Future of Labour
The Automation Paradox: Why Robots Keep Creating More Jobs
Every time a machine has replaced a human worker throughout history, the total number of jobs has gone up — and nobody has a satisfying explanation for why this keeps happening.
The Idea
The intuitive fear about automation is straightforward: machines do the work, humans become redundant. It has a name — the lump of labour fallacy, the belief that there is only a fixed amount of work to go around. Economists have spent decades pointing out that this fear has never materialised at civilisation scale. The handloom weavers displaced by the power loom didn't vanish into permanent unemployment; their grandchildren became factory supervisors, train drivers, typists. And yet the fear resurfaces with each new wave of technology, and each time it feels more plausible, because this time the machines are smarter. What actually happens is subtler than either the optimists or the pessimists admit. Automation doesn't eliminate demand for human labour — it shifts it, and in doing so, it tends to lower the cost of goods and services, which frees up spending power that creates demand elsewhere. An ATM does the cash-dispensing work of a bank teller, but banks responded by opening more branches, not fewer, because cheaper transactions made branches profitable in smaller locations. The tellers who remained shifted toward sales and relationship work. The genuine anxiety isn't about total job numbers. It's about transition costs — who bears them, how painful they are, and whether the new jobs are as good as the old ones. A coal miner retraining as a data analyst at fifty is not simply a data point in a smooth economic adjustment. The macro story and the human story can both be true simultaneously, and keeping that tension in view is what separates clear thinking about automation from either naive optimism or apocalyptic dread.
In the World
In 2013, two Oxford economists, Carl Benedikt Frey and Michael Osborne, published a paper estimating that 47 percent of US jobs were at high risk of automation within two decades. The paper went viral in a way academic economics rarely does. Politicians quoted it. Consultants built entire practices around it. A minor industry of anxiety emerged. What happened next is instructive. Unemployment in the US and across most wealthy economies fell to historic lows in the years that followed — not because the automation didn't happen, but because the methodology had a blind spot. Frey and Osborne had assessed whole occupations, not tasks. Most jobs are bundles of tasks, some automatable and some not. A radiologist's job includes reading scans, which AI now does with impressive accuracy, but it also includes communicating difficult diagnoses to frightened patients, managing a department, and making judgment calls that require institutional context. The job didn't disappear; it mutated. The OECD later re-ran the analysis at the task level and found that only around 9 percent of jobs were at genuine high risk. Not nothing — 9 percent represents millions of people — but a very different civilisational picture. The gap between the two estimates wasn't a matter of better data. It was a matter of asking a more precise question. In labour economics, as in most things, the framing determines the answer you get.
Why It Matters
How you think about automation shapes decisions you're probably already making — about which skills to invest in, which industries to build a career around, whether to retrain or sit tight, how worried to be about the next round of AI announcements. The more useful mental model isn't 'will my job exist?' but 'which parts of my job are hardest to replicate, and am I spending my time there?' Routine cognitive work — processing, pattern-matching, rule-following — is genuinely under pressure. Work that requires contextual judgment, physical dexterity in unpredictable environments, emotional attunement, or creative synthesis in messy real-world conditions is proving stubbornly resistant to automation, often in surprising places. There's also a structural point worth carrying: the workers who lose most from automation tend not to be the ones at the top or the bottom of the wage distribution, but those in the middle — the routine clerical and semi-skilled manufacturing roles that provided stable, dignified middle-class incomes for generations. That hollowing out of the middle is already well underway, and it's a better frame for thinking about economic anxiety than the blunter question of whether robots are coming for all of us.
A Question to Ponder
If you stripped your current work down to its individual tasks, which ones would be genuinely difficult to automate — and are those the tasks you're actually spending most of your time on?
Get a new one of these every morning.
Start learning with Thinkable