More work than it looks: the new jobs coming from AI adoption
Published here:
From synthesisers to assembly lines, technological change has threatened familiar jobs while creating unexpected forms of work. Research into AI-related job vacancies is revealing strong demand for the people who help organisations adopt AI. Ed Teather argues that the workforce is not merely adapting to this transformation, it is driving it, and explains how policies that support workers can also accelerate adoption.
It can often be easier to imagine what we might lose from technological change than what we could gain.
In May 1982 the Central London branch of Britain’s Musicians’ Union voted to call for a ban on synthesisers, like the one pictured above, and other electronic instruments. The motion was inspired by American crooner Barry Manilow, who had toured with electronic keyboards instead of an orchestra, alarming professional string players.
The ban never became national union policy. Yet New Order, Eurythmics and the Pet Shop Boys soon helped define an era of electronic music. By 1988, Yazz’s The Only Way Is Up was topping the charts.[1]
The fears were real. But the technology’s creative possibilities – and the work for producers, programmers and sound engineers – were harder to anticipate.
Nobel laureate Daniel Kahneman and Amos Tversky called our tendency to weigh losses more heavily than equivalent gains loss aversion . It is a useful lens for today’s debate about artificial intelligence. We can name the jobs at risk, but struggle to picture those that do not yet exist.[2]
The work that hasn’t been invented yet
In a landmark study , economist David Autor and his co-authors estimated that more than 60 per cent of US employment in 2018 was in job specialties that did not exist in 1940. New work emerged as technologies complemented existing occupations and created new specialisms.[3]
That doesn’t mean that AI will automatically create more jobs than it displaces. The transition could be unusually fast, and the people who gain may not be those who lose out. Our challenge is to identify opportunities early and build routes into them.
There are already positive signals. In a US Federal Reserve survey , 81 per cent of workers who used generative AI at work said it saved them time. Research on job vacancies in Britain found a 23 per cent advertised wage premium for AI skills. And Ramp’s study of the companies investing most heavily in AI adoption finds that they are also reporting stronger headcount growth, suggesting that the technology is raising these business’ ambitions and the speed with which they can scale.[4][5][6]
AI’s missing middle
Some jobs will emerge from products and services we can barely imagine. AI-enabled drug discovery could create new medical specialisms. Advances in AI systems that generate interactive virtual worlds could make new immersive entertainment experiences dramatically easier to produce. Another source is already visible: the people turning AI breakthroughs into useful tools.
In research with the AI Adoption Initiative and Carnegie Mellon University, we developed the AI Labor Stack . It has three layers: innovators who build frontier AI systems; users who apply them in everyday work; and facilitators who help organisations deploy, integrate and govern them.[7]
Facilitators might connect AI to hospital software, adapt factory processes or help teams redesign workflows. They bridge invention and widespread adoption.
The lesson from earlier technologies
History suggests this middle layer matters enormously.
Consider the early automobile industry. Pioneers such as the Italian Ceirano brothers used electricity in relatively straightforward ways such as installing light bulbs so their workshops could keep making cars after dark. This extended production hours without changing how work was done.
At Ford’s factory in Highland Park, electricity was put to different use. Electric motors powered conveyors and tools, letting engineers rethink production itself. Rather than workers moving between stationary cars, components moved through an organised sequence of tasks. The moving assembly line, introduced in 1913, transformed the economics of manufacturing.[8]
The difference was not simply using electricity, but redesigning work around it. That demanded engineering skills and organisational know-how.
We saw something similar with personal computers. Initially, many people used them as little more than electronic typewriters, much as many people today use AI like a search engine. Bigger gains came from connecting databases and redesigning workflows.
As Jeffrey Ding argues in Technology and the Rise of Great Powers , America’s rise during the Second Industrial Revolution reflected its capacity to spread technologies such as machine tools through a broad base of practically trained engineers. Invention mattered, but so did the people who put it to work.[9]
What employers are looking for now
Our latest paper, Mapping the Missing Middle in the AI Workforce , launched at the LSE’s inaugural Global Forum for AI and Social Sciences , examined around 94,000 AI-related job postings among S&P 500 companies.[10][11]
Facilitators accounted for approximately 60–80 per cent of AI-related postings across sectors. Demand was substantial in manufacturing, financial services, technology and healthcare, and has grown strongly since 2024.
These aren’t all coding jobs. We identified three overlapping types of facilitator:
Technical, the people who connect AI models, data and software. Social, thepeople who help organisations redesign processes, build trust and govern AI responsibly. And Infrastructural, the people who provide the computing, power and connectivity on which AI deployment depends.
Among facilitator postings, 39 per cent were exclusively technical, 32 per cent combined technical and social roles and nearly 20 per cent were exclusively social. Infrastructure appeared less frequently, although we likely undercount roles in energy and connectivity that aren’t advertised as AI jobs.
Adverts aren’t hires, and large American companies don’t represent every labour market. But a substantial share of AI-related demand is for people connecting frontier developers with everyday users.
A policy agenda for participation and resilience
Many national AI strategies prioritise frontier researchers or basic AI literacy. Both matter, but neither is sufficient.[12]
First, governments need better evidence of facilitator demand across regions and sectors. Manufacturing clusters won’t need precisely the same skills as healthcare or finance. Second, training should be modular and practical, shaped by employers and linked to real jobs.
There are promising approaches across the three domains. Singapore’s AI Apprenticeship Programme combines technical training with industry projects. Ireland’s telecommunications and data-network apprenticeship develops connectivity skills through paid, workplace-based learning. And Germany’s Mittelstand-Digital helps smaller businesses develop the skills and organisational practices needed to adopt new technologies.[13][14][15]
Institutions matter just as much as courses. Regions with strong universities can bring together researchers, employers and educators, as Pittsburgh’s Advanced Robotics for Manufacturing Institute (ARM) illustrates. For countries building training provision at scale, India’s FutureSkills Prime offers another model.[16][17]
Displacement still matters. Workers need support as jobs change or disappear. Participation and resilience must go together: creating pathways into new work while helping people through technological transitions.
Not everywhere needs to become Palo Alto
This summer, we explored these questions with more than 70 experts at Carnegie Mellon University in Pittsburgh, a city with a thing or two to teach us about industrial transitions. Once defined by coal and steel, Pittsburgh experienced profound economic upheaval before developing new strengths in robotics, advanced manufacturing, healthcare innovation and education.
Its reinvention was neither quick nor painless. But it illustrates a lesson from our discussions: successful transitions depend not on technology alone, but on skills, institutions and local capacity. Training needs to respond to the demands of local industries, with universities, employers and educators working together to create pathways into new work. Pittsburgh’s ARM offers one model.
The lesson travels. Sheffield, a British city with a similar industrial heritage to Pittsburgh for example, need not become a frontier AI hub like Palo Alto or London’s King’s Cross to prosper in the AI economy. Instead, it can build on its engineering and manufacturing strengths, by investing in the facilitators who can help local firms adopt AI and modernise their industries.
The synthesiser revolution reminds us that technology can threaten familiar work while enabling opportunities we cannot foresee. But those opportunities won’t come automatically. They depend on investing in the people and institutions that help communities navigate change and share in the benefits.
This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment.