Fourteen taxes in total. Three in the moment, five across your career, two on each piece of work you sign, four on the institution, compounding across years.

 

Quadrant 1: Individual, Per-interaction

Taxes paid the moment you skim.

• Under-challenge, the first plausible answer accepted; the better one was three prompts away.

• Under-verification, different reliability thresholds, all treated the same.

• Premature convergence, the first answer was adequate, you stopped.

The board paper errors paid all three at once.

 

Quadrant 2: Individual, Compounding

Taxes that arrive over months. The expensive ones, easiest to miss.

• Context debt, prompts and standards never captured.

• Judgement atrophy, the muscle that tells good from plausible goes quiet.

• Jagged frontier invisibility, no picture of where AI is reliable; blindsided by the spikes.

• Fragile trust calibration, drifting to blind trust or blanket distrust.

• Self-knowledge gap, every honest AI session is a mirror; skimming closes it.

Six months in, the leader who skims has none of these; the leader who loops has all five.

​

Quadrant 3: Organisation, Per-interaction

Taxes others pay on the work you sign.

• Reader disenfranchisement, the colleague, junior or customer who reads the skim and quietly draws down the trust they had in your name.

• Under-capture at team level, the insight your session produced that the team never sees.

 

Quadrant 4: Organisation, Compounding

Taxes the institution pays over time. The most expensive ones, nearly always missed until they are irreversible.

• Loss of distinctive voice, every output drifting toward the statistical centre of the model.

• Downstream teaching failure, leaders who skim produce teams that skim.

• Institutional context never builds, the curated-context advantage that no model can substitute for.

• External trust erosion, customers, partners and regulators noticing, slowly.

 

Genesis

Most of the taxes themselves are known. Ethan Mollick named the Jagged Frontier, the uneven shape of where AI is reliable. Kate Niederhoffer and Jeff Hancock at BetterUp and Stanford named workslop, confidently produced AI output that does not stand up. HBR asked last year whether AI had ended thought leadership. What is new in the Skim Tax is to total the bill.

 

The refund

Every one of the fourteen taxes is paid back by the same hour of effort: run the Investment Loop, challenge, check, capture, once per meaningful AI session. Challenge prevents under-challenge and premature convergence; Check builds trust calibration and maps the jagged frontier; Capture prevents context debt and builds the institutional library. What the Skim Tax costs, the Investment Loop buys back. Companion piece: virenlall.com/investment-loop.

​

How ChangeSchool applies it with executives

We run senior teams through the Skim Tax Audit in three-hour cohort sessions. Bring a real AI-assisted output that failed. Map every cost that flowed from the failure onto the four quadrants on a blank page. Overlay the canonical fourteen and notice which quadrants the team under-populated.

Tested with chairs and governors in further education, deep-tech founders at the Royal Academy of Engineering, manufacturing leaders in the UK government’s Made Smarter programme, and C-suite cohorts in corporate populations.

​

​

“The Skim Tax: fourteen costs paid invisibly when AI output is skimmed instead of worked through.”

 

Viren Lall, Managing Director,

ChangeSchool LDN (2026).

virenlall.com/skim-tax