AI may be creating more jobs than you think

Hello! 👋

It's Thursday, 10^th September 2026. Hello and welcome back to Bold Efforts. This week, two stories about AI appeared within a few days of each other.

Before I continue, some announcements first:

Anyways. This week, two stories about AI appeared within a few days of each other.

Story 1

On September 8, OpenAI announced that an internal AI system had produced a proposed solution to the Navier–Stokes existence and smoothness problem, one of mathematics’ Millennium Prize Problems. The question has remained unresolved for roughly 90 years.

I actually remember learning about this equation around 12 years ago while studying chemical engineering. At the time, I would never have imagined that I would have to encounter it again. But here we are.

OpenAI says around 10,000 agents worked on the problem concurrently, exchanging 2.7 million messages and producing roughly 130 billion output tokens. They reached their result in 88 hours, followed by another 17 hours of formal verification.

Whether the proof survives the scrutiny of mathematicians matters enormously. But even before that process is complete, the experiment tells us something about where AI capabilities are heading.

A machine was not merely summarising existing knowledge or writing another piece of software. Thousands of artificial agents were coordinated against a research problem that generations of very smart humans had failed to resolve.

Story 2

Four days earlier, The Economist published a different story.

Its analysis estimated that AI has so far been associated with roughly one million new jobs in America, compared with around 200,000 layoffs attributed to the technology since mid-2023.

The estimate is necessarily fuzzy, but the direction is interesting. Despite extraordinary improvements in what AI can do, there is little evidence yet of an economy simply running out of work.

Put these stories next to each other.

AI may be getting capable enough to help solve one of mathematics' hardest problems. And the jobs apocalypse keeps getting postponed.

I don't think these two things are contradictory.

Most arguments about AI and employment start with an intuitive model.

There is a certain amount of work that needs to be done. Humans currently perform it. Machines become capable of performing some percentage of it. Therefore, there must eventually be less work left for humans.

The arithmetic makes sense but the assumption driving it does not. Economies do not contain a fixed quantity of useful work. When something important becomes dramatically cheaper, we usually consume much more of it.

Computers did not merely automate calculations that people were already doing by hand. They made it economical to perform calculations that nobody would previously have bothered doing.

The internet did not simply replace letters with email. It increased communication by orders of magnitude.

Software did not eliminate the need for businesses to build things. It allowed businesses to build millions of things that would previously have been uneconomical. AI may do something similar to intelligence.

A company that once commissioned ten pieces of analysis may commission a thousand. A scientist who could investigate three hypotheses may investigate three hundred. A small business that could never afford a lawyer, designer, analyst, or software team may suddenly consume some version of all four.

As the cost of producing cognitive work falls, the amount of cognitive work we demand can rise.

The interesting question is therefore not simply how much existing work AI can automate. It is what we choose to do once intelligence becomes much cheaper.

And, to be fair, anyone who has worked in a large organisation already knows that work possesses a mysterious ability to expand to fill whatever time is available. Some managers appear deeply committed to preserving this law of nature.

When people imagine AI increasing productivity, they often picture the same company producing the same output with half the employees.

That will happen in some places.

But productivity can also mean producing far more with roughly the same resources. AI coding tools may reduce the effort required to build software, but that also makes far more software worth building. The same could happen with research, analysis, design, education, and scientific exploration.

So perhaps the better question is not how many people it will take to perform today's work. It is how much more work becomes worth doing.

None of this means disruption is imaginary. Some jobs will disappear, some teams will shrink, and some skills will lose value surprisingly quickly. The people displaced will not necessarily be the people who benefit from whatever comes next.

But capability also does not translate instantly into adoption.

A model might be able to perform part of a lawyer's, analyst's, engineer's, or consultant's job. Actually replacing that work requires companies to redesign workflows, connect data, manage errors, assign responsibility, and make the whole system reliable.

I have seen this myself while building with AI. Increasingly, the model is not the hardest part. The difficult part is deciding what should actually change, where AI improves the economics, where humans remain involved, and how everything works together.

Even OpenAI's Navier–Stokes experiment makes this visible. The interesting part was not just a powerful model. It was the system around it: thousands of agents exploring different directions, exchanging information, discarding dead ends, and consolidating promising approaches.

The models will improve. But so will everything we build around them.

Last month, I wrote about what becomes scarce when intelligence becomes abundant.

There is another consequence of cheaper intelligence that deserves attention.

The frontier of worthwhile activity expands.

When something becomes 10x or 100x cheaper, we stop merely asking how to perform today's work more efficiently. We start doing things that were never worth doing before.

That is why I would be careful with both extremes in the AI jobs debate. AI may destroy some jobs. It may create others. The transition could be messy.

But the belief that increasingly capable AI must eventually leave humans with nothing valuable to do rests on a strangely unimaginative view of the future.

It assumes we have already thought of everything worth doing. We clearly have not.

For most of history, progress has not meant finishing the world's list of tasks faster. It has made the list longer. AI might do the same. Except this time, it may also help us figure out what to add to it.

Thank you for reading.

Best,
Kartik

I write Bold Efforts every week to think clearly about where work and life are actually headed. If you want these essays in your inbox, you can subscribe here.