How Soon Until AI and Robots Handle Most Human Work?

The question keeps turning up on Reddit, usually from someone who has had enough vibes: how soon until AI and robots handle most human work? Skip the TED talk. Give me a date.

Fair enough. The short answer: by the end of September 2026, US employers had blamed AI for 120,136 job cuts, which is a lot for one technology and almost nothing for a labor market. McKinsey has 29.5% of American work hours automated by 2030. The forecasts that mean “most” land ten to twenty years out. Office work goes first, the warehouse second, and the fitted sheet last.

The count so far: 120,136

Challenger, Gray & Christmas tracks US layoff announcements and the reason each employer gives. Through the first nine months of 2026, employers named AI in 120,136 cuts. That is about 21% of the 573,195 cuts Challenger tracked, and more than double the 54,836 blamed on AI in all of 2025. In March, AI became the top stated reason for the first time in the firm’s history. In May it was named in 38,579 cuts, 40% of the month and the largest monthly AI figure Challenger has recorded. The technology sector, less than 2% of US employment, announced 165,925 cuts, 29% of the national total.

Now the part the doom threads skip. The country had 7.1 million unemployed people in September, so 120,136 is small change. AI’s five-month run at the top of the table ended in August, and by September it ranked fifth, with 3,961 cuts. Total layoff announcements were running 39% below 2025’s pace. Anyone who tells you most work is already gone is selling something.

The gauge that can’t see it

Unemployment should settle the argument, and in the September jobs report it was 4.2%. Over the twelve months to August the economy added an average of 45,000 jobs a month, against 166,000 a month in 2023 and 2024. The rate stayed put because the labor force has nearly stopped growing: the St. Louis Fed now puts the monthly gain needed to keep unemployment steady between 15,000 and 87,000, down from roughly 150,000. A flat rate used to be evidence that displaced workers were finding new jobs. With almost no new entrants to absorb, it has stopped being evidence of anything.

In August the acting Labor Secretary, Keith Sonderling, admitted that “the government does not have the data” on AI displacement. New York put an automation checkbox on its layoff-notice form in March 2025; more than 160 companies filed in the first year, and not one ticked it. Worldwide, the ILO counts 186 million unemployed people and a jobs gap of 408 million, meaning people who want paid work and can’t get it. The headline rate was missing more than half the problem before a single robot showed up.

Ten years, where the last one took thirty

Speed is why the date is close. Mechanization gave workers about 60 years to adapt, electrification about 40, computing about 30. Cognitive automation is on pace to give them about 10, because frontier AI keeps improving on a cycle measured in months. Community colleges still run on semesters.

McKinsey projects 29.5% of US work hours automated by 2030, and 12 million Americans needing a different occupation. In Mercer’s 2026 survey of 825 executives, 99% expect AI to reduce headcount within two years, and 43% plan to cut junior roles, up from 17% a year earlier.

The best counterargument deserves its numbers too. Indeed found US software development postings up roughly 15% since early 2025. But 71% of that net increase is senior roles, and in the most AI-exposed occupations, entry-level jobs fell from 29% of postings that list a salary to 10% between 2021 and 2026. The people already inside are doing fine. The door they came in through is closing.

Where these numbers live: two pages we keep current

Every figure above comes from two research notes on unscarcity.ai, both current through the September reports. The 2025-2030 Labor Cliff carries the layoff side: Challenger’s month-by-month counts and the argument for why this transition runs on a ten-year clock. Global Employment Statistics 2026 carries the jobs side: every BLS print with its revisions (July 2026 has been reported as a loss, then a gain, then a loss again), the Fed’s breakeven math, and the ILO’s 408 million.

On the podcast: Microsoft asks 60,000 people to volunteer

The October 28 episode of Minds, Bodies & Terawatts, “Microsoft Asked a Quarter of Its Workforce to Volunteer to Leave”, is this whole article inside one company. In the last week of October, Microsoft asked about 60,000 employees, a quarter of its workforce, to volunteer to leave. The company is growing 18% a year. It says AI isn’t the reason.

Most of us picture job loss as a failing business turning off the lights. Microsoft is the opposite case: a company that keeps growing while needing fewer of the people who got it there. Ask 60,000 people to decide for themselves whether they are surplus, and the company never has to say who it no longer needs, or why. Challenger records a voluntary offer only when someone accepts it, so none of the 60,000 appeared in October’s layoff total, a hole we describe in anticipatory displacement.

The trend line is already on the record: about 15,000 cuts in 2025 beside $80 billion invested in AI infrastructure, buyout offers to as many as 8,750 US staff in April, then 60,000 in October.

In the book: three engines and the last line standing

Why is the curve this steep? Chapter 1 of Unscarcity, “The Foundation,” answers with an invoice. Maria, the book’s house cleaner from Detroit, lives in an apartment that would cost about $300,000 to build today: roughly 40% labor, 35% materials, 10% energy, and 15% permits, profit and overhead. The section called “Three Engines” takes those lines away one at a time.

The brain: the Metaculus forecast for artificial general intelligence moved from 2057 to the early 2030s in six years, and the big cloud providers are on track to spend more than $700 billion on data centers and chips in 2026. The body: 1X’s Neo humanoid rents for $499 a month, about 69 cents an hour around the clock, while factory pay in Bangladesh averages well under a dollar an hour. The fuel: fusion ignition in December 2022, and a binding contract for 50 megawatts of fusion power, with the plant due to begin operating in 2028. The buyer is Microsoft.

The chapter counts the brakes too. Tesla missed its first-quarter 2026 production target for Optimus, deployed humanoids number in the low tens of thousands, and the fitted sheet “remains, for now, beyond them.” In the chapter’s words: “The curve is real. The curve is also slower than the slide decks.”

“The Last Line Standing” follows the invoice to the bottom. Materials turn out to be “energy and labor wearing a costume,” which leaves permits and land. “Robots can pour the concrete. They cannot attend the planning commission.” This is the hopeful half of the timeline: the curve that erases the paycheck also erases most of what the apartment costs. The chapter argues that housing, food and transportation can become infrastructure “cheap enough that metering it costs more than providing it,” and that the open decision is “whether we design a drainage system or let it flood the basement.”

The rest of the invoice, and the plans for the drainage system, are in the book.

So, how soon?

Nobody with a number says “most” by 2030. The forecasts that do say it land inside twenty years, which is less time than computing gave workers to adjust and a third of what the Industrial Revolution allowed.

The unemployment rate won’t be what tells you it has started. Watch the entry-level share of job postings, the twelve-month hiring average, and the buyout offers nobody is forced to accept. In October, one of those went to 60,000 people.

More questions like this one are collected on the blog page. Follow the series on Substack.