Britain is building the AI infrastructure. Now it needs the people

The UK does not need to win every part of the global AI race. But it can become the country that sets the standard for putting AI to work responsibly in the world’s most demanding industries. Compute give us capacity. A specialist, trusted and widely accessible workforce will determine whether we can convert that capacity into lasting economic value.

The appointment of Kanishka Narayan as the UK’s Minister for Artificial Intelligence comes at an important moment. The first chapter of AI was dominated by proving what the technology could do. The next will be defined by whether organisations can deploy it safely, responsibly and at scale.

The government has rightly invested in the foundations of an AI economy, from compute capacity to wider skills programmes. But infrastructure alone will not create a competitive advantage. Britain must now invest just as seriously in the specialist workforce required to turn capable models into trusted, accountable services.

The real AI bottleneck

For many businesses, access to capable AI is no longer the principal constraint. Models can be licensed or adapted far more quickly than organisations can redesign the processes around them. The bottleneck has moved to the governance layer: the people, controls and rules connecting a model to a real-world decision.

This is particularly visible in regulated sectors. Banks are applying AI to fraud detection and credit assessments, while healthcare organisations are exploring its use in triage and operational decision-making. These are not environments in which an AI system can simply be switched on. Its decisions must be explainable, auditable and open to challenge, with clear human accountability for the outcome.

That requires more than general AI literacy. It requires professionals who understand both how a model behaves and the regulatory duties governing a particular sector. The person supervising an AI-supported credit decision needs knowledge of financial regulation and model risk. Someone overseeing AI in healthcare must understand clinical safety, patient rights and operational realities.

Some businesses are already building this expertise themselves. However, we need stronger education and training pathways that equip people across the workforce to govern AI effectively. Without deliberate investment, demand will outstrip supply and access to these careers will be restricted to a narrow group who entered the field early.

Agentic AI raises the stakes

This gap will become more pressing as enterprise AI adoption becomes more common. Unlike a conventional tool that waits for each instruction, an agentic system can plan tasks, take actions and make or recommend decisions across a process.

The valuable human skill therefore shifts from operating a tool to supervising a decision-making system working at a speed and volume no human team could review manually. Organisations will need people who can determine where human intervention is required, test the system’s behaviour, investigate failures and take responsibility when outcomes are challenged.

Countries that treat this simply as another productivity software rollout risk missing the larger change. Agentic AI is as much a governance and workforce issue as a technology issue.

The UK now has an opportunity to take a leadership role in defining what good AI governance looks like by developing the skills, training and professionals needed to put it into practice. However, left alone, agentic AI may create a relatively small number of highly paid supervisory positions. Building accessible routes into those roles is therefore both an economic necessity and a question of opportunity.

A vertically specialised talent strategy

The UK has already introduced broad AI training and apprenticeship initiatives. That is a useful foundation, but the next step must be more targeted. We need nationally recognised apprenticeships, standards and credentials for professionals who can deploy and govern agentic AI in sectors such as banking and healthcare.

These pathways should combine technical understanding with sector regulation, ethics, operational risk and practical accountability. They should also be accessible to people entering from compliance, operations, clinical, legal and risk backgrounds—not reserved for computer scientists.

Crucially, the standards should be nationally governed and independent of the companies supplying the models. If technology providers both rent out the systems and define who is qualified to oversee them, then Britain will surrender control of the layer where it is best placed to lead.

‍