Edward Egan
Headlines warn of a looming ‘jobpocalypse’, but the reality is more complex. Rather than simply causing a wave of job losses, the economic literature suggests generative AI could influence the labour market through several – potentially offsetting – channels: productivity gains, job displacement, new job creation, and compositional shifts. The balance between these effects, rather than displacement alone, will shape AI’s aggregate impact on employment. The latest research suggests that overall effects remain limited so far, but there are some early signs of AI’s impact. I find that, since mid-2022, new online vacancies in the most AI-exposed roles have decreased by more than twice as much as the least exposed group. This highlights the need for ongoing monitoring as AI adoption accelerates.
How will AI affect employment?
To help us think through this complex question, we can use a ‘task-based’ framework (Acemoglu and Restrepo (2019)). This approach stems from the idea that jobs are made up of a defined set of tasks. Rather than looking at broad occupations or industries, it’s more useful to understand how particular tasks can be automated, augmented or created by new technologies like AI. The impact on any given job will then depend on the mix of different tasks within that role.
For example, in finance, AI could help automate data collection and reporting, which is a large part of a junior analysts’ role, while senior portfolio managers might use AI to scan market sentiment or simulate risk scenarios – hence using AI to streamline decision-making. This can help explain why some roles may be displaced by AI while others may become more productive, despite being in the same industry.
We can broadly simplify this framework into four key channels through which AI may affect the labour market:
- Productivity (Augmentation): AI can make workers more productive by automating repetitive tasks, freeing workers up for other higher-value activities. If firms use gains to expand production, this can increase demand for labour in non-automated tasks.
- Displacement (Automation): AI could automate a large share of (if not all) tasks in some roles, reducing demand for labour in certain jobs.
- Reinstatement (New Tasks): Historically, technological innovations create new tasks that we could not have imagined before. For example, in an AI context, this could mean the emergence of new roles which help customise and integrate AI tools into firms’ workflows. Since the start of 2023, there has been a significant increase in demand for these workers (known as Forward-deployed Engineers).
- Compositional (Reallocation): Even if aggregate employment does not change significantly, AI is likely to reallocate jobs between sectors. Some industries might shrink, others grow, and some workers will need to retrain to adapt their skills accordingly.
Most of the public debate focusses on the evidence around the ‘displacement’ channel. But perhaps the most important message to take away from this post is that the long run net impact of AI on employment will depend on the balance of these effects, as well as the speed of AI development and adoption. Since these forces may also unfold over different time horizons, understanding how they ultimately balance out remains highly uncertain at this stage.
What does the evidence say so far?
Despite widespread speculation about AI-driven job losses, the aggregate evidence for the UK remains limited. A recent Decision Maker Panel Survey found that AI has had little effect on employment so far, with only a minor reduction expected in coming years. Similarly, the Business Insights and Conditions Survey reports just 4% of AI-using firms (23% of all firms) reduced their workforce due to AI, while only 7% of future adopters expect reductions. Meanwhile, data from Indeed shows that demand for AI-related skills has increased in the UK recently (Chart 1), suggesting some early evidence for the ‘reinstatement’ effect, as new tasks that require AI-related skills are becoming more common.
Chart 1: Share of Indeed job postings referencing AI skills (per cent)
Source: Indeed. Data to October 2025.
Evidence from the US also suggests the story is more nuanced. Researchers at the Yale budget lab find no significant aggregate labour market disruption so far, noting that shifts in job composition began before AI’s widespread adoption. While some have attributed the rise in youth unemployment to be due to AI, analysis from the Economic Innovation Group and the Financial Times finds that broader macroeconomic factors are still likely to be more important. Encouragingly, survey data from the Federal Reserve Bank of New York shows most AI-using firms are currently retraining staff rather than cutting them. This underscores that displacement is only one channel of AI’s labour market impact, with upskilling and new job creation also playing an important role in future dynamics.
Digging deeper: slowing in AI-exposed occupations and for junior workers
While overall employment effects seem muted, there may be some early signs of impact in more AI-exposed occupations. My analysis of UK data finds a negative relationship between posting of new online job vacancies and AI occupational exposure. In other words, the more exposed a job is to AI, the less likely a firm is to post a new vacancy in that position. This relationship is even more pronounced if we group jobs into AI exposure quintiles (Chart 2). Here, I find that new online job postings in the most AI-exposed roles have dropped by almost 40% relative to mid-2022, more than double the fall in the least exposed group. While these findings corroborate similar work by McKinsey, it could be the case that these occupations are simply more exposed to a cyclical slowing in the economy, so this evidence suggests correlation rather than proving any causation.
Chart 2: Percentage change in new online job postings since mid-2022 by AI occupational exposure quintile
Notes: ONS online vacancy data by SOC is experimental so should be treated with caution and is likely subject to future revisions. Six-month averages are used to smooth volatility and missing data. Department for Education (DfE) use Felten et al (2021) measure of AI occupational exposure and map this to UK labour market data.
Sources: DfE (2023) and Experimental ONS online vacancy data.
Recent academic research also finds faster falls in vacancies and employment in AI-exposed occupations, particularly concentrated in junior positions. Henseke et al (2025) find that, by mid-2025, UK job postings were 5.5% lower in AI-exposed occupations than they would have been if pre-ChatGPT trends had continued. Similarly, Teeselink (2025) finds that highly exposed UK firms reduced employment by 4.5% (concentrated almost entirely in junior roles) and were 16 percentage points less likely to post new vacancies. In the US, research finds early-career workers in the most AI-exposed occupations have experienced a 13% relative decline in employment, while less exposed and more experienced workers in the same roles were largely unaffected (Brynjolfsson et al (2025)). Research from Hosseini Maasoum and Lichtinger (2025) largely corroborates this, finding that the adjustment has largely taken place via reduced hiring rather than increased layoffs.
But despite growing evidence, AI likely remains an amplifier rather than the sole driver of the slowing in youth employment. Most studies acknowledge that there is a lack of high-quality data and significant challenges with disentangling explicit causality, especially given the tightness (and subsequent loosening) of the labour market since ChatGPT’s release in November 2022. So, while AI may be amplifying effects for hiring of new entrants in AI-exposed sectors, the broader slowdown appears to also reflect typical labour market downturns, where younger and less experienced workers are disproportionately affected.
What about longer-term forecasts?
Forecasts vary significantly, but most suggest the outlook is less severe than headlines imply. Scenarios of UK job displacement due to AI range from zero to around eight million over the long run (IPPR (2024), Tony Blair Institute for Global Change (2024), PwC (2018)), but most analysis expects this to be largely offset by the creation of new roles and higher productivity, in line with historical evidence from previous technological advances (Hötte et al (2023)).
The key risk is if productivity gains are more limited than expected and if new jobs and tasks are not created quickly enough to offset those lost to automation. This could lead to a temporary rise in unemployment, though the magnitude would depend heavily on the speed of AI adoption and size of the displacement effect (Goldman Sachs (2025)).
Another risk to the long-term outlook stems from the development of more advanced forms of AI (such as ‘Artificial General Intelligence’). This post does not explore what this could mean for the labour market, but some suggest the impacts could be more severe (Restrepo (2025)).
Conclusion
Current evidence suggests AI has had little effect on overall labour market dynamics so far. However, my analysis and other research finds signs of AI amplifying the slowdown in hiring in AI-exposed occupations. Looking ahead, the impacts could be broader if AI’s productivity gains disappoint or if new roles do not emerge quickly enough. This could pose a risk of higher unemployment which could take some time to unwind as the labour market adjusts. Therefore, it’s essential to monitor not only displacement effects, but also how AI is impacting productivity, job creation rates and compositional shifts. Developing more sophisticated metrics for monitoring these factors will be key to understanding the transition to an AI-augmented economy. Ultimately, the long run net impact of AI on employment will depend on the balance of the effects outlined in this blog and the speed of AI development and adoption.
Edward Egan works in the Bank’s International Surveillance Division.
If you want to get in touch, please email us at bankunderground@bankofengland.co.uk 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.