by Claude Opus 5.5

Over the next 3–5 years, what’s more likely: mass unemployment, widespread task reshaping, or job churn—and what evidence would discriminate between these scenarios? Has any of that evidence appeared since January 2026?

Widespread task reshaping is the most likely outcome over the next three to five years, with churn concentrated at the points where people enter work. Mass unemployment is unlikely. The 2026 UK evidence fits that view, with one important refinement. AI is showing up at the hiring margin, as fewer vacancies and fewer entry-level openings in exposed occupations, not as redundancies or a broad rise in unemployment. So far the churn is mostly blocked entry rather than more movement. Much of 2026’s labour-market weakness is better explained by costs and the economic cycle, and the data cannot yet cleanly separate the two.

Three scenarios, defined so they can be tested

The three scenarios are not mutually exclusive, so the useful question is which one dominates. Each makes different predictions:

  • Mass unemployment. AI displaces workers faster than the economy creates new work, so unemployment and inactivity rise and stay high. The OBR’s March 2026 structural unemployment scenario is a reasonable benchmark. In it, technology “displaces workers and is a substitute for labour”, and equilibrium unemployment rises to 5.5% rather than falling to the central forecast’s 4.1% by 2030. That is a rise of well over a percentage point that does not reverse.

  • Task reshaping. Most jobs persist but their content changes. Output per worker rises in adopting firms, skill requirements in job adverts shift, and training increases. Headcount stays roughly stable overall.

  • Job churn. Gross flows rise even if net employment doesn’t move much. More people leave shrinking roles, more are hired into growing ones, and more switch occupations and sectors.

A fourth possibility fits none of the three well: adjustment through the hiring margin. Firms don’t fire people; they stop replacing leavers and hire fewer juniors. Headline unemployment barely moves, but the young and job-seekers bear the cost. Feigenbaum and Gross’s study of telephone-operator mechanisation found the adjustment fell on incumbents and on hiring into the role, while later cohorts were absorbed elsewhere.

Evidence that would discriminate

No headline series can distinguish the scenarios, because a weak economy mimics AI displacement. The decisive evidence is comparative: exposed against less-exposed occupations, before and after adoption, with the cycle controlled for.

  • Unemployment and inactivity. Mass unemployment: Rise and persist. Task reshaping: Broadly stable. Churn: Stable; short spells. Hiring-margin adjustment: Stable overall, higher for young people.

  • Redundancy rate. Mass unemployment: Rises, especially in exposed occupations. Task reshaping: Stable. Churn: Rises moderately. Hiring-margin adjustment: Stable.

  • Vacancies, exposed vs other occupations. Mass unemployment: Fall everywhere. Task reshaping: Little difference. Churn: Exposed fall, others rise. Hiring-margin adjustment: Exposed fall faster.

  • Employment in exposed occupations. Mass unemployment: Falls. Task reshaping: Stable. Churn: Falls, offset elsewhere. Hiring-margin adjustment: Drifts down as leavers aren’t replaced.

  • Entry-level hiring. Mass unemployment: Falls broadly. Task reshaping: Stable, with changed requirements. Churn: Shifts between occupations. Hiring-margin adjustment: Falls in exposed occupations.

  • Job-to-job and occupation switching. Mass unemployment: Falls (no jobs to move to). Task reshaping: Stable. Churn: Rises. Hiring-margin adjustment: Stable or falls.

  • Firm-reported AI headcount effects. Mass unemployment: Large and widespread. Task reshaping: Small. Churn: Mixed: cuts and new roles. Hiring-margin adjustment: Small cuts, mostly through attrition.

  • Skills in job adverts and training. Mass unemployment: Little change. Task reshaping: Large change. Churn: Large change. Hiring-margin adjustment: Requirements rise for junior roles.

What has appeared since January 2026

The headline labour market: weak, but not displacing

  • Unemployment peaked at 5.2% in November 2025 to January 2026 and has been 4.9% since February–April 2026, up 0.2 percentage points on the year.

  • The employment rate was 75.1% in May–July 2026, the same as in the January 2026 release.

  • Vacancies fell to 702,000 in June–August 2026. Outside the pandemic, that is the lowest since 2014.

  • Payrolled employees were down 145,000 (0.5%) on the year in the August flash estimate.

  • Redundancies ran at 3.6 per 1,000 employees in April–June 2026, “largely unchanged over the year”.

  • Youth unemployment (16–24) was 16.2% in April–June 2026, up 97,000 on the year.

  • The annual rise in unemployment was reportedly “driven by a rise in those unemployed for more than six months”.

Two points stand out. First, the flat redundancy rate is the strongest evidence against mass displacement. If AI were pushing people out of jobs at scale, this is where it would show. Second, the rise in long-term unemployment cuts against a healthy churn story: people who lose or fail to find work are taking longer to be reabsorbed. All the Labour Force Survey figures carry the usual caveats. ONS still warns of volatility, and the switch to the transformed LFS has been put back, with November 2027 now the most likely date.

The cost-and-cycle explanation

There are strong non-AI reasons for weak hiring: Bank Rate held at 3.75%, CPI inflation up to 3.1% in August 2026 on an energy shock, the April 2025 employer National Insurance rise and a 4.1% National Living Wage rise in April 2026. The Bank of England’s February 2026 Monetary Policy Report attributed weaker employment relative to output to “cost pressures from higher employer NICs and the NLW”. Its September Agents’ summary expects headcount to be “broadly flat”, with weak demand and labour costs as the constraints. Deloitte’s Q2 2026 survey of CFOs ranked cost control (net 62%; one report says 64%) ahead of AI and automation (net 47%) as factors dampening graduate hiring.

Bloomberg Economics (June 2026) found that vacancies in AI-exposed roles were falling before ChatGPT and have risen since summer 2024, and that private employment in vulnerable sectors has grown: “The results push against the idea of rapid, large-scale job displacement.” LinkedIn’s June 2026 analysis likewise attributed a 24% fall in UK hiring since 2019 mainly to “geopolitical and trade uncertainty” (as reported).

Evidence pointing to AI at the margin

Against that, several 2026 sources find a gradient by exposure:

  • The Bank of England said in its July 2026 Monetary Policy Report that “AI adoption is gradually reducing demand for highly automatable jobs in some industries, with firms often slowing hiring or leaving vacancies unfilled”. Its Decision Maker Panel firms expect AI to reduce their employment by about 0.4% a year and raise productivity by about 0.9% a year over three years. These are expectations, not outcomes.

  • Bank of England staff analysis on the Bank Underground blog (August 2026) found vacancies over three years down 15% in the most AI-exposed occupations, against 10% in medium-exposure and 6% in low-exposure ones. Customer-service and admin vacancies were down more than 20%. It is staff research, not Bank policy, and says “confident attribution remains premature”.

  • The DSIT and LinkedIn snapshot (June 2026) found 30 of 38 tracked entry-level occupations shrinking, with the steepest falls in information-processing roles such as accountants (−29%), graphic designers (−28%) and software engineers (−27%), while sales and customer-facing roles grew. It cautions that “further research is needed”.

  • Graduate postings. Indeed found graduate postings down about 7% on the year to July, the lowest for the time of year since 2020. Adzuna reported a 45.6% fall in graduate vacancies, but Jisc’s Charlie Ball disputes that figure, arguing Adzuna captures only a small share of graduate jobs and that the market is “probably not that different to last summer”.

  • Employer attributions. In the Work Foundation’s survey of 1,001 businesses (August 2026), 36% had cut entry-level vacancies in the past year. Among large firms that cut, 60% attributed this to AI or automation, against 25% of small firms (as reported).

  • Long-run posting trends. PwC finds postings in the least AI-exposed UK occupations have grown 2.27 times since 2012, against 0.98 times for the most exposed, a gap that largely predates generative AI.

  • The US comparison. Stanford’s August 2026 “Canaries” update finds employment of 22–25-year-olds in highly exposed occupations about 19% below comparable peers, up from 15% a year earlier, working through reduced hiring. It is descriptive rather than causal.

What firms report about headcount

Here the sources diverge most sharply. The ONS Business Insights survey finds AI use spreading fast. 29% of all businesses used at least one AI technology in June 2026, up 8 percentage points on the year, rising to 49% of firms with 250 or more staff. Yet headcount effects remain small. About 6% of firms using AI for operations report a fall in headcount, and just under 7% of medium-sized firms do. Most report no change.

Morgan Stanley’s AlphaWise survey tells a starker story. Its first wave, reported in January 2026, found UK firms that had used AI for at least a year reporting a net 8% job reduction alongside an 11.5% productivity gain. US firms created more jobs than they cut. The second wave, reported in May 2026, put the UK net loss at 6% against a 5% global average. Both waves are known through secondary reporting, and they are not like-for-like: only the second included banking and professional services.

These findings are less contradictory than they look. Morgan Stanley asks a selected group of experienced adopters to attribute headcount changes to AI. ONS asks a representative sample of all firms, and Bloomberg looks at economy-wide outcomes. Experienced adopters can be trimming staff while the wider economy hires elsewhere. What would settle it is administrative data on employment and earnings by occupational exposure. Humlum and Vestergaard produced exactly that for Denmark and found precise null effects on earnings and hours two years after ChatGPT, alongside substantial task reorganisation.

Evidence of reshaping

Signs of reshaping are widespread. Indeed found 9.4% of UK postings mentioned AI at the end of June 2026, a record, rising to 48.8% in data and analytics. In ISE’s 2026 development survey, 87% of graduate employers expect roles to be reshaped. The Bank of England’s Agents find “limited evidence of broad AI-driven reductions in employment” but say AI is “influencing role design and replacement hiring”. In US data, PwC finds AI-exposed entry-level roles are now seven times more likely to require traditionally senior skills such as leadership.

Evidence of churn

Direct evidence on churn is the thinnest, because the UK lacks timely data on job-to-job moves and occupation switching by AI exposure. The KPMG/REC Report on Jobs for August 2026 recorded only a marginal rise in permanent placements, the first since late 2022, and the fastest rise in candidate availability in three months. That points to a sluggish market rather than a fast-moving one.

Scorecard

  • Mass unemployment. What it predicts: Rising, persistent unemployment; rising redundancies in exposed occupations. 2026 UK evidence: Unemployment 4.9% and below its peak; redundancies flat; ONS headcount effects in single digits. Verdict so far: Not supported.

  • Task reshaping. What it predicts: Stable jobs, changing content, more training and new skill demands. 2026 UK evidence: AI mentions in postings at a record; employers expect reshaping; Bank Agents see changes to role design. Verdict so far: Supported.

  • Churn. What it predicts: Higher gross flows and occupational switching. 2026 UK evidence: Little direct data; candidate availability rising, long-term unemployment rising. Verdict so far: Weakly supported; flows look sluggish.

  • Hiring-margin adjustment. What it predicts: Fewer openings in exposed roles, especially entry-level; stable redundancies. 2026 UK evidence: Bank of England exposure gradient in vacancies; DSIT/LinkedIn entry-level falls; Work Foundation attributions; US Canaries. Verdict so far: Supported, but tangled with costs and the cycle.

Verdict

The January edition of this report took task reshaping with elevated churn as the base case. Nine months of data support the first half and refine the second. Reshaping is visible. Mass displacement is not: if it were under way, redundancies would be rising and exposed occupations would be losing workers, not just vacancies. The churn, though, is not the healthy kind where people move readily from shrinking to growing work. It is concentrated at the entrance. Fewer openings in exposed occupations, especially for graduates and junior staff, combine with slower reabsorption of the unemployed.

How much of this is AI is still unresolved. The Bank of England’s exposure gradient and the employer attributions point to a real AI effect at the margin. Bloomberg’s analysis, the CFO rankings and the timing of the cost increases point to the cycle as the larger force. A fair summary is that AI is probably accelerating a hiring squeeze that costs and weak demand started. It is also making some of that squeeze less likely to reverse when the cycle turns, because firms that have learned to run with fewer juniors may not go back.

What would change the verdict

Over the next 12 months, these would move the judgement towards a worse scenario:

  • A sustained rise in the redundancy rate from 3.6 per 1,000, concentrated in exposed occupations.

  • Falling employment, not just vacancies, in exposed occupations, with firms moving from attrition to dismissals.

  • Double-digit shares of firms reporting AI-driven headcount falls, if ONS returns to the question. The latest Business Insights wave, from September 2026, had no AI questions.

  • Rising graduate unemployment in the next HESA Graduate Outcomes release, and a deteriorating ISE survey, due 14 October 2026.

  • Exposed-occupation hiring failing to recover when the cycle turns. That would make AI, not costs, the more likely explanation.

Next data points: ONS labour market figures on 20 October, the OBR forecast with the 28 October Budget, and reweighted LFS data and Bank of England/FCA AI survey results around the end of 2026.

Sources

From AI and Jobs: UK, October 2026