AI exposure does not mean automatic job loss. It means that parts of a job could be done faster, differently or with less direct human effort. The ILO (International Labour Organization) (2023) argues that generative AI is more likely to augment occupations than fully automate them, although clerical work is highly exposed. Goldman Sachs (2023) similarly estimates major exposure across occupations but warns that not all automated work will translate into redundancies.
The practical question is therefore: which parts of the job create human value, and which parts are mainly repeatable information processing?
The World Economic Forum (2025) also identifies technological change as a major driver of labour-market transformation to 2030, making workforce redesign and reskilling part of responsible automation.
Administrative and data-entry roles are among the clearest examples of likely task elimination. AI and robotic process automation can extract information from forms, check records, route documents, produce standard emails, update systems and flag missing data. The human role may shift towards exception handling, process improvement, data quality and customer support. Where the role is almost entirely routine processing, headcount risk is higher.
Customer service and contact-centre roles are likely to be both enhanced and reduced. Routine queries, order updates, password resets and basic policy questions can be handled by chatbots or self-service tools. More complex, emotional or high-risk conversations still need human judgement. Brynjolfsson, Li and Raymond (2025) found that a generative AI assistant increased customer-support productivity, particularly for less experienced workers, showing how AI can improve performance rather than simply remove people.
Software, data and digital roles are also exposed because AI can generate code, write documentation, test scripts, summarise requirements and analyse datasets. This does not remove the need for technical professionals. It changes the skill emphasis towards problem framing, architecture, testing, security, ethical judgement and explaining outputs to users. Entry-level tasks may be under pressure if organisations use AI to complete basic coding or analysis that previously helped juniors learn.
Marketing, content and communications jobs will be reshaped by drafting tools, image generation, audience analysis, search optimisation and campaign testing. Standard copy, summaries and first drafts may be cheap to produce, so value will move towards brand judgement, audience insight, originality, evidence checking and campaign ethics. Similar changes apply to finance, legal, compliance and HR roles: AI can review documents, summarise rules, compare options and identify patterns, but high-stakes advice still needs accountability and context.
Project, operations and management support roles are likely to be enhanced through meeting summaries, risk logs, plans, dashboards, scheduling and stakeholder updates. However, AI cannot own relationships, negotiate trade-offs or take responsibility for performance.
Frontline roles in care, hospitality, construction, engineering and trades may be less exposed to full replacement because they require physical presence, social trust and adaptation to unpredictable environments. Even these jobs may still change through scheduling tools, diagnostics, reporting, stock control and remote support.
The biggest mistake is to treat job titles as fixed categories. Eloundou et al. (2024) show that large language models (LLMs) affect task content across many occupations, while the OECD (Organisation for Economic Co-operation and Development) (2023) stresses that AI brings both opportunity and risks around privacy, bias and work intensity. UK Government analysis also warns that exposure is not the same as adoption and that early labour-market signals do not prove AI is the sole cause of hiring changes (DSIT, 2026). Responsible organisations should therefore map tasks before making workforce decisions.
A useful approach is to classify each task as enhance, eliminate, redesign or keep human-led.
- Enhance tasks where AI improves speed, quality, access or learning while a person remains accountable.
- Eliminate tasks where the work is repetitive, low-risk, measurable and better completed by a system.
- Redesign tasks where AI changes handovers, controls, skills or job boundaries.
- Keep tasks human-led where empathy, ethics, legal accountability, physical dexterity, professional judgement or trust are central.
AI will not affect every person equally. It may help newer workers learn faster, reduce boring administration and improve access to information. It may also remove entry-level practice tasks, increase monitoring, intensify workloads or make poor decisions appear more objective than they are.
The best answer to “enhance or eliminate?” is therefore evidence-based job redesign: understand the task, test the tool, involve affected workers, protect fairness and measure whether performance, quality and job value improve.
Action Point
Select one job in your organisation and break it into ten tasks. For each task, decide whether AI is most likely to enhance it, eliminate it, redesign it or leave it human-led. Add evidence for your judgement, including risk, data quality, customer or employee impact, skills needed and how human accountability will be protected.