Metacognition Is a Practiced Habit

Edmundo Barrientos · 2026-07-30 · 3 min read

TL;DR: Getting value from AI requires people to monitor and evaluate their own thinking. Those metacognitive habits are built by practice, in simulated scenarios that force the thinking into the open, and the right scenario is different for a junior analyst than for a senior executive.

The gains followed the habit, not the tool

Across 250 employees at a technology consulting firm in China, MIT Sloan professor Jackson Lu and colleagues found that chatbot use made people's work more creative, but only for those who recognized what they did not know, reflected on their own thinking, and changed approach when results came back weak. Everyone else had the same tools and less to show. Vipin Gupta wrote it up for MIT Sloan Management Review in April 2026.

Those three moves are the whole habit. Gupta's advice is to audit yourself, which is where most organizations stop. Self-auditing is a skill, and skills are built.

You cannot instruct a habit into place

The usual answer is a two-hour prompt session, a slide that says "always verify AI output", and no change on Tuesday. That is the knowing-doing gap in a new outfit. Thinking about your own thinking stays invisible until something makes you put it in writing, which is what a simulated scenario does: a real decision, the reasoning demanded out loud, and feedback on the reasoning rather than the answer.

Three practice scenarios, one per move

Build each from material you already have, plus a deliberately flawed version where the exercise needs one.

  • Find what is missing (recognizing what you do not know). An AI draft of something the person knows well, with a piece quietly absent: the policy answer without the exception. They mark the hole and say how they caught it.
  • Commit before you ask (reflecting on your own thinking). A question your team answers every week, pulled from a ticket queue or a call recording. They write their own answer and how sure they are, then see the model's, then explain the difference. The score is the explanation, not the answer.
  • Recover a weak result (changing approach when results are poor). The scenario hands back a bad output. They diagnose why, change the approach instead of asking again, and state what they changed. Deciding to drop the tool passes.

Rebuild the scenario for the person in it

Take the first one. For a junior analyst, use work they already know well, because you cannot notice an omission in a field you have never worked in. For a senior leader, it is the recommendation going to the board; the missing piece is the risk nobody surfaced, and the judgment being practiced is which analysis may be delegated at all.

Industry sets the artifact: for a claims adjuster, a file whose tidy summary drops a coverage exclusion; for a software engineer, code that passes every test and misses the edge case. Same three moves, different trap, and only the specific version transfers.

The design rule holds either way: scenarios that adapt to what each person actually needs, built from work the organization already does.

How you know it worked

The scenario is also the instrument. Because people put their reasoning in writing, you can watch the habit itself: whether they name what is missing before being told, whether the explanation sharpens over reps, and whether they change approach instead of asking again. Read that next to the work, rework on AI-assisted drafts and QA findings, and you are measuring skill rather than completion. Courses and coaching still do their job; this is the layer they were never built to carry.

References

  • MIT Sloan Management Review. Audit Yourself to Get More From GenAI (Vipin Gupta), 2026.

About the publisher: this article is published by Grasply (grasply.ai), a capability builder. Grasply turns what a company already knows, its docs, calls and recordings, into skills people use at work, for employees, partners and customers. Practice turns knowing into doing, and Grasply shows proof that the work changed.