Three Million Quiet Goodbyes
Yesterday, a number finally stuck. Buried in a Guardian exclusive, the National Foundation for Educational Research’s five‑year Skills Imperative 2035 programme put a time‑boxed estimate on what many employers have been hinting at in earnings calls and org charts: between one and three million lower‑skilled UK jobs could vanish by 2035 as AI and automation absorb tasks that once made the workday tick. It is not a prophecy about a jobless future; the same modelling projects total employment up by roughly 2.3 million. But the growth is in professional and associate professional roles, and the attrition is in the rungs where people have traditionally entered the labour market—administration and secretarial work, customer support, trades adjacent to plant and machine operation. The escalator still runs, just with fewer steps at the bottom.
What changed isn’t the technology—it’s the timetable
We’ve had years of sweeping claims about AI and work. This is different because it narrows the aperture: a decade, a set of occupations, and a displacement band large enough to shape wages, training systems, and local economies. NFER’s headline lands alongside something more unnerving than the raw job count: a skills bottleneck that is already here. The report highlights six essential employment skills—clear communication, collaboration, problem‑solving, organising and planning, creative thinking, and information literacy—and estimates that shortfalls affect 3.7 million workers now, rising to around seven million by 2035 without intervention. Translate that into throughput and you get the scale of the task: on the order of hundreds of thousands of workers per year need to cross a skills threshold that many entry‑level roles didn’t previously require.
Exposure is not destiny, but it is pressure
The counterpoint came inside the same coverage: UK government analysis finds the highest AI exposure in professional jobs—finance, law, management, teaching—while research from King’s College London shows higher‑paying firms shed roughly 9.4% of roles from 2021 to 2025. These aren’t contradictions so much as a reminder that “exposed” and “displaced” are different metrics. In some sectors, AI amplifies output per worker and shifts headcount rather than cutting it. In others, it shaves labour out of routine layers where customer expectations, compliance, and process standardisation make automation easier to deploy. The result is not a single wave, but overlapping currents pulling different cohorts in different directions.
The first rung is moving
On the ground, the signals rhyme. Clifford Chance cut about a tenth of London business‑services roles, with AI explicitly in the frame. PwC scaled back a splashy hiring plan even as it invests in higher‑skilled talent. Employers aren’t necessarily swinging axes; many are simply hiring less into entry‑level and support functions while adding headcount in roles that demand stronger judgment and tool‑use. For a displaced call‑centre agent or a back‑office administrator, that subtle shift changes everything. It turns an application into an audition for skills that weren’t prerequisites a few years ago and inserts a time cost that many workers can’t absorb. For young people and mid‑career switchers, it introduces a new kind of experience tax: the first job expects you to have already automated the first job.
Low‑skill is becoming misnamed
“Low‑skilled” work has never actually been simple; it has been a bundle of predictable tasks that could be learned quickly. AI unbundles those tasks and keeps the parts it can handle. What remains skews towards human judgment, context, and coordination—ironically, the very “soft” capabilities that are hardest to teach at scale and slowest to credential. As NFER defines them, those essential skills are no longer a nice‑to‑have set of traits. They are the hiring firewall for the new entry level, and they will decide whether displaced workers move sideways into adjacent roles or drift into inactivity.
The arithmetic governments can’t duck
The modelling is blunt about direction: fewer roles at the base, more roles above it, and a widening skills gap over a ten‑year horizon. The policy implication is not just curriculum reform; it is capacity. Further education, workplace learning, and assessment pipelines need to process millions, not thousands. Even assuming overlap between the at‑risk cohort and those with essential‑skills deficits, closing gaps for a workforce measured in millions means sustained training volumes in the high hundreds of thousands per year, every year, with accreditation that employers actually reward. That requires money, instructors, digital infrastructure, recognition of prior learning that shortens pathways, and incentives that make it rational for firms to convert productivity gains into internal mobility rather than external shedding.
Uneven impacts, uneven preparedness
The story’s real novelty is not the topline number. It is the message that the next decade will be less about whether jobs exist in aggregate and more about who can cross the widening skills threshold fast enough. Regions concentrated in administrative services or routine plant operations will feel stress earlier. Firms with strong internal academies will rotate people; firms without them will rely on the market and pause hiring. Workers with the time and cash to retrain will find the new professional roles that the models say are coming; those without will bounce between short‑term gigs and training that doesn’t map cleanly to vacancies.
What to watch next
NFER’s final report launches in London with a simple framing—cradle‑to‑grave reskilling—that sounds humane until you price it and schedule it. The hard question is no longer whether AI takes work; it is whether the UK can build a skills‑conversion machine that runs faster than adoption. Yesterday gave us the number. The next few months will tell us whether anyone is willing to build the apparatus that makes it survivable.
