Reading time: 40 minutes
Artificial intelligence
andfuture of trades
Automation, work transformation, new occupational balances and coping strategies
This dossier analyses the impact of artificial intelligence on all major trade families, without favouring a particular profession. It distinguishes technological exposure, effective automation, job transformation and the real risk of job reduction.
Employment projections should be interpreted with caution. A profession called « Exposed » artificial intelligence is not necessarily expected to disappear. The exposure indicates that a significant proportion of its tasks can be accelerated, assisted or automated. The real effects will depend on the maturity of the technologies, their cost, regulation, business organisation, social acceptability, data quality and management choices.
The file focuses on a task-based approach. The same profession can contain highly automated activities and others that require judgment, responsibility, physical presence, negotiation, empathy or contextual understanding. This distinction is essential to avoid two opposing errors: to announce the premature disappearance of entire professions, or to minimize the downsizing that may result from partial automation.
Summary of Article
Understanding breakup and mapping the exhibition
Chapter I Executive summary
Artificial intelligence is entering a general diffusion phase. After mainly automated industrial operations and specialized calculations, it is now able to produce text, code, images, synthesises, recommendations and analyses. It thus penetrates the heart of many intellectual professions: law, finance, health, education, computer science, marketing, media, administration and engineering.
According to the International Labour Organization, about one in four jobs worldwide belong to a profession with a certain degree of exposure to general AI. Exposure is higher in high-income economies, with more office jobs and cognitive activities. The International Monetary Fund estimates that about 40 per cent of global employment is exposed to AI, with nearly 60 per cent in advanced economies. These figures do not constitute forecasts of job losses: they measure the potential for transformation.
The strongest conclusion is that the AI will first transform tasks before removing complete jobs. However, this transformation may reduce the number of staff required. A team of six people assisted by software agents can, in some processes, produce what previously required ten people. The loss of jobs may therefore result from an increase in productivity, a decrease in recruitment or a non-replacement of departures, without any occupation being legally or technically abolished.
The most exposed functions combine several characteristics: digital work, large volumes, relatively stable rules, standardized production, strong documentary component, availability of historical data and ability to measure the quality of the result. Conversely, tasks based on physical dexterity, human relationship, intervention in an unpredictable environment, negotiation or ultimate responsibility are more resistant in the short term.
| Finding | Likely consequence |
|---|---|
| IA automates cognitive tasks | The professions qualified as offices are directly concerned. |
| Automation is partial | The trades recompose before disappearing. |
| Productivity increases | Companies can produce more with smaller teams. |
| Junior tasks are the easiest to delegate | Entry into certain professions is becoming more difficult. |
| Quality depends on data and supervision | Human control, audit and governance are gaining value. |
| Tools become commonplace | The competitive advantage moves towards expertise, judgment and relationship. |
Chapter II Understanding rupture: from conventional automation to general AI
1. An extension of automation to intellectual activities
Traditional automation was mainly applied to explicit rules: calculating a pay, registering an order, moving an industrial part or executing a series of instructions. The contemporary AIA deals with less structured information. It recognizes patterns in an image, summarizes a file, classifies a mail, formulates an answer, generates a computer program or proposes a differential diagnosis.
This changes the boundary between automated work and human work. The supposed occupations protected by the diploma or intellectual complexity become exposed when a part of their activity can be described in the form of documents, examples and evaluation criteria.
2. The four forms of impact
| Form of impact | Description | Illustration |
|---|---|---|
| Substitution | The machine performs a previously human task. | Automatic transcription of a meeting. |
| Increase | The machine accelerates or improves human work. | Radiologist assisted to detect an anomaly. |
| Recomposition | Tasks change and the job is re-focusing. | Jurist moving from research to strategy. |
| Establishment | New functions appear. | Algorithm Auditor or Governance Officer IA. |
3. Why the effects will be fast but uneven
The speed of dissemination results from the low marginal cost of the software, its integration into the desktop suite and its ability to be deployed simultaneously to thousands of users. Inequality in adoption, on the other hand, comes from regulatory constraints, old computer systems, confidentiality, data quality and organizational resistance.
Chapter III Method of assessing trade exposure
The assessment is based on five dimensions. It does not seek to predict mechanically a number of job deletions, but to assess the vulnerability of tasks and the transformation capacity of the trade.
| Dimension | Question of analysis |
|---|---|
| Digitization | Is the work mainly done on computer and producing usable data? |
| Standardization | Can the expected results be described by rules, models or examples? |
| Repeatability | Are the same operations carried out in large volumes? |
| Error tolerance | Can an error be detected and corrected before causing serious damage? |
| Human or physical dimension | Does the profession require presence, empathy, negotiation, dexterity or direct responsibility? |
A highly exposed occupation on the first three dimensions but strongly constrained by responsibility or human contact will probably be increased rather than eliminated. Conversely, a standard, digital and low-risk administrative process can be automated from end to end.
Chapter IV General mapping of the professions concerned
| Level | Characteristics | Typical trade families |
|---|---|---|
| Very high exposure | Digital, repetitive, documentary, standardized tasks. | Seizure, back office, call centres, standard translation, secretariat, codification. |
| High exposure | Structured intellectual analysis and production, with human validation. | Law, finance, computer, marketing, media, human resources. |
| Intermediate exposure | Combination of automated tasks and responsibility/contextualization. | Medicine, teaching, engineering, management, consulting. |
| Low short-term exposure | Physical presence, dexterity, unstable environment, direct relationship. | Crafts, personal assistance, maintenance, construction, restoration. |
| Future exposure by robotics | Standardised physical activities in a controlled environment. | Logistics, warehouses, industrial production, transport. |
A distinction must be made between occupation, position and task
A profession is an assembly of tasks. A teacher prepares courses, explains, motivates, corrects, evaluates, manages a group and accompanies courses. It can produce exercises and correct certain works, but it does not necessarily replace the social and pedagogical function. Similarly, a physician may delegate documentation and certain tests without delegating the clinical examination, the announcement of a diagnosis or the final decision.
Sector by sector: which is concerned, and how
Chapter V Detailed analysis by major sectors
Fifteen large families of trades are reviewed, from support functions to crafts, following the same reading grid: what the AI already knows how to do, and what remains human.
Administration, secretariat and support functions
Administrative functions are among the most directly exposed. AI agents can read an email, identify the subject of the request, extract data from an attachment, provide software, generate a response, create a task and archive the file. Thus, automation is no longer confined to an isolated operation, but to a complete chain.
What remains humanJobs will not disappear entirely. The functions will evolve into exception management, coordination, hosting, data quality and process control. The most vulnerable items are those whose value is based almost exclusively on the circulation of standardized information.
Customer service, call centres and commerce
Voice and text conversational agents can process common requests continuously, in several languages, with access to the client's file. They can explain an invoice, modify an option, register a complaint, propose a solution and summarize the exchange.
What remains humanHuman advisors will focus on conflict situations, negotiations, complex sales, vulnerable customers and exemptions. The main social risk is a sharp reduction in first-level employment volumes.
Law, justice and conformity
The legal professions use considerable volumes of normative texts, contracts, decisions and correspondence. The AIA is particularly suitable for research, summary, comparison of clauses and production of first projects.
What remains humanThe litigation strategy, negotiation, context interpretation, pleading and the responsibility of counsel will remain human. The junior research and literature review functions are nevertheless highly exposed.
Banking, insurance and financial services
Banking and insurance have numerous, structured and historically modeled data. Generative L的IA adds a ability to read and interact with the client.
What remains humanSenior analysts and managers will see their activity reduced or refocused. The trades of structuring, atypical decision-making, heritage management and complex relationships will be mainly increased.
Accounting, audit and management control
These activities combine structured data, rules, reconciliations, controls and document production. AIA can automate collection, classification, detection of anomalies, management comments and some of the documentation.
What remains humanThe value will move to analysis, judgment, certification, risk management, interpretation and advice. Manual production volumes and execution stations are expected to decrease.
Health and medicine
Imaging, digital pathology, genomics, clinical writing, surveillance, and pharmaceutical research are the main areas of activity. Specialties based on image or signal analysis are highly exposed to algorithmic assistance.
What remains humanClinical responsibility, physical examination, therapeutic relationship, shared decision and uncertainty management limit complete substitution. Tomorrow's doctor will be more supervisor, data integrator and referee.
Computer, development and cybersecurity
Programming assistants can generate code, testing, documentation and prototypes. They greatly increase the productivity of experienced developers and reduce the time required for standardized tasks.
What remains humanJunior developers are particularly exposed. Architecture, security, integration, professional understanding and quality control are gaining more value. Cybersecurity is becoming increasingly important simultaneously, as AI also increases the ability of attackers.
Marketing, advertising, design and communication
It produces texts, visuals, videos, voices, segmentations and campaign variants. It significantly reduces the cost of producing standard content.
What remains humanBrand strategy, artistic direction, cultural understanding, originality and reputation management remain different. Performance and mass production are at greatest risk.
Media, journalism, publishing and translation
Systems can summarize documents, transcribe, translate, title, reformulate and produce structured dispatches. The boundary between human production and automated synthesis becomes more difficult to identify.
What remains humanInvestigation, terrain, independent verification, source protection, analysis and editorial responsibility remain central. Specialized and cultural translation retains a higher value than standard translation.
Education, training and research
IA can offer an individualized tutoring, create exercises, adapt content, correct certain productions, translate materials and assist in document search.
What remains humanThe teacher will focus on accompaniment, motivation, critical thinking, socialization, authentic evaluation and pedagogical design. Evaluation systems will need to be thoroughly rethought.
Human resources and recruitment
The AIA is involved in the drafting of announcements, screening of applications, mapping of skills, preparation of interviews, personnel administration and training.
What remains humanThe risks of discrimination, opacity and surveillance require strict supervision. Sensitive decisions should not be abandoned to an automated score.
Architecture, engineering, construction and industry
Generative AIA and simulation allow to produce variants, optimize structures, detect conflicts, predict faults and monitor installations. Robotics and artificial vision extend automation to the physical world.
What remains humanStandard drawing and calculation tasks are described. Technical responsibility, coordination, safety, on-site intervention and management of unplanned situations limit substitution.
Logistics, transport and distribution
Warehouses, road planning, demand forecasting and inventory management are already highly algorithmized. Mobile robotics and assisted driving continue this transformation.
What remains humanThe full range of vehicles progresses more slowly than expected in the open environment. In controlled warehouses and sites, automation will be faster. Trades will move to maintenance, supervision and incident management.
Agriculture, environment and energy
Precision farming combines imaging, sensors, drones and predictive models. The AIA can optimize irrigation, detect diseases, predict yields and reduce some inputs.
What remains humanThe impact will depend heavily on access to capital, data and infrastructure. Field expertise remains essential, but the best-equipped farms will gain productivity.
Crafts, care, catering and local services
These occupations are less exposed in the short term because they require dexterity, presence, confidence and adaptation to varying physical environments. However, they will use the AI for planning, forecasting, training, diagnosis and customer relationship.
What remains humanRobotics could automate some repetitive tasks, but their cost and difficulty of integration slow down substitution. These occupations can become relatively more attractive as office jobs change.
Employment, generations and new professions
Chapter VI Effects on employment, wages and work organisation
1. The decline in the number of people can precede the disappearance of the trades
The most likely scenario is a gradual reduction of labour requirements on certain functions: recruitment freeze, non-replacement of departures, reduced outsourcing, concentration of teams and increase of production targets. This will be less visible than a sudden wave of redundancies, but it can profoundly change the labour market.
2. Polarisation of jobs
The AIA can promote a polarization between experts who can design, supervise and take responsibility for decisions and, on the other hand, physical service jobs that are difficult to automate. The intermediate functions of standardized intellectual production are the most vulnerable.
3. Wages and earnings sharing
When AI complements a rare and skilled worker, it can increase productivity and pay. When it makes competence easily accessible, it can, on the contrary, trivialize that competence and put pressure on prices and wages. Gain sharing will depend on the bargaining power of employees, competition, regulation and ownership of tools and data.
4. Intensification and monitoring
The AIA can alleviate difficult tasks, but also increase job measurement, monitoring, pace and standardization. An organization can use time savings to improve quality and reduce burden, or to significantly increase targets. Technology alone does not determine the effect on working conditions; the management model is decisive.
Chapter VII The critical case of young graduates and entry jobs
The tasks entrusted to beginners are often precisely those that the AI automates best: document research, synthesis, initial analysis, project writing, simple testing, editing and data control. The removal of these tasks creates a paradox: organisations want to recruit experienced professionals, but reduce the steps that allow them to gain this experience.
This can lead to a contraction of junior positions in law, computer science, finance, marketing, media and consulting. Training will therefore have to include more situations, direct supervision, project learning and progressive responsibility.
| Risk | Required response |
|---|---|
| Reduction of entry points | Create safe learning pathways and rotations. |
| Excessive dependence on tools | Require exercise without assistance and justification. |
| Insufficient technical expertise | Train in validation, data and model boundaries. |
| Diplomas disconnected from needs | Review programs and certifications quickly. |
| Inequalities between students | Ensure equitable access to tools and guidance. |
Chapter VIII New trades and emerging skills
In addition to creating specialists in machine learning, the IA does not create only specialists. It generates interface functions between technology, trade, law and risk.
| Emerging occupation | Main mission |
|---|---|
| Head of governance IA | Define rules, responsibilities, inventories and controls. |
| Auditor of algorithmic systems | Evaluate performance, bias, safety and compliance. |
| Business data engineer | Prepare and rely on sectoral data. |
| Product owner IA | Translate an operational need into a controllable solution. |
| Human Validation Specialist | Organize reviews, climbing thresholds and controls. |
| Model safety expert | Protecting models, data and use chains. |
| Manufacturer of agents and process | Assemble agents capable of performing workflows. |
| Literacy trainer IA | Train professionals for critical and responsible use. |
| Legal Officer IA and data | Framework contracts, liability, intellectual property and compliance. |
| Specialist in human-machine experience | Design interfaces that avoid errors and overtrust. |
The most sustainable skills
- Deep professional expertise
- Critical reasoning and verification capacity
- Communication and pedagogy
- Conflict negotiation and management
- Process design
- Data control
- Legal and ethical culture
- Creativity and problem formulation
- Ability to work with evolving tools
- Responsibility and decision-making in uncertainty
Global inequalities and future scenarios
Chapter IX Inequality, territories and developing countries
1. Different exposure according to employment structure
Advanced economies are more vulnerable because they have more office and cognitive jobs. Low-income countries may be less exposed statistically to general AI, but are more vulnerable to indirect effects: outsourcing of services, pressure on outsourced jobs, technological dependency and difficulty in accessing advanced tools.
2. Catch-up opportunity
The AIA can also reduce skill shortages, improve access to training, support isolated professionals, facilitate translation, and allow small organizations to produce services previously reserved for large structures. The potential for catching up, however, depends on electricity, connectivity, local data, training and digital sovereignty.
3. Risk of new addiction
When models, cloud infrastructures and data are controlled by a few foreign actors, user countries may become dependent on tariffs, standards and contractual conditions that they do not control. An independent data, training and evaluation policy becomes an issue of economic sovereignty.
Chapter X Scenarios for 2030-2040
Inclusive increase
The AIA complements the workers, the earnings finance training and reduction of penitibility. Dominant effects: productivity, quality, professional mobility.
Defensive automation
Companies use AI as a priority to reduce costs. Dominant effects: downsizing, intensification, polarization.
Two-speed market
Major organizations have advanced tools, while others are lagging behind. Dominant effects: concentration, dependence, productivity differences.
Restrictive regulation
Risks and responsibilities slow down some uses. Dominant effects: slower diffusion in sensitive sectors.
Robotic failure
IA joins robotics on a large scale. Dominant effects: rapid extension to standardized physical occupations.
The actual scenario will likely combine these trajectories across sectors. Digital services will evolve faster than physical activities. Regulated professions will progress by successive validation, while content and back office activities will be subject to rapid reorganization.
Strategies, governance and recommendations
Chapter XI Business and government strategies
- Mapping processes by task rather than by job titles.
- Identify activities to automate, increase and maintain human control.
- Evaluate the full cost: licensing, integration, data, control, security and training.
- Conduct limited experiments with quality and productivity indicators.
- Define responsibilities, validation thresholds and incident procedures.
- Engage employees in designing new processes.
- Maintain learning pathways for young professionals.
- Measure bias, drift and effects on working conditions.
- Predict reversibility and avoid excessive reliance on a supplier.
- Reinvest part of the earnings in training and quality of service.
IA does not automatically correct an inconsistent organization. Automation of an unnecessary, poorly controlled or data-based process can speed up errors. Transformation must begin with clarifying the need, simplifying the process and defining controls.
Chapter XII Individual strategies and training policies
1. For professionals
- Learn to use multiple tool families without dependent on a brand
- Developing sector expertise difficult to reproduce
- Checking results and documenting decisions
- Strengthen communication, negotiation and client understanding
- Mastering data, confidentiality and cybersecurity
- Building a portfolio of achievements rather than mere theoretical knowledge
2. For training institutions
- Review programs annually
- Integrate AI into business courses and not into an isolated module
- Evaluate reasoning, sources and capacity for criticism
- Teaching basic tasks without assistance to preserve autonomy
- Develop project learning and alternance
- Train teachers and organize equitable access to tools
3. For public authorities
- Anticipating sectors exposed by task observatories
- Financing conversions before massive abolitions
- Adapting unemployment insurance and continuing training
- Framework algorithmic monitoring at work
- Support SMEs in assessment and cybersecurity
- Preserving human access to essential services
- Developing infrastructure and data of general interest
Chapter XIII Governance, social law and accountability
The adoption of the AI raises issues of data protection, intellectual property, accountability, discrimination, transparency, occupational health and consultation with staff representatives. Organizations must maintain an inventory of systems, qualify their level of risk, frame authorized data and maintain traceability of important decisions.
Purely formal human supervision is not sufficient. The professional must have the time, information and authority to challenge the recommendation. When the pace makes verification impossible, human responsibility becomes fictitious.
| Governance risk | Control measure |
|---|---|
| Discriminatory decision | Subpopulation testing, human control and recourse. |
| Data leak | Authorized environment, minimization and encryption. |
| Persuasive error | Validation, sources, thresholds and control of sensitive cases. |
| Supplier dependency | Reversibility clauses, data export and audit. |
| Excessive surveillance | Proportionality, employee information and usage limits. |
| Loss of skills | Continuing training and exercises without assistance. |
| Model drift | Periodic monitoring, logging and withdrawal procedures. |
Chapter XIV Operational recommendations
1. Decision matrix for a task
| Question | Answer | Consequences |
|---|---|---|
| Is the task repetitive and digital? | Yes | Candidate for automation. |
| Can the result be easily controlled? | Yes | Safer automation. |
| Can an error cause serious damage? | Yes | Enhanced human validation. |
| Does the task require empathy or negotiation? | Yes | Maintenance of human intervention. |
| Are the data reliable and authorised? | No | Do not deploy before correction. |
| Is the time saved more than the cost of control? | No | Economically questionable project. |
2. Short-term priorities
- Train all employees at a minimum level of AI culture.
- Secure spontaneous use and prohibit the sending of sensitive data to unauthorized tools.
- Choose some high volume and low risk processes to experiment.
- Create cross-cutting governance involving trades, information technology, legal, security and human resources.
- Measure actual gains and effects on quality, not just speed.
Chapter XV · General conclusion Automatize without losing judgment
Artificial intelligence will affect almost all trades, but according to different mechanisms. In administrative and documentary activities, it can replace complete task chains. In skilled and regulated professions, it acts primarily as a system of assistance, calculation, detection and preparation. In physical and relational occupations, its impact will first be indirect, before gradually spreading through robotics.
The main economic risk is not the instantaneous disappearance of entire professions. He resides in the reduction in the number of people needed to produce the same service, the contraction of junior positions and the polarization between supervisors and service functions that are difficult to automate.
The answer cannot be limited to learning a few orders or buying licenses. It involves transforming organizations, training pathways, responsibilities, control systems and sharing productivity gains.
The actors that will succeed will be those who can automate without losing judgment, increase productivity without degrading quality and integrate technology without making human work accessory.
Synthetic table of exposure by trade family
| Family | Exposure | Dominant effect | Sustainable human skills |
|---|---|---|---|
| Administration and seizure | Very high | Process automation | Exception management, data quality |
| Call centres | Very high | Reduction of first level volumes | Conflicts, complex sales, vulnerability |
| Law | High | Automated research and writing | Strategy, negotiation, accountability |
| Finance and insurance | High | Standard analysis and treatment | Atypical decision, relationship, control |
| Accounting and audit | High | Automated production and controls | Judgement, certification, advice |
| Health | High on certain tasks | Assisted diagnosis and documentation | Examination, relationship, clinical decision |
| Information technology | High | Code and automated tests | Architecture, security, integration |
| Marketing and design | Very high for execution | Production of low cost content | Strategy, creativity, leadership |
| Journalism and Translation | High | Synthesis, translation, standard writing | Investigation, verification, grade |
| Education | Intermediate to high | Tutoring and automated content | Support, motivation, socialisation |
| Human resources | High | Sorting and administration | Decision, social dialogue, equity |
| Engineering and architecture | Intermediate to high | Assisted design and simulation | Responsibility, coordination, field |
| Logistics | High | Optimization and robotics | Maintenance, incidents, supervision |
| Agriculture | Intermediate | Accuracy and prediction | Land, investment, adaptation |
| Crafts and care | Low in the short term | Peripheral assistance | Dexterity, confidence, presence |
Indicators to be followed in an organization
- Average time per folder before and after deployment
- Error and recovery rate
- Volume of cases climbed to a human
- Customer and employee satisfaction
- Trends in junior staff and recruitment
- Security or confidentiality incidents
- Performance gaps between populations
- Full cost per process
- Level of dependency on supplier
- Time spent on training and monitoring
References Selected Bibliography
- International Labour Organization Generative AI and Jobs: A Refined Global Index of Occupational Exposure, Working Paper 140 — 20 May 2025 ilo.org — Refined Global Index of Occupational Exposure
- International Labour Organization Generative AI and jobs: a 2025 update — 20 May 2025 ilo.org — Generative AI and jobs: a 2025 update
- International Monetary Fund Gen-AI: Artificial Intelligence and the Future of Work, Staff Discussion Note — January 2024 imf.org — Gen-AI and the Future of Work
- International Monetary Fund New Skills and AI Are Reshaping the Future of Work — 14 January 2026 imf.org — New Skills and AI Are Reshaping the Future of Work
- OECD AI and Work oecd.org/en/topics/ai-and-work
- OECD Who Will Be the Workers Most Affected by AI? — 2024 oecd.org — Who Will Be the Workers Most Affected by AI?
- OECD Artificial Intelligence and the Changing Demand for Skills in the Labour Market — 2024 oecd.org — Changing Demand for Skills in the Labour Market
- OECD Employment Outlook 2023: Artificial Intelligence and the Labour Market oecd.org — OECD Employment Outlook 2023
- World Economic Forum Future of Jobs Report 2025 weforum.org — Future of Jobs Report 2025
- World Economic Forum Four Futures for Jobs in the New Economy: AI and Talent in 2030 — 2025 reports.weforum.org — Four Futures for Jobs in the New Economy
Sources consulted for this edition: June 2026. Foresight data should be updated periodically, taking into account the rapid development of technologies and uses.
