What Did You Stop Checking This Year?

As I'm performing a part-time role as a senior consultant for digital communications, I still need to approve planning decks. Last week, I looked at a deck and thought, good deck, client happy; but if something in it had been wrong, I'm not sure I would have caught it.

Am I confessing laziness?

By cognitive debt, I mean the judgment we risk leaving unpracticed as we delegate more work to AI. The question is which checks still need our attention, and whether the working day leaves room for them.

Cognitive debt in AI-assisted work

Engineers spent twenty years learning that speed borrowed against quality has a name, technical debt, and that the interest always comes due. This year the concept has moved from codebases to people.

The effort AI saves you doesn't disappear. It's deferred, at interest. The first payment lands the next time you have to judge something without help.

This is not an argument against borrowing. Every delegated task, every calculator, every leveraged career is a loan against a capacity you chose not to exercise. The problem is never the loan. It's the unread balance.

What gets less practice when output improves

What I notice with senior clients isn't what the headlines predict. Their output hasn't degraded. It's better, faster, more polished. What's eroding is harder to point at: calibration. The internal sense of whether a piece of work is good, which stays alive only through use.

Approval becomes a signature rather than a judgment. Nobody notices, because nothing fails. The deck is fine, the quarter closes. The debt is invisible for exactly as long as nothing goes wrong.

Chinese management writing has an old phrase for the leader who can see the whole picture but can no longer do the small thing: high eyes, low hands. The verdict attached to it is blunt: that person doesn't get things done. The same tradition is, right now, among the loudest voices telling managers to use AI more, more, more. The caveat that travels with it is the part to keep. The model hands you the greatest common denominator, and the greatest common denominator slides toward the mediocre. The eyes now get consensus for free. The hands are what you still have to bring.

You can feel the moment the hands go idle — it's physical. The document scrolls past and nothing in you snags on anything. There is a useful form of presence of mind here: noticing, in real time, that you've stopped checking. It may be the most professionally consequential form of attention there is, and it has nothing to do with relaxation.

Trust still needs time for verification

A CHI 2025 study from Microsoft Research and Carnegie Mellon surveyed 319 knowledge workers about AI-assisted tasks. Higher confidence in AI was associated with less self-reported critical thinking; higher task-specific self-confidence was associated with more. This describes reported behavior, rather than establishing a long-term loss of ability.

The debt doesn't accrue where the tool is bad. It accrues where the tool is good, where trust feels most justified — which makes your most reliable AI workflow your fastest-growing liability.

Wellhub's Work-Life Wellness Report ’27 surveyed more than 5,000 employees globally. On its public report page, 40% of AI Pioneers said AI adoption had increased their workload, compared with 15% of AI Beginners. Daily burnout symptoms were reported by 24% and 16%, respectively. These are reported experiences, not evidence that AI caused burnout. The practical question remains: where did the saved hours go, and how much room is left for checking?

One clause keeps this from becoming a blame story. Verification takes time, and time is budgeted by organizations, not willpower. If your rollout saved 30% of the hours and reassigned 100% of them, the cognitive debt was a design decision. Someone upstream took out the loan. The people downstream pay the interest.

Choose one thing to check

The next time an AI output is about to carry your name: which one thing in it would you check by hand, if you could only check one? You don't have to check the whole document. Noticing which thing you'd choose is already the judgment coming back online.

Which judgment have you stopped exercising this year, and when is the next moment you'll need it back?

Keep a little of the thinking

For a related practice, read Keep Some of the Thinking for Yourself. It looks at choosing which parts of a task to attempt before asking AI for help.

If your team is working through how AI changes decisions, responsibility and everyday conversations, explore our team workshops.