London's AI problem is a white-collar problem

The capital’s strengths make it unusually exposed to generative AI

London has spent decades building an economy around words, numbers, code, advice and deals. That now makes it unusually vulnerable to a technology that is good at manipulating words, numbers, code, advice and deals. A new report by GLA Economics estimates that at least 46 per cent of London’s workers, about 2.4m people, are in jobs where generative AI could automate or transform some tasks. That is well above the UK average of 38 per cent. About 313,000 Londoners are in the highest-exposure group, where most tasks could in theory be automated with current generative AI tools. A further 748,000 are in roles with high, though more varied, exposure.

The caveat matters. The report is not a forecast of mass unemployment. It is a map of exposure, not a redundancy notice. GLA Economics adapts a task-based framework from the International Labour Organisation, using early-2025 generative-AI capabilities to ask which occupations contain tasks that machines can already perform. That makes the work useful, but also limited. It cannot capture seniority, workplace culture, regulation, customer trust or whether a business actually redesigns jobs around AI.

Office jobs, not factory jobs

Still, the map is revealing. The jobs most exposed are not factory jobs, but office jobs. Administrative and clerical occupations sit at the sharpest edge. All roles in this group show some exposure; 61 per cent are in the highest-exposure category and another 27 per cent in the next one down. Administrative workers account for almost 90 per cent of Londoners in the highest-exposure band. This is not surprising. Much admin work consists of drafting, filing, summarising, scheduling, processing emails and entering data. These are precisely the dull but necessary tasks at which generative AI is improving fastest.

Professionals should not feel too smug. The largest absolute concentrations of exposed workers are in professional and associate-professional occupations. IT, finance and business services are full of tasks that involve research, coding, analysis, documentation and client communication. Many of these tasks can be accelerated by AI. Fewer can be handed over entirely. Judgement, accountability, ethics and context still matter. The likely effect is therefore not a clean replacement of people by software, but a messier reshaping of jobs: fewer routine tasks, more checking, more supervision of automated output and, in some cases, fewer junior posts.

Why London is more exposed

This is why London is more exposed than the rest of Britain. Its economy is tilted towards finance, technology, professional services and other knowledge-heavy industries. GLA Economics estimates that 77 per cent of workers in information and communication are exposed to some degree. Finance and insurance show the same overall exposure, but with a larger share of workers in the highest bands. Construction, hospitality, healthcare and education are less directly exposed because much of their work is physical, in-person or highly context-dependent. But even they contain back-office and customer-contact roles that AI can reach.

Awkward distributions

The distributional politics are awkward. Graduates are more exposed than non-graduates, because they are more likely to work in knowledge-intensive jobs. Younger Londoners are more exposed than older ones: 52 per cent of workers aged 16-29 are in substantially exposed occupations, against 39 per cent of those aged 50 and over. Women are more likely than men to be in the highest-exposure roles, largely because of their concentration in administrative and customer-service work. The report also warns that AI could hit the lowest and highest earners in different ways: lower-paid clerical workers may face displacement, while higher-paid professionals may capture productivity gains.

The most important labour-market question may not be whether AI destroys jobs, but whether it destroys the first rung of the ladder. Junior staff often learn by doing the routine work that senior staff later stop doing. If AI takes over the drafting, coding, formatting, checking and summarising that once trained graduates, firms may become more productive in the short run while quietly weakening their own talent pipelines. This is already a worry in software, where AI coding tools can speed up routine tasks but may also change how new developers acquire experience.

Task change, not wholesale replacement

So far, the machine has not eaten the labour market. The evidence collected by GLA Economics suggests that AI adoption is rising, but that its first effect is mostly task change rather than wholesale replacement. UK business use of AI, or uncertainty about whether AI is being used, rose from 9-16 per cent in late 2023 to 26-35 per cent by March 2026. Adoption is higher among large firms and in sectors such as ICT, professional services, education and the creative industries. The common uses are predictable: text generation, visual content, data processing and business-process improvement.

But the early signs are not wholly benign. About 5 per cent of UK businesses using AI in March 2026 said it had allowed them to reduce headcount. A larger share said they were unsure. Recruitment has softened in some highly exposed roles, though the report is careful not to overclaim. The post-pandemic labour market was already cooling, and it is hard to isolate the effect of AI from the broader slowdown. The prudent conclusion is not that AI has already caused a jobs crisis. It is that the places where weakness is emerging broadly match the places exposure models told policymakers to watch.

City Hall responds

City Hall’s response is to create a London AI and Jobs Taskforce, chaired by Baroness Martha Lane-Fox. It is meant to examine how AI is changing work, where the near-term risks and opportunities are, and what practical interventions are needed to protect routes into work and support broad-based growth. Its membership includes figures from business, trade unions, the skills sector and AI policy.

That is sensible, as far as it goes. But taskforces do not retrain workers. Nor do press releases redesign jobs. The report’s policy recommendations point in the right direction: basic AI literacy across the workforce; stronger human skills such as judgement, problem-solving and critical thinking; deeper specialist AI capability; targeted support for exposed administrative workers; and help for small firms that lack the capacity to adopt AI safely and productively. The harder task is execution. Training must reach the people whose jobs are most exposed, not merely the people already best placed to benefit.

A paradox of success

London’s AI problem is therefore a paradox of success. The capital is exposed because it is rich in the kinds of work that modern AI can touch. That exposure may bring faster growth, better services and more productive workers. It may also bring weaker entry-level hiring, hollowed-out admin roles and a bigger divide between those who use AI well and those whose tasks are used to train it, benchmark it or replace them.

The future of work in London will not be decided by the technology alone. It will be decided by whether firms use AI mainly to cut costs or to raise output, whether workers are trained to use it rather than compete blindly against it, and whether young people still get a way into good jobs once the easy tasks have been automated. The report’s warning is not that the robots are coming for everyone. It is that they are coming first for the paperwork.

The numbers

The scale of the exposure is large. GLA Economics estimates that 2.4m London workers, or 46% of the capital’s workforce, are in jobs where generative AI could automate or significantly change at least some tasks. That compares with 38% across the UK.

Within that, 313,000 Londoners are in the highest-exposure category, where a large share of tasks could in theory be automated by current generative-AI tools. A further 748,000 workers are in a high-exposure group. Taken together, that means just over 1m London workers are in occupations where AI is likely to matter soon.

The exposure is unevenly spread. In administrative and clerical work, 61% of workers are in the highest-exposure category and another 27% are in the next-highest group. Administrative roles account for almost nine in ten of the London jobs in the top exposure band. Younger workers are also more exposed: 52% of Londoners aged 16-29 are in substantially exposed occupations, compared with 39% of those aged 50 and over.

By sector, the capital’s knowledge economy stands out. Around 77% of workers in information and communication are exposed to generative AI to some degree. The same broad exposure rate applies in finance and insurance, though with a bigger concentration of workers in the highest-risk bands. This is the central irony of the report: London is vulnerable to AI not because its economy is weak, but because so much of it is built around the office work AI is best suited to change.