The AI Engagement Trap

The Engagement Matrix, two axes (ownership claimed/disclosed, evaluated on merit/provenance) and four positions: Calibrator (target), Critic (half-way), Co-Author (trap), Rejector (NIH trap).

​

The Engagement Matrix holds the four positions.

• Calibrator (target), you evaluate the output on merit before deciding what to do with it, and when you send it on you disclose what AI contributed; the colleague reading downstream can tell what they are looking at.

• Critic (half-way), you read carefully and pick the output apart, but you never integrate it into your own work; useful as a check on someone else’s draft, not as a working posture for your own.

• Co-Author (trap), you accept the output because it reads well and it came out of your prompt, then you sign it as your own work without separating what AI did from what you did; the people downstream of you cannot tell what they are looking at, and over time neither can you.

• Rejector (NIH trap), you dismiss the output before reading it because it came from AI; you forfeit the merit assessment and any value that was in the draft.

​​

Most senior leaders default to one of the two trap poles depending on domain. NIH-style rejection in the areas where their professional identity is invested, ownership blur in the areas where it is not. The pattern is rarely uniform across a leader’s week, and that domain-specificity is what makes it hard to see.

​

Genesis

Two older frames sit underneath. Ralph Katz and Thomas Allen (1982), in R&D Management, studied fifty research-and-development project groups and found that highly cohesive teams systematically rejected good ideas from outside the group. The rejection was not on merit; it was on provenance. They called it Not-Invented-Here (NIH) syndrome, and the mechanism has replicated across four decades because the pattern is about how humans evaluate ideas whose provenance threatens their professional identity.

​

The second frame is the workslop literature. Kate Niederhoffer and Jeff Hancock at BetterUp and Stanford (fall 2025), through their year-long research, report that 53% of surveyed knowledge workers send AI-generated work they know is below standard, typically believing at the moment of sending that they are the authors of what they are sending. Harvard Business Review (24 September 2025) published AI-Generated Workslop Is Destroying Productivity, naming the pattern in the at-work literature. Ownership has blurred without anyone deciding to blur it.

​

Why it matters now

AI is the ultimate outside source: a non-human category of origin that the senior leader’s professional identity was built without reference to. Four forces pull a leader toward the Rejector pole, each rational in the moment: identity threat, expertise stake, fluency read as a provenance tell, and a missing provenance signal the filter cannot adjudicate. The Co-Author pole is pulled by the opposite asymmetry. AI output is fluent, ready-to-send, and absorbs into the leader’s voice without resistance. Both reflexes arrive faster than the merit evaluation that should govern engagement.

​

The moves

Three moves shift a leader from reflex to calibration.

​

Run the Blind Evaluation. A colleague sends three short documents on a common topic, one AI-authored, one human-authored, one mixed, without disclosing which is which. Rate each on merit and rank them, then ask for the reveal. The rating shift between blind and disclosed is the leader’s pole.

​

Install the phrase that catches the dominant pole. For a Rejector: ‘Before I reject this, have I read it on merit?’ For a Co-Author: ‘Before I claim this, have I done more than write the prompt?’ The reflex is fast, and only a sentence structured against it catches it.

​

Put three named pieces of work through the Calibrator filter this week. A board paper drafted with AI assistance, a memo from a direct report, an email reply AI helped draft. Merit evaluation first, ownership disclosure second.

​

How ChangeSchool applies it with executives

We run the Blind Evaluation as a live cohort exercise in three-hour cohort sessions, then map each leader’s week of AI engagements onto the Engagement Matrix by intent rather than by output.

​

The discipline

Three habits embed the practice. A weekly Calibrator review, five minutes on a Friday, assigning each significant AI engagement to a position on the Engagement Matrix and noticing the domain-specific split. An attribution line added to AI-assisted work the leader signs or sends, written for the leader rather than the recipient, because the line cannot be written without first separating merit from provenance. A team-level normalisation move, asking one inquiry-mode question of any direct report sending AI-assisted work, ‘walk me through how you put this together’, so disclosure becomes the precondition for team-level calibration.

​

The discipline is to make merit evaluation the default mode of engagement with AI output, against the gravitational pull of the two cheaper modes that arrive faster.

​

​

​

‘The leaders who reject AI and the leaders who claim it as their own are practising the same failure.’

 

 

Viren Lall, Managing Director,

ChangeSchool LDN (2026).

virenlall.com/ai-engagement-trap

​

​

​