Glossary

This glossary collects the vocabulary that AI Boss uses: the terms of algorithmic management and the AI boss, the five management functions, the degrees of delegation and usage modes, the artefacts an organisation keeps under version control to make management by or with AI accountable, the terms of oversight and contestability, the vocabulary of labour relations and co-determination, the European instruments that constrain the AI boss, and the management theory the book draws on. Terms used in the earlier volumes of the series keep the meaning given there; where this book relies on such a term, the entry names the volume and chapter. Systems are described by what they do (they rank, assign, flag, score, recommend or draft), not by intentions. Statements about law are attributed to the instrument and article and are organisational and architectural implications, not legal advice.

A

ADR (Architecture Decision Record): A short, numbered, immutable record of a significant architectural decision, its context and its consequences. The Management Decision Record and the AI Decision Record keep its discipline of one decision per record and supersession rather than editing. Architecture as Code (Chapter 4).

Agent: A system in which a model directs its own sequence of tool calls towards a goal, acting with an identity and permissions of its own. In this book agents are directed by people under a mandate manifest and may never hold M3 Consequence over a person. AI Platform and Security, Chapter 11; Chapter 16.

"Agent boss": Microsoft's term, in its Work Trend Index, for an employee who directs AI agents. The book writes it in quotation marks and does not adopt it as its own term; it separates the vendor's claim from what is documented about people who manage agents. Chapter 16.

AI Act: Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence. For work it classifies AI used in employment and workers' management as high-risk (Annex III, point 4), prohibits emotion recognition in the workplace (Article 5(1)(f)) and places duties on the employer as deployer (Articles 26 and 27). Chapter 21.

AI boss: Any AI system or agent that performs part of a management function for or over a person. It acts for a person when it assists a manager who remains the one acting, and over a person when its output reaches the worker directly or takes effect without a manager acting on it. Written in lower case in prose. Chapter 4.

AI Decision Record (AIDR): An Architecture Decision Record with the fields a decision about AI needs, such as usage mode, degree of delegation, evaluation evidence and the obligations it touches. Adopting or changing a delegation profile is itself an AIDR. This book uses AIDR-0600 to AIDR-0649 for Nordvik Region and AIDR-0650 to AIDR-0699 for Hansa Mutual. AI Platform and Security, Chapter 5; Chapter 1.

Algorithmic management: The practice of performing management functions wholly or partly by automated means, whether the software uses AI or fixed rules. The term entered the research literature through studies of ride-hailing drivers and now covers platforms, warehouses, hiring and office work. Chapter 4.

Annex III, point 4: The AI Act's high-risk category for employment, workers' management and access to self-employment. Point 4(a) covers recruitment and selection; point 4(b) covers decisions affecting terms of work-related relationships, promotion or termination, task allocation based on individual behaviour or personal traits, and monitoring and evaluation of performance and behaviour. Chapters 7 and 21.

Anti-pattern checklist (AP): The twelve recurring failures of the AI boss, AP-01 The delegation nobody recorded to AP-12 The register that is never reviewed, each with its symptom, why it looks like success, the functions it concerns and its remedy. It is kept as code and used to review an organisation's delegation register. Chapter 31.

Appeal path: The route by which a worker contests a management decision, with five parts: receipt, a first look by the decider (who may only grant), independent review, a written answer and escalation. Each part has an owner and a time limit, and the whole path must work without a lawyer. Chapter 27.

Article 5(1)(f): The AI Act's prohibition on placing on the market, putting into service or using AI systems to infer the emotions of a natural person in the workplace and in education institutions, except for medical or safety reasons. It has applied since 2 February 2025. Chapter 21.

Article 26(7): The AI Act's duty on a deployer that is an employer to inform workers' representatives and the affected workers, before putting a high-risk AI system into service at the workplace, that they will be subject to it. The worker information notice is drafted against it. Chapters 21 and 28.

Article 86: The AI Act's right of a person affected by a decision based on the output of an Annex III high-risk system, with legal or similarly significant effects, to obtain from the deployer clear and meaningful explanations of the system's role and the main elements of the decision. How its date of application interacts with the deferred classification rules is labelled speculation. Chapters 21 and 27.

Automation bias: The tendency to accept the output of an automated system without sufficient scrutiny, producing errors of omission (missing what it did not flag) and of commission (following it when it is wrong). Research found that training is necessary but does not prevent it by itself, so oversight must be designed against it. Chapter 26.

B

Betriebsrat: The works council of an establishment in Germany, elected by the employees under the Works Constitution Act (BetrVG). Above it may sit a Gesamtbetriebsrat for a company and a Konzernbetriebsrat for a group. Chapter 23.

BetrVG (Betriebsverfassungsgesetz): The German Works Constitution Act. BetrVG §87(1) no. 6 gives the works council co-determination over the introduction and use of technical equipment suited to monitor employees' behaviour or performance, and §§ 94 and 95 over personnel questionnaires and selection guidelines, which is why almost every AI boss in Germany needs a works agreement (Betriebsvereinbarung), an agreement between works council and employer. Chapter 23.

C

Campbell's law: Donald Campbell's statement that the more a quantitative social indicator is used for social decision-making, the more it is subject to corruption pressures and the more it distorts the processes it was meant to monitor. Read with Goodhart's law, it explains why per-person figures computed by an AI boss drift from aid to target. Chapters 3 and 15.

Chain record (CHN): The versioned record of an agent chain (CHN-*), listing every link, its mandate and owner, which output each link consumes, where a person checks the output, the stop path and its holders, the maximum depth of sub-delegation and the rule for using the figures each link produces. Its owner is the chain owner. Chapter 17.

Change classes: Three classes, agreed with the representatives in advance, that decide what a change to a management system requires. Class 1 (new system, function, unit or higher degree) needs a new notice and information, consultation or negotiation before use; Class 2 (changed input, objective, threshold or recipient) needs an updated notice; Class 3 (retraining within tested bounds, defect fix) needs only a change-log entry. Chapters 23 and 28.

Charter, Article 31: The provision of the Charter of Fundamental Rights of the European Union that gives every worker the right to working conditions which respect their health, safety and dignity, and to limitation of maximum working hours, rest periods and paid leave. It is the fundamental-rights anchor of M5 Care. Chapters 14 and 24.

Collective agreement: An agreement between a union and an employer or employers' association on pay and conditions. In the Nordic model collective agreements fill the frame that law sets and often regulate in more detail than any statute; GDPR Article 88 allows them to set specific rules for employee data. Chapter 23.

Co-determination: The right of workers' representatives to take part in the employer's decisions, ranging from information and consultation through a duty to negotiate to consent rights. It is a European constant with national forms, and the delegation profile is in practice a negotiated document. Chapter 23.

Competence record (CMP): The versioned evidence of one manager's competence (CMP-*), recording judgement exercised against systems' proposals and assessed against a competence framework such as Nordvik's NR-MGR-CF. It is reviewed half-yearly and before a delegation that requires the competence. The Transformed Security Architect, Chapter 21; Chapter 19.

Constructive debate: The method of Part III, taken from AI Threat: the case for concern, the case for calm, where the positions meet and what can be done, followed by a verdict. Each position is stated at its strongest, what a source reports is kept apart from interpretation, and effects on people are described by what studies measured. Chapters 11 to 15.

Contestation record (CNT): The record of a worker's challenge to a management decision and its outcome (CNT-NR-*, CNT-HM-*), owned by a reviewer outside the decider's line. It cites the contested Management Decision Record and the correction, and contests in aggregate are reported to the representatives. Chapter 27.

Correction: What follows a successful contest, at up to four levels: the decision in the case, the data about the person, the rule or model that produced the proposal, and the delegation register. A corrected decision receives a new Management Decision Record that supersedes the old one. Chapter 27.

Counterfactual: The part of an explanation that tells a person what would have had to be different for the outcome to change. In a management decision it names an input the person or the organisation could actually change, never a protected characteristic. Chapter 27.

D

Decision-rights file (DRF): The versioned file management/decision-rights.yaml (DRF-NR-01, DRF-HM-01) that assigns each class of management decision to a decider, sets the ceiling of delegation per function, names the stop and pause rights and records the rights that law and agreement fix. It says what may be delegated and by whom; the delegation register says what has been. Chapter 20.

Decisive in practice: The state of an output that is advisory on paper but in fact determines the decision, shown when decisions match the proposal almost always and are taken in seconds. The observed degree of delegation is then higher than the recorded one. Chapter 8 sets out the practice test; Chapter 26 applies it in the oversight record.

Degrees of delegation: The series' five autonomy levels, Level 0 Advisory, Level 1 Assisted, Level 2 Supervised, Level 3 Bounded and Level 4 Autonomous, applied per management function and action class. The line between Level 2 and Level 3, where per-decision human approval stops, is the most consequential choice. Reversibility is judged from the position of the person affected. AI Red Team, Chapter 6; Chapter 4.

Delegation ceiling: The highest degree of delegation a decision-rights file allows for a function or decision class. The book's defaults are Level 2 Supervised for M3 Consequence and Level 1 Assisted for adverse consequence decisions; lowering a degree may be done at once, raising a ceiling is a decision for the management body. Chapters 4 and 20.

Delegation freeze: The hold on further delegation in a function and team after a reversibility test fails or passes only with conditions: no new action class, no higher degree, no extension to a new unit or entity. A freeze is not a stop; the system keeps running at the degree already agreed. AI Threat, Chapter 26; Chapter 25.

Delegation profile (DLG): A short, versioned statement, for one system or team, of the degree of delegation per management function and action class, the named decider, the representatives' agreement, the stop path, the records the system must produce and the next review (DLG-NR-*, DLG-HM-*). It is the management form of the mandate manifest. Chapters 4 and 25.

Delegation register of management functions: The file management/delegation-register.yaml that holds all delegation profiles and shows the sum of delegation per function and team, the reversibility tests, the oversight-quality metrics and the representatives' position, including features discovered rather than decided. It extends the delegation register of AI Threat, Chapter 26, and is what the management body receives. Chapter 25.

Demand-control model: Robert Karasek's model in which mental strain arises from the interaction of a job's psychological demands and the worker's decision latitude. Algorithmic direction that raises pace while removing discretion produces the high-strain combination the model predicts. Chapters 11 and 14.

Deployer: Under the AI Act, a natural or legal person that uses an AI system under its authority, except in a personal non-professional activity (Article 3). An employer that uses an AI boss is its deployer, with the duties of Article 26; the supplier that places the system on the market under its own name is its provider. Chapter 21.

Deviation rate: The share of proposals from which overseers depart, recorded with reasons and broken down for the groups concerned. A rate near zero signals rubber-stamping; a high rate signals an unfit system. It is one of the six conditions of real oversight. Chapter 26.

Digital Omnibus: Regulation (EU) 2026/1744, in force since 27 July 2026, which amended the AI Act and deferred the high-risk requirements, including Articles 26 and 27, to 2 December 2027 for Annex III systems. The prohibition of Article 5(1)(f), AI literacy, national information and consultation duties and the GDPR apply now. Chapter 21.

Disparate-impact test (DIT): A test, run with the representatives and the data protection officer, of whether a management system's outputs put a group at a particular disadvantage (DIT-NR-ROS-01, DIT-HM-CTA-01). It is run before go-live, on a new model version and periodically, and its results are evidence under the burden of proof in equality law. Chapter 12.

DPIA (data protection impact assessment): The assessment required by GDPR Article 35 for processing likely to result in a high risk to people's rights. For an AI boss it records human involvement and cross-refers to the fundamental rights impact assessment. Chapter 22.

E

Equality directives: Directive 2000/43/EC (racial or ethnic origin), Directive 2000/78/EC (religion or belief, disability, age and sexual orientation) and Directive 2006/54/EC (sex), which prohibit direct and indirect discrimination in employment. Indirect discrimination through apparently neutral criteria is the main risk of an AI boss. Chapters 12 and 24.

European Works Council: A body for information and consultation on transnational questions in a group operating in several EU and EEA states, established under Directive 2009/38/EC. Hansa Mutual's European Works Council covers its EU and EEA entities. Chapter 23.

Evaluation plan (EVP): The plan (EVP-NR-ROS-2027-01, EVP-HM-AB-2027-01) for evaluating an AI system that performs management functions over years rather than quarters, with outcomes for performance, quality, health, fairness and skill and a comparison group. It is written before a rollout wave and updated at each evaluation point. Chapter 15.

Explanation: What a worker is entitled to learn about a management decision, in four elements: the role of the system and its logic, the inputs about the person, an actionable counterfactual, and the human decider. It is answered from the Management Decision Record in the person's own terms. Chapter 27.

F

Feature register (FTR): The register management/office-feature-register.yaml (FTR-*) of features in office assistants and analytics that could perform a management function, with their status. When a vendor release adds or enables a feature, the entry is updated and the change classified. Chapter 10.

Five management functions: The book's division of management, each with its own decisions, law, failure and way of being delegated. M1 Direction is assigning, sequencing, pacing and instructing work; M2 Evaluation is measuring, rating and judging performance and conduct; M3 Consequence is hiring, pay, promotion, discipline, dismissal, deactivation and access to work, and begins wherever a person's legal or economic position changes; M4 Development is feedback, learning, coaching and career; M5 Care is safety, health, wellbeing, voice and the duty of care, and overlaps all the others. Every chapter in Parts II to VI names the functions it concerns. Chapter 2.

Framework Directive: Council Directive 89/391/EEC on safety and health at work, which places a general duty on the employer to ensure workers' safety and health in every aspect related to work, to assess risks and to consult workers. The employer cannot discharge the duty to a supplier or a system. Chapters 14 and 24.

FRIA (fundamental rights impact assessment): The assessment required by AI Act Article 27 before first use of an Annex III high-risk system by bodies governed by public law, private entities providing public services and certain deployers in credit and insurance. Nordvik Region's assessment for its rostering system is FRIA-2026-007. Chapters 21 and 22.

Function map (FMAP): The file management/function-map.yaml (FMAP-*) that maps, for one team, which management decisions are taken, by whom and with which systems, function by function. It is the starting point for delegation profiles. Chapter 2.

G

GDPR Article 22: The data subject's right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects or similarly significantly affects them. Where an exception applies, Article 22(3) requires at least human intervention, the right to express a point of view and the right to contest. Chapter 22.

GDPR Article 88: The provision that lets member states, by law or by collective agreement, set more specific rules for processing employee data, with suitable and specific measures to safeguard dignity, legitimate interests and fundamental rights, particularly for monitoring systems at the workplace. Chapters 13 and 22.

Goodhart's law: The principle, usually given in Marilyn Strathern's wording, that when a measure becomes a target it ceases to be a good measure. Read with Campbell's law, it is the book's explanation of gaming and of quality crowded out by metrics. Chapter 15.

H

Hansa Mutual: The fictional mutual insurance group headquartered in Hamburg, with about 9,500 employees in several European countries inside and outside the EU, that runs through every chapter. Its AI boss directs office work, including a claims-triage agent that allocates claims to handlers, under a Konzernbetriebsrat and six other national arrangements. Chapter 1.

High-risk AI system: An AI system classified as high-risk under Article 6 of the AI Act, through Annex I or Annex III, and subject to the requirements of Chapter III of the Act. AI used in employment and workers' management is high-risk under Annex III, point 4. Chapter 21.

Horizon register (HZN): A versioned record of trends, each labelled observable, expected or speculation, with the decision it affects and a review date. Hansa Mutual's HZN-HM-MGR-01 keeps one for the role of the transformed manager. AI Governance, Chapter 22; Chapter 29.

Human in the chair: The book's name for real human oversight of management decisions, as opposed to a human in the loop who approves without influencing the outcome. Oversight of the system, which can be sampled, is added to the decision in the case about one person, which cannot; for M3 Consequence a named person decides every case with time, authority, independent information and a stop path. Chapter 26.

I

Independent reviewer: The person who reviews a contested decision and is neither the decider nor the system, normally outside the decider's line. The reviewer owns the contestation record and may correct the decision and refer the rule or model for correction. Chapter 27.

Information and consultation: The floor of worker participation set by Directive 2002/14/EC, which requires information and consultation on decisions likely to lead to substantial changes in work organisation or contractual relations. National law and collective agreements build stronger rights on it. Chapter 23.

Ingrid Halvorsen, Mikael Strand, Jana Petersen and Thomas Brandt: The four fictional people whose year Chapter 30 follows. Ingrid Halvorsen is Nordvik Region's HR director and owns its commission AIDR-0600; Mikael Strand manages the Strandvik home-care unit of about 35 staff; Jana Petersen leads about 20 claims handlers in Hamburg and, from January 2027, the agents of Hansa Mutual's pilot; Thomas Brandt chairs Hansa's Konzernbetriebsrat and sits on its European Works Council. Chapters 1 and 30.

Ironies of automation: Lisanne Bainbridge's observation that automation leaves people the hardest tasks while removing the routine practice that kept them able to do them, and asks them to monitor a system they no longer operate. For management it applies to junior management work and to oversight. Chapters 15 and 19.

J

Job demands-resources model: The model of Arnold Bakker and Evangelia Demerouti in which any feature of work that requires sustained effort counts as a demand and any feature that helps achieve goals or reduce demands counts as a resource. It is used to assess how an AI boss changes the balance of strain. Chapters 11 and 14.

K

Konzernbetriebsrat: The group works council in Germany, which faces systems used across a group. Hansa Mutual's Konzernbetriebsrat negotiated the group works agreement on IT systems KBV-IT-2024. Chapter 23.

L

Level 4 rule: The book's form of the series' rule that Level 4 Autonomous is never used for irreversible actions: Level 4 Autonomous is never used for M3 Consequence, and every consequence decision about a person has a named human decider who can be identified, contested and held to account. AI Red Team, Chapter 6; Chapter 4.

M

Management Decision Record (MDR): A short, numbered, immutable record of one management decision about one or more people that an AI system made or shaped, holding the decision class, the person by pseudonymous reference, inputs, the system's proposal and version, the degree of delegation, the named decider, reasons, information given and the route to contest. Nordvik Region numbers its records MDR-NR-0001 onwards and Hansa Mutual MDR-HM-0001 onwards. Chapter 25.

Mandate manifest (MND): The written mandate of the agents a manager directs: tasks, degree of delegation per action, tools, data, budgets, a never list, approvals and the stop path (MND-HM-CLM-01). It reuses the mandate manifest of The Transformed Security Architect, Chapter 13; Chapter 16.

MBL (Medbestämmandelagen): The Swedish Co-determination in the Workplace Act, which obliges an employer bound by a collective agreement to negotiate with the union before deciding on an important change in its activity or in working or employment conditions (MBL §11). It gives no veto over technology, but no decision may be taken before the negotiation. Chapter 23.

Mixed team: A team in which people and agents work under the same manager. It needs work divided by what each does best, agents' figures kept apart from people's, designed handovers and learning work reserved for people. Chapter 18.

Moral crumple zone: Madeleine Clare Elish's term for the misattribution of responsibility to a human operator who had limited control over an automated system, so that the person, not the system, absorbs the blame when it fails. In an agent chain, which multiplies what Dennis Thompson called the problem of many hands, it is usually the person at the end who approves. Chapter 17.

N

Named human decider: The identifiable person who takes, with their own authority, every M3 Consequence decision in which AI took part, and who can be contested and held to account. The decider needs authority, information, time and a record. Chapters 4 and 20.

Nordvik Region: The fictional regional public authority in a Nordic EU member state, with about 6,000 employees, two local hospitals and home care for five municipalities, that runs through every chapter. Its AI boss directs shift work in care through a procured, high-risk rostering system under Swedish-style co-determination. Chapter 1.

O

Observable, expected, speculation: The three labels for any statement about the future: observable (visible now in adopted law, published guidance or production systems), expected (strongly evidenced but uncertain in timing or form) and speculation (plausible but not yet evidenced). The author's own projections are confined to Chapters 29 and 33 and labelled as such. AI Governance, Chapter 22.

Ondernemingsraad: The works council in the Netherlands, which under the Works Councils Act (WOR) has a consent right over regulations on processing employees' personal data and over facilities suited to observing or checking their attendance, behaviour or performance. Chapter 23.

Orchestrator: An agent that directs other agents, assigning, sequencing and checking their work. It performs M1 Direction and sometimes M2 Evaluation for its agents, may never hold M3 Consequence over a person at any depth of the chain, and answers to a named chain owner. Chapter 17.

Oversight record (OVR): The record, for one delegation profile and one period (OVR-NR-ROS-2027-1), of who oversaw, what they saw, what they did and with what authority, together with the classification of the observed degree of delegation and the findings. It holds the evidence of the observed degree, as the profile holds the intended one. Chapter 26.

Oversight-quality metric (OQM): A small set of figures agreed in advance (time to decision, departures with reasons, seeded-case catch, reversals on contest, stops and pauses) that shows whether oversight is real (OQM-HM-CLM-01, OQM-NR-ROS-01). It is reported at group level and never used to evaluate the overseer. Chapter 26.

P

Platform Work Directive: Directive (EU) 2024/2831 on improving working conditions in platform work, to be transposed by 2 December 2026. Its Chapter III on algorithmic management limits data processing, requires information on automated systems, human oversight and review, and requires that any decision to restrict, suspend or terminate a person's account or contract be taken by a human being. Chapters 5 and 22.

Procurement requirement set: The requirements a public authority sets for an AI system that performs management functions, tested at acceptance and on every release, keeping logs, the stop path, constraint tests, outcome data and the configuration of delegation in the authority's hands. Nordvik Region's set is NR-PRQ-MGT-01. Chapter 24.

Psychological safety: Amy Edmondson's term for a shared belief that a team is safe for interpersonal risk-taking, associated with learning behaviour and team performance. In a mixed team it includes being able to say that an agent was wrong. Chapters 18 and 29.

R

Reversibility test: The test, recorded in the delegation register (RVT-*), of whether a unit could perform a management function without the system, at an acceptable level, for a defined period, and whether its managers still can. The result is passed, passed with conditions or failed; the latter two freeze further delegation. AI Threat, Chapter 26; Chapter 25.

Role charter (ROLE): The versioned definition of a line manager's role (ROLE-NR-MGR-01, ROLE-HM-MGR-01): decision rights per function, the systems overseen, the agents directed, what may not be delegated, escalation and a time budget, which plans how time freed by systems is given back to the people rather than absorbed by a wider span. It binds the manager's other artefacts to a person. The Transformed Security Architect, Chapter 4; Chapter 29.

Rubber-stamping: Approval that does not influence the outcome, shown by approvals given in seconds, departures near zero and seeded errors that pass. It turns a recorded Level 1 or Level 2 into Level 4 by accident. AI Governance, Chapter 21; Chapter 26.

S

Safety representative: A worker elected to represent colleagues on safety and health, who in Sweden may require the employer to investigate a risk and, in case of immediate danger, stop work. Safety representatives take part in work risk assessments of AI systems that set pace. Chapters 6 and 14.

Scientific management: Frederick Winslow Taylor's system of managing by measurement, in which planning was separated from execution and work was timed and standardised into one best way. The book reads algorithmic management as its descendant and asks what is new. Chapter 3.

Seeded error: A known fault placed in synthetic cases or in drafts that are never acted on, to measure whether overseers find it. A gate removes any seed that is missed, so no real person is affected; the catch rate is part of the oversight-quality metric. The Transformed Security Architect, Chapter 14; Chapter 26.

Selective adherence: The finding that decision-makers follow algorithmic advice selectively, more often when it matches their prior beliefs or stereotypes about a group. It is why departures are broken down by group in the oversight-quality metric. Chapter 26.

Self-determination theory: Richard Ryan and Edward Deci's theory of three innate psychological needs, competence, autonomy and relatedness, whose satisfaction yields self-motivation and wellbeing and whose frustration diminishes them. Algorithmic direction that sets sequence and pace without reasons bears on all three. Chapter 11.

Span of control: The number of people a manager directly manages. Widening spans when systems take the planning, up to the vendors' image of one human and twenty agents, is labelled speculation; the evidence available sets the number by the human attention needed to verify and to care. Chapter 29.

Stop path: The means of stopping or pausing a management system that lies outside the system it stops and does not depend on its cooperation, with named holders and a stated time. For a procured system, a public authority holds its stop path outside the supplier's system. AI Platform and Security, Chapter 19; Chapters 17 and 20.

Survey plan (SVY): A survey of autonomy, perceived fairness and strain (SVY-NR-STV-01), agreed with the union and the safety representative, run before go-live and twice after, reported at unit level only and compared with a unit not yet using the system. Chapter 11.

T

Team charter (TCH): The charter of a mixed team (TCH-HM-CLM-01), recording the division of work between people and agents, separated figures, handovers and the learning work reserved for people. It is owned by the line manager and changed when an agent joins. Chapter 18.

Transparency register (TRG): The register of management systems and their worker information notices (TRG-NR-SW-HC, TRG-HM-CTA-DE), giving workers' representatives, supervisors, auditors and, for a public authority, the public a single view of what is in use. Chapter 28.

U

Usage modes: The six ways in which AI is used: Mode 0 Ambient AI, Mode 1 Conversational assistants, Mode 2 Embedded workplace assistants, Mode 3 Agentic work tools, Mode 4 Built AI applications and Mode 5 Autonomous production agents. A Mode 0 feature in licensed software can become an AI boss through a decision about data. AI Governance, Chapter 2; AI Platform and Security, Chapter 2.

W

Work-environment committee: A joint body of employer and workers' representatives that oversees the work environment. Nordvik Region's central work-environment committee meets quarterly, treats every change to rostering routes as a change in the activity that needs a risk assessment, and its chair reviews the Region's commission. Chapters 14 and 24.

Work risk assessment (WRA): The assessment of a pace, rate or routing system as a work-environment risk (WRA-NR-HC-01), made by the line manager with the safety representative when a system is introduced, when contests show a pattern and after an incident. Its findings become constraints in the optimiser, enforced and tested. Chapters 6 and 14.

Worker: The book's broader term for a person who works, covering employees, agency workers and people who perform platform work, some of whom are self-employed. "Employee" is written only where a statement depends on an employment contract, and "person performing platform work" where the Platform Work Directive's scope matters. Chapter 4.

Worker information notice (WIN): The notice that tells workers and their representatives what a management system does, which functions and degree it has, what data it uses, who decides and how to contest (WIN-NR-SW-HC-01, WIN-HM-CTA-DE). It is issued before use, reissued at each Class 1 or Class 2 change, and extracted into a decision notice at each decision. Chapter 28.

Workers' representatives: The AI Act's term for the bodies that represent workers, named precisely in this book as works council, European Works Council, trade union or union. They are informed under Article 26(7) and take part in agreeing delegation profiles, change classes and oversight figures. Chapters 21 and 23.

Working Time Directive: Directive 2003/88/EC, which sets numerical minimums for daily and weekly rest, breaks and an average maximum of 48 hours a week. An optimiser enforces them as feasibility conditions, but a compliant roster is not thereby healthy. Chapters 9 and 24.

Works agreement: An agreement between a works council and an employer (Betriebsvereinbarung), distinct from a collective agreement with a union. Hansa Mutual's group works agreement KBV-IT-2024 governs the claims-triage agent's figures and recruitment screening. Chapter 23.

Works council: An elected body representing the employees of an establishment, company or group towards the employer, with rights of information, consultation and in some countries co-determination or consent. In Germany it is the Betriebsrat; in the Netherlands the ondernemingsraad. Chapter 23.