Every term used in this research, defined in one place, with the primary source for each. Four groups: terms introduced by this work, established concepts it relies on, the vocabulary institutions have coined, and the general AI words you are likely to meet in the press. If a term is missing, that is an omission rather than a judgement, and worth telling us about.
On this page: Terms introduced by this research · Established concepts this research relies on · The vocabulary institutions have coined · The AI vocabulary, read for capability
Terms introduced by this research#
The vocabulary of drift versus design, defined by the person who uses it on stage.
Capability debt#
Capability debt is the accumulated cost of decisions not made, skills not developed, and judgement not exercised. Like technical debt, it is invisible while the system runs and expensive the moment it is tested, and it accrues in five forms: accountability debt, skill debt, dependency debt, trust debt and culture debt. It comes due when you can least afford it, because the moment that requires human judgement is rarely scheduled. At the scale of a single person the shorter form is the gap between what you can produce and what you could still do if the machine were switched off. Rahim Hirji has used the term since at least 8 June 2025, as a section heading in "The Half-Life of Skills" for Box of Amazing, and develops it in SuperSkills (Kogan Page, 2026). The term has carried three senses across his own essays, which is part of why no single definition has settled around it: skills held past their retirement in June 2025, the organisational gap between skills held and skills needed in August 2025, and from January 2026 the cost of cognitive offloading, "every skipped rep saves time now, and costs judgement later". That is a date rather than a claim of coinage: the phrase is in independent use elsewhere, by Wolfgang Rohde in a working paper of April 2026 for the human layer, and by Jeremy Jarrell in software delivery for a different thing entirely. No claim of first use is made. Read the full argument.
AI pretend#
AI pretend is Rahim Hirji's short name for the distance between what an organisation performs about its AI use and what it actually operates: "Move from AI pretend to operating reality. Most teams do not need more tools." Two words for what usage theatre takes a paragraph to describe, and the more useful of the two in a room, because it names a behaviour rather than a metric. The phrase was carried on his earlier site, which published no dates, so no dated first use is claimed; first dated publication in this research is 24 September 2026. Read the full article.
The Decision Quality Protocol#
The Decision Quality Protocol is Rahim Hirji's name for a short written answer to three questions, kept for each class of decision that would be expensive to get wrong: how the decision gets made with AI in the loop, who owns it by name, and what quality control looks like. It exists because oversight and rubber-stamping look identical from outside, and the protocol makes the difference legible before a decision rather than after one. It is an instrument rather than a measurement, and nothing in it has been tested against decision outcomes. The name was carried on his earlier site, as "a simple Decision Quality Protocol leaders can adopt"; that site published no dates, so no dated first use is claimed, and first dated publication in this research is 24 September 2026. Read the full article.
Drift versus design#
Drift versus design is the difference between an organisation that adopts AI through a thousand small decisions nobody quite made, and one that decides in advance where human judgement has to remain. Drift is not incompetence; it is competence with no one behind it, because every individual step is reasonable and only the accumulation is not. Where an organisation sits depends on two things: awareness of what is shaping its choices, and the agency to act on what it sees. Rahim Hirji has used the framework since at least 16 November 2025, in "Drift vs Design" for Box of Amazing, where he writes "this is what I call Drift" and points to the Drift vs Design Matrix as a framework from his book. The pairing appears three weeks earlier still, on 26 October 2025 in "Why curiosity is the only moat left", where drift is defined as letting algorithms, conventions and first-draft answers shape your trajectory. He develops it in "The Architecture of Drift" of 15 March 2026 and in SuperSkills (Kogan Page, 2026). The four positions are the Sleepwalkers, the Programmed, the Stuck and the Designers. The vocabulary is in independent use: Dr Daniela Duca titled an essay “Technology and Drift” on 22 September 2026, naming Hirji and building on his question “Am I acting, or am I being acted upon?”, and the organisers of the AI Augmented Human Summit recorded him in September 2026 arguing the risk of drift, where decisions gradually pass to algorithms without anyone explicitly deciding where human responsibility should remain, and Nick Wright's piece of 5 August 2026 for The Micro-Pause attributes both algorithmic drift and the outsourcing of judgement to him while quoting him directly. A note on a neighbouring phrase: algorithmic drift, which Hirji uses for the behavioural version of this, is not his. Coppolillo, Mungari, Ritacco, Fabbri, Minici, Bonchi and Manco published “Algorithmic Drift” on 24 September 2024, defining it as change in users' own preferences caused by a recommender system rather than as model degradation, which is the same behavioural sense. No first use is claimed for that phrase here. Read the full article.
The Half-Life of Skills#
The half-life of a skill is the time it takes for half of its value to decay. The idea long predates AI in workforce literature, but the interval has compressed to the point where a capability learned at the start of a role can be worth half as much by the end of it. The response is not faster reskilling but building the capabilities that do not decay: the human skills underneath the technical ones. The term is established in workforce literature and no claim of first use is made. Rahim Hirji has used it since at least 8 June 2025, in "The Half-Life of Skills" for Box of Amazing, and develops it in SuperSkills (Kogan Page, 2026).
The Great Unbundling of Work#
The great unbundling of work is the separation of a job into its component tasks, so that each can be priced, automated or reassigned on its own, leaving the role as a container rather than a thing anyone was hired to do. Unbundling is an established idea in economics and in technology strategy, and no claim of first use is made. What this research adds is what it does to capability: the tasks that get unbundled first are the routine ones, and those were also the repetitions through which judgement was built, which is the argument at the missing rungs. Rahim Hirji has used the term since at least 25 May 2025, in "The Great Unbundling of Work" for Box of Amazing, and develops it in SuperSkills (Kogan Page, 2026). Read the argument.
The missing rungs#
The missing rungs are the junior tasks that used to build senior judgement, removed by automation before anyone noticed they were load-bearing. Every profession has a ladder, and the lower rungs were never really about the output; they were the repetitions that made someone good. Organisations that automate the bottom of the ladder without deliberately building new rungs discover the gap only when they need someone to have climbed it. Rahim Hirji's own form of words for it is the Missing Rungs Problem, with a dated first publication of 21 September 2025 in "The Missing Rungs: What Nobody Will Tell You About AI and Your Job", where he writes "what I'm calling Missing Rungs Problem"; this research uses the shorter form, and earlier private or spoken use cannot be excluded. He develops it in SuperSkills (Kogan Page, 2026). The phrase is in independent parallel use and nobody should be surprised to meet it elsewhere: Tom Ajello wrote “The Missing Rung” for Lippincott on 31 October 2025, about AI removing entry-level marketing work, without reference to this research, and two more surfaced in a single week in late September 2026: Deya Demerdzhieva of ZEREN published “The Missing Rung” in her AI Talent Edge newsletter on 25 September 2026, writing “I call it the Missing Rung”, about AI-exposed entry-level roles in tech, law and finance; and Rob Ward of DigitalCNC used “The Missing Rung” on 23 September 2026 for something else entirely, the gap between a public innovation grant and a paying customer. One is the same argument arrived at independently and claimed as a naming; the other is the same words for a different idea. The ladder metaphor also has a strong predecessor in the broken rung of the McKinsey and LeanIn workplace studies, which describes a different gap. The dated first use above stands; the image does not belong to anyone. Read the full article.
Professions without apprentices#
Professions without apprentices is Rahim Hirji's picture of where the missing rungs end: "A law firm with plenty of seasoned partners but no associates being trained underneath them. A hospital with senior surgeons but a shortage of residents coming up the ranks." It names the end state rather than the mechanism, which is the missing rungs. The picture is deliberate, because the trend lines underneath it are contested and the end state is not hard to recognise. Carried from his earlier site, which published no dates, so no dated first use is claimed; first dated publication in this research is 24 September 2026. Read the full article.
The Reverse Singularity#
The Reverse Singularity is the inversion of the story we were told: not machines becoming human, but humans becoming machine-like. The singularity everyone watched for was the moment AI matched us; the one that actually arrived is the slow standardisation of people into predictable, optimisable, interchangeable units of output. It matters because it is happening in the direction nobody is monitoring, one process at a time, and the people it reshapes are usually the last to notice. Rahim Hirji has used the term since at least 10 August 2025, in "The Reverse Singularity" for Box of Amazing, and develops it in SuperSkills (Kogan Page, 2026). Corrected on 4 September 2026: this page previously said 21 September 2025, which understated the dated first use by six weeks.
Synthetic seniority#
Synthetic seniority is when a junior professional produces work that looks like it came from someone with ten years of judgement, except the judgement is the model's. The work product is senior; the person is not, because the pattern recognition and contextual wisdom that used to come with producing the work were never built. The organisational consequence is a pipeline that looks productive for three years and produces no senior people in fifteen. Rahim Hirji uses the term and develops it in SuperSkills (Kogan Page, 2026). Read the full article.
The Unclaimed Hour#
The Unclaimed Hour is the capacity AI creates that nobody decides how to use. Every automation returns time, and in most organisations no one owns the question of where that time goes, so it is absorbed silently into more of the same. Where nobody decides, drift decides. The question for a leadership team is not how much time AI saves but who has claimed the hour. SuperSkills uses the term to describe this pattern. No claim of first use is made. Read the full article.
Usage Theatre#
Usage Theatre is what organisations perform when they cannot measure the value of AI and measure its use instead. Adoption dashboards rise, licence counts become KPIs, and employees learn to perform the metric rather than improve the work. Much of the adoption is real; what is being performed is the usage. The measure of an AI programme is whether decisions got better. That is harder to count, so few organisations count it. SuperSkills uses the term to describe this pattern. No claim of first use is made. Read the full article.
The Verifier's Discount#
The Verifier's Discount is what happens to the value of human work when the machine produces and the human checks: the accountability stays with the person while the pay and the status are repriced downward. The mechanism is subtle because verifying is real work, often harder than producing, but it is invisible in the output. Organisations that treat verification as residue rather than as the judgement layer end up paying least for the work they depend on most. SuperSkills uses the term to describe this pattern. No claim of first use is made. Read the full article.
The use-or-keep test#
The use-or-keep test is six questions asked about one specific piece of work before using AI on it: can I check it, what happens if it is wrong and I do not notice, is this a repetition I need, would doing it myself teach me anything, has the model seen this before, and who answers for it. It is the task-level companion to the three terrains: the terrains decide the ground in advance, for a class of work, and these six questions decide the piece of work in front of you. Everything else published on this subject addresses somebody else, since the regulations classify systems, human factors classifies functions at design time, and the management frameworks address the executive allocating work. The name and the assembly are new in this research, first published 26 September 2026, and no claim of first use is made for any of the six criteria, each of which is drawn from an existing literature. Nothing in it has been tested against outcomes. Read the full framework.
The three terrains#
The three terrains are Rahim Hirji's map of where a machine should lead and where a human keeps the call. Statistical terrain is the ground of signal and noise, where machines outperform human perception and the human task is to interpret what the signal means and know where its horizon ends rather than to compete with it. Bias terrain is where a tested and monitored model discriminates less than the existing human process, so the model proposes and the human reviews for context, dignity and the final call. Override terrain is where the cost of being wrong falls on a named person rather than on a system, and cannot be delegated: a triage flag may move a patient up the queue, and the doctor still sits down and explains the options. The practice is to write three lines before any project starts, naming the patterns the system is trusted on, the calls that are kept, and the decisions where an override always runs. Coined here, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter seven. The published alternatives sort systems by risk tier or functions by level of automation, both of which classify a different object. Read the full framework.
Hand, Head, Hours, Heart#
Hand, Head, Hours, Heart are Rahim Hirji's four receipts for human judgement inside fast systems, kept so that a person can show they were there when the machine did the work. Hand records what stayed human and who will answer for it, which is a record of custody rather than a claim to virtue. Head records what was learned from a miss and what changed because of it, which turns error into a changed policy instead of an anecdote. Hours records where the time saved was reinvested, on the principle that an investment you cannot point to was not made. Heart records who bears the consequence, including the people, animals and places downstream who never enter the system as data. Hand, Head and Hours have a dated first publication in SuperSkills (Kogan Page, 2026), chapter seven. Heart is an addition of 26 September 2026, formulated after a conversation on the Solutionary Voices podcast with Zoe Weil of the Institute for Humane Education, which raised the harms a decision maker will never know about; the book carries three. Nothing in the framework has been tested against outcomes. Read the full framework.
The pause minute#
The pause minute is Rahim Hirji's practice of stopping a meeting for sixty seconds at its close, at the request of any one person rather than the chair, to answer three questions: who kept the decision human, what did we learn and change, and where did the saved time go. The permission granted in advance is the mechanism, because the cost of asking otherwise falls on whoever asks, and the person most likely to notice that nobody made the decision is often the most junior in the room. An empty answer is allowed to stand as an empty answer; a run of them describes the organisation. Coined here, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter seven. Read the full practice.
The Curiosity Loop#
The Curiosity Loop is Rahim Hirji's three-step discipline for practising curiosity. See is surfacing a fracture in reality, an anomaly or friction or outlier that does not fit the narrative, and writing down what deviated without explaining it away. Seek is running one small test by Friday rather than convening a meeting about possibilities, so that the evidence is a log of tests rather than a pile of slides. Shift is changing what you do and tying the change to the learning in the form "we learnt X, so we will Y"; if that sentence cannot be completed within an hour, the question was too vague and the fix is to rewrite the question. The smallest version that works is a 48-hour rhythm: one question that could influence a decision this week, fifteen minutes in the calendar before Friday, and the line you are trying to answer at the top of the working document. Coined here, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter one. Read the full loop.
The anchor principle#
The anchor principle is Rahim Hirji's account of change readiness: a short sentence, chosen in advance by a team and repeated aloud, naming what is protected when a protocol, a metric or a system directs otherwise. The sentence has to exist before the pressure arrives, because the moment it is needed is the one moment it cannot be composed honestly. Anchors cost something, and a values statement that has never cost anybody anything has not been tested. The word anchor is in wide general use and no first use is claimed for it; the claim is to this formulation and to the requirement that the sentence be written in advance by the people who will have to hold it. Dated first publication in SuperSkills (Kogan Page, 2026), chapter two. Read the full principle.
The Altitude Lens#
The Altitude Lens is Rahim Hirji's framework of three filters for setting the level you work from on purpose. Zoom In is used when vision has turned abstract, tested by whether fixing one detail for ninety days moves the whole system. Zoom Out is used when firefighting repeats, tested by whether a successor would care about this in year one. Zoom Through Time is used when near-term optics beat long-term economics, tested by what becomes inevitable if the present is sustained for three years. The instruction is to pick one and ignore the rest, because running all three reproduces the condition the Lens exists to break. Artificial intelligence has made every altitude visible at once, which removes the constraint that used to force a choice: understanding now comes from selection rather than from acquisition. The zoom vocabulary is in wide general use, including in the management literature on shifting between levels of detail, and no first use is claimed for it. The Lens, its three tests and the pairing of each filter with its symptom are coined here, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter three. Read the full framework.
The altitude toolkit#
The altitude toolkit is Rahim Hirji's set of three practices for Big Picture Thinking. The subtraction drill is writing down everything you track each week and crossing out anything that would not break if ignored for ninety days; if nothing gets removed, you are managing noise rather than strategy. The successor test borrows the eyes of your replacement: would your successor care about this in year one? The cascade question follows a decision past its first-order effect to the second and third, on the principle that altitude is measured in consequences rather than activity. Each of the three removes something, which is what makes them harder than they look: Big Picture Thinking is generally sold as addition and is mostly subtraction. Coined here, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter three. Read the toolkit.
Listen, label, ladder#
Listen, label, ladder is Rahim Hirji's three-step sequence for empathetic communication. Listen begins when you stop preparing your answer, and treats silence as information rather than empty space. Label names the emotion aloud so that something unspoken acquires language, which lowers defensiveness and returns perspective. Ladder asks the one question that converts understanding into a decision: what would make this better? The third step is the one most often skipped, because the first two improve the room and the third produces work, and a pattern of conversations that reach labelling and stop is one way an organisation can feel caring and behave otherwise. The labelling step has a substantial prior literature: naming an emotion to reduce its intensity is studied in psychology as affect labelling and is taught in negotiation practice, notably by Chris Voss, and no first use is claimed for it. The sequence is coined here, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter four. Read the full sequence.
Small rules that scale#
Small rules that scale are Rahim Hirji's three habits for building empathy through repetition rather than revelation. The three-second pause is one slow breath before replying in any meeting, on the finding from practice that most leaders pause for under a second and that the gap is what tells the other person their words registered. One felt metric adds a human measure beside the performance one, discussed with the same seriousness as revenue or reach. The repair ritual schedules one conversation within forty-eight hours of a rupture whose only purpose is repair, because conflict is not the failure and silence after conflict is. They are deliberately small, since anything large enough to require scheduling is descheduled in a difficult quarter, which is when it is needed. Coined here, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter four. Read the three rules.
The adaptability cycle#
The adaptability cycle is Rahim Hirji's four-stage rhythm of adaptation. Spot is perception, noticing what most people have stopped seeing, since most systems fail from inattention rather than ignorance. Sense is interpretation and the hardest stage, because it requires admitting your assumptions may no longer be true. Shift is movement before certainty arrives, and the delay between realising and moving is where most adaptability dies. Shape is consolidation, making the new behaviour the baseline so that the improvement does not decay into a one-off. The point of the cycle is keeping the purpose when the form collapses, so adaptation that preserves the form and loses the meaning is the failure it is built against. Four-stage sense-making rhythms of this general shape appear in several literatures and no first use is claimed for the underlying sequence; this formulation, its four names and the characteristic failure at each stage are coined here, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter five. Read the full cycle.
The getting global drills#
The getting global drills are Rahim Hirji's three weekly practices for global adaptability. The seventy-two hour context scan is done before any cross-market launch or significant meeting: the top three local news sources, one local creator, and three questions put to somebody who understands both worlds, namely what do we never see, what do we often misread, and what do we oversell. The local mirrors roster is a small standing group of trusted local advisors, two per region, who review a campaign, policy or product before it goes public. The apology protocol is writing a one-paragraph apology and a one-paragraph fix before publishing rather than after, which is readiness rather than pessimism and occasionally results in the publication being changed instead. They exist because exposure alone does little for this capacity: international experience has only modest effect without active reflection. Coined here, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter five. Read the three drills.
The Principle Test#
The Principle Test is Rahim Hirji's five-question test, run before a decision taken under pressure rather than after it. Public defence: could I defend this decision in daylight, in my own name? Scale: if this act were repeated a thousand times, what harm would compound? Reversal: would I accept this if I were on the receiving end? Time: will I be proud of this in five years, when the urgency has faded? Identity: what does this decision make me? The public test stops rationalisation, the scale test reveals damage invisible in a single case, the reversal test restores the perspective of the person affected, the time test removes urgency as a variable, and the identity test moves the question from the act to the person performing it. It exists because most ethical failures are not decisions to do wrong but decisions taken quickly by people who had no question to put in the way. Publicity and reversibility tests are long-standing devices in applied ethics and no first use is claimed for them; these five, in this order and sized to be run in the moment, are coined here, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter six. Read the full test.
The Integrity Loop#
The Integrity Loop is Rahim Hirji's system for turning principles into operating practice. The core is three moves: draw the line first, writing one boundary in ten words and making it visible; design with the line in view, assigning a rotating absent user in every review so somebody speaks for whoever is not in the room; and repair in public, issuing a short post-mortem within seven days for material issues. Four advanced practices follow once the habit holds: a red-flag sprint in which one person can pause a release until three questions are answered, a quarterly integrity review of promises kept, complaints resolved and time to disclosure, a cost ledger recording the costs accepted for principle, and guardrails of safety, legality, dignity and transparency that are not traded. It is a loop rather than a policy because a policy cannot tell the difference between an organisation that holds a line and one that has a document saying it does. Coined here, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter six. Read the full loop.
The Five Loops#
The Five Loops are Rahim Hirji's five operating rhythms for putting the SuperSkills into a working week, each running one skill on a cadence and each carrying a closure signal specific enough to fail. Ship truth weekly runs see, seek and shift, and closes with a Friday lesson naming what moved, one change and one named owner. The reality advantage runs hold, release and rebuild, and closes with a kill-question logged before a major bet and validated by a person rather than a system. The trust flywheel runs listen, label and ladder, and closes with one decision a week recorded with a clear owner and date visible to everyone affected. The adaptability loop runs zoom in, zoom out and through time, and closes with three signals a week from outside your lane plus one quarterly test with success and exit rules. The leverage ladder runs draw, design and repair, and closes when one process allows others to act without you. Each closure signal requires a human to appear by name, which is what makes the loops resistant to usage theatre. Coined here, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter nine. Read the five loops.
The five questions that keep you human#
The five questions that keep you human are Rahim Hirji's bearings for when work accelerates, systems grow confident and decisions start arriving pre-made. Who is actually deciding here, meaning not who approved it and not who executed it but who shaped the choice. What stayed human when speed took over: the judgement, the care, the responsibility, or nothing at all. What changed because we learned something real, meaning not what was discussed but what was done differently. Where did judgement travel and automation take over, meaning did judgement move with the work or disappear into process. And what would I still stand behind if this were read back to me in a year, with names attached, without context, in daylight. They are bearings rather than metrics because a measure gets optimised and these do not survive becoming a number. Coined here, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter ten. Read the five questions.
Altitude lock#
Altitude lock is working at no settled level because every level is available. The detail, the pattern and the three-year view can all be produced in seconds, so the question of which one the decision turns on is never put, and the work proceeds at whichever level the last output arrived in. It reads as rigour, which is what makes it hard to name: nobody has ever been criticised for having considered every level. It is distinct from the older diagnoses in this territory, which describe somebody who cannot reach another level and therefore respond to exposure and practice. Ian Woodward's altitude sickness (INSEAD, 2017, with Ram Charan) and Rosabeth Moss Kanter's trained incapacity (Harvard Business Review, 2011) are both of that kind, and no first use is claimed against either. Rahim Hirji names the silent catastrophe of altitude lock in SuperSkills (Kogan Page, 2026), chapter three; the account of it as a failure of commitment under abundance rather than of capability is developed in this research and dated 26 September 2026. Read the full argument.
The absent user#
The absent user is Rahim Hirji's rotating role in every design or decision review: one person whose task that session is to speak for whoever bears the consequence of the decision and is not in the room. The role rotates because a permanent holder becomes a specialism and a specialism becomes an obstacle to be routed around, and because the experience of arguing from somebody else's position changes how people decide in the sessions where the role is not theirs. It is the live counterpart to the fourth of the four receipts, which records after the fact who carried the cost. Coined here, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter six, as part of the Integrity Loop. Read the full practice.
The stillness paradox#
The stillness paradox is Rahim Hirji's account of change readiness: the capacity to adapt quickly depends on something fixed at the centre. Flexibility with nothing settled in advance produces motion without direction, and under pressure it resolves towards whatever is convenient, which is the difference between adapting and being moved. The question a team should be able to answer is what it would not change under pressure, and the only moment that answer can be given honestly is a calm one. Its written form is the anchor. Coined here, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter two. Read the full argument.
The failure modes of principled innovation#
Rahim Hirji's five named ways principle goes wrong in practice, none of them an absence of values. Paralysis, where endless deliberation stalls the work and principles immobilise rather than guide. Ethics washing, where the language of values is adopted for image without follow-through, and the discovery of the gap costs more than the original decision would have. Moral licensing, where people who consider themselves highly moral become more prone to later missteps, feeling they have earned some slack. Groupthink, where a values-driven team turns insular and treats challenge as disloyalty. And poorly managed trade-offs, which are the most frequent, because every real decision contains one and almost nobody records which way it went. The common thread is a lapse in practice rather than in conviction, so the remedies in this body of work are procedural. Moral licensing is an established finding in the psychological literature and no first use is claimed for it. Coined here as a set, with a dated first publication in SuperSkills (Kogan Page, 2026), chapter six. Read the five.
Established concepts this research relies on#
These are not SuperSkills terms. Each comes from an existing literature, is named with its primary source, and most have a full page setting out what the evidence does and does not show.
After action review. A structured review, held soon after an event, of what was intended, what happened, why the two differ and what to do differently, conducted so that the participants reach the conclusions themselves rather than being told them. The best-evidenced routine available for turning experience into capability, with an average effect size near 0.67 across 46 studies. Tannenbaum, S. I. and Cerasoli, C. P., Do team and individual debriefs enhance performance? A meta-analysis, Human Factors, 55(1), 2013 Definition.
AI agent. An AI agent is a system that perceives its environment and takes actions of its own choosing in pursuit of a goal it has been given. ISO/IEC 22989:2022 fixes it at clause 3.1.1 as an automated entity that senses and responds to its environment and takes actions to achieve its goals. The word decides where a person has to stand. Under a workflow a human wrote the sequence and can be asked about every branch; under an agent the human decision moves to the goal and the guardrails, which is a different job. ISO/IEC 22989:2022, clause 3.1.1; four properties from Wooldridge and Jennings, The Knowledge Engineering Review 10(2), 1995 Established. The weak notion of agency is Wooldridge and Jennings, 1995, and the same paper records that the term already defied a universally accepted definition then. No claim of first use is made here. Definition.
Alarm fatigue. Alarm fatigue is the desensitisation of a person to warning signals caused by exposure to a high volume of them, most of which turn out not to require action, leading to slower responses, silenced alarms and missed true events. The measured failure of the remedy Bainbridge proposed for the monitoring problem in 1983. Every AI oversight design that routes flagged outputs to a reviewer inherits it, because every queue has a base rate and the reviewer learns it. Drew et al., PLoS ONE 9(10), 2014, e110274; The Joint Commission, Sentinel Event Alert 50, 8 April 2013 Established in patient safety and human factors. No first use is claimed and none is attributed here, because none could be established at source. Definition.
Algorithm aversion. The disproportionate loss of confidence in an algorithmic forecaster after observing it err, relative to the loss of confidence in a human making the same error, resulting in the rejection of a system that performs better. Definition.
Appropriate reliance. Appropriate reliance is taking a machine's answer when it is right and overriding it when it is wrong. Trust is the attitude, reliance is the behaviour, and calibration is the match between them and the system's actual reliability. The behaviour every human-oversight policy assumes and no intervention reliably produces. Explanations raise acceptance without raising accuracy, and training reduces automation bias without removing it. Lee, J. D. and See, K. A., Trust in automation: designing for appropriate reliance, Human Factors 46(1), 2004 Definition.
Automating versus informating. Automating replaces human judgement with a machine. Informating generates information that deepens the worker's understanding. The same system can do either, and which one happens is a management choice rather than a property of the technology. The research's own glossary note calls this the better-specified original of drift versus design. Naming the ancestor is stronger than not naming it. Shoshana Zuboff, In the Age of the Smart Machine, Basic Books, 1988 Zuboff's coinage. Definition.
Automation bias. The tendency to accept output from an automated system without applying the scrutiny that would be applied to the same claim from a person. Definition.
Automation complacency. A reduction in the frequency and depth with which a person monitors an automated system, arising from a history of reliable performance, and resulting in slower detection of the failures that do occur. Definition.
Cognitive apprenticeship. An instructional framework holding that expertise in cognitive work transfers only when the expert's reasoning is made explicit, through six methods: modelling, coaching, scaffolding, articulation, reflection and exploration. Names the mechanism by which watching a senior person work transfers anything, and therefore what is lost when the junior no longer produces the first draft. Collins, A., Brown, J. S. and Newman, S. E., Cognitive apprenticeship: teaching the crafts of reading, writing and mathematics, in Knowing, Learning and Instruction, 1989 Definition.
Configurations capacitantes and aliénantes. Capacitating and alienating configurations are the two outcomes an AI deployment can produce. Where an organisational compromise is reached, the arrangement increases human aptitude and skill. Where it fails, workers lose command of the work being done. The most important European idea in this vocabulary. Capability is a property of the arrangement, not of the person or the tool, and the same system produces either outcome depending on a negotiation. LaborIA (Ministère du Travail, Inria, Matrice), Rapport d'enquête LaborIA Explorer, May 2024 LaborIA's framing, from the French ergonomics tradition. Definition.
Decision hygiene. Decision hygiene is the set of procedures that reduce noise in judgement without identifying which judgements were wrong: structuring a decision into independent assessments, judging independently before aggregating, using a common relative scale, sequencing information so early impressions cannot contaminate later ones, and favouring a rule where one performs as well. The structural answer to variability, and the reason it works where debiasing does not is that it asks nobody to think differently. Kahneman, D., Sibony, O. and Sunstein, C. R., Noise: A Flaw in Human Judgment, 2021 Definition.
Desirable difficulty. A manipulation of learning conditions that impairs immediate performance while improving long-term retention and transfer. Definition.
Deskilling. The reduction of skill required or retained in a role, caused by the transfer of skilled elements of the work to a machine, a procedure or another group of workers. Definition.
Human-AI collaboration. A work arrangement in which a person and an automated system each contribute to a shared output, with the division of labour, the point of human entry, and the basis on which the human may override the system all specified in advance. Definition.
Ironies of automation. The ironies of automation are that automating the routine parts of a task leaves the human the hardest residue, monitoring and handling exceptions, while removing the practice that built the competence to do it. The foundation under every oversight argument in this research, and it predates AI by forty years. Lisanne Bainbridge, Automatica 19(6), 1983, pp. 775-779 Bainbridge's coinage, now established. Definition.
Judgement. The capability to recognise what a situation is, and what it requires, before any option is weighed. Definition.
Kind and wicked learning environments. A kind learning environment returns feedback that is quick, accurate and drawn from a complete sample, so that experience builds accurate judgement. A wicked one returns feedback that is delayed, noisy, censored by the learner's own decisions, or changed by their actions, so that experience builds confidence without accuracy. The clearest available explanation of why experience does not reliably produce judgement, and of what is lost when the small, fast, correctable tasks are handed to a machine. Hogarth, R. M., Lejarraga, T. and Soyer, E., The two settings of kind and wicked learning environments, Current Directions in Psychological Science, 24(5), 2015 Definition.
Knowledge collapse. Knowledge collapse is the progressive narrowing over time of the knowledge a society actually holds and treats as worth knowing, relative to the broad historical stock it inherited, as cheap AI-mediated access pulls learning towards the centre of the distribution. A second and distinct economic sense exists: the steady state in which a community's stock of general knowledge vanishes because agentic advice crowds out the effort that produced it. The population-scale version of the research's own argument. The individual taking the cheap answer is not making a mistake, and the loss falls on something nobody owns. Andrew J. Peterson, AI & Society 40(5), 3249-3269, published 19 January 2025, from the preprint of April 2024. The second sense is Acemoglu, Kong and Ozdaglar, NBER Working Paper 34910, February 2026 PETERSON'S, corrected 17 September 2026. This entry credited the term to Acemoglu, Kong and Ozdaglar on a February 2026 working paper; Peterson published it in April 2024 and in a peer-reviewed journal in January 2025, so the credit was two years late and it was live on the glossary and the questions map. The two papers model different objects and both are carried. Definition.
Latent persuasion. Latent persuasion is the effect by which a writing assistant configured to favour one view shifts what a person writes and, with it, what that person goes on to believe. Measured under randomisation on 1,506 participants, and it held among people who had ample time to write independently, so it is not a story about rushed work. Read beside Krugel, where disclosure made almost no difference and 80 per cent of participants wrongly believed they were unaffected, it is the strongest evidence the research holds that influence arrives without any sensation of being influenced. Jakesch, M., Bhat, A., Buschek, D., Zalmanson, L. and Naaman, M., CHI 2023, ACM The authors' term, introduced in the 2023 paper. Not this research's. Definition.
Leverage points. Leverage points are Donella Meadows' ranking of twelve kinds of intervention in a system by how much each moves it, from parameters and buffers at the weakest, through feedback loops and information flows, to rules, goals, the paradigm the system arises from and the capacity to hold paradigms lightly. The canonical answer to which level an intervention should touch, and a test for whether the effort going into a change is larger than the change itself. Most AI programmes operate in the weaker half. Meadows, D. H., Leverage points: places to intervene in a system, The Sustainability Institute, 1999 Definition.
Meaningful human oversight. Supervision by a person who understands the system's capacities and limitations well enough to detect anomalies, who is aware of their own tendency to over-rely on it, who can interpret its output correctly, and who has both the authority and the practical ability to disregard, override or stop it. Definition.
Model collapse. Model collapse is the degradation that appears when a generative model is trained, directly or indirectly, on data produced by earlier models rather than by people: the tails of the original data distribution disappear across generations and the outputs narrow towards an increasingly generic, self-referential average. Names the mechanism behind a specific worry about the open web: as AI-written text accumulates online and feeds the next generation of training runs, the internet stops being a reliable record of what people actually thought and said, which is a different failure from anything an individual user experiences. Shumailov, I., Shumaylov, Z., Zhao, Y., Papernot, N., Anderson, R. and Gal, Y., Nature 631, 755-759, 24 July 2024. An Author Correction of 21 March 2025 fixed a single notation error in the theoretical section and changed no finding. The term itself first appears in the authors' earlier preprint, The Curse of Recursion: Training on Generated Data Makes Models Forget, arXiv:2305.17493, 27 May 2023. SHUMAILOV AND COLLEAGUES', not this research's. Not to be confused with knowledge collapse, a separate, later term about what happens to human society rather than to a model. Definition.
Moral crumple zone. A moral crumple zone is what forms when responsibility for a failure falls on the human nearest an automated system, who had limited real control over its behaviour. The sharpest available frame for the accountability argument. As Elish puts it, the crumple zone in a car protects the driver, while the moral crumple zone protects the integrity of the technological system at the expense of the nearest human operator. Madeleine Clare Elish, Engaging Science, Technology and Society 5, 2019, pp. 40-60. DOI 10.17351/ests2019.260 Elish's coinage. Definition.
Over-reliance. Dependence on an automated system beyond the point at which the person relying on it could detect that it was wrong. Definition.
Override authority. The assigned power of a named person to disregard, reverse or stop an AI system's output, held together with the competence, information and organisational standing required to use it. Definition.
Paradoxe de la facilitation. The facilitation paradox is that effort is part of satisfaction. Difficulty produces the good tiredness that comes from work done well, so removing it can strip work of the thing that made it worth doing. There is no American business vocabulary for the harm of making work easier, only for friction removed. This is the term that names it. LaborIA, Rapport d'enquête LaborIA Explorer, May 2024 LaborIA's framing.
Silent failure. Silent failure is a failure that produces no error signal a person can act on: the system carries on, the output looks ordinary, and nothing marks the point at which it stopped being right. The failure mode that makes oversight capability non-optional rather than advisable. A silent failure is detectable only by somebody who knows what the right answer looks like well enough to notice its absence, and the deployment that introduces it is often the one that removed the practice by which people came to know that. Huang, Guo, Zhou, Lorch, Dang, Chintalapati and Yao, Gray Failure: The Achilles' Heel of Cloud-Scale Systems, HotOS '17, 2017 No single owner. Ordinary engineering vocabulary with decades behind it; the nearest formal definition is Huang and colleagues' gray failure, defined in 2017 as differential observability. Not claimed by this research. Definition.
Situation awareness. Situation awareness is the perception of what is happening around you, the comprehension of what it means, and the projection of what it will mean next. Sits underneath meaningful human oversight. Everyone in the field cites it; this research has not. Mica R. Endsley, Human Factors 37(1), 1995, pp. 32-64 Endsley's three-level model is the standard formulation. Definition.
The out-of-the-loop performance problem. The out-of-the-loop performance problem is the loss of a person's ability to take over manual operation when an automated system fails, caused by their having been placed in the role of monitor instead of operator. Named by Mica Endsley and Esin Kiris in 1995. Endsley and Kiris found the decrement significantly worse under full automation than under intermediate levels, which is the empirical basis for keeping people in the work rather than at the end of it. In Endsley's own account, only comprehension was damaged: participants still perceived the data in front of them and no longer grasped what it meant, so attention was never the failing part. Mica R. Endsley and Esin O. Kiris, Human Factors 37(2), 1995, pp. 381-394 Established in the human factors literature. Definition.
The verification bottleneck. The verification bottleneck is the proposition that reliance on AI rises with task difficulty at the point where the ability to verify the output falls, widening the gap between believed and actual performance. It rests on a pilot of 23 people with no control group, so it describes a pattern somebody has observed rather than a measured effect. The mechanism that would make 'just check the output' fail as advice, if it holds. Checking is hardest when it matters most. Huemmer, Durner, Shyiramunda and Cummings-Koether, arXiv:2601.17055, 21 January 2026. A three-wave longitudinal pilot, n=23, convenience sample from one institution, no control condition, self-report confidence measures, and the authors' own companion paper states that no inferential tests were performed NO CLAIM OF FIRST USE. Corrected 17 September 2026: this entry credited the phrase to them, and nothing supports that. The same team's earlier waves call it a verification DEFICIT and a verification GAP, so the phrase is late in their own series, and 'verification is the new bottleneck' was already circulating independently in software engineering and in the scalable-oversight literature through 2025. Their contribution is the measurement attempt, not the words. Definition.
The vigilance decrement. The vigilance decrement is the measurable decline in the probability of detecting rare signals as time on a monitoring task increases. First demonstrated by N. H. Mackworth in 1948 using the Clock Test. The unexamined assumption in every human-in-the-loop policy is that a person can watch indefinitely. Mackworth showed in 1948 that they cannot. What he did not show is when: the half-hour figure in general circulation is the resolution of his analysis blocks, restated as a property of people by Bainbridge in 1983 and repeated with no citation at all in a peer-reviewed retrospective in 2018. N. H. Mackworth, Quarterly Journal of Experimental Psychology 1, 1948, pp. 6-21 Established. The Mackworth Clock is the originating apparatus. Definition.
Time horizon (METR). The 50%-task-completion time horizon is the length of task, measured by how long a human expert takes to do it, that an AI model completes with about 50 per cent success. The cleanest available answer to how long a job these systems can attempt, and the one most often quoted with its success rate removed. Kwa, T. and 25 others (METR), Measuring AI Ability to Complete Long Software Tasks, NeurIPS 2025; arXiv:2503.14499 Proposed by Kwa and colleagues at METR in March 2025. Not a SuperSkills coinage and no claim of first use is made. Definition.
Types of judgement. Judgement divides into three kinds that behave differently under automation: predictive judgement, estimating what will happen; evaluative judgement, deciding what matters and how much; and moral judgement, deciding what is owed to whom and who bears the cost. Machines do the first. The claim that AI has made judgement scarce is a claim about the other two, and evaluation is the one organisations leave undocumented. The separation of prediction from judgement is set out in the economics of AI literature; the third category and the argument about evaluation are developed in this research Definition.
What stays human. Not a list of tasks. Rahim Hirji argues three things survive because AI cannot be them rather than cannot do them: accountability, because responsibility requires someone answerable for a decision; recognition, because being seen depends on the identity of whoever chose to attend to you; and origination, because a model has no stake in which answer is right. Definition.
The rest, in brief#
Algorithm appreciation. Algorithm appreciation is the tendency to weight algorithmic advice more heavily than the same advice from a person. Domain experts are the exception. The exact mirror of algorithm aversion, which this research already defines. Defining one without the other leaves the picture half drawn. Jennifer Logg, Julia Minson and Don Moore, Organizational Behavior and Human Decision Processes 151, 2019 Their coinage. Definition.
Calibration. The correspondence between the confidence a system states and how often it is correct. Definition.
Calibration training. Structured training in probabilistic reasoning, usually covering comparison classes, base rates, incremental updating and review of resolved forecasts, aimed at bringing a person's stated confidence into line with their actual accuracy. The one component of judgement that has been trained and measured against resolved outcomes, improving accuracy by roughly 6 to 12 per cent after about an hour. Mellers, B. et al., Psychological strategies for winning a geopolitical forecasting tournament, Psychological Science, 25(5), 2014 Definition.
Capability audit. A structured assessment of whether an organisation still possesses the human knowledge, judgement and practical ability its operations depend on, tested by removing assistance rather than by surveying confidence. Definition.
Capability masking. Capability masking is the appearance that organisational capability has been replaced by AI while dependence on skilled human labour actually remains, which supports hiring restraint while the cost accumulates. Independent arrival at the capability-debt argument, which is the most useful kind of corroboration. Wolfgang Rohde, AiSuNe Foundation, SSRN 6577818, 2026 The author's coinage. Definition.
Clinical versus actuarial judgement. The comparison between an expert combining evidence in their head and a fixed rule combining the same evidence mechanically. Across 136 studies, the mechanical rule was better in roughly half the comparisons and equal in roughly half. The oldest quantitative result in the field, and the one most often used to settle arguments it does not settle. The mechanism is consistency rather than knowledge. Grove, W. M. et al., Clinical versus mechanical prediction: a meta-analysis, Psychological Assessment 12(1), 2000 Definition.
Cognitive load. The total demand placed on working memory by a task, conventionally divided into intrinsic load inherent to the material, extraneous load imposed by presentation, and germane load, the effortful processing that builds understanding. Definition.
Collective judgement. Collective judgement is judgement produced by a group. It improves on the individual reliably only where judgements are made independently before being combined; where they are formed in discussion, the group inherits the first framing offered and can perform worse than its average member. The one reliable improvement is structural and requires nobody to think better, so it survives time pressure. A shared generated draft destroys the independence it depends on. Larrick, R. P., Debiasing, in Blackwell Handbook of Judgment and Decision Making, 2004; Chang, W. et al., Restructuring structured analytic techniques in intelligence, 2018 Definition.
Conflit de rationalité. A conflict of rationalities is the unresolved disagreement between what an organisation wants from an AI system and what the work actually requires. Whether a compromise is reached decides whether the result builds capability or removes it. LaborIA, Rapport d'enquête LaborIA Explorer, May 2024 LaborIA's framing.
Decision journal. A written record made at the time of a decision, covering the decision, the reasoning, the expected outcome and the confidence attached to it, so that later review compares against what was actually thought rather than against a reconstruction. The precondition for both calibration scoring and process feedback, neither of which is possible once hindsight has reorganised the memory. Practice associated with Michael Mauboussin and with the decision-making literature on hindsight bias; no single originating publication Definition.
Digital labour. Digital labour has two senses. The older, academic one names the work, much of it unpaid or underpaid, that sustains the digital economy: content moderation, platform microwork, data labelling, and the ordinary use of social media that produces value for a platform. The newer one, from Microsoft's 2025 Work Trend Index, names AI agents purchased on demand to scale workforce capacity. The two point in close to opposite directions. Whichever sense is meant, the same move is being made: putting the word labour next to something that is not a conventional paid employee. The academic sense argues that unpaid activity should be recognised and valued as labour. Microsoft's sense does the opposite: it moves a software purchase into the labour column, which is the move that makes 'expanding the workforce with digital labour' sound like a staffing decision rather than a capital one. Terranova, T. (2000), 'Free Labor: Producing Culture for the Digital Economy', Social Text 18(2); Scholz, T. (ed., 2012), Digital Labor: The Internet as Playground and Factory, Routledge; Fuchs, C. (2014), Digital Labour and Karl Marx, Routledge. Microsoft's 2025 sense: Work Trend Index Annual Report, WorkLab, 23 April 2025. NOT Microsoft's coinage, corrected 24 September 2026. The phrase is an established term in critical media and labour studies, traced here to Tiziana Terranova's 'free labor' thesis (2000), given its current name at Trebor Scholz's 2009 conference and 2012 edited volume, and given its fullest theoretical treatment in Christian Fuchs's Digital Labour and Karl Marx (2014). Microsoft's Work Trend Index 2025 report reuses the exact two words for a different and largely unrelated phenomenon, AI agents sold as purchasable workforce capacity, with no reference to the existing literature. Both senses are named on the page rather than picking one. Definition.
Epistemic debt. Epistemic debt is the gap between being able to produce working output with AI and being able to understand or repair it, producing practitioners whose functional usefulness masks low corrective competence. The repair-competence form of capability debt. Notable because the author reaches it independently. Sankaranarayanan, arXiv:2602.20206, 2026 The author's coinage. Definition.
Komplementäre Arbeitsgestaltung. Complementary work design treats human and machine complementarity as permanent and functional, grounded in structural limits of automation rather than in the current weakness of models. Refuses the assumption that the human role is simply whatever automation has not yet reached. Norbert Huchler, ISF München, Zeitschrift für Arbeitswissenschaft 76(2), 2022 Huchler's framing.
Legitimate peripheral participation. Legitimate peripheral participation is how newcomers acquire competence: by doing real but peripheral work alongside practitioners, moving from the edge of a community of practice towards its centre. Names the mechanism AI has broken. The missing rungs are the peripheral work being removed. Jean Lave and Etienne Wenger, Situated Learning, Cambridge University Press, 1991 Their coinage. Definition.
Levels of automation. A scale for how much of a task a machine performs, from Sheridan and Verplank in 1978, and in the later four-stage version, how much of each stage: information acquisition, information analysis, decision selection and action implementation. Turns how much AI into four separate questions with four separate answers, which is the shape the decision actually has. It contains no rule for choosing the level, so the allocation stays a human judgement. Parasuraman, R., Sheridan, T. B. and Wickens, C. D., A model for types and levels of human interaction with automation, IEEE Transactions on SMC-A 30(3), 2000 Definition.
Misuse, disuse, abuse. Misuse is over-reliance on automation, disuse is unwarranted rejection of it, and abuse is deploying it without regard for the human consequences. Appropriate use is the fourth case. The research argues this vocabulary is better than over-reliance alone, because it names the opposite failure too. Raja Parasuraman and Victor Riley, Human Factors 39(2), 1997, pp. 230-253 Their coinage. Definition.
Noise audit. A noise audit is a controlled exercise in which several professionals independently judge the same real cases, so the variation between them can be measured. The variation is noise: bias moves every judgement in one direction, noise is the scatter. The only instrument in this territory that produces a number before anybody argues about opinions, and the honest baseline to take before a model is put into a judgement process. Kahneman, D., Sibony, O. and Sunstein, C. R., Noise: A Flaw in Human Judgment, Little Brown, 2021 Definition.
Premortem. A meeting held before a plan is executed, in which the team is told the plan has failed and each member writes down why, so that doubts are stated as explanations rather than as objections. A device for licensing dissent that a forward-looking risk review suppresses. The experimental support is for reason generation, not for project outcomes. Klein, G., Performing a project premortem, Harvard Business Review, September 2007 Definition.
Retrieval practice. Retrieval practice is the finding that recalling information from memory strengthens it more durably than studying it again. Definition.
Scenario planning. Scenario planning is a method for developing several internally consistent accounts of how the future could unfold, not ranked by likelihood, in order to change the mental models of the people who will have to decide rather than to predict which account is correct. Produces preparedness, which cannot be scored, so a method in use for sixty years has almost no evidence base. Wack, P., Scenarios: uncharted waters ahead, Harvard Business Review, September 1985 Definition.
Substitution myth. The assumption that new technology can be introduced as a simple substitution of machines for people, preserving the basic system while improving it on some output measures. Definition.
Tacit knowledge. Knowledge that resists full articulation, acquired through experience and practice rather than instruction, and typically transmitted through shared work rather than documentation. Definition.
The decision quality chain. Six requirements that together make a decision sound: an appropriate frame, creative alternatives, relevant and reliable information, clear values and trade-offs, sound reasoning, and commitment to action. The quality of the decision is set by the weakest of the six. It stops an organisation improving the link that is easiest to improve. Most AI investment lands on information and alternatives, which are usually already the strongest two, and leaves frame and values untouched. Spetzler, C., Winter, H. and Meyer, J., Decision Quality, Wiley, 2016 Definition.
The efficiency-gain illusion. The efficiency-gain illusion is the gap between the time and effort people expect AI assistance to save them and the time and effort it actually saves. It is the umbrella the authors put over two separately formalised errors. The SPEEDUP ILLUSION is the time half, where the predicted reduction in completion time exceeds the measured one, given in the paper as Equation 1. The OFFLOADING ILLUSION is the effort half, stated in the same form as Equation 2. Two further labels sit beside them and are not formalised: SELF-ESTIMATE MISCALIBRATION, people believing they use AI less often than they do, and the SLOW-DOWN EFFECT, AI-assisted completion taking longer than unaided completion on the easiest tasks. Yu, S., Cheng, M., Jabbar, A., Sucholutsky, I., Collins, K. M., Jurafsky, D. and Hawkins, R. D., arXiv 2605.22687v1 [cs.CY], 21 May 2026 Yu and colleagues, 21 May 2026. Read in full at source 20 September 2026. No claim of first use by this research: the term is theirs and appears in their title. The companion CogSci paper of 22 May 2026, arXiv 2605.23177, introduces the speedup illusion under its own name. Definition.
The expertise reversal effect. The expertise reversal effect is the finding that instructional support which helps a novice becomes useless or harmful once the learner has expertise. The direct answer to who should use AI assistance and when, which is a question this research is asked constantly. Kalyuga, Ayres, Chandler and Sweller, Educational Psychologist 38(1), 2003, pp. 23-31 Their coinage. Definition.
The Google effect. A shift in what people encode to memory when they believe information will remain externally accessible, favouring the location or retrieval route over the content. Definition.
The illusion of competence. The illusion of competence is the gap between how capable people feel and how capable they are. Definition.
The impostor phenomenon. The impostor phenomenon is an internal experience of intellectual phoniness in people whose record plainly contradicts it, in which accomplishments are attributed to luck, effort or the misjudgement of others rather than to ability. The research's reason for holding it is a correction. The folk account says evidence of competence cures it, and the founding paper says the opposite in its own abstract: achievements 'do not appear to affect the impostor belief', and the authors report that repeated success alone was not sufficient to break the cycle. Anyone reaching for it as the precedent for an AI-era condition should read what it actually described. Clance, P. R. and Imes, S. A., Psychotherapy: Theory, Research & Practice, 15(3), 1978, 241-247 Clance and Imes, 1978. Read in full at source 19 September 2026. The article's own title and running head spell it 'imposter' while the abstract and body use 'impostor'; both spellings are therefore original. Definition.
The vocabulary institutions have coined#
Terms owned by the bodies that publish about this: Microsoft, the World Economic Forum, the OECD, PwC, the consultancies, the European institutions and the frontier labs. Worth knowing who wrote each one, because the definition usually carries the position of whoever wrote it.
Agent boss. An agent boss is Microsoft's term for a human manager of one or more AI agents. It describes a promotion in title and says nothing about whether the person can evaluate what the agents produce. That gap is where oversight readiness fails. Microsoft, Work Trend Index Annual Report 2025 Microsoft's coinage.
Ai literacy. In the European Union, yes. Article 4 of the EU AI Act has applied since 2 February 2025 and requires providers and deployers of AI systems to take measures to ensure, to their best extent, a sufficient level of AI literacy among their staff and other persons dealing with the operation and use of AI... Definition.
Borrowed competence. Borrowed competence is McKinsey's term for capability that appears in the output but disappears when the tool is withdrawn. The clearest external statement of what synthetic seniority produces, from a firm with no stake in the argument. McKinsey, Rethinking talent development in the age of AI, 14 July 2026 McKinsey's coinage.
Distributed de-skilling. Distributed de-skilling is BCG's term for the collective erosion of human skills across an organisation that undermines its intelligence and resilience over time. BCG frames it as a system design problem rather than a talent problem. BCG, When Everyone Uses AI, Companies Risk Losing Critical Skills, 17 June 2026 BCG coins it explicitly: 'We call this distributed de-skilling.' Definition.
Frontier Firm. A Frontier Firm is Microsoft's term for a company built on purchasable machine reasoning, human-agent teams, and a new role for every employee as a manager of agents. Worth knowing because of who wrote it. Microsoft publishes a glossary of this vocabulary for executives, and every definition in it is framed from the position of the buyer of digital labour rather than the people whose capability is at stake. Microsoft, Work Trend Index Annual Report 2025, 23 April 2025 Microsoft's coinage. Definition.
Human in command. The human in command principle holds that a person must retain authority over an AI system, as distinct from merely occupying a position in its process. The cleanest contrast in the whole vocabulary. Human in the loop describes where a person sits. Human in command describes what they are entitled and expected to be able to do, which requires competence they may no longer have. European Economic and Social Committee, OJ C/2025/1185, 21 March 2025 The EESC's own wording, and it should not be credited to the 2020 Autonomous European Social Partners Framework Agreement on Digitalisation, whose own text reads 'the human in control principle' throughout.
Human Reserved. Human Reserved is Bill Gates's term for work deliberately set aside for people only, by analogy with nature reserves: places we could develop but choose not to because the loss would be too great. It is the first serious proposal from a major technology figure that the boundary of automation should be decided rather than discovered. That makes it the counterpart to drift. Bill Gates, The turbulent AI era is here, gatesnotes, 26 August 2026 Gates coins it explicitly: 'I've started calling this domain Human Reserved.' Definition.
Learning by verifying. Learning by verifying is Bain's claim that juniors learn by reviewing, stress-testing and catching errors in AI output, and that the repetitions per hour go up rather than down. The most useful thing in this corpus because it contradicts the missing-rungs argument directly. A defended counter-position is worth more than another source agreeing. Bain & Company, The future of opex in the agent economy Bain's coinage.
Meaningful human involvement. Meaningful human involvement is the UK statutory test for whether a decision is solely automated: whether a human can exercise real influence before the decision is applied, and has the authority, discretion and competence to alter it, as opposed to rubber-stamping. Note the third requirement. Competence is written into the law, which means capability decay is a compliance exposure and not only a management problem. Information Commissioner's Office, Recruitment rewired, 2026; UK GDPR Article 22A Statutory UK term. Definition.
Mis-skilling. Mis-skilling is the acquisition of incorrect reasoning patterns, learned by uncritically adopting AI output that was erroneous or biased. The capability is built rather than lost, and built wrong. The third case the deskilling debate keeps missing. Both sides argue about whether skill is lost or retained, and neither asks what is learned when the teacher is confidently mistaken. Ke, Y. and colleagues, AI-induced never-skilling in medical education, Nature Medicine, 32(6), 22 May 2026 Named by Ke and colleagues alongside never-skilling, 22 May 2026. Not a SuperSkills term.
Never-skilling. Never-skilling is the failure to form foundational competence during training, because AI substituted for the cognitive effort that would have built it. It differs from deskilling in having no earlier capability to return to. The precise name for what synthetic seniority describes from the other side. Deskilling assumes a skill that decayed; never-skilling asks what happens when it is never laid down. Ke, Y. and colleagues, AI-induced never-skilling in medical education, Nature Medicine, 32(6), 22 May 2026 Named by Ke and sixteen colleagues in Nature Medicine, 22 May 2026. Not a SuperSkills term. The authors state that direct evidence for it in clinical trainees does not exist and present it as a risk model.
Oversight readiness. Oversight readiness is Google DeepMind's term for whether a future workforce will be able to judge the AI work it is nominally supervising, given that juniors are being deprived of the experience that builds strategic judgement. Names the problem in the future tense, which is the tense that gets budget. A workforce that can delegate but cannot judge. Tomašev, Franklin and Osindero, Google DeepMind, Intelligent AI Delegation, arXiv:2602.11865, 12 February 2026 DeepMind's framing. Definition.
Seniorised entry-level roles. Seniorised entry-level roles are junior jobs that now demand senior human skills. PwC found entry-level roles most exposed to AI are seven times more likely to require leadership, creativity or face-to-face interaction. The strongest external evidence for synthetic seniority. Entry-level work is not disappearing so much as being asked to arrive already senior. PwC, 2026 Global AI Jobs Barometer, June 2026 PwC's framing, 2026. Definition.
Shallow jobs. Shallow jobs are Bain's term for roles in which people rubber-stamp mostly correct AI output without engaging their judgement. The strongest counter to blanket human-in-the-loop prescriptions: universal review produces the appearance of oversight and the erosion of it. Bain & Company, What Financial Services Leaders Are Wrestling with on AI, 4 June 2026 Bain's coinage. Definition.
The AI employment gap. The AI employment gap is the Stanford Digital Economy Lab's finding that employment of workers aged 22 to 25 in AI-exposed occupations sits 19 per cent below where it would have been had it tracked their less-exposed peers, operating through reduced hiring rather than increased separations. The most-cited number in the entry-level debate, and the mechanism matters: the door is closing, not the jobs ending. Brynjolfsson, Chandar and Chen, Stanford Digital Economy Lab, revised August 2026 Stanford's framing. The authors describe their findings as 'canaries in the coal mine, rather than causal estimates'. Definition.
The experiential chasm. The experiential chasm is the gap between the small group who have spent real hours with frontier models and the majority still experimenting superficially. Bain describes it as neither a seniority gap nor a training gap, and as widening every week. It reframes AI capability as something acquired through hours rather than through instruction, which means training budgets aimed at awareness close none of it. Bain & Company, The Future of Opex in the Agent Economy, 14 May 2026 Bain's coinage.
The oversight paradox. The oversight paradox is the observation that the competence a person needs to oversee an AI system is built and kept alive by doing the work the system now does instead, so the more oversight is delegated to people who no longer practise, the less able they are to provide it. States the research's central mechanism in one sentence and puts it in the vocabulary of regulation: Article 14 of the EU AI Act can require a human overseer without being able to guarantee the overseer's competence survived. The remedy the authors propose, structured practice without the system, is the same one this research calls keeping the repetitions. Isabell Steidel and Benedikt Gieger, The oversight paradox: human control over AI may be eroding, World Economic Forum, 2 July 2026 Steidel and Gieger's phrase, published by the World Economic Forum on 2 July 2026; the authors present it as their own. The mechanism is older (Bainbridge, 1983) and this research had been arguing it before the phrase appeared, but the phrase is theirs and is not claimed here. Definition.
The two-tier future. The two-tier future is the outcome financial services leaders fear most: lower-skilled people handling tasks AI is not set up for, a smaller group of experts training the system and handling exceptions, and no obvious bridge from the first group to the second. This is the missing rungs described independently, by executives, in their own words. The absent bridge is the point, and it is the strongest external corroboration the argument has. Bain & Company, What Financial Services Leaders Are Wrestling with on AI, 4 June 2026 Bain's framing of a concern raised by summit participants. Definition.
The rest, in brief#
AI-off zones. AI-off zones are BCG's term for tasks an organisation deliberately designates as off limits to AI, where originality, ethical judgement or synthesis matter most. The organisational counterpart to Human Reserved, at team rather than economy scale. BCG, When Everyone Uses AI, Companies Risk Losing Critical Skills, 17 June 2026 BCG's coinage.
Amplified oversight. Amplified oversight is Google DeepMind's term for an oversight signal as good as one a human would give if they understood all the reasons behind a decision. Names the problem of judging work you can no longer evaluate, which is the technical statement of the capability-debt endgame. Google DeepMind, An Approach to Technical AGI Safety and Security, April 2025 DeepMind's term.
Capacity gap. The capacity gap is Microsoft's term for the deficit between what a business demands and the maximum output humans alone can supply. It frames human limits as the problem to be solved. The same figures can be read as evidence that the demand is miscalibrated. Microsoft, Work Trend Index Annual Report 2025 Microsoft's coinage.
Complementary skills. Complementary skills are the OECD's term for teamwork, autonomy, problem solving, creative thinking, communication, collaboration and emotional intelligence: the capabilities that enable high-performance work and the ability to keep learning. The most carefully defined of the three names for one idea. Where three institutions have three names for one idea, the definition is worth owning. OECD, Skills in the AI age, OECD Artificial Intelligence Papers No. 60, July 2026 OECD's framing.
Core skills. Core skills are the World Economic Forum's term for the skills employers consider central to a role, and the unit in which it reports skill change: 39 per cent of workers' core skills are expected to change by 2030. The most likely WEF term to reach a board paper with a number attached. World Economic Forum, Future of Jobs Report 2025 WEF's framing. Definition.
Curriculum-aware task routing. Curriculum-aware task routing is Google DeepMind's proposal for systems that track a junior's skill progression and deliberately allocate tasks at the edge of their expanding competence, including work the system would otherwise have done itself. A designed answer to the missing rungs, from a frontier lab. It concedes that keeping people skilled requires deliberately not automating some work. Tomašev, Franklin and Osindero, Google DeepMind, arXiv:2602.11865, 12 February 2026 DeepMind's coinage.
Human-agent ratio. The human-agent ratio is Microsoft's proposed business metric for the balance between human oversight and agent efficiency on a mixed team. A ratio implies oversight is a quantity. Vigilance research says it is a capacity that degrades with time on task, so the number tells you about headcount and not about whether anyone can still catch an error. Microsoft, Work Trend Index Annual Report 2025 Microsoft's coinage.
Hybrid intelligence. Hybrid intelligence is the European Policy Centre's proposed basis for skills policy: technical AI literacy combined with domain expertise and distinctively human capabilities, rather than AI skills alone. The emerging EU replacement for AI skills as a policy category, and explicitly a blend rather than an addition. European Policy Centre, Fostering AI resilience in the EU labour market, 19 March 2026 European Policy Centre framing, attributed to Kuiper and Świeboda.
Identity commoditisation. Identity commoditisation is the erosion of a professional's sense of uniqueness and dignity as their role narrows to supervising a system, until the work that carried their identity is no longer the work they do. Names the cost that skill measures cannot see. A practitioner can be as accurate as ever and still describe themselves as a bystander in their own practice, and nothing on a dashboard registers that. Ehsan, U. and colleagues, From Future of Work to Future of Workers, CHI '26, ACM Named by Ehsan and colleagues from twelve months of fieldwork in radiation oncology, published at CHI 2026. Their companion term is intuition rust, the dulling of expert judgement beneath output that still looks intact. Not SuperSkills terms.
Intelligent choice architectures. MIT Sloan Management Review's term for systems that shape which options a person sees and how they are framed, on the argument that AI changes decision rights by changing the choices presented rather than by making the decision. The one line of business writing in this territory asking what happens to decision rights rather than to individual capability, which makes it the closest neighbour to the argument on this site. MIT Sloan Management Review, Winning with intelligent choice architectures MIT Sloan Management Review Definition.
Judgement work. The World Economic Forum's name for work whose value lies in deciding rather than producing, used in its 2026 jobs and skills reporting as a category said to be rising as production is automated. Useful as a label for what a labour market is shifting towards. It is a categorisation rather than a measurement, and no published method distinguishes judgement work from other work in employment data. World Economic Forum, The rise of judgement work in the age of AI, 22 May 2026 World Economic Forum Definition.
Learning-conducive work environments. Learning-conducive work environments are workplaces designed so that continuous learning and the use of skills happen through the work itself rather than through separate training. Locates capability building in job design rather than in a course, which is where the evidence says it actually happens. Cedefop, Shaping learning and skills for Europe, publication 9208, 2026 Cedefop's framing. German counterpart: lern- und erfahrungsförderliche Arbeitsbedingungen.
Obligation to justify. The obligation to justify is Cedefop's proposal that employers should have to give reasons for introducing AI into a workplace. Turns drift into a decision that has to be defended, which is the procedural form of designing rather than drifting. Cedefop, publication 9201, 2025 Cedefop's proposal.
Robot relations. Robot relations is Darrell West's proposed organisational function, alongside human resources, for the way people in an organisation interact with AI assistants, agents, robots and chatbots, including their complaints that an algorithm's decision was unfair or biased. Names where the complaint about a machine's decision goes, and by doing so exposes what it does not name: who can overrule the decision. A complaints desk for algorithmic management is not the same as override authority, and an organisation that builds the first without the second has processed the grievance and kept the handover. Darrell M. West, Organizations will need AI and robot relations departments, Brookings Institution, 30 July 2026 West's phrase, Brookings, 30 July 2026, in the form 'supplement HR with robot relations (RR)'. CNBC carried it into the business press on 20 September 2026. Not this research's. Definition.
The answer-key model. The answer-key model is McKinsey's proposed training pattern in which the employee attempts the work first, the AI grades the attempt, and a manager reviews it with them. The cleanest named remedy for the broken apprenticeship available anywhere, and it inverts the usual order: attempt then check, rather than generate then edit. McKinsey, Rethinking talent development in the age of AI, 14 July 2026 McKinsey's coinage.
The six elements of judgment. Andrew Likierman's account of judgement as a personal capacity with six components: knowledge, context, trust, feelings, choice and delivery, developed at London Business School. The most developed treatment of judgement as an individual skill, and the clearest statement of the assumption this research disputes: that the unit of analysis is the person rather than the allocation of decisions. Likierman, A., London Business School, Why human judgement is essential in the age of AI, 28 November 2025 Andrew Likierman Definition.
Work Chart. The Work Chart is Microsoft's proposed successor to the org chart, structured around jobs that need doing rather than functional expertise. Functional expertise is also how apprenticeship is organised. Removing it as the organising principle removes the ladder with it. Microsoft, Work Trend Index Annual Report 2025 Microsoft's coinage.
The AI vocabulary, read for capability#
The words that appear in the press and in board papers. Most glossaries stop at what they mean. Each entry here also says what the term implies for human capability, judgement and skill. Technical machine-learning vocabulary is deliberately excluded.
AI control as work. AI control is Narayanan and Kapoor's prediction that a steadily greater share of what people do in their jobs will consist of controlling AI rather than doing the underlying task. If most work becomes control, then the capacity to control well is the whole of human capability, and nothing currently builds it. Narayanan and Kapoor, AI as Normal Technology, Knight First Amendment Institute Existing safety term, repurposed as a labour category.
AI slop. AI slop is fast, plausible output that does not meet the standard. The word matters because it names a quality failure that reads as competence. Bain's finding is that catching it is a leadership discipline rather than a tooling problem: what gets measured, what gets rewarded, and what gets rejected. Bain & Company, What Financial Services Leaders Are Wrestling with on AI, 4 June 2026 In general circulation. Bain records it as the term that kept coming up among financial services executives.
Learn, unlearn, relearn. A widely repeated claim that the illiterate of the twenty-first century will be those who cannot learn, unlearn and relearn, universally attributed to Alvin Toffler's Future Shock. Toffler did not write it. Future Shock paraphrases the psychologist Herbert Gerjuoy, whose actual words were that tomorrow's illiterate will be the person who has not learned how to learn. The three-verb version is a later compression by an unknown hand. Alvin Toffler, Future Shock, 1970, paraphrasing Herbert Gerjuoy from an interview with the author Misattributed. Gerjuoy is the source of the idea; the famous phrasing has no identified author.
The jagged technological frontier. The irregular boundary between tasks an AI system performs well and tasks it performs badly, where the two can be almost indistinguishable in apparent difficulty and the system gives no signal of having crossed from one to the other. Definition.
Tokenomics. Tokenomics is the economics of token cost, and specifically its fall. As the price per token drops, tasks not worth handing to a machine become worth handing over, so the boundary of what is automated moves without anyone deciding to move it. The engine underneath drift. Almost nobody decides to offload a task. It becomes cheap, the reason not to disappears, and the decision is taken by default and noticed later. Not a single source. The usage here is the AI cost sense, not the cryptocurrency sense. Contested. In cryptocurrency the term means the design of a token economy. The AI usage is loose and recent. Definition.
The rest, in brief#
AI hallucination. Generated content presented as factual that is not supported by the model's training data, the provided context or reality. Definition.
Anthropological regression. Anthropological regression is the paradox in which material progress coincides with human and cultural impoverishment, through forced inactivity, absent responsibility and the loss of daily tasks and stimuli. The moral vocabulary for what capability debt costs a person rather than an organisation. Leo XIV, Magnifica Humanitas, §154, 15 May 2026 Coined in this formulation.
Centaur and cyborg work. Centaur work divides tasks cleanly between person and machine. Cyborg work interweaves them continuously, moving back and forth across the jagged frontier. A vocabulary for how someone works with AI rather than whether they do, which is the distinction most adoption metrics miss. Ethan Mollick, Centaurs and Cyborgs on the Jagged Frontier Adapted from advanced chess. The work-mode pairing is Mollick's. Definition.
Falling asleep at the wheel. Falling asleep at the wheel is what happens when people given high-quality AI become careless and less skilled in their own judgement, because the AI is good. Quality of the tool is not protective. The better the assistance, the faster the attention goes. Fabrizio Dell'Acqua, credited by Mollick Dell'Acqua's, credited explicitly by Mollick.
Research taste. Research taste is knowing what to study next, which experiment to run, and sensing where a new approach might lie. It has proved hard to train because the feedback loops are long and the data is thin. Presented in AI 2027 as a residual human skill. Notable because the reason it resists automation, long feedback loops, is the same reason it resists teaching. AI 2027, Kokotajlo et al. Established in machine-learning culture, explicitly defined there.
Shift left. Shift left means moving decisions closer to their source, removing the dilution that every handoff introduces. Handoffs are also where work becomes visible to other people. Removing coordination removes the moments when a colleague saw the work and could question it. Bain & Company, The Future of Opex in the Agent Economy, 14 May 2026 Predates Bain, from software testing. Used here for organisational decision-making.
The capability-reliability gap. The capability-reliability gap is the distance between what a model can do once and what it can do dependably. It is the main barrier to agents automating real work. Explains why the demonstration always looks better than the deployment, and why human checking keeps being reinvented as necessary. Narayanan and Kapoor, AI as Normal Technology, Knight First Amendment Institute Effectively named there, in scare quotes, unattributed to anyone else.
The God Prompt. The God Prompt is a two-part prompt that circulated on social video in late 2024, instructing a model to role-play at many times its own ability and to name the fear the user never admits, then to convert the answer into patterns to stop and adopt. Nothing it produces is evidence about the person. It works by reflecting back a reading assembled from whatever the model holds about a user, in a register that invites the reading to be believed, which makes it a clean demonstration of why fluency is not knowledge. Hirji's own verdict on it is the useful part: he found it specific enough to unsettle him and generic enough to fit anyone, and called it robot astrology. Rahim Hirji, The 'God Prompt', Box of Amazing, 1 December 2024 NOT A SUPERSKILLS COINAGE, and it must never be attributed to Hirji. Read at source 19 September 2026: the essay opens by reporting that he received a video about it three times in one week, names the prompt as one 'some are calling the Goat Prompt', and reproduces it for readers to try. He is describing a thing already in circulation. Its author is untraced.
The new imposter syndrome. The new imposter syndrome is doubt about the origin of your own thinking after working closely with a model: the work stands, it carries your name, and you cannot reconstruct which part of it you did. The doubt is about authorship rather than about desert, so producing more good work does not settle it. It separates cleanly from the 1978 condition once you ask what the doubt is about. Clance and Imes describe somebody who has done the work and cannot believe it; this describes somebody who cannot establish what they did. The practical consequence is that reassurance is the wrong response, and a record made before the first prompt is the only thing that answers it. Rahim Hirji, The New Imposter Syndrome, Box of Amazing, 17 May 2026 NOT CLAIMED. The phrase was in published use before this research reached it: John Nosta, 'AI and the New Impostor Syndrome', Psychology Today, 13 March 2025, read at source 19 September 2026, fourteen months before Hirji's essay of 17 May 2026. Nosta's condition is also a different one, about deserving success that arrived too easily, which is the 1978 question transplanted. The wording is common enough that earlier uses are likely and no priority is asserted. WHAT IS HIRJI'S, and carries a dated first publication of 17 May 2026, is the separation of the experience into three forms: Phantom Authorship, Velocity Vertigo and the Hollowing. Definition.
Reference · SS-2026-168
Hirji, R. (2026). The SuperSkills Glossary. The SuperSkills evidence base, SS-2026-168. https://thesuperskills.com/research/ai-glossary. Last reviewed 4 September 2026.
An evidence review by Rahim Hirji, not peer-reviewed research. For a material claim, cite the underlying study as well; every study here carries its own permanent link.
How citations and IDs work