by Claude Opus 5.5

What mechanisms typically drive labour-market change from new technology: substitution, complementarity, new demand, new products, or changes in firm organisation?

All five operate at once. What differs is where and when each shows up. Substitution and organisational change act first, inside firms, and are what workers feel. New demand and new products act later and across the economy, and are what usually rescues aggregate employment. That timing gap is why early evidence on any technology tends to look worse than the long-run outcome. It is a historical tendency, though, not a law, and it depends on choices firms and governments make.

One framework that holds the five together

The most useful way to connect the mechanisms is the task framework of Acemoglu and Restrepo (2019). Technology changes which tasks are done by people and which by capital, and three forces then act on labour demand:

  • a displacement effect, as machines take over tasks people did;

  • a productivity effect, as cheaper production raises output and with it demand for people in the tasks that remain;

  • a reinstatement effect, as new tasks are created in which people have the advantage.

Complementarity sits inside the productivity effect. New demand and new products drive it and the reinstatement effect. Organisational change determines how large each force turns out to be in practice.

Substitution: the first thing to move

Substitution means capital doing a task a person used to do. In 2026 some UK-headquartered firms say so openly. Standard Chartered’s chief executive described part of its planned 7,800 back-office reductions (it is unclear how many would be in the UK) as “replacing, in some cases, lower-value human capital with the financial capital and investment capital we’re putting in” (as reported). More commonly, though, substitution shows up as hiring that doesn’t happen rather than redundancies. The Bank of England’s July 2026 Monetary Policy Report found that “AI adoption is gradually reducing demand for highly automatable jobs in some industries, with firms often slowing hiring or leaving vacancies unfilled”. Its Agents had earlier reported firms planning to meet demand “by investing in automation and AI, rather than by raising headcount”.

Relative prices matter as well as capability. From April 2025, employer National Insurance rose from 13.8% to 15% and the threshold at which it starts fell from £9,100 to £5,000 a year. That raised the cost of a job, especially a low-paid one, relative to software. Acemoglu and Restrepo warn about “so-so automation”: technology adopted because it is marginally cheaper than labour, not because it is much more productive. It displaces workers without generating the productivity gains that create jobs elsewhere. Cost-driven substitution carries that risk.

Complementarity: making people more productive

A technology complements labour when it raises the value of what a person does. Brynjolfsson, Li and Raymond found that an AI assistant raised customer-support agents’ productivity by 14% on average and by 34% for novices. Noy and Zhang found that ChatGPT cut the time professionals took on writing tasks by 40% and raised assessed quality by 18%. Complementarity is strongest where a person keeps the judgement and the accountability and the tool does the drafting, retrieval or first pass.

Complementarity alone does not guarantee more jobs. If each worker produces more and demand for the output is fixed, a firm needs fewer workers.

New demand: the scale effect

This is the mechanism people most often forget. If technology cuts the cost of a service and its price falls, people may buy so much more that total employment rises. James Bessen’s study of US bank tellers is the standard example. As ATMs spread, tellers per urban branch fell from about 20 to 13, but banks opened 43% more urban branches between 1988 and 2004, and teller numbers did not fall.

The condition is elastic demand: there must be large unmet or price-sensitive demand. Plausible UK candidates include legal help for small claims, where the SRA authorised Garfield.Law as the first AI-driven law firm in May 2025, as well as tutoring, software for small firms and some health diagnostics. Unlikely candidates are services where demand is fixed by regulation or need, such as statutory audit, payroll or standard conveyancing. There, cheaper production mostly means fewer people.

New products and new tasks: the long-run rescue

Autor, Chin, Salomons and Seegmiller estimated that about 60% of US employment in 2018 was in job specialties that did not exist in 1940. Most work today is “new work”. Their research also shows that new work has increasingly appeared at the top and bottom of the pay distribution rather than the middle, which matters for who benefits.

Early signs of reinstatement are visible in the UK. PwC found UK postings for specialist AI roles rose 61% in 2025, from 112,000 to 180,000, or 2.2% of all postings. LinkedIn reports about 95,000 AI roles created in the UK since 2023 (as reported). The government-commissioned Warwick projections expect jobs involving AI activities to grow from about 158,000 in 2024 to about 3.9 million by 2035, though that is a projection rather than evidence. The new work is concentrated in technical and London-heavy roles, while the displaced work is spread more widely.

Organisational change: the amplifier

Bresnahan, Brynjolfsson and Hitt showed that IT raised demand for skilled labour mainly when it came with reorganisation: decentralised decisions, redesigned jobs and new incentives. UK evidence points the same way. Bloom, Sadun and Van Reenen found that US-owned firms in Britain got more productivity from the same IT than other firms in Britain, because of how they managed it.

A modern UK example of the missing reorganisation is the Department for Business and Trade’s 2024 Copilot trial. Staff saved time, but the evaluation “did not find evidence that time savings have led to improved productivity”. ONS finds that around 62% of businesses citing a lack of AI expertise retrain existing staff, but only 11% of businesses with 10 or more employees have trained more than half their workforce. Organisational change is where most firms still are.

Timing and visibility

  • Substitution. Where it shows: Within firms, at the hiring margin. Speed: Fast. What 2026 UK evidence shows: Slower hiring in exposed roles; a few explicit corporate statements.

  • Complementarity. Where it shows: Output per worker. Speed: Fast but uneven. What 2026 UK evidence shows: Strong in trials, weak in measured productivity.

  • New demand. Where it shows: Prices and volumes in a sector. Speed: Medium. What 2026 UK evidence shows: Hard to see yet.

  • New products and tasks. Where it shows: New occupations and firms. Speed: Slow. What 2026 UK evidence shows: AI specialist postings rising from a small base.

  • Organisational change. Where it shows: Management practice. Speed: Slow. What 2026 UK evidence shows: Most firms still at shallow adoption.

Bottom line

Ask four questions of any technology. What is being displaced? What becomes more valuable? How elastic is demand for the output? And are firms redesigning work or just adding tools? In 2026 the UK is seeing the fast mechanisms, substitution at the hiring margin and complementarity in pockets, more clearly than the slow ones. That is normal for this stage, but it means the costs are visible before the benefits.

Sources

From AI and Jobs: UK, October 2026