Every technological revolution replaces the one before it. The loom replaced the spinning wheel. The excavator replaced the shovel. The electric vehicle is replacing the combustion engine. In each case, humans were still the users. Still essential, still employed, still in the room.

This time is different. The thing being automated is not a physical tool but knowledge itself, commodified, democratised, and available to anyone with an internet connection. Small wonder people fear artificial intelligence (AI) is coming not just for their work but for their identity.

The AI ‘panicians’ are not wrong to be alarmed. I know, because I used to be one of them. But the alarm has produced the wrong diagnosis. This is not a story of replacement. It is a story of displacement. That distinction matters more than the AI doomers and AI accelerationists are willing to admit.

Lay-offs make headlines, but what they obscure is the more important story: the roles that disappear are not being refilled, and the junior positions that once brought new people into these industries are simply not being created.

A gradual tightening of the white-collar labour market produces no single moment of crisis, just a generation of young professionals who cannot get a foothold.

There are currently only 1.6 job openings per 100 employees in white-collar service roles, the lowest level since 2015. For new graduates it’s the worst job market in 37 years.

Jack Dorsey, CEO of Block, recently cut 40 per cent of the company’s workforce, predicting most companies would reach the same conclusion within the year.

Amazon Web Services recently cut hundreds of technical specialist roles, then announced it was developing AI agents to perform the same functions. Fire first, automate second.

It may be scary, but there is a silver lining. The central mistake in most AI and labour market analysis is conflating tasks with jobs.

Consider the radiologist. In 2016, experts predicted AI would soon outperform them at reading scans. Ten years later, there are more radiologists than ever, and they earn more than they did then. This is because a radiologist does not just sell image classification. They triage cases, communicate with physicians, train residents, make the difficult calls, and sign off on diagnoses.

The labour market is buying a bundled service. AI may be able to perform one or more tasks inside that bundle but that does not devalue the job, in fact it often makes it more valuable.

The question AI poses is not whether it can perform a task. It is whether that task can be extracted from the bundle. In human-centric jobs – roles defined by presence, trust, accountability, and judgment – the tasks are so deeply interwoven that extraction is impossible without destroying the product entirely.

You cannot shake an AI hand when closing a deal. You cannot wine and dine a chatbot to get your lobbying firm’s amendment through a Senate committee. Factories automate. Societies, with their cultures, rivalries, hierarchies and vanities, do not.

When AI enters an organisation, it raises the floor by automating the grunt work junior employees once performed, and raises the ceiling by amplifying the output of experienced professionals.

What it hollows out is the developmental path between the two.

AI can already write code, draft legal memos and analyse spreadsheets. These are precisely the tasks that graduates once performed to learn their craft. The medical specialist, the experienced lawyer, the seasoned executive: none of them arrived fully formed. They were built, slowly, through the accumulation of exactly the kind of work AI is now doing instead.

The irony is that the very capacity AI cannot replace – human judgment – is the one that AI use is quietly degrading. Outsourcing thinking to a machine is Ozempic for the mind. The weight comes off, but so does the muscle.

Researchers at the Wharton School found that even when AI provides incorrect answers, people still follow it roughly 80 per cent of the time. Worse still, when given bad AI advice, their accuracy falls below baseline, and they perform worse than if they had no AI at all. More troublingly, access to AI increases confidence even when the answers are wrong.

It is a finding every CEO currently replacing their workforce with AI should understand.

The researchers call it cognitive surrender. You do not decide to stop thinking critically. You simply stop noticing that you have.

This cleaves the workforce in two: the AI super-users, who wield these tools as force multipliers, with a single person now capable of doing the work of a small team, and the passive consumers, who use AI casually without understanding what they are losing in the process. The divide between these two groups will be one of the defining labour market cleavages of the next decade.

But outsourcing one’s thinking in the workplace is a secondary concern for those worried about being able to get into the workforce at all. For a growing cohort of young Australians, that first step is no longer available. And you cannot defend against AI eroding your judgment if you never had the chance to build it in the first place.

The people deploying AI to eliminate junior roles are, by and large, not junior themselves. They are the senior managers, the executives, the partners. People whose positions depend on relationships, institutional knowledge, and tacit authority accumulated over decades.

They are pulling the ladder up behind them. The rungs they are removing are the ones that allowed them to climb in the first place. These are not inefficiencies to be optimised away. They are how professions reproduce themselves. Eliminate them and you do not just harm a generation of young workers. You hollow out the pipeline that produces the next generation of senior ones. And you create exactly the kind of grievance that populist politics feeds on.

For most of the past century, the existence of meritocracy was defensible. Study hard. Acquire knowledge. Be useful. Get ahead. AI has not killed that illusion so much as exposed it. Knowledge was never the moat we thought it was.

What remains, once the knowledge premium evaporates, is something far older and considerably less fair. It will not be what you know that determines your prospects in the AI economy. It will not even be what you can do. It will be who you know, now more than ever.

Displacement is not elimination. But it is not painless either. And the people who will feel it least are, with a grim predictability, the ones who were already most comfortable to begin with.