Anthropic shines a light on an uncomfortable truth, how AI is really eating into the labour market

More than simple automation, AI as the new colleague

Anthropic’s report argues that AI is not yet wiping out entire professions en masse. What it is doing, with the appetite of a hungry predator, is swallowing individual tasks. In sectors where moving, processing and reshaping information is the core business, artificial intelligence already took over as much as 30 per cent of routine daily work.

That matters because it changes the shape of entry level employment. Companies once hired younger staff to sort data, draft basic copy and handle the repetitive groundwork. Now Claude, or one of its many cousins, can do much of that in seconds. The human role does not disappear, but it changes. The worker becomes the reviewer, the editor, the one who checks the machine’s output and decides whether it deserves to live.

That, in turn, demands a very different kind of skill. Execution on its own carries less weight. Judgement, context and the ability to steer the machine now matter more.

Wage pressure and the productivity paradox

The report also lands on a less cheerful point. Higher productivity does not automatically mean higher pay for workers. Anthropic found that companies tend to channel the gains from AI into profits or further investment, rather than into employees’ pay packets.

That is how the labour market starts to split in two. On one side sit AI capable specialists, people whose value rises because they can direct these systems with the confidence of a seasoned rally driver threading a car through a wet forest stage. On the other side are traditional executors, whose market value slips because a machine can now produce similar output faster and more cheaply.

This is the paradox at the centre of the current AI boom. Businesses get more efficient. Output increases. Costs fall. Yet for many workers, that shiny productivity story feels suspiciously like a pay freeze wearing a smarter suit.

Are we heading towards technological unemployment

Unlike the more theatrical prophets of doom, Anthropic’s analysis remains measured. The report points out that technology has historically created new kinds of work even as it destroyed old ones. The difference this time is speed.

The internal combustion engine replaced the horse over decades. Large language models, or LLMs, stormed into offices in a matter of months. That pace is what creates the real social strain. Workers, employers and education systems are being told to adapt almost overnight, which is a neat trick in theory and a messier one in practice.

In industrial terms, the shift is easy to grasp. An engineer no longer needs to calculate aerodynamics on paper. The real task is telling AI what the new body line should express, what feeling it should carry and what constraints it must obey. The machine calculates. The human curates.

That sounds elegant enough, until you remember that not everyone is paid to curate. Plenty of people were paid to do the work that now vanishes quietly into a prompt window.

Adapt or get left under the wheels

Anthropic’s study feels less like a forecast and more like an alarm clock. AI is no longer a novelty, nor a toy for technology enthusiasts. It is an economic force, and one already reshaping the labour market with a bluntness that recalls the aftermath of a financial shock.

The winners will be those who learn to master these models, bend them to useful ends and build judgement on top of automation. The losers will be those who assume the old routines still hold value simply because they did yesterday.

One day they may look down and realise the old rails are still there. It is just that the train left some time ago.