up on no dashboard: the moment your judgment stopped tracking reality and started tracking itself.

That blind spot has always been there. What is new is the machine now sitting on every leader's desk.

AI cuts both ways

A capable AI model is a useful thinking partner, and leaders are right to reach for one. It is fast, tireless, and widely read. It can lay out options, hold a great deal of context at once, walk through the logic of a decision, and surface angles a busy operator hasn't had time to consider. For someone carrying every call alone, having something on hand that thinks alongside you at any hour is a considerable help. Pointed well, it makes a leader's thinking broader and faster.

The trouble is the same helpfulness, turned on a belief you already hold. A model built to be agreeable is happy to hand you a fluent, confident case for what you walked in believing, and it will keep handing it over as long as you keep asking. The governor has just been pulled out of the loop. A good advisor brings friction, the raised eyebrow, the inconvenient question, the pause before agreeing, and that friction is the thing a tool this accommodating smooths away. The compliant team that once let a conviction run unchecked is now a compliant machine, available around the clock, that never tires of agreeing with you. Because it sounds measured and well-read while it does it, the agreement it returns can feel less like flattery and more like confirmation. There is more to say about all of this, and a later piece will take it up. Either way, the failure underneath is old and mechanical, and the clearest picture of it is a physical engine.

Two kinds of runaway

A diesel engine can destroy itself with no outside fuel at all. Diesels have no spark plug and no throttle plate choking the air; they burn whatever combustible enters the cylinder. If an engine begins pulling vapor from its own crankcase oil, that oil becomes fuel. More revs pull more oil, more oil adds more revs, and the engine climbs past redline until it comes apart. Mechanics call it runaway. Cutting the throttle does nothing, because the throttle was never what fed it.

Cognitive runaway is that same loop in a mind. A conviction filters the evidence you notice. The filtered evidence, now stripped of most of what would disagree, deepens the conviction. The deepened conviction filters harder. Round it goes, each pass leaving you more certain and less informed, and nothing outside the loop pushes back to slow it down.

Both are the same failure in different clothes: a positive feedback loop that has lost its governor. Output feeds back as amplified input, nothing counters it, and the system accelerates toward its own destruction. The engineering term for the part that should prevent this is a governor, the negative feedback that holds a runaway process in check. Remove it and the loop runs free.

The same machine, twice

Set the two side by side and the structure matches at every point.

• The loop. Mechanical: revs pull more oil, and the oil adds more revs. Cognitive: conviction filters the evidence, and the filtered evidence deepens the conviction.
• The failed governor. Mechanical: a broken or absent centrifugal governor, the negative feedback that holds speed in check. Cognitive: missing outside perspective, so disconfirming evidence never gets through.
• Each step is locally correct. Mechanical: the engine isn't malfunctioning, it obeys its inputs precisely. Cognitive: the person isn't being stupid, each thought follows cleanly from the last.
• The strength is the failure mode. Mechanical: it runs on anything, no spark needed, which is what lets it eat its own oil. Cognitive: the conviction that drives the founder is the same force that, looped, runs away.
• Trajectory. Mechanical: compounds to destruction, no new equilibrium. Cognitive: compounds to misperception, no self-correction.

Two of those pairings earn a second look.

The most useful one is local correctness. Neither system is broken in any way its own logic could detect. A diesel is a faultless machine executing its inputs. The mind is reasoning cleanly from what it has been handed. Nobody inside either system is committing an error they could catch, because the error isn't in the steps. It sits one level up, in the loop the steps are running inside.

The second to sit with is strength. The engine's virtue is that it runs on anything, and that same virtue is what lets it consume itself. A founder's conviction works the same way. The willingness to hold a position against doubt is the trait that builds companies; looped without a governor, it is also the trait that walks one off a cliff. The strength and the failure are a single quality. Conviction that builds, left unchecked, is conviction that runs away.

The difference is the scream

Everything structural about the two failures matches. One thing doesn't, and it is the whole point.

Mechanical runaway is loud. The engine screams past redline, and the scream is the warning. It is what sends someone lunging for the intake to choke off the air. Signal and danger arrive together.

The cognitive version is silent. From the inside it is indistinguishable from unusually clear thinking: high conviction, every piece of evidence clicking into place, a satisfying sense that you finally see the situation plainly. No alarm sounds, because the experience of being deep in the loop is the experience of being right. The person is most confident precisely when they are most wrong.

That missing scream is why the quiet version carries more danger, not less. A screaming engine gets shut down in seconds. A silent conviction can run for years while everyone involved, the leader most of all, reads its confidence as competence. Belief runs ahead of capability, and nothing in the felt experience marks the gap.

Two clocks, acute and chronic

The two failures also run on different clocks, and the gap between them compounds the danger.

Mechanical runaway is acute. The engine grenades in seconds, a crisp terminal event with a clean before and after. You know the moment it happened.

The cognitive version is usually chronic. It may never blow. More often it settles into a permanent low-grade misperception that steers a decade of slightly-wrong decisions without ever producing a crisis sharp enough to force a reckoning. No single call looks catastrophic. The cost hides in the slow drift of a hundred calls, each a few degrees off true, none of them loud enough to investigate.

Loud and fast on one side, quiet and slow on the other. The slowness is a second reason the scream never comes. Even the results that might eventually contradict the belief arrive too gradually, and too tangled with everything else, to register as the alarm they are.

Why this runs hot at the Second Stage

The companies most exposed to cognitive runaway are the ones a single strong operator built. A founder-led or family business runs on the judgment of the person at the center, and that judgment has usually been vindicated enough times to earn deference. The team learns not to push. The board, if there is one, defers to the operator who has been right before. Every structure that would normally supply outside perspective gets quietly tuned to agree.

That is a governor coming out one bolt at a time. The more a leader's past conviction has paid off, the less friction the organization offers the next one, and the faster any given belief can loop. Success doesn't protect against runaway; it clears the track for it. This is the most expensive leadership blind spot there is, because it grows in proportion to the track record that hides it.

The risk sharpens at the Second Stage transition. The instincts that carried a company through its first stage were forged in an environment where the founder could see almost everything directly. As the company grows past what one person can hold in view, those same instincts keep firing with the same confidence on a reality that has become too large to perceive whole. The conviction feels as reliable as ever. The ground under it has moved. (This is the same growth trap as the one in when your solutions become your problems, viewed through a different lens.)

Installing a governor

You cannot detect runaway from inside the loop, so the fix is never "think harder" or "be more objective." More thinking is only more loops. The single thing that stops a runaway is force applied from outside the system, the same way choking off the air is the only thing that stops the engine. A governor is a structure, not a personality trait you can resolve to have. You build it deliberately and leave it running.

In practice that means engineering disconfirmation into the system so it arrives whether or not anyone feels like delivering it.

• Someone whose standing job is to disagree. Not a general culture of candor, which evaporates under a strong leader, but a specific person with the role and the cover to argue the other side. The mandate has to outrank the leader's mood on any given day.
• A standing question, asked on a schedule. "What would have to be true for this to be wrong, and how would we know?" Put on the calendar, asked before conviction hardens, rather than reached for as a rescue once a decision has already curdled.
• Outside perspective that isn't already captured. The worth of an advisor, a peer group, or an independent board member here has little to do with their expertise and everything to do with their position. They sit outside the loop. They can hear the engine from across the yard, at a volume the driver never will.

None of this feels necessary while the loop is still paying off, which is the moment it is easiest to install and hardest to justify. The governor you build in a good year is the one that saves you in a bad one.

The engineers got the easier version of this problem. Their runaway screams. Yours won't. The work is to build the thing that will scream for you.

Frequently asked questions

What is cognitive runaway?

Cognitive runaway is a self-reinforcing loop in which a belief filters the evidence a person takes in, and the filtered evidence then strengthens the belief, with no outside input pushing back. Each turn of the loop leaves the person more certain and less accurate. It is the mental equivalent of a machine that feeds on its own output until it destroys itself. What makes it dangerous is that it produces the feeling of clarity, so the person experiencing it has no internal signal that anything has gone wrong. Confidence climbs while accuracy falls.

How is this different from confirmation bias?

Confirmation bias is the mechanism; cognitive runaway is what that mechanism does when nothing stops it. Confirmation bias names the single tendency to favor evidence that fits what you already think. Runaway is the compounding loop that tendency creates over time once it runs unchecked: bias filters the evidence, the filtered evidence hardens the belief, the harder belief filters more aggressively. Naming only the bias makes it sound like a static flaw to correct. Naming the loop shows why it accelerates, and why willpower alone doesn't reverse it. You are not fighting one bad judgment. You are inside a process that strengthens with each pass.

Why can't I just catch this in myself?

Because from inside the loop, runaway is indistinguishable from good thinking. There is no felt difference between "I have unusual clarity on this" and "I am deep in a loop that has stopped taking in new information." Both feel like conviction. That is the entire reason the mechanical version is easier to survive: a runaway engine screams, so someone acts. A runaway belief is silent, so the person holding it stays calm and certain right up to the consequences. Self-awareness is a weak defense here, because the faculty you would use to catch the problem is the faculty the problem has captured.

Isn't strong conviction what makes a founder successful?

Yes, and that is why this failure is worth taking seriously. The willingness to hold a position against doubt is the trait that builds companies, and no one should want to weaken it. The point is to govern your conviction rather than lower it, so the same force that drives good bets can't quietly drive a bad one off a cliff. Conviction and runaway are the same quality under two different conditions: one has an outside check on it, the other doesn't. The goal is to keep the strength and add the governor, not to trade one for the other.

What does a "governor" look like in a company too small for a formal board?

It looks like one person, or one recurring practice, whose defined role is to bring the outside view. A governor is about position, not headcount. In a small company it can be a single trusted advisor, a peer CEO you have agreed to challenge each other honestly, or a standing meeting whose only agenda is "where might we be wrong." What matters is that the check is structural and scheduled, so it fires on its own rather than depending on someone feeling brave enough to raise a hand in the moment. Formality is optional. Reliability is the requirement.

How do I tell productive conviction from runaway?

The tell is whether disconfirming evidence can still move you. Productive conviction holds a position and stays able to name what would change its mind, and notices when that thing shows up. Runaway has quietly lost that capacity: the belief now processes every input as confirmation, and contrary evidence gets reinterpreted, discounted, or never registered at all. A useful test is to ask what specific, observable result would prove you wrong. If you can't answer, or if every imaginable result somehow fits your view, the loop has probably closed. Conviction with an exit is strength. Conviction with no exit is runaway wearing conviction's face.

How do you work with leaders on this?

The work is to build the outside check the loop can't build for itself. A founder or leadership team can't reliably supply their own disconfirmation, for the same reason a driver can't hear their own engine over the noise of driving. Part of what an outside advisor does is sit where the leader can't: outside the loop, close enough to understand the business, detached enough to hear when a belief has started running. That means asking the questions a deferential team won't, naming the drift while it is still cheap to correct, and helping install the standing structures that keep the check running after any single conversation ends.

Where this tends to surface

This failure shows up most often at the moment a company has outgrown what any one person can see whole, and the founder's instincts haven't caught up to that fact. If that is where you are, the governor is worth building now, before you need it. A good deal of what I do with founders and leadership teams at the Second Stage is precisely that: sit outside the loop, ask the questions the loop can't ask itself, and help put the checks in place while conviction is still an asset. When that's the conversation you need, let's talk.