Why the AI Skill Centre Exists

The adoption race

The container and the lesson it left behind

In April 1956, a converted oil tanker called the Ideal X sailed from Newark to Houston carrying 58 metal boxes. Malcom McLean, the trucking entrepreneur who chartered her, was not trying to change the world. He was trying to cut the cost of getting cargo from a lorry onto a ship. At the time, loading a ton of loose cargo cost $5.86. Loading a ton inside one of McLean's boxes cost 16 cents.

The container was an American invention, built by an American, first sailed between American ports. But much of the wealth it created landed elsewhere. The great winners of containerisation were Singapore, Rotterdam, Hong Kong and, later, Shenzhen: places that looked at the box and reorganised everything around it. Ports, customs, labour and logistics rebuilt to exploit a technology none of them invented. The inventors changed the technology. The adopters captured much of the economic upside.

Artificial intelligence looks set to follow the same pattern, and the question facing every country is not who will build it, but who will reorganise around it first.

This paper argues that the UK should not try to win AI by owning the technology stack, that its strongest route to advantage runs through its services economy, and that one constraint deserves far more attention than it gets: whether anyone can tell who in the workforce is competent with these tools and who is not.

By AI competence we mean something specific: the demonstrated ability to select, direct, check and govern AI tools in real workplace tasks, while protecting data, recognising the tools' limits, and keeping a human accountable for the outcome. Not familiarity with one product. Not attendance on a course. Demonstrated ability, assessed independently, with a public way to prove it.

The hidden bottleneck

Competence nobody can verify

Start with a picture that will be familiar to anyone running a mid-size services firm. A 200-person insurance broker. Half the staff are already using AI, mostly well, some recklessly, much of it on personal accounts the firm cannot see. The board signed an AI policy eighteen months ago that nobody can enforce. A junior employee pasted client claims data into a free chatbot in a moment of deadline pressure. The firm does not know. A senior manager's email was convincingly spoofed with AI-generated voice and video, requesting an urgent transfer. It very nearly worked. A tender response arrived from a competitor that was visibly better-drafted than anything the firm could produce with its current tools. The gap was real and everyone felt it.

The scale of this is not speculation. Microsoft and LinkedIn's global survey of knowledge workers found 75% already using generative AI at work, and 78% of those users bringing their own tools rather than using anything their employer provided — a figure that rises in smaller firms. Adoption has outrun governance almost everywhere.

Notice what the bottleneck is not. It is not access to AI: the tools are on every desk already, by subscription or by stealth. It is not enthusiasm: staff adopted these tools faster than their employers could write policies about them. Three different things are getting confused, and separating them is the heart of this paper.

Workforce competence is not the only constraint on adoption. Procurement, data governance, cyber security, legacy systems and liability all matter. But competence is the constraint that touches every desk, every day, in every services firm, and it is the one for which the market currently has no trusted measure. That is the gap the Institute exists to close.

Why certification

And not just more training

A fair challenge: why does this need a certificating body? Firms can train internally. The big AI vendors run their own academies and badges. Universities and online platforms issue credentials by the million. Sector regulators may eventually define their own requirements. All true. None of it solves the problem in the picture above, because all of it answers a different question.

Training teaches; it does not verify. A completion certificate proves attendance, not ability, which is why nobody in our composite firm believes theirs. Vendor badges prove familiarity with one company's product, issued by a party with an obvious interest in the answer, and they travel poorly between tools and jobs. Internal assessment can be excellent, but it helps one employer and is invisible to clients, insurers and future employers. University credentials can be valuable, but they are slow, expensive and calibrated to a general workforce, not the specific demands of mid-size services firms.

Markets have faced this structure before — where individual capability creates risk that buyers cannot inspect — and they have converged on the same machinery each time: a published standard, independent assessment, and a public register. Drivers, electricians, accountants, pilots.

We do not claim the analogy is perfect. Those are regulated roles where incompetence causes visible harm, and AI competence is broader, more contextual and faster-moving. Which is exactly why the assessment has to be renewed every year, and why the standard has to be updated twice a year. But the basic structure — assessed against a published standard, recorded on a public register, renewable — is the right one for this problem.

The UK's position

Services make the UK unusually exposed to the upside

Why does this matter beyond any single firm? Because of what the UK economy is made of. The UK is the world's second-largest exporter of services — roughly £508bn of them in 2024, behind only the United States. Exports to the US alone were worth around £137bn. Services account for about four fifths of the UK economy and a majority of its exports. In financial services specifically, the UK runs the largest trade surplus of any country.

Look at what that economy is made of and the AI exposure becomes obvious. A commercial law firm drafting and reviewing contracts. An insurance broker summarising risk submissions. An accountancy practice reconciling and reporting. A consultancy synthesising research into recommendations. An architecture practice producing specifications and tender documents. These businesses sell thinking: drafting, analysis, review, judgement and advice. And drafting, analysis and review are precisely what the current generation of AI tools does.

For some economies, the AI productivity story begins with manufacturing, logistics or industrial automation. For the UK, the opportunity lands unusually directly on its largest export industries. Few countries have more to gain from getting workplace AI right, and few have more to lose from getting it wrong — because those same industries run on something AI can quietly destroy: client trust.

National strategy

The UK's advantage is adoption, not ownership of the stack

The national debate keeps asking how the UK catches up on AI infrastructure. The more useful question is whether that is the right race to enter at all. UK industrial electricity — the raw feedstock of AI data centres — was the most expensive of any IEA member country reporting data in 2024, and by mid 2025 large industrial users were paying around 125% more than the EU-14 median. Meanwhile the largest US cloud and platform companies reported AI-related capital expenditure of roughly $300bn in 2024 alone.

None of that means the UK has no AI capability. It plainly does: world-class research universities, DeepMind's London roots, a serious startup base, national supercomputers, and strengths in fintech, legaltech, insurtech and life sciences. The precise claim is narrower and more useful. The UK cannot rely on owning the full AI stack — the models, the data centres and the chips — as its route to national advantage. Frontier-scale infrastructure is increasingly a capital and energy game that favours the United States and, to a lesser degree, China.

What the UK can do is reorganise around the technology faster than anyone else. That is not the consolation prize. As the container showed, it is where much of the value goes.

The five layers

What a credible standard has to do

The route from here to there runs through five layers, and it is worth being explicit about them.

A certification body sells trust before it sells anything else, so it is reasonable to ask how the standard will earn its legitimacy. We think a credible workplace AI standard has to meet seven tests, and we are building the Institute to meet all of them.

What we are building

The Institute, the standard, and the library

First, what we are and what we are not. Our mission is to equip the UK workforce as quickly as possible, so we start broader than a pure certificating body. Alongside the standard, the Institute publishes a learning library: open learning and practice materials mapped to every competency we assess, across all five levels. We do this because speed matters more than tidy boundaries; the training market that should exist around a standard like this does not exist yet, and waiting for it would be wrong.

That breadth creates an obvious objection: a body that trains and certifies is marking its own homework. We answer it with separation, and the separation is structural. The learning materials are open to everyone, including candidates who train elsewhere or not at all. Practice questions never appear in certification papers; the assessment draws on a separate, independently reviewed question bank and is taken under controlled conditions against the published standard. Holding our pass rate to a meaningful failure rate is the structural answer to the conflict of interest, and we will be judged on the published numbers.

Certification comes at five levels, each answering a different question an employer needs answered.

On rigour, one design commitment matters above the rest: the assessment can be failed. We set our design targets expecting a meaningful share of first-attempt failures at Foundation, and we will publish actual pass-rate data from the first cohorts onwards, because a credential that everyone passes is not a credential. Where this paper's claims rest on our own intentions rather than evidence, we have tried to say so plainly, and we expect to be judged on the published numbers.

For organisations, the Institute will offer a workforce mark for firms where at least 80% of in-scope staff hold current certification, with the scope, exclusions and headcount basis published on the register and checked annually. The intended uses are concrete: a line in a tender response that a client can verify independently, an answer to the due diligence questionnaire that is more than assertion, and evidence that insurers could use when assessing professional indemnity risk.

What success looks like

How we will measure ourselves

A founding paper should say what would count as failure, so here is what we will measure ourselves against.

Failure would be a register that nobody checks, a pass rate of 95% or above, a workforce mark that organisations display without anyone verifying it, and a standard that drifts behind the technology because revision is too slow. We are aware of these failure modes and the structure of the Institute is designed against each of them.