The AI jobs miracle? Don't bet your wages on it

We are fans of artificial intelligence. We want it to make businesses faster, services better and tedious work less tedious. But there is a rather awkward question behind all the excitement: what happens to the people whose jobs it replaces?

Apparently, they can all become AI specialists. Splendid. And when those specialists have finished installing the AI agents, what exactly are they supposed to do next?

Meet your new colleague. It doesn't need a salary.

Imagine a company employing twenty customer service staff. It installs an AI agent that handles routine questions, tracks orders and processes simple requests. The company may still need people for difficult cases, but it could need fewer of them. That's part of the attraction.

Those customer service workers are often on modest wages. They may have little financial cushion, limited access to retraining and responsibilities that cannot be put on hold while they study machine learning. Announcing that the AI industry is hiring does not give them a new job.

Nor does it help the plasterer's administrator. The plasterer still needs to turn up, mix the plaster and finish the wall. The person who answered calls, booked appointments and chased invoices may discover that a monthly software subscription now does most of her work. We can applaud the efficiency and still regard the loss as a mortal shame.

Get on the AI bandwagon. Just don't buy a lifetime ticket.

There is money to be made building and installing AI systems today. Upwork reported annual growth of 178% in demand for AI integration skills, 71% for chatbot development and 154% for AI data annotation and labelling in its 2026 skills report. Those figures show a booming market. They do not show that the boom will last.

We've been here before. Y2K kept armies of specialists busy until the deadline passed. Windows rollouts filled diaries until deployment became more automated. Exchange Server experts had plenty to do until hosted email changed the business. Cloud office migrations created another profitable wave, then became increasingly routine.

AI agents may follow the same pattern, only faster. Once an agent has been built, tested and packaged, the next business may simply buy a version of it. Better still, from the customer's perspective, AI may help build and configure the next agent. A market can sell more AI while employing fewer people to install it.

But businesses are always changing!

Of course they are. Yet most changes are small: a new form, a different approval limit or a revised invoice rule. An agent designed to streamline a process may handle many of those adjustments without a consultant arriving with a day rate and an impressive diagram.

There will still be difficult projects, regulatory changes and strategic decisions. There may also be entirely new kinds of work. But nobody has demonstrated that these will absorb everyone displaced by automation. Hoping that they will is not an employment strategy.

Who is safer? The plumber or the payroll clerk?

A plumber must reach the leaking pipe. A carpenter must fit the awkward door. A plasterer must finish the uneven wall. Robotics may one day take on more of these jobs, but the physical world is a rather less obliging workplace than a computer screen.

Digital administrative work is easier to automate. That does not mean every office job disappears, or that trades are permanently immune. It means the risks are uneven, and the people most exposed are not necessarily the people best placed to profit from the AI boom.

We want AI. We also want an honest conversation.

The International Labour Organization found in 2025 that one in four workers worldwide held occupations with some exposure to generative AI. Exposure is not the same as losing a job, and many occupations will change rather than vanish. But the uncertainty cuts both ways: there is no evidence that new AI jobs will automatically replace every job lost.

So by all means learn to work with AI. If you're a consultant, get involved while businesses are buying implementations. Just remember that the point of a successful implementation is often to reduce the amount of human work needed afterwards.

And when the next technology evangelist assures us that the displaced customer service worker can simply become an AI architect, perhaps we might ask whether there's actually a vacancy, how long it will last and who's paying for the training.

AI may be wonderful for productivity. That does not make it a miracle cure for unemployment.

Read the evidence

For the historical comparisons, evidence, limitations and fuller argument, read our companion white paper, The AI Employment Mirage.