Indeed's Hiring Lab has documented a structural split in the UK labor market: occupations that can absorb AI tools are adding jobs and wage premiums, while occupations that cannot are stagnating or contracting. The productivity gains are real and measurable. Who captures them — and who carries the cost of transition — is a question the UK government's current policy framework has not answered.
Indeed did not describe a success story. They described a divergence.
That distinction matters. When a jobs platform — one whose revenue depends on hiring volume — flags a two-speed labor market, the commercial logic is transparent: the top lane generates premium listings and higher recruiter fees. The bottom lane churns lower-margin placements or exits the formal hiring process entirely. Read Indeed's findings with that incentive structure in mind. They are not wrong. They are not disinterested.
The UK labor market right now is producing two simultaneous realities. In professional services, software, finance, and media, AI tools are lifting individual output and, in some cases, salary. In retail, logistics, administrative support, and entry-level customer service, AI is doing the same work without the employee. The result is not a single labor market adapting to technology. It is two markets, operating at different speeds, with different rules about who gets paid.
The phrase that circulates in labor economics is "AI-complementary skills." It sounds neutral. It is not.
An AI-complementary role is one where a human still needs to be in the loop — for judgment, accountability, client relationship, legal liability, or creative direction. These roles are, almost exclusively, concentrated in the upper half of the wage distribution.
A copywriter who uses a large language model to produce first drafts faster is AI-complementary. The administrative assistant whose job was to produce those drafts is not. The lawyer using AI to review contracts retains the engagement. The paralegal who did that review is the one whose headcount gets cut in the next budget cycle.
Indeed's Hiring Lab has tracked this divergence in UK job postings consistently since 2023. The pattern is unambiguous: postings mentioning AI skills are rising and concentrated above the median wage. Below the median, demand for routine cognitive roles is contracting faster in the UK than in comparable European labor markets. That compression at the bottom is the story. The growth at the top is the headline. The gap between them is the policy problem.
The OECD put a number on the exposure. Their 2023 Employment Outlook estimated that roughly 27% of jobs in OECD countries face high AI exposure — meaning a significant share of their task content is automatable with tools already commercially available.
The UK scores above the OECD average, partly because of its outsized financial and professional services sector. These industries are both highly productive and highly susceptible to the routine cognitive tasks that generative AI handles competently: drafting, summarizing, classifying, retrieving.
The Bank of England's 2024 analysis sharpened that picture. Around 30% of UK job tasks are automatable using current AI. That is not a forecast. The tools exist. The deployment is happening now, even if official labor market statistics have not yet caught up to what firm-level data is already showing.
Where the Bank of England is careful — and rightly so — is on the distributional question. Automation of tasks does not automatically mean job loss. Workers reallocate. Firms restructure. But the Resolution Foundation's research on UK low-to-middle income workers documents something the aggregate productivity numbers obscure: the workers most exposed to AI substitution are the workers least equipped — by income, geography, and employer investment in retraining — to reallocate quickly.
A logistics warehouse worker in the West Midlands and a financial analyst in Canary Wharf are both facing AI-driven change. The analyst gets a faster workflow and a retention bonus. The warehouse worker gets a robot.
One variable that most policy models have not fully priced in: AI tool costs are falling fast. The rapid succession of cheaper, more capable models means automation is economically viable for tasks that were marginal at higher cost points 18 months ago. As documented in analysis of recent AI market dynamics, each significant price drop in frontier AI models opens a new tier of previously uneconomic automation use cases, effectively pulling forward the timeline on labor market disruption.
This matters for policymakers because retraining programs, social safety nets, and regulatory frameworks are calibrated around transition timelines. If the timeline compresses, the programs do not automatically scale up. They lag. The people in the gap between the old job and the new one pay the difference.
Let me be specific, because "stakeholders" is a word that lets everyone off the hook.
For journalists: When you write about AI productivity gains, the first question is where the gains go. A firm reporting 20% output-per-worker improvement is not the same as a firm reporting 20% wage growth. The gap between those two numbers is the story. The Reuters Institute's research on newsroom AI adoption has documented exactly this pattern inside journalism — editorial productivity metrics improving while freelance rates stagnate and staff headcounts shrink. The industry covering AI's labor impact is experiencing that impact in real time. That is not a conflict of interest to disclose and move past. It is a subject to report.
For policymakers: The UK government's AI Opportunities Action Plan projects aggregate economic gains from AI adoption using models that assume gains will distribute across the economy through labor demand and wages. The evidence from previous automation cycles — documented in Journal of Labor Economics research — suggests wage gains lag productivity gains by 18 to 24 months, and the lag is systematically worse for workers in the bottom wage deciles. A 24-month wage lag at the bottom of the UK earnings distribution, during persistent real-wage weakness, is not an acceptable transition cost. It is a political failure dressed as a market outcome.
The EU AI Act, which took effect in stages through 2024 and 2025, explicitly requires high-risk AI system deployers to assess impacts on workers and notify authorities. The UK, having left the EU framework, has no equivalent obligation. That is a regulatory gap with a measurable cost — workers displaced faster than public retraining infrastructure can absorb them.
For labor economists: The methodological challenge is separating AI adoption from the broader automation trends already underway. The UK's labor market has been bifurcating on skills and wages since at least the 2010s — a pattern the Resolution Foundation and the LSE Centre for Economic Performance documented extensively before generative AI existed. AI is accelerating a pre-existing trajectory, not creating a new one. The empirical question is whether AI changes the speed of bifurcation, changes which occupations are exposed, or does both. Early evidence suggests the answer is both.
The divergence is not abstract. Specific tools are being deployed right now in UK workplaces. Here is where adoption is most active, and what it means for the workers in those sectors.
| Tool / Category | Primary UK Deployment | Who Benefits | Who Bears the Cost |
|---|---|---|---|
| Generative AI writing (Jasper, Writer) | Marketing, media, PR agencies | Senior editors, strategists | Junior copywriters, content associates |
| AI contract review (Harvey, Luminance) | Legal, financial services | Partners, senior associates | Paralegals, contract administrators |
| AI customer service (Intercom Fin, Zendesk AI) | Retail banking, telco, e-commerce | Product and CX managers | Frontline support agents |
| AI-assisted coding (GitHub Copilot, Cursor) | Software, fintech | Senior engineers | Junior developers, QA testers |
| Document processing automation (Hyperscience) | Insurance, public sector | Operations leadership | Data entry, claims processing staff |
The pattern is consistent across every row. The person setting direction retains the role. The person doing the volume work does not.
Before you cite a productivity figure, a hiring trend, or an economic projection about AI's impact on work, ask these questions.
The entry-level job will become the leading indicator. Junior roles disappear before aggregate unemployment figures move. The most reliable early signal of structural displacement is contraction in entry-level job postings — not unemployment rates, which lag by quarters. Indeed's own data should show this before ONS does. Journalists covering the labor market should be tracking posting volumes at the bottom of the wage distribution every month, not every quarter.
Union responses will produce the first firm-level data. The UK Communications Workers Union, Prospect, and Unite have all begun tracking AI deployment in member workplaces. Their data, collected at the shop-floor level, will be more granular and more honest about displacement than any industry-commissioned survey. Watch what they publish, not what company earnings calls say.
The EU AI Act's worker provisions will create pressure on UK policy. Multinationals operating across UK and EU jurisdictions face worker notification obligations under the EU framework. How they implement — and whether the UK government responds with equivalent domestic rules — will determine whether British workers have any visibility into the automation decisions affecting them. The absence of regulation is also a policy choice. It has consequences.
AI tool cost curves will keep pulling the timeline forward. The economic viability threshold for automating a task falls every time a cheaper, more capable model enters the market. Policy designed around a 2023 timeline is already operating on outdated assumptions.
The redistribution question will not resolve itself. Every previous automation wave eventually generated net new employment. The transition periods lasted a decade or more. The new jobs were often in different geographies and required different skills. In the meantime, workers carried costs that policies were not designed to absorb. That is the pattern. There is no reason to believe it will not repeat unless policy explicitly intervenes to change it.
Doesn't AI create new jobs to replace the ones it eliminates?
It does create jobs. Whether those jobs replace eliminated ones in quantity, quality, wage level, and geographic distribution is unresolved. Previous automation waves generated net new employment — eventually. Transition periods lasted a decade or more, and new jobs often required different skills and were located in different regions. For a 45-year-old administrative worker in Sunderland, "the economy will eventually adjust" is not a policy. It is an abdication.
Is the UK more vulnerable than other advanced economies?
More exposed than most, yes. The combination of a large professional services sector, relatively weak union density compared to continental Europe, light-touch employer obligations, and historically low investment in workforce retraining creates fewer structural buffers than comparable economies. The OECD data confirms the exposure level. The institutional framework for managing it is thinner than in France, Germany, or the Nordics.
Why does Indeed have an incentive to flag this now?
Indeed's revenue depends on hiring volume. A contracting labor market at the bottom of the wage distribution is a business problem for them before it is a social problem. Naming the trend publicly positions them as an authoritative data source — commercially valuable — and may signal to policymakers that intervention, which would stimulate formal hiring, is warranted. That does not make their data wrong. It makes their motivation worth noting.
What does the EU AI Act actually require for workers?
The EU AI Act classifies AI systems used in employment decisions — recruitment, performance evaluation, task allocation — as high-risk. High-risk systems require transparency documentation, human oversight mechanisms, and worker notification in some interpretations. The UK has no equivalent framework in force as of mid-2025. The government's official position is described as "pro-innovation" — which in regulatory terms means permissive until a harm is large enough to demand response.
Can retraining programs actually work at the scale this requires?
The evidence is mixed, and the timescale is everything. Short-cycle digital skills programs — the kind typically funded by government initiatives — show modest wage returns and high dropout rates when participants lack educational foundation or are managing financial precarity during training. Effective retraining requires sustained income support during transition, not just course fees. Current UK provision does not meet that standard, and the Skills England budget allocation, relative to AI investment announcements, reflects that gap.
Who is producing the most credible policy analysis on this right now?
The Resolution Foundation on UK-specific distributional impacts. The OECD Employment Outlook for cross-country comparison. The LSE Centre for Economic Performance for firm-level empirical work. The Institute for the Future of Work on regulatory framing. What is absent: a UK government white paper that treats this as a distributional problem rather than a growth opportunity.
What should journalists actually do with Indeed's data?
File a right-to-information request to the Department for Work and Pensions asking for internal modeling on AI's impact on benefit claimant projections. Compare aggregate AI investment announcements against the Skills England budget allocation. Ask every company announcing AI adoption how many roles it has eliminated or not backfilled in the same period. The PR-facing narrative and the operational reality are rarely the same document. The gap between them is, as always, the story.
College graduate unemployment has climbed and underemployment has reached multi-decade highs, but AI automation is not the proximate cause — not yet. The contraction in white-collar hiring is mostly explained by...
The tech industry frames AI's labor market impact as a paradox: jobs disappear and new ones appear, net effect uncertain but probably fine. The OECD and IMF data tell a different story — the jobs that disappear and...
The labor market is being restructured in real time, and the people absorbing the shock are not the ones writing the press releases. The IMF estimated in January 2024 that nearly 40% of global employment is directly...
AI is not emptying the labor market. It is squeezing the people left in it. Productivity gains are real, documented, and concentrated — overwhelmingly at the firm level, not the worker level. The question...
Institutions that measure the labor market and companies deploying AI are looking at the same transformation and reaching incompatible conclusions. The IMF says roughly 40 percent of jobs globally sit inside AI's...
A growing body of evidence from MIT, Stanford, Goldman Sachs, and the IMF is shifting the labor market consensus: AI isn't purely a job-destroyer, and in controlled experiments, it measurably raises productivity and...
BestAIFor