Artificial intelligence is beginning to alter work well beyond offices and software teams, but training systems are not adapting at the same pace for the technicians, tradespeople, and service workers keeping hospitals, factories, and infrastructure running, according to new research from Pearson. The company describes the problem as a “triple capability gap” spanning AI literacy, human skills, and practical occupational knowledge.

Those demands are arriving as experienced workers leave many of the same occupations. Pearson estimates that 99% of UK openings across the skilled roles it examined over the coming decade will arise through replacement demand caused by retirement, career changes, and people leaving the workforce rather than by net employment growth.

The combination makes workforce planning more complicated than teaching employees how to use a generative-AI interface. Organisations have to introduce new technology while preserving the tacit knowledge that allows experienced workers to recognise when an automated recommendation does not fit the machine, patient, building, or operating environment in front of them.

That distinction becomes particularly important in jobs where an incorrect output has consequences beyond a badly written document. Healthcare, maintenance, manufacturing, and technical services all contain tasks in which judgement and physical context remain difficult to separate from the formal procedure.

AI reaches jobs where errors have physical consequences

Early generative-AI adoption concentrated heavily on knowledge work because writing, software development, research, administration, and customer service offered immediate applications. Pearson instead examines occupations including pharmacy technicians and industrial machinery mechanics, where software assistance meets physical systems and regulated processes.

A model can retrieve information, suggest a diagnostic step, summarise instructions, or identify a pattern, but a worker still needs enough occupational knowledge to judge whether the recommendation is safe and appropriate. That makes AI literacy inseparable from the competence required to challenge the system.

Pearson reports that fewer than half of surveyed industrial machinery mechanics felt their training was preparing them for AI-augmented work, while many early-career pharmacy technicians said AI had not been covered explicitly during training. The precise adoption rate will vary by employer, but the gap matters because workplace software can change considerably faster than formal qualifications.

Accuracy also carries different consequences in these occupations. A hallucinated sentence can be edited; an incorrect medicine, repair, or technical intervention can cause harm before anybody has an opportunity to revise it.

Experience becomes harder to replace

Practical expertise is often transferred through observation, supervision, repetition, and exposure to unusual situations rather than through a manual that can simply be incorporated into a model. When experienced employees retire or change career, organisations can lose that context even if every formal procedure remains documented.

Pearson argues that training therefore needs more deliberate mentoring, simulation, observation, feedback, and hands-on practice alongside AI skills. Technology may make information easier to retrieve, but access to information is not the same as knowing when it applies.

The problem echoes patterns already appearing in office work. Techopia recently examined IBM research on the additional checking and exception handling created by enterprise AI, which can increase the value of judgement even where some of the original task has been automated.

In practical occupations, that effect can be stronger because the final output is physical. A worker who receives a technically plausible recommendation without enough experience to challenge it may become faster without becoming safer.

Training cycles trail technology cycles

Formal occupational standards can take years to revise, whereas AI functionality can be added to workplace software within months. Equipment suppliers, manufacturers, health systems, and other employers may therefore deploy automated assistance before qualification frameworks or college programmes have incorporated the same tools.

Workplace learning has to absorb more of the gap, which means employers need clear rules around acceptable AI use, mandatory verification, product-specific training, and the role of experienced workers in mentoring. A general AI-awareness course is unlikely to transfer automatically into a specialised technical occupation.

Entry-level work also needs attention because routine tasks have traditionally provided some of the repetition through which people learn patterns and accumulate judgement. Automating too much of that work before training is redesigned can remove part of the mechanism through which the next generation becomes experienced.

Well-designed systems could also support learning by making simulations, examples, procedural guidance, and expert knowledge easier to access. Whether AI accelerates competence or hollows it out will depend partly on how organisations redesign work around the technology.

Workforce AI becomes an operating issue

Pearson’s research combines labour-market and AI-exposure analysis with surveys, interviews, and roundtables across several countries and occupations. Its conclusions therefore go beyond a conventional employer-sentiment poll, although future impact will still depend on how quickly particular sectors adopt the technology.

The report does not argue that skilled workers are about to disappear. Instead, it suggests that readiness is changing because workers increasingly need to combine occupational competence with judgement about where an AI system can be trusted.

That has practical consequences for workforce planning because training investment has to sit closer to technology investment, while experienced staff need time to transfer knowledge before they leave. Implementation teams also need input from people performing the work rather than designing automation solely around what a model appears able to do.

Britain’s replacement-demand figures raise the stakes because organisations have to absorb large numbers of new entrants while accumulated experience is leaving. Hospitals, industrial plants, infrastructure operators, and technical services may therefore become some of the most important tests of whether AI improves work without weakening the knowledge needed to catch its mistakes.