More Problems, Not Fewer People

The AI debate keeps asking the wrong question.

For decades, in both our utopias and our dystopias, the question has been the same: which humans can the machine remove? It’s a tempting frame, and I understand its pull. Machines don’t get sick, don’t need breaks, don’t quit. Some low-value tasks can be, have been, and will be automated, and pretending otherwise helps nobody, least of all the people doing those tasks.

But replacement is a small ambition. At its very best, it gets you the same output, cheaper. The more interesting question, the one almost nobody in the debate is asking, is this: what could these people do that we couldn’t afford before?

What IKEA did instead

IKEA’s parent company, Ingka Group, gave us a real answer. In 2021 they launched a chatbot called Billie across their customer service channels. Within two years it was resolving nearly half of all inquiries, the routine volume: order status, delivery tracking, returns. On paper, that is the setup for a layoff announcement. A bot absorbs half the workload; the 8,500 people who handled that workload look redundant.

Ingka looked at the other half of the calls instead. Buried in what Billie couldn’t handle was demand for something different: people asking for help designing their rooms. How do I make a small apartment feel warm. What works in this awkward corner. That demand had been invisible for years, because agents drowning in routine queries had no capacity to notice the pattern, let alone act on it.

So they retrained those 8,500 workers as remote design advisors. People who had spent years absorbing the catalog and the customers were suddenly doing consultative work no chatbot could touch. The remote design channel is now credited with revenues around 1.3 billion euros a year. This figure is for the whole channel and not the reskilled workers alone, but it still speaks. The bot took the routine. The humans moved up.

Notice what happened there. The value wasn’t in the automation. The automation saved a respectable 13 million euros. The value was in the capacity it released, and in an organization curious enough to ask what that capacity could reach.

We’ve seen this movie before

If “the work moves up” sounds like wishful thinking, consider that we have already lived through exactly this cycle, in my own field.

When the web began, nobody knew what the medium was. People reached for analogies to make sense of it: a website was like a car, like a radio station, like a brochure. Every project meant discerning basic questions from scratch. How should a person select one option among several? How do they know something is clickable? It took years of collective trial, error, and study to distill those questions into solved conventions: radio buttons, checkboxes, navigation patterns, the components now embedded in every framework.

That solved layer did not eliminate web work. It moved it. There’s always room to make something more polished, a bit more intuitive. But the frameworks answer most of the questions that consumed the early years, and building a simple website stopped being a big project. You follow conventions, and you get a satisfying result. The human work migrated upward, to harder, more abstract problems the conventions can’t answer.

I recently lived the next turn of that cycle myself. I built my own website with Claude Code: custom-made, following conventions, no template and no developer. The solved layer plus an AI assistant extended my reach into territory that used to require a specialist. And here is the part that matters: for a complex project, I would still want the developer, even though the AI could produce the code. He has the knowledge. He knows which conventions apply, where the edge cases hide, what will break at scale. The AI should extend his reach, not stand in for his judgment.

I won’t pretend this transition is painless. Some simple development work genuinely is disappearing, the way template-website shops faded before it. The point is not that nothing changes for anyone. The point is the shape of the change: the solved problems get absorbed, and the human work moves up to the unsolved ones. That is what technological maturity has always done. AI is now doing to many more domains what frameworks did to the web.

The capacity horizon

Which brings me to the frame I keep coming back to: AI as a reach extender, not a replacement. I’ve written before that these tools extend what you can do while leaving you responsible for what comes out. The IKEA story and the web story show the same principle operating at the scale of a workforce and a profession.

And notice what the reach frame reveals that the replacement frame hides. We are not running out of work. We are surrounded by problems nobody solves because nobody has the capacity. Services never offered because the routine ate every hour. Demand that stays invisible because no one has room to notice it. Every hour of routine the machine absorbs is an hour of human judgment, care, and skill that becomes available for something previously out of reach.

The organizations that treat automation as a headcount question will get a cheaper version of what they already had. The ones that treat it as a capacity question will get things that didn’t exist before. IKEA found a billion-euro service hiding inside its call centers. What’s hiding inside yours?

So when the machine takes half the work, the question isn’t who to let go. It’s what we finally get to take on.