Glossary

The discipline of asking, before you optimise a solution with AI, whether there is a fundamentally different approach that makes optimisation unnecessary.

Six cognitive biases that distort AI use, grouped by where they bite; biases of the prompt (anchoring, availability, satisficing) and biases of the response (automation, fluency, confirmation).

Discovery exercise from 2.1A. Compare your original prompt for a real task with a revised prompt written after restating the problem from two different vantage points; the distance between them is your prompt gap.

A three-move sequence (diagnose the AI-shaped challenge; filter by unfair advantage; concentrate the moves) for narrowing AI-expanded option space into focused, defensible action. Anchored on Rumelt's kernel of strategy.

A three-move practice (Challenge, Check, Capture) for treating AI engagement as an investment activity rather than a maintenance one. Runs inside a session and as a loop across sessions. Benefits-framing companion to the Skim Tax.

The cumulative cost of dip-in/dip-out AI use, organised on a 2×2 of level (individual / organisation) × time horizon (per-interaction / compounding). Fourteen named risks. Costs-framing companion to the Investment Loop.

The deepest form of Capture - a persistent proxy that carries the leader's judgment, values and voice forward. Block 4's constructive close; anchored in Tapscott's Digital Don exemplar, extended into a disciplined practice.

The governance principle that each leader must retain ownership of the identic AI that represents them. The sovereignty question spans model, data, training signal, values, override and exit rights.

The framing piece's organising axis, running from identity absent (workslop) to identity total (identic AI). Every article in the series names a specific move along this axis.

The three observable behaviours that signal cognitive surrender (cannot summarise from memory; accepted first answer; signs undefended work) paired with their three interventions (summarise-from-memory test; three-question rule; pre-mortem aloud). From 2.2.

The dual failure mode around AI output provenance, and the calibrated middle-band position that exits it. From 2.3.

The five-step practice that scales AI capability from a leader's personal use into a team's durable capability - leader delegates to AI → refine and encapsulate → transfer the skills project → team member delegates to AI → leader builds quality-check process. From 3.1.

Five named roles AI can take in a leader's work, ordered from most-supervised to least-supervised. Authority-position taxonomy; complements Daugherty and Wilson's Six Hybrid Roles (direction-of-help taxonomy). From 3.2.

The deliberate thirty-second pause before any AI interaction to choose between input-first (leader has context) and suggestion-first (leader wants new framings), or mixed mode (suggestion-first for framing, input-first for execution). From 3.3.

The pattern of using AI for the cheapest-available tasks rather than the highest-value ones, because the tool's low cost of invocation biases allocation toward whatever is nearest to hand. Sits in the maintenance × dip-in/dip-out quadrant of the Newport–Saunders stack. Block 1 opener.

The fixed-mindset trap in which a senior leader's expertise becomes the identity obstacle to practising the AI use that would extend that expertise. Block 1 closer.

The antidote to the Expert's Handicap. The deliberate willingness to pay, in visible slowness and early clumsiness, for a capability that does not arrive without that cost.