PUBLIC LEXICON
The vocabulary of work nobody counts
AI entered your workflows faster than the words to describe what it changed. This lexicon gathers those words, with the source of each one, its date, and what that source is actually worth. It is free, and it stays free.
Why the status matters
The three most quoted figures in this market are not peer-reviewed, and one is offered as the explanation of another in a third source that has not been empirically tested either. Reading them without their method is how a decision gets made on a number that does not hold. Every entry here carries its standing.
Peer-reviewed foundationConvergent sourcesRecently formalisedMethod contestedNot yet formalised
The phenomenon
7 entriesWhat happens when AI enters a workflow nobody mapped. None of these terms were coined here.
Measured adoptionadoption mesuréeThe European statistical office counts AI use by enterprise size. In 2025: 17% of small enterprises, 30.4% of medium ones and 55% of large ones, for an EU average of 19.95%, u…From executing to instructingde l'exécution à l'instructionIn a field experiment on 2,310 participants, people working with an AI agent made 71% fewer direct edits and sent 63% more messages than those working with another human.Invisible worktravail invisibleWork that is done, that the organisation depends on, and that appears in no count. The founding result is counter-intuitive: invisibility is not a property of the work itself,…Articulation worktravail d'articulationThe unseen labour of coordinating tasks, people and breakdowns: what people do to link tasks that are otherwise separate. It appears in no job description because it is the co…Shadow AIIA de l'ombreAI use that runs inside an organisation without having been declared, mapped or contracted. The founding observation is that employees keep using personal AI tools even when t…Agent sprawlprolifération d'agentsAgents created, deployed or connected across systems faster than the organisation can inventory them, assign ownership, control their permissions, monitor their behaviour and…Silent scope creepdérive silencieuse de périmètreAn agent quietly starts handling edge cases it was never approved for, and nobody tracks the drift. The machine-side counterpart of human role drift.
The costs
10 entriesThe family of terms that formed around one idea: AI produces faster than it validates, and someone pays the difference.
Role debtdette de rôleThe accumulated obligation to redefine roles that drifted faster than the contracts describing them. A person's work moves toward AI engineering without anyone having decided…AI debtdette d'IAVerbatim definition: the accumulated obligation to remediate AI-generated artefacts that were produced faster than they were validated. Like technical debt, it has a principal…Verification burdencharge de vérificationThe human hours spent chasing AI errors that nobody counts: the cumulative effort of reviewing, validating, repairing and integrating AI-generated artefacts, inspection time a…Rework externalityexternalité de repriseThe cost is displaced from the immediate user onto downstream recipients: reviewers, managers, clients, colleagues, administrative staff. The person who gains the time is not…Asymmetric adoptionadoption asymétriqueBenefits accrue to the people using AI while the correction work falls on others in the workflow. The academic name for what is usually described, loosely, as uneven and undoc…WorkslopworkslopAI-generated content that looks like good work but lacks the substance to move the task forward. Reported figures: 41% of workers received some in the preceding month, at an a…Knowledge decaydégradation de la connaissanceWhen errors compound across teams and departments, the organisation's collective knowledge base degrades. The debate moves away from individual execution speed toward whether…Prompt debtdette d'instructionThe accumulation of unmanaged, inconsistent AI instructions that produces risk, rework and non-auditable outputs: undocumented tweaks, stacked quick fixes, no version control,…Behavioural lock-inverrouillage comportementalThe switching cost accumulated when a persistent agent learns how your organisation communicates, decides and operates, and that learning is not exportable in any meaningful w…Portable versus captive skillcompétence portable et compétence captiveThe literature settles one question and opens another. Prompt and context engineering skills are portable across tool boundaries; specific tool portfolios are not. There is th…
Breakpoints
5 entriesWhere chains break, and in what proportion. This is the part most often assumed rather than measured.
ReconstructionreconstructionEstablishing what actually happens in a workflow from sources that are not people's recollection, then setting that against what is declared. The object produced is not the li…Hand-offpoint de passageThe moment a task, a context or a responsibility passes from one holder to the next: person to tool, tool to person, person to person, tool to tool, and the chains that combin…Failure taxonomy at hand-offtaxonomie de défaillance au point de passageFour error types account for hand-off failures: signal corruption (36.8%), data gap (29.1%), referential drift (27.3%), capability gap (6.8%).Structural incompatibilityincompatibilité structurelleTwo tools adopted independently by two departments, each locally coherent, that will produce friction when their data meet at a higher level of the organisation. The incompati…First-pass acceptance ratetaux d'acceptation au premier passageHow often an AI output is usable without correction. Paired with its companion measure, rework minutes, meaning the human time elapsed between the first « it's done » and actu…
The framework
5 entriesWhat the regulation actually requires, and the precise point where the market gets it wrong.
Human in the Mapl'humain dans la carteThe established models describe where a person sits inside a decision: in the loop when the system waits for their approval, on the loop when they watch and may intervene, in…AI literacymaîtrise de l'IA, article 4Every deployer, not only those running high-risk systems, must ensure a sufficient level of AI literacy among staff and anyone operating AI on their behalf, calibrated to each…Human oversight and authoritycontrôle humain, article 26(2)The deployer assigns human oversight to natural persons who have the necessary competence, training and authority, as well as the necessary support. In practice: identifying s…What Article 99 actually coverssanctions, article 99Dozens of sites announce €15M or 3% of worldwide turnover for a breach of the AI literacy obligation. Reading Article 99 in full: Article 4 is named nowhere in it. The provisi…Use with no contractprocessor without an agreement, sous-traitant sans contratA tool brought in without going through the organisation may fall under article 28, which requires a contract with any processor acting on your behalf. Where it does, the orga…
The survey these terms describe How the analysis works →