The first generation of AI recruitment was relatively easy to recognise. Technology companies competed for machine-learning engineers, data scientists and researchers capable of building increasingly powerful models. Those specialists remain valuable, but they no longer represent the whole AI labour market.

A second category is now emerging inside banks, manufacturers, professional-services firms, retailers, healthcare groups and public institutions. These employees may not train foundation models or hold computer-science degrees. Instead, they determine where AI belongs in the organisation, connect it to existing work, test whether its output can be trusted and decide when a person must remain in control.

Their titles vary because the operating model is still taking shape. Companies are recruiting AI product managers, model-risk specialists, automation architects and forward-deployed engineers. Others are assigning AI responsibilities to lawyers, compliance officers, communications directors, researchers and operational managers without changing their job titles at all.

That distinction matters. The future of work is unlikely to be divided neatly between people who “work in AI” and everyone else. AI capability is becoming a layer within established professions, much as digital fluency, data analysis and internet research became expected across roles that were never classified as technology jobs.

The Labour Market Is Changing By Task

Predictions about AI employment often alternate between mass redundancy and an abundance of newly invented careers. Both narratives miss how organisations usually change.

Companies rarely automate an occupation in one step. They reorganise the tasks within it.

A communications manager may spend less time compiling media coverage and more time interpreting narrative risk. A lawyer may use AI to compare clauses but remain responsible for legal advice. A junior analyst may produce a first financial model more quickly, while being expected to interrogate assumptions earlier in their career. A customer-service employee may handle fewer routine enquiries and more complex cases involving judgement, emotion or financial consequences.

The International Labour Organization estimates that one in four jobs worldwide has some exposure to generative AI. Its more consequential finding is that transformation is more likely than outright replacement because relatively few occupations consist entirely of tasks that current systems can automate.

Exposure should not be mistaken for disappearance. It means that the composition of the work may change.

The World Economic Forum expects labour-market transformation to create 170 million roles and displace 92 million by 2030, producing a net increase of 78 million jobs under its survey assumptions. Technology-related roles are among the fastest-growing in percentage terms, but much of the absolute growth is expected in fields such as care, education, delivery and agriculture.

AI is therefore reshaping employment within a broader set of demographic, economic and technological pressures. Not every hiring slowdown is caused by automation, and not every new position will carry AI in its title.

AI Product Managers Are Translating Possibility Into Work

One of the most important emerging roles is the AI product manager.

A conventional product manager determines which customer or operational problem a product should solve, coordinates technical and commercial teams and measures whether the result creates value. An AI product manager performs the same core function but must also account for probabilistic output, model limitations, data provenance and the need for human oversight.

Consider an insurer introducing an AI assistant for claims handlers. The technical team may be able to summarise case files and suggest next steps. Someone must still determine whether the system should merely retrieve information, recommend an action or make a decision. The answer affects legal exposure, customer treatment, workflow design and the evidence required before launch.

The AI product manager does not need to be the organisation’s best model developer. The role requires enough technical understanding to know what is feasible, combined with a strong grasp of users, operations and risk.

This is why domain expertise is becoming more valuable rather than less. A technically impressive system can fail because it misunderstands how employees actually work, introduces delays at the wrong point or produces an answer that cannot be defended to a customer or regulator.

Forward-Deployed Engineers Bring AI Into The Organisation

The forward-deployed engineer sits closer to implementation.

Rather than developing a general model in isolation, these specialists work directly with clients or internal business teams to connect AI systems to data, processes and software. They may configure retrieval systems, build agent workflows, integrate models with enterprise applications and adapt the system as real users expose its weaknesses.

The title has become closely associated with enterprise AI companies, but the underlying function is broader. Organisations need people who can bridge the distance between a compelling demonstration and a dependable production system.

That distance is often underestimated. A chatbot that performs well with a clean example may struggle with outdated documents, conflicting permissions, specialist terminology and incomplete records. It may also require monitoring, fallback procedures and escalation to a human operator.

A forward-deployed engineer therefore solves an organisational problem as much as a technical one. The role involves spending time with users, observing exceptions and modifying the workflow rather than assuming that the first model configuration is sufficient.

For smaller companies, this may not justify a permanent job. The capability might be provided by a consultancy, software vendor or technically fluent operations manager. The requirement nevertheless remains: someone must understand both the AI system and the environment into which it is being introduced.

AI Operations Will Become A Permanent Function

Once AI moves from experimentation into recurring work, organisations need people to operate it.

AI operations, sometimes described as machine-learning operations or agent operations, includes monitoring performance, controlling model versions, managing access and responding when output deteriorates. It also covers the less glamorous work of maintaining data connections, investigating incidents and documenting changes.

An organisation deploying AI agents faces additional complexity because the system may do more than generate text. It could retrieve a customer record, update a database, draft a response and trigger another action. The potential productivity gain is greater, but so is the consequence of a mistake.

This is creating demand for professionals who can decide which permissions an agent receives, how its activity is recorded and where approval is required. They will need to understand security and process design, but also how people behave around automated systems. Employees may accept an AI recommendation too readily, bypass it entirely or invent informal workarounds when it slows them down.

The strongest AI operations teams will therefore combine technical monitoring with operational judgement. Their job is not simply to keep the model running. It is to ensure that the complete system remains useful, controlled and aligned with the purpose for which it was approved.

Model Risk Is Moving Beyond Banks

Financial institutions have long employed model-risk specialists to challenge the assumptions, data and limitations of quantitative systems. Generative AI is extending similar disciplines into a wider range of businesses.

A company using AI to screen job applications, prioritise customers or recommend insurance decisions needs to know whether the system behaves consistently across relevant groups. A professional-services firm must determine whether generated analysis is grounded in approved sources. A communications team using AI to monitor public sentiment should understand whether the underlying dataset overrepresents particular platforms or audiences.

This work may be performed by an AI assurance manager, model validator, responsible AI lead or existing risk professional. The title is less important than the independence of the challenge.

The person who built or commissioned a system is not always best placed to assess it objectively. Organisations need a credible process for testing failure modes, documenting limitations and stopping deployment when the evidence is insufficient.

The role should not be confused with writing an ethics statement. It is closer to quality assurance, risk analysis and internal audit. The work includes practical questions: Which claims can the system make? How often is its output wrong? What happens when source information conflicts? Can a user contest the result? Who investigates an incident?

AI Governance Requires More Than A New Committee

Demand is also growing for professionals who translate regulation and policy into operational controls.

An AI governance lead may maintain an inventory of systems, classify their risk, define approved uses and coordinate legal, security, procurement and compliance teams. The role becomes particularly important when employees adopt public tools independently or when AI functionality arrives through software the company already owns.

Europe’s AI Act is reinforcing this need. Its requirements are being introduced in stages, including provisions concerning prohibited practices, AI literacy, general-purpose models and high-risk systems. Organisations must determine which obligations apply to their activities rather than assuming that every AI tool carries the same regulatory treatment.

Governance professionals consequently need enough legal knowledge to interpret requirements and enough operational understanding to make them usable. A policy stating that “AI output must be reviewed” is inadequate unless the organisation defines who reviews it, what they check and what evidence is retained.

The most effective practitioners will not attempt to prevent every experiment. They will create proportionate rules that distinguish a low-risk drafting assistant from a system influencing employment, credit, healthcare or access to essential services.

The AI Trainer Is Becoming A Domain Expert

The phrase “AI trainer” once referred mainly to people labelling data or rating model responses. That work continues, but a more specialised version is appearing inside organisations.

A legal AI assistant must learn which sources have authority and how the firm structures advice. A customer-service system needs examples of acceptable language, escalation rules and prohibited commitments. A communications tool must distinguish between a factual discrepancy, an emerging narrative and a genuine reputational threat.

People who teach and evaluate these systems therefore need subject expertise. They may design examples, review responses, construct evaluation criteria and identify recurring failures. In many cases, the best candidate will be an experienced employee who understands the work deeply and develops additional AI skills.

This creates a career path for professionals who do not wish to become engineers. A senior claims handler, policy analyst or communications specialist can become responsible for the quality of AI-assisted work in that domain.

The contribution is not merely producing prompts. It is converting tacit professional judgement into instructions, examples and tests that a system can use.

AI Literacy Is Entering Ordinary Job Descriptions

The largest employment shift may not involve new titles at all.

LinkedIn reported in January 2026 that US job advertisements requiring AI literacy had grown 70 percent year on year. The demand extends beyond technical positions because employers increasingly expect people to use AI within established functions.

AI literacy does not mean accepting every generated answer or knowing the vocabulary of model architecture. At a practical level, it means being able to define a task, provide relevant context, verify evidence, recognise sensitive information and decide whether the output is suitable for use.

For a public affairs manager, that may involve comparing proposed legislation, mapping stakeholder positions and producing a first briefing while checking every consequential claim against primary material. For a researcher, it may mean using AI to organise literature without presenting generated synthesis as empirical evidence. For a manager, it may involve redesigning a workflow rather than simply asking employees to complete the same work faster.

These abilities will increasingly sit alongside writing, presentation and spreadsheet skills. Their value comes from application within a profession, not from AI fluency in isolation.

Some Proposed AI Jobs Will Not Become Large Professions

Not every fashionable title will survive.

Prompt engineer became shorthand for the idea that companies would employ specialists whose main skill was composing instructions for language models. Some organisations still need advanced prompt and evaluation expertise, particularly when building repeatable systems. Yet prompting is increasingly becoming one component of product design, software development and professional AI use rather than a standalone department.

The same caution applies to titles such as chief AI officer. A large bank, pharmaceutical group or multinational manufacturer may need an executive coordinating investment, governance and implementation. A smaller organisation may create unnecessary bureaucracy by appointing a senior AI figure before identifying a worthwhile use case.

Businesses should recruit around enduring responsibilities rather than fashionable labels. They need someone to own the outcome, someone to operate the system, someone to challenge its risks and domain experts who can judge its work. Whether those responsibilities require four jobs or parts of two existing roles depends on scale and exposure.

Entry-Level Work Needs To Be Redesigned Carefully

AI creates a particular problem at the beginning of the career ladder.

Many junior roles traditionally combine routine production with learning. A young consultant prepares research, a communications executive compiles coverage, and a trainee lawyer reviews documents. These tasks may be inefficient, but they also expose employees to the material from which professional judgement develops.

If AI removes the routine component without replacing the learning mechanism, organisations may produce fewer experienced professionals in future. Junior employees could be asked to evaluate work they have never learned to create.

Companies need to separate low-value repetition from developmental experience. A graduate can use AI to accelerate initial research while still being required to inspect sources, explain the reasoning and defend the recommendation. Managers may need to provide more deliberate coaching because learning will no longer occur automatically through volume of work.

This also changes recruitment. Entry-level candidates who can combine AI use with critical thinking may contribute to more complex work earlier, but employers should not mistake fast production for mature judgement.

Human Skills Become More Visible, Not Less Important

As AI handles more drafting, retrieval and routine analysis, the remaining human contribution becomes easier to identify.

Employers still need people who can make decisions with incomplete information, persuade sceptical stakeholders, detect political sensitivities and take responsibility for consequences. Adaptability, analytical thinking, resilience, leadership and social influence remain prominent in employer surveys alongside AI and data skills.

This does not mean “soft skills” will automatically protect every role. A personable employee performing work that can be fully standardised may still face pressure. The stronger combination is domain knowledge, AI capability and human judgement.

For communications and public affairs professionals, this is particularly relevant. AI can produce a stakeholder list or simulate hostile questions. It cannot be held accountable in front of a minister, advise a chief executive during a reputational crisis or reliably interpret the emotional dynamics of a difficult meeting. Those activities require credibility, restraint and institutional understanding.

What Employers Should Build Now

Before creating new AI positions, leaders should identify the work that is changing.

Which decisions will AI influence? Which tasks are being automated? Where must a person approve the result? What expertise is required to recognise a serious error? These questions reveal the capabilities the organisation actually needs.

Some gaps should be filled through recruitment. Others are better addressed by retraining people who already understand the company’s customers, systems and risks. A technically capable external hire may take months to acquire knowledge that an experienced employee already possesses.

Training should therefore be role-specific. A general introduction can establish basic rules, but employees develop useful competence by applying AI to real work, reviewing failures and learning the boundary between assistance and delegation.

The new class of AI jobs will include engineers and specialists. It will also include lawyers who can test an automated decision, managers who can redesign a process and communicators who know when an efficient answer would be the wrong one to publish.

The most resilient career strategy is not to chase every new title. It is to become the person who understands a valuable field of work, knows how AI changes it and can remain accountable for the result.