待翻译：Canaries in the column? AI exposure and the UK's hiring slowdown

AI 服务暂时不可用，以下为来源正文，待恢复后补全翻译。

Haley Schlicht From Silicon Valley executives promising to automate white-collar work to headlines claiming AI is foreclosing the graduate entry market, the strained ‘low fire, low hire’ environment has increasingly been ascribed to technological transformation. UK vacancies nearly halved since their 2022 peak – a contraction so sustained it has reshaped the British hiring market for the better part of three years. This post examines how evidence of AI-driven transformation at the hiring margin is proving considerably more tenuous than the headlines suggest. Brynjolfsson et al (2025) at Stanford University champion vacancy compression in AI-exposed occupations as a canary, the labour market’s early warning system and a leading indicator of technological displacement in adolescence. Lambert and Schindler (2026) counter with a discomfiting reframe. Rather than AI, the hiring market is shifting due to the structural upheaval that remote working visited upon firms’ internal labour dynamics, breaking the lower rungs of the career ladder. The UK makes for an exigent test case. The vacancy retrenchment here has been severe even compared to peer economies, compounded by the previous post-pandemic over-hiring, an energy shock from a land war in Europe, cyclical deterioration, and additional pressures on labour demand. This post builds off other monitoring efforts and this broader framework for tracking how AI may diffuse through the economy. This article focuses on the labour-demand channel within that framework, seeking to disentangle AI’s contribution from the surrounding storms to reveal whether the hiring market is beginning to display the kinds of patterns we might expect during the early stages of a general-purpose technology transition. What looks like weather damage to the labour market may, beneath the surface, already be a shifted shoreline. The vacancy retreat The fall in vacancies was not spawned from a single event. The labour market, which was exceptionally tight when vacancies peaked in 2022, gradually loosened in the following years as the cycle unwound. Pandemic over-hire met its correction as firms confronted the scale of labour they had banked against demand that never fully materialised. Remote working may have played a role too, raising the cost of training junior staff and prompting firms to withhold intake. National Living Wage upratings and changes to employer National Insurance contributions formed part of the wider labour-demand environment. Simultaneously, businesses weighing AI investment may have become more reluctant to refill headcount. Occupational signals Identifying technology’s footprint in the UK labour market first requires a credible measure of where AI capabilities augment labour. To address divergence between individual measures, this analysis coalesces five leading indices from the academic and industry literature, each capturing a different dimension of occupational exposure. Sourcing the original works’ task-level ‘AI susceptibility’ assessments, we reconstruct the AI exposure scores using UK occupational and industry employment data aggregated with pre-treatment employment weights. As such, the exposure scores are constructed to reflect the structure of the British labour market rather than a US-derived benchmark. By benchmarking multiple exposure frameworks and validating the resulting scores against reported AI adoption in the Bank’s Decision Maker Panel and ONS Business Insights and Conditions Survey, the measure aims to provide a more robust signal of technological exposure than any single index alone. Chart 1 shows the strongest signal of technology disruption to hiring appears at the occupational level where occupation-indexed AI exposure exhibits a strong, monotonic correlation with online-vacancy contraction across UK occupations. Chart 1: Growth in advertised vacancies falls as occupational AI exposure rises Notes: Spearman p = -0.70, p <0.001, n = 26. The composite score is validated against surveyed businesses’ reported Al adoption from the Bank of England’s DMP (p = +0.68, p = 0.006, n = 15) and ONS BICS (p = +0.85, p < 0.001, n = 14). Sources: ONS Online Job Adverts via Textkernel; internal calculations. Composite Al exposure score is a weighted average of percentile ranks across four measures (Felten et al (2021) and (2023), Henseke et al (2026), Anthropic Economic Index (March 2026 release) and Eloundou et al (2023) included as a robustness measure. Sorted into terciles by exposure percentile in Chart 2, high-exposure groups lost 15% of online adverts, mid-exposure 10%, low-exposure 6%. The sharpest declines land where the task-based account predicts; customer service down 23%, administrative occupations down 22%. These are the task bundles – scheduling, correspondence, routine information processing – that generative systems can now credibly substitute. Chart 2: Growth of online job adverts fell most in high-exposure occupations Notes: Pre-period uses valid 2019 year-on-year observations because OJA starts in January 2018. Post-period covers available observations in 2023–26 Q1; March 2026 occupation cells are suppressed in the source. Terciles reflect SOC two-digit sub-major groups grouped by composite Al exposure score. Sources: ONS Online Job Adverts via Textkernel and internal calculations. In Chart 3’s right panel, the correlation is not clearly visible at the Industry level. This is an industrial aggregation artefact. For example, Professional services employ accountants and building services staff under the same SIC code, blurring the key occupational signal. Unpack industries into their occupational composition in Chart 3’s left panel and the pattern re-emerges. Finance, ICT and Professional services, where high-exposure occupations account for 84%, 81% and 63% of employment respectively, show deep vacancy reversals. The compositional channel, rather than the aggregate sector, indicates where industries may be pinching hiring even when headline sectoral vacancy data obscure the adjustment. Chart 3: Industry exposure and vacancy growth Sources: ONS VACS02; Annual Population Survey 2022 SIC-SOC employment weights; internal calculations. Composite Al exposure score is a weighted average of percentile ranks across four measures (Felten et al (2021) and (2023), Henseke et al (2026), Anthropic Economic Index (March 2026 release) and Eloundou et al (2023) included as a robustness measure. (a) Shows the share of each sector’s workforce employed in occupations in the top exposure tercile, calculated using APS SIC-SOC employment weights (2022). (b) Plots each SIC section’s employment-weighted mean composite Al exposure against its mean VACS02 vacancy-rate YoY growth over 2023–26 Q1 (Spearman p = +0.08, showing that aggregate sector data blur occupational composition). Broken ladders and hollowing pyramids Occupational signals alone cannot tell us whether the weakening in hiring can be partially explained by AI, remote working, or both. As Lambert and Schindler (2026) evidence, the industries carrying the sharpest occupational signal – Finance, ICT and Professional services – are also those that reorganised most radically around remote working during 2020–22. Both AI and WFH adoption are concentrated in highly digital, capital-intensive occupations, making the exposures difficult to differentiate. But while the infrastructural transition to support hybrid work has largely been completed, AI is still in its nascent stages and may have more persistent labour market implications. Friebel et al (2026) provide a compelling framework for how labour markets may be shifting. The traditional pyramid structure, built on large cohorts of junior workers, may be oscillating toward a diamond, hollow at the base and centred on experienced staff. Remote working may have jumpstarted the transition by raising the cost of on-the-job training and AI strengthens the economic incentives to cement it. As generative systems increasingly absorb routine information-processing tasks, senior workers can perform more of the work that once formed the bedrock of junior hiring. A shifted shoreline Once the weather clears, the shoreline may look different. The UK vacancy puzzle is a picture of overlapping shocks: pandemic over-hire followed by cyclical loosening, but also remote-working adjustment, labour-cost pressures, and AI arriving close enough together that confident attribution remains premature. What the evidence does support is that UK firms are adjusting hiring mainly through the occupational channel, with an indicative signal that this adjustment is happening prevalently in roles with a strong technology exposure. The series to watch are therefore narrow: whether employment follows vacancies down, entry-level hiring continues to weaken, and measured AI exposure translates into realised adoption. The same logic applies to productivity. As discussed in a companion article, industries reporting higher AI adoption are also showing tentative signs of improving productivity performance. If AI is genuinely behaving like a general-purpose technology, these productivity and labour-market signals should ultimately be interpreted in coordination rather than in isolation. Thus, disentangling AI’s role from these concurrent influences remains challenging, but crucial to appraising these evolving dynamics in the UK labour market. Haley Schlicht works in the Bank’s Data and Statistics Division. If you want to get in touch, please email us at [email protected] or leave a comment below. Comments will only appear once approved by a moderator, and are only published where a full name is supplied. Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees. Share the post "Canaries in the column? AI exposure and the UK’s hiring slowdown" Twitter Facebook LinkedIn Reddit Tumblr Email Bluesky