AI Is the First Machine That Follows Us Up the Stairs

You have a job today. The numbers say that has never been safer and AI will create more jobs than it destroys. They are not looking at what I am about to show you.

Global unemployment sits at just 4.9%, the lowest on record. That is about 186 million people without work. Widen the lens to everyone who wants work but cannot get it, the figure more than doubles to 408 million. Same source, same year, the ILO. One number fits on a poster. The other is 408 million people standing outside the building.

I hear this in conversations with friends. Capable people who want to move, who should be moving, who are staying frozen in roles they have outgrown because the door they expected to find open simply is not. The official story says the market is healthy. The people living in it tell me something quieter, and far less certain. That gap has been bothering me all week, because the headline numbers are built to hide what AI is doing to work.

How a falling rate hides the reality

The unemployment rate has a flaw most people never think about. It is a ratio, which can be misleading when the denominator is quietly growing.

In 2010, global unemployment ran at about 6.3%. That worked out to roughly 198 million people out of work. By 2026 the rate had fallen to 4.9%, which sounds like genuine progress, the kind of drop you would expect from a healing world.

So the rate fell and the actual number of human beings without work dropped by around twelve million over sixteen years. During which the global labour force grew by more than half a billion people, from roughly 3.15 billion to 3.7 billion.

The rate did not fall because joblessness collapsed. It fell mostly because the denominator ballooned. More people poured into the workforce, so the same rough quantity of unemployment became a thinner slice of a much bigger pie.

And those 408 million in the jobs gap? The unemployment rate cannot see most of them. To be counted as unemployed you have to be actively looking and available to start. The moment you give up looking, or you are caring for a relative, or the survey simply cannot reach you, you vanish from the figure. You are not employed. You are not unemployed. You are not counted.

This is the metric people lean on when they tell you the labour market is healthy. It is not lying, exactly. It is just answering a far narrower question than the one we think we are asking. Which is why it can say "record low" while the people actually trying to move tell you something completely different.

Let me give the optimists their best argument

I want to be fair to the other side, because their case is genuinely strong and worth stating at full strength.

The optimist points to history, and history is on their side. Research by the economist David Autor and his colleagues at MIT found that around 60% of the jobs Americans did in 2018 didn't exist in 1940. Sixty per cent! Whole categories of work invented inside a single lifetime. Nobody in 1940 was an App Developer, a Social Media Manager, or a Cloud Architect.

So the argument writes itself. Technology has always destroyed jobs and always created more. The people who panicked about the tractor, the spreadsheet were all wrong. The work changed; it did not vanish. Bet against that pattern, the optimist says, and you are betting against two hundred years of evidence.

It is a good argument. I take it seriously. And I think there is a flaw in it that most people repeating the 60% figure never mention.

The flaw in the optimist's favourite statistic

The same body of research that gives us the cheerful 60% figure has a second finding attached, and it is far quieter.

Autor's team also found that automation eroded twice as many jobs between 1980 and 2018 as it had between 1940 and 1980. And while new technology kept adding roles, the creation no longer kept pace. It added fewer jobs than automation took away.

Read that slowly, because it changes the whole picture. The engine the optimist is counting on, the one that has reliably produced new work to replace the old, was already losing power decades before anyone typed a prompt into a chatbot. The replacement machine was sputtering in the 1980s. The reassuring statistic and its inconvenient footnote come from the very same study.

So the honest version of the optimist's claim is not "new jobs always replace the old ones". It is "new jobs still replace the old ones, just slower with each passing decade". That is a different sentence. It points somewhere darker.

Why this time might actually be different

Now I have to deal with the phrase that should make all of us suspicious of ourselves.

"This time is different".

It is the most dangerous sentence in two of my favourite topics, economics and investing. People said it before every crash. They said it about electricity, which a surprising number of people feared and resisted. They said it about computers. Every time, the doom was overstated, and the sceptics ended up looking foolish.

So when I feel the urge to say AI is different and will create more jobs, I check myself. The base rate on that claim is terrible. Most people who have ever said it were wrong.

But here is what I keep coming back to, and I would genuinely like to be argued out of it.

Every machine we have ever built took over something physical or routine. The tractor took over the digging. The factory robot took over the bolting. The spreadsheet took over the adding up. And every single time, the same escape hatch opened. The machine did the muscle work or the repetitive work, and humans climbed upstairs into the work that needed judgement, taste, and decisions. That upstairs room, the thinking room, is where most good jobs live now. It is why the electricity sceptics were wrong. Electricity never came for anyone's judgement. It just powered the tools they used to apply it.

AI is the first technology pointed directly at the upstairs room.

It does not just dig or bolt or calculate. It drafts, analyses, designs, advises, decides. For the entire history of automation, "move up into thinking work" was the escape hatch humans used. AI is the first machine that follows us up the stairs.

That is the argument. And notice what it is not. It is not a feeling that things seem frightening, the same vague dread the electricity sceptics had. It is a specific claim about how this differs: every other technology pushed us up the stairs to safer, smarter work. AI is the first one following us up.

The track record of "this time is different" is genuinely awful, and I might be the latest person about to look foolish for saying it. But this is the first time the claim comes with a mechanism rather than just a mood. So I am holding two things at once: the history says I am probably wrong, and the logic says this time might be the exception. That tension is where I actually sit, and I am not going to pretend it away to sound more certain than I am.

Where I sit, and why it is personal

I have spent a decade inside agile teams. I work on transformations for a living. And I have already said in public, that half of what my role looks like today is being competed away by AI agents. The synthesis, the documentation, the pattern-spotting across a wall of workshop notes. No need to spend half a day on that anymore. Now I photograph the wall, drop it into a tool, and the patterns surface in minutes. That work did not vanish. It just stopped needing me.

But the other half, the facilitation, the judgement, the hard human conversations a model cannot have for you, is becoming more important, not less. My role is not disappearing. It is splitting, and the difference between the professionals who thrive and the ones who fade is whether they can tell the two halves apart.

So I am not studying this from a safe distance. I am one of the people standing in the upstairs room, watching which parts of it the machine can now reach.

So I am not going to hand you a tidy forecast. I will not tell you the jobs are not coming back, because I cannot prove it, and an honest voice does not dress a worry up as a prediction.

But here is what I can prove, with numbers you can check yourself. The one figure we use to tell ourselves everything is fine is built to hide the exact damage we should be watching for. And the one pattern we lean on for comfort, that new work always replaces the old, was already breaking down before AI showed up.

The question underneath the question

Some of the people building this technology already talk about universal basic income. They are not being generous. They are quietly conceding that the work may not come back, and a monthly payment is the patch.

But a payment was never the point of work. Work is structure. It is identity. It is the feeling that someone, somewhere, needs what only you can do. Strip that out and hand people a cheque, and you have solved the rent while gutting the reason.

So no, I cannot tell you how many jobs disappear, or how fast. Anyone who gives you that number is guessing. But I can tell you what we should actually be measuring, and it is not the unemployment rate.

Because that number will keep looking healthy for a while. It was built to. By the time it finally moves, the damage it was supposed to warn us about will already be done.

Two questions worth sitting with. Which half of your job can the machine already do? And if that half keeps growing, what is left that is only yours?

I know which half is mine. I am less sure it stays that way.