Essay · July 2026

Cognitive Debt at the Top

Output grew cheap. Judging whether it is right did not.

AI made producing things almost free. A report, a model, a deck, a plan, all of it arrives in seconds now. What stayed expensive is the work of judging whether any of it is right, because that still runs on human attention, and human attention did not get any faster.

This is the bill most leaders have not seen yet. Every artifact AI produces comes with an obligation attached: read it, question it, reconcile it, decide. When output was slow, the obligation was small, because there was not much of it. Now output is a flood, and the obligation compounds. Researchers at MIT have a name for the version that shows up in the brain, cognitive debt, the quiet cost of leaning on the machine to think until your own engagement fades. Inside an organization the debt is structural. Volume grows faster than the capacity to check it, and the gap is where bad decisions hide.

I watch this on my own calendar. The day was already full before AI started adding a layer of artifacts that all look finished, a polished draft, a confident analysis, a recommendation with its reasoning laid out cleanly, each one asking for a slice of judgment I have to find somewhere. The trouble is exactly that they look finished. A clean surface invites you to skip the question underneath it, and at senior altitude, the questions you skip are the expensive ones.

The damage compounds in a specific way. Output rises, verification falls behind, and people begin to assume that whatever reached them was already checked. It rarely was. Each person in the chain trusts that someone earlier did the work, and the unverified conclusion travels up with all the confidence of a finished product. By the time it reaches a real decision, nobody remembers what was tested and what was guessed. That is decision drift, and it is harder to catch than a hallucination, because nothing on the surface looks wrong.

The instinct is to verify everything, which is its own kind of failure. You do not have the hours, and treating every artifact as suspect paralyzes a team as surely as trusting all of them. The discipline is to decide in advance where verification is mandatory, where good enough will do, and where AI output cannot be used at all without human proof. I think of it as a verification budget. Like any budget, it forces a choice about where the scarce resource goes, and here the scarce resource is attention.

In my own function this is not optional. We built verification into how the work flows: what gets checked, by whom, against what evidence, before a finding turns into a decision. The aim is to make the speed produce sound conclusions, not just fast ones. A team that ships ten times as much and verifies none of it is only more confident about work nobody checked.

AI did not remove the work. It moved it, from producing the artifact to judging it, and then it hid the meter. The leaders who pay the debt down are the ones who can see the meter running, who treat every clean draft as an invoice for a piece of their attention, and who spend that attention on purpose. The ones who cannot see it go on accepting finished-looking work until a decision built on something nobody checked goes wrong, in a way that turns out, in hindsight, to have been entirely visible.

What did you take on faith this week because it arrived looking finished?