Lord Stockwood, the minister for investment, recently floated the idea of universal basic income to cushion AI-driven job losses. Last month, the European Central Bank published a study of 5,000 eurozone firms showing that companies which adopt AI are 4 per cent more likely to hire. Something doesn’t add up. So what’s going wrong?
In 2013, a widely cited Oxford study told us that 47 per cent of American jobs were at high risk of automation. Since then, every serious forecast has done the same thing: decompose a job into tasks, score which tasks machines can do, announce a crisis. The renowned VC investor Marc Andreessen refined the point this January: it’s tasks, not jobs, that get replaced. He’s right. But he doesn’t explain why the job survives once the tasks are gone.
For an answer to that question, look to Adam Smith. It should come as no surprise that just when we think the rules of economics are about to be utterly changed, that the core insight comes from the father of the subject.
The pin factory is the most famous anecdote in economics. Ten men can create 48,000 pins a day performing 18 operations between them. Every textbook reads it as a story about task decomposition. But Smith was describing something else. The wire drawer’s output has to fit the straightener’s technique. The pointer’s work must be compatible with the header’s. Each worker depends not just on their own skill but on the relationships between them: timing, trust, the tacit knowledge that builds over months and years of working side by side.
Smith said the division of labour is limited by the extent of the market. That’s the line everyone quotes. But the extent of the market isn’t a task list. It’s a web of human relationships. And it’s the nature of those relationships that determines what a job looks like in the age of AI.
The best granular data comes from across the Atlantic. America’s Bureau of Labor Statistics projects every occupation out to 2033. They say software developer jobs will increase by 18 per cent, financial advisors by 17 per cent and lawyers by 5 per cent.
Personal financial advisors are heavily exposed to AI. Robo-advisers already handle portfolio rebalancing and tax-loss harvesting. On paper, doomed. In reality, this profession is booming. And that’s because the financial adviser’s real job isn’t rebalancing a portfolio but rather being the person a client trusts enough to ring.
Software engineer job postings hit multi-year highs in early 2026, even as AI writes more of the code. When something gets cheaper to produce, you produce more of it. And the software developer isn’t writing code in isolation. Instead she is coordinating with product managers, designers, and other developers all day, every day. AI can take the tasks. It can’t take her position in the network.
I run a creative and digital agency that has worked on more than 25 election campaigns across 40 countries. Matthew Kilcoyne, who has done additional reporting for this piece, has spent a decade in technology policy. Between us, we’ve seen the AI-replaces-everything thesis tested in both directions: in client relationships and campaign war rooms, and in the policy frameworks supposed to prepare workers for what comes next.
The honest answer, at least in the knowledge economy, is that about 40 per cent of what an organisation does is automatable right now. Research synthesis, first-draft copy, scheduling, data processing are just some of the tasks that AI is brilliant at. But the other 60 per cent is relationships, judgment and taste: knowing when the machine’s output is brilliant and when it’s rubbish. That scales at the speed of human trust.
The anthropologist Robin Dunbar showed that humans can maintain roughly 150 stable relationships. The limit is neurological, and shows up across cultures throughout history. Hutterite farming colonies have split at that number for centuries. Gore-Tex caps its factory units at 150 employees. The relational capital in any job builds slowly, can’t be parallelised, and is destroyed in an instant when someone is displaced.
The anthropologist Robin Dunbar showed that humans can maintain roughly 150 stable relationships
That’s the real problem with retraining. Every programme focuses on skills. But a displaced worker’s actual deficit isn’t skills. It’s that she doesn’t know who to ring or whose emails matter. Anyone who has started a new job knows the first six months aren’t about learning the tasks. They’re about learning the network. Policy that ignores this will keep failing.
If the evidence points to augmentation, why does the public assume displacement? Look at the name.
We call it artificial intelligence. ‘Artificial’: the GMO of cognition. Almost fraud-coded before the conversation begins. And then ‘intelligence’ is coming for your livelihood. No wonder people are frightened!
What if we’d called it what it actually does? Not artificial intelligence. Augmenting intelligence. One name says the machine is coming for you. The other says it’s working alongside you. The evidence supports the second. But we’ve spent three years trapped by the first.
The question was never whether the machine is coming for us. It’s what we become alongside it.
Additional reporting by Matthew Kilcoyne. Matthew is a policy analyst at the Centre for Data Innovation.
Comments