UK AI JobsExpected · in use · happened
Grow vs shrink · the split

Same AI exposure.
Opposite futures.

At AI exposure 7 or higher, one group of roles grows because cheaper AI output creates more demand (software, engineering, creative tech). Another group shrinks because the output volume is fixed (admin, customer support, medical secretaries). Same exposure score, different demand dynamic, opposite 2035 curves.

Quick glossaryHow AI is used, measured: augmentation (AI assists) vs automation (AI does the task).AI exposure 0–10, scored: how much of the work AI can do.Elasticity 0–10, scored: whether cheaper supply grows demand.Full methodology →
Jevons wages at AI 7+
£92B
109 roles · 3.6M workers
Anti-Jevons wages at AI 7+
£56B
49 roles · 3.5M workers
Total exposed wage pool
£269B
169 roles at AI 7+
Anti-Jevons is 69% female
69%
1.0M women in high-exposure female-dominated roles
Exposure × outlook · all 738 occupations

The fork opens past exposure 6.

Each bubble is a SOC occupation. Area ∝ UK employment. Colour by zone. The fork appears past AI exposure 6: the upper-right is where Jevons pays off, the lower-right is where Anti-Jevons bites.

Zonejevonsanti-jevonsambiguousbaumolsafebubble area ∝ workers
The Jevons cliff · UK developer headcount, 2024 → 2035

Jevons gives. Then agentic systems take it back.

A single scenario illustrating how demand elasticity can reverse. Under this model, UK developers boom from 741.6K to a peak of 867.7K by 2027 as cheap AI-assisted code makes new projects viable, then compress to 111.2K by 2035 as agentic systems remove the human-in-the-loop. That's a 87% displacement from peak: the Jevons paradox playing out on a time delay.

Phase toneBoom · Peak · PlateauTurn · Cliff · Collapse · Remnant
2025–27Boom · Peak · Plateau

Latent demand opens up. Cheap AI assistance expands the addressable surface.

2028The Turn

Zero-knowledge entry: non-coders spin up production code. Junior layer cut first.

2029–30Cliff · Collapse

Direct business-to-machine. The IDE-as-interface dissolves; the model is the build pipeline.

2031–35Hyper-Verifier · Equilibrium

Small senior-verifier remnant. In this scenario, around £41B/yr in UK coding wages no longer goes to humans.

Where the split matters most

Top 10 of each branch by wages at stake.

Anti-Jevons · declining
£42B combined
#RoleWorkersAIOutlookWages
01Admin assistant473.3K9-3%£11.6B
02Customer service assistant337.1K7-6%£8.4B
03Office manager125.1K7-3%£4.1B
04Customer services manager90.4K7-6%£3.5B
05Executive assistant69.2K7-3%£3.0B
06Purchasing manager62.7K7-3%£2.8B
07Bilingual secretary69.2K9-3%£2.4B
08Sales administrator80.3K8-6%£2.0B
09Retail buyer52.4K7-6%£2.0B
10E-commerce manager42.7K8-6%£1.8B
Jevons · expanding
£48B combined
#RoleWorkersAIOutlookWages
01Network manager231.1K7+20%£11.6B
02Artificial intelligence (AI) engineer94.4K9+20%£5.2B
03Software developer94.4K9+20%£5.0B
04Computer games developer94.4K9+20%£4.6B
05App developer94.4K9+20%£4.5B
06Web developer94.4K9+20%£4.1B
07Data scientist68.8K9+20%£3.9B
08IT support technician110.1K7+20%£3.3B
09Business analyst75.3K8+20%£2.9B
10Technical architect43.0K8+20%£2.8B
The gender gap

AI displacement is not gender-neutral.

Twentieth-century occupational sorting routed women into clerical, administrative, and customer-service work, the exact mid-skill roles most amenable to LLM automation. Men clustered into trades (Baumol-protected) and engineering (Jevons-expanded). So the same AI exposure score predicts very different labour outcomes by sex.

Anti-Jevons (AI 7+, declining)
1.3M women · 69%596.0K men · 31%
Jevons (AI 7+, growing)
759.0K women · 32%1.6M men · 68%
Female-dominated high-exposure roles
RoleWomen% FAIOutlook
Admin assistant365.0K77%9-3%
Office manager95.0K76%7-3%
Bilingual secretary67.0K97%7-3%
Medical secretary43.0K95%8-3%
Health records clerk51.0K77%8-3%
Payroll administrator118.0K70%9+3%
Bookkeeper118.0K70%9+3%
Scenario lab · wage inequality under AI

Pull the levers. Watch the Gini move.

Four dials, four AI-exposure zones, one national wage pool. Every number recomputes live across all 738 occupations. Base pay, UK employment and per-occupation zone are held constant.

Scenario inputs

Drag to model the split.

Four levers, applied by AI-exposure zone. Everything else stays fixed: occupations, employment, base pay.

20% cut
AI 7+ roles with inelastic demand: admin, customer support, medical secretaries
25% boost
AI 6+ roles where cheaper output expands demand: software, creative tech
10% boost
AI 4–6 roles: assisted, not replaced (managers, technicians, analysts)
5% boost
AI 0–3 trades and care: wages rise with surrounding inflation
Total UK wage pool
£893.6B
base £845.6B
Δ vs baseline
+£48.0B
+5.68%
Gini (employment-weighted)
0.252
base 0.226
Gini shift
+2.63 pp
more unequal
Wage pool by zone, before → after
Anti-Jevons (AI 7+, declining)£163.3B£130.7B (-20.0%)
Jevons (AI 7+, growing)£180.8B£226.0B (+25.0%)
Mid-range (AI 4–6)£224.2B£246.6B (+10.0%)
AI-safe (AI 0–3)£260.9B£274.0B (+5.0%)

Gini is computed per-occupation weighted by UK employment (738 roles, 24.7M workers). The baseline Gini (0.226) reflects pay inequality across occupations, not individuals. A positive shift means inequality rises.

Already visible · UK Big Four graduate intake

The entry-level layer is being cut first.

Graduate programmes are the Anti-Jevons canary. AI has replaced the analytical scaffolding (spreadsheets, first drafts, decks) that used to justify large cohorts. Cuts range from −11% to −29% across the Big Four between 2023 and 2024.

Firm20232024ΔStated reason
KPMG1,399942-29%AI + low attrition
Deloitte1,7001,400-18%Restructuring + offshore
EY1,8001,600-11%Partner cull + consulting cuts
PwC1,6001,500-6%AI + offshoring to acceleration centres
Job postings · 2022 → 2025

High-exposure postings fell almost twice as fast.

High AI exposure-38%
McKinsey UK analysis 2022-2025
Low AI exposure-21%
McKinsey UK analysis 2022-2025
Wage effects in academic literature
Moderate (10%)+4%
High exposure-35%

IMF / academic meta-analysis, pooled effect across 2022–2025 studies.

Next

The regional picture: where the split hits hardest.