AI postings grew 61% in a year when UK vacancies fell 6.6%. So do you buy those skills, or build them?

Most AI workforce research is either vendor marketing or a survey of 400 executives telling you what they intend to do. This one is different, and it is worth ten minutes of your attention. PwC's 2026 AI Jobs Barometer, published on 15 June 2026, analysed more than a billion job advertisements across 27 countries. It measures what employers actually advertised and paid, not what they said they were planning.

One note before the numbers, because it trips people up. Everything below is the UK cut. PwC also publishes a global figure, and the two are very different: the global average AI wage premium is 62%, against 34.2% in the UK. If you see 62% quoted somewhere, that is the world, not Britain.

What PwC found in the UK

Read the first two together, because that is the whole story in two numbers. In a shrinking job market, one category grew by nearly two thirds. Specialist AI roles now make up 2.2% of all UK postings, up from 1.3%.

The ratio that should decide your strategy

Here is the finding almost every write-up buried, and it is the one that matters most if you are trying to work out what to actually do.

PwC splits AI roles in two. AI developers build the technology. AI users apply it effectively within a field of expertise, so a marketer, a finance analyst, an operations manager or an HR lead who knows how to get real work out of these tools.

User roles grew 65.8% and are now the majority of all AI-related demand in the UK. Developer roles grew 21.6%. The pattern holds across sectors, and in government and the public sector 97.3% of AI roles are user roles.

It is worth saying what an AI user role actually is, because the phrase sounds vaguer than the job. It is the finance analyst who automates a reconciliation that used to eat two days a month. The bid writer who builds a reusable process instead of starting every submission from nothing. That is the job the AI & Automation Practitioner Level 4 exists to produce, and it is the standard we build almost everything else around.

PwC's Chief Technology and Innovation Officer, Claire Reid, put the implication plainly.

She also noted that "the experimentation phase is over and businesses want to scale and embed the technology properly". That is a different job from the one most firms have been doing for two years, and it needs different people.

Build or buy, with actual numbers

Put the ratio and the premium together and the commercial question resolves itself for most employers.

We should be straight that we sell the build option, so weigh that accordingly. But the argument does not rest on our opinion. It rests on PwC's ratio between the roles employers are actually advertising for and the roles most people assume the shortage is in.

What "building" actually looks like

"Build the capability" is easy to write and vague to act on, so here is the concrete version. Every route below is fundable through the Growth and Skills Levy, and none of them requires a coding background.

If you are not sure which of those fits, our guide to AI apprenticeship levels compared puts them side by side with entry requirements.

The two-track market, and which track your roles sit on

Every role you have sits on one of two tracks, and which one decides what you should train for.

• AI amplifies the expertise. It removes routine work so the person spends more time on judgement and decisions. These roles grew 39% since 2018.
• AI simplifies the task. The job becomes easier for non-experts to do. These roles grew 17%.

Both are growth, but they are opposite propositions for the person doing the job. The first track pays more for judgement. The second track invites more people to compete for the work.

That distinction should decide what you buy. People on the amplified track need judgement, governance and the confidence to decide, which is what the Level 5 leadership units are built for. People on the democratised track need something that differentiates them from everyone else who can now do the same job, which is the practical, build-something capability of the Level 4.

What it is doing to entry-level roles

The most uncomfortable finding concerns young people, and it is short.

Entry-level jobs most exposed to AI are now seven times more likely to require traditionally senior skills such as leadership and judgement. Those roles grew 35% since 2019. Other entry-level roles declined 10%. This is from PwC's analysis of 2.4 million US roles, so treat the multipliers as directional rather than a British measurement.

The UK implication is one sentence: the bottom rung is not disappearing, it is being rebuilt higher up the wall.

PwC's own Global Workforce Leader names the mechanism, and uses our word for it.

If the routine work that used to teach people their trade has gone, something has to replace it deliberately. A structured route where someone earns while building both domain knowledge and judgement is one of the few mechanisms that puts the rung back.

In practice that is AI Champion at Level 3 for a first-jobber, stepping up to the Level 4 practitioner. And for apprentices aged 16 to 24 the training is fully funded even once your levy account is exhausted, which removes the budget objection entirely. We covered the other end of the same problem when GCSE resit pass rates fell again this month.

Where the premium is highest, and lowest

Do not take 34.2% as your number. The variation is enormous.

PwC's explanation is neat: premiums run high both where AI skills are scarce and specialised, such as energy and consumer markets, and where they are deeply embedded and strategically vital, such as technology and financial services. Technology, media and telecoms leads on overall AI intensity, followed by financial services and the public sector.

Two readings, depending on where you sit.

Technology, media and telecoms, and financial services. These are the most AI-intensive sectors in the UK, and premiums are high because the capability is embedded and strategically important rather than merely scarce. TMT also has the highest concentration of genuine developer roles, at 11.3%. If you are here, you are competing hardest on price for the smallest part of the market, and your user gap is going unaddressed while you do it.

Public sector. The 12% premium alongside 97.3% user roles tells you something useful. Buying is comparatively cheap, but almost nobody is buying developers. Your gap is capability across existing teams, which is exactly what our public sector AI guide covers.

What to do with this

1. Work out which of your roles are users, not developers

   Do: list the jobs where AI would remove routine work, then ask whether you need someone to build a tool or use one.
   Why: almost every employer overestimates how much genuine developer capability they need, and underestimates the user gap.

2. Price the buy option honestly before you dismiss build

   Do: take a real salary from a real role, add your sector's premium, and add recruitment cost and time to competence.
   Why: "training is expensive" survives mainly because nobody costs the alternative properly.

3. Check what your levy would actually cover

   Do: look at your Growth and Skills Levy position, including that 16 to 24 year olds are fully funded even once an account is exhausted.
   Why: most employers underspend the levy while paying market premiums for the same skills. Both problems have one solution.

Two smaller things while you are at it. Decide which track each role is on, because amplified roles need judgement and simplified roles need differentiation. And if your junior roles now demand senior-sounding skills, decide how anyone is supposed to acquire them at your organisation, because "three years' experience" on an entry-level advert is not a hiring standard.

What we would not claim from this data

Three honest caveats, because a report this useful gets misquoted quickly.

• These are job postings, not hires. They measure advertised demand and advertised pay. Some of those roles will not have been filled, and posted salaries are not the same as paid salaries.
• The 34.2% premium is an average across wildly different sectors. Quoting it at a public sector employer where the figure is 12% would be misleading.
• The entry-level analysis is US data. The direction is instructive. The precise multipliers are not a UK measurement.

None of that undermines the core finding. It just means you should use the numbers for your own sector rather than the headline.

The honest summary

The AI skills market did something unusual last year. It grew 61% inside an economy where vacancies fell, and it repriced the people who have those skills by a third. That is what a genuine shortage looks like when it arrives.

What it is not is a shortage of people who can build models. The shortage is people who can apply AI properly inside a real business, with the domain knowledge to know what is worth applying it to. Those people are mostly not on the market. They are on your payroll, doing a different job, and they are cheaper to develop than to replace.

The employers who work that out first will spend the next two years hiring less and getting more out of who they already have.

Sources: all figures from PwC's 2026 AI Jobs Barometer, UK findings, published 15 June 2026, based on analysis of more than one billion job advertisements across 27 countries and territories. Entry-level findings are from PwC's analysis of 2.4 million US entry-level roles and are indicative rather than a UK measurement. Figures describe advertised roles and advertised pay. TESS Group provides levy-funded AI apprenticeships, so we have a commercial interest in the build option; the underlying data is PwC's and linked above so you can check it.