Are tech workers cooked?What seven years of data say about AI and tech jobs

The cleanest way to see what killed tech hiring is to look at demand before employment. Employers post jobs months before the hires show up in payroll counts, so postings move first.

Indeed's index of US software-development postings peaked in February 2022 at 2.3 times its pre-pandemic level. The Federal Reserve raised interest rates for the first time in that cycle on 16 March 2022. By November 2022, the month ChatGPT launched, software postings had already fallen 36%. They kept falling for another two and a half years and bottomed in May 2025 at 63, more than a third below where they stood before the pandemic.

Employment followed with the usual lag. Information-sector payrolls peaked in November 2022, nine months after postings did, and IT services peaked in March 2023. The coincidence with ChatGPT is a coincidence of lag. The hiring pipeline had been shut off by rates in the spring; the payroll count simply took that long to turn.

The year-by-year scorecard shows the sequence.

2021 and 2022 were the boom. The information sector added 380,000 jobs in two years and IT services added 261,000, financed by near-zero rates and a pandemic surge in digital demand. 2023 and 2024 were the correction: 197,000 information jobs and 52,000 IT-services jobs gone, with announced tech layoffs peaking at roughly 262,000 in 2023. All of this happened while national unemployment sat between 3.4% and 4.2%, the lowest in half a century. Tech cut while everyone else hired, because tech had over-hired more than anyone else.

If AI had been the driver, the losses would track AI capability, which improved steadily from 2023 onward. Instead they tracked the cost of money. That is the first finding, and it is uncomfortable for both camps. The people who say AI destroyed tech jobs are pointing at a rates story. The people who say AI has had no effect need to explain the next three years.

Corrections end. Once the excess hiring has been worked off, demand recovers, employers start posting again, and payrolls turn up. Every line of that script played out in 2025 and 2026 except the last one.

The Fed began cutting rates in September 2024. Across the whole economy, job openings returned to their pre-pandemic level: Indeed's all-jobs index stood at 104 in October 2026, and the BLS count of openings, 7.1 million, is roughly where it was in early 2020. Software-development postings bottomed in May 2025 and have risen 25% since. CompTIA counted 625,633 active tech postings in September 2026, the most in more than three years and 22% more than a year earlier.

And yet the information sector shed jobs faster in 2026 than in any year since the pandemic. It lost 103,000 jobs in the first nine months of this year, against 53,000 in all of 2025 and 68,000 in 2024. The pace, about 11,000 jobs a month, is slightly faster than in 2023, the year of the great tech layoffs. The year-on-year decline, 120,000 jobs or 4.2%, is the steepest since 2020.

That is the decoupling in one picture. Since January 2025, demand for software developers is up 16% and the number of people employed in the information sector is down 5%. Employers are posting more jobs and employing fewer people. In a normal recovery those two lines move together.

Two explanations are available. The first is macro. Inflation was back at 3.7% in September 2026 and the Fed had begun raising rates again, which squeezes the same rate-sensitive companies that cut in 2023. The second is that employers are hiring for different jobs than the ones they stopped filling, so headcount keeps shrinking even as postings rise. The macro story explains some of the weakness. It cannot explain why postings and payrolls point in opposite directions, or why the same divergence shows up in India, where US rates are beside the point. For that you have to look at what the new postings ask for.

Three independent sources, in three countries, using three different methods, describe the same change in the composition of tech demand.

In the United States, CompTIA's count of postings that ask for AI skills reached 349,821 in September 2026, up 282% since May 2024. That is 56% of all active tech postings. Indeed's economists looked at the rise in software-development postings between May 2025 and May 2026 and found that 37% of it came from roles with AI in the title or core requirements, and 71% from senior roles. Roles that require AI skills carry a wage premium of around 56%.

In India, Naukri's September report shows hiring for IT and software services down 4% on a year ago while AI and machine-learning roles are up 20%. Inside IT companies the split by experience is stark: AI/ML hiring rose 54% for people with 13 to 16 years of experience and did not rise at all for people with four to seven.

In China, Maimai's recruitment data shows AI postings up roughly eightfold in the first seven months of 2026, led by ByteDance, Ant Group and Tencent, while Alibaba ended 2025 with a third fewer employees and Baidu with 7% fewer. A summary of Maimai's 2026 report uses one word for what is happening to entry-level roles: de-juniorization.

The mechanism matters as much as the direction. The shrinkage is happening through the hiring door, not the firing door. The BLS layoffs rate was 1.0% in August 2026, the lowest since March 2025. The hires rate was 3.3%, close to levels last seen in 2010, when unemployment was near 10%. Announced tech layoffs this year, about 129,000 through 10 September, are half the 2023 total. Companies are not firing engineers in large numbers. They are not replacing the ones who leave, and when they do post a role it is a different role. That is what you would expect if productivity tools let a team absorb the work of a departed colleague, and it is why the losses show up as a steady drip rather than a wave.

Look at where the US losses sit and the pattern sharpens.

Over the twelve months to September 2026 the economy added 496,000 jobs. Health care alone added 372,000, and health care and social assistance together added more than the whole economy, which is why Moody's Mark Zandi describes payroll growth outside health care as "essentially zero". Construction added 109,000, much of it among the specialty contractors who build data centres. On the other side, insurance carriers lost 76,000, software publishing 33,000, IT services 32,000, telecoms 31,000, and data processing and hosting 23,000. Finance, information and computer-design services together shed 259,000.

The split is not tech against everything else. It is work that has to happen in person against work that happens in text, forms and queues. Claims processing, customer service, content production, routine software maintenance and back-office finance are the first places large language models are deployed, and they are the places where headcount is falling. Nursing, building and shelf-stocking are not.

If the new demand is for senior people with AI skills, the people shut out are the ones who have neither yet. The data on that is unambiguous, and it is international.

It is worth being precise about the direction, because it is the opposite of what a normal recovery looks like. Usually hiring returns at the bottom first: entry-level roles are cheap, and firms rebuild from there. This time the recovery is senior and the shortfall is junior. In the United States, 71% of the rebound in software-development postings since May 2025 went to senior roles. In India, hiring for people with 13 to 16 years' experience is up 7% while fresher hiring is up 1% and hiring for the four-to-seven-year band is down 2%. In China, the word Maimai's analysts use is de-juniorization. The jobs coming back are the ones that require judgment. The jobs that used to teach judgment are the ones not coming back.

In the United States, unemployment among 16-to-19-year-olds is 14.5%, three and a half times the national rate. In India, urban unemployment among 15-to-29-year-olds reached 19.0% in September 2026, against 6.9% for all urban workers, and 26.6% for young urban women. In China, unemployment among 16-to-24-year-olds who are not students was 18.9% in August, matching the highest level since the statistics bureau changed its method in 2024; for 25-to-29-year-olds it is 7.5%, and for 30-to-59-year-olds 3.9%. Three economies with different growth rates, demographics and monetary policy, and in each one young people are out of work at three to four times the headline rate.

The experience data points the same way. Naukri's September figures show fresher hiring in India up 1% while hiring for people with 13 to 16 years' experience is up 7%. Indeed's finding that 71% of the recovery in US software postings came from senior roles is the American version of the same fact.

The people who already have jobs are, for now, protected. US tech unemployment is 3.3%, below the national rate. But they are also stuck.

The quits rate is the best gauge of whether people believe they can find something better. It peaked at 3.0% in late 2021, during what was called the Great Resignation. In August 2026 it was 1.9%, lower than before the pandemic and the joint lowest since June 2020. Long-term unemployment, people out of work for 27 weeks or more, has risen from 21% of the unemployed in January 2025 to 27% in September 2026. Wages are growing 3.0% a year against inflation of 3.7%. Nobody is moving, and the people who fall out of the market are finding it harder to get back in.

The 2022 bust was global. Indeed's software-development postings index fell between 60% and 72% from its peak in every one of the six countries where it is published. What is different in 2026 is that the recovery is not.

The United States is the only one of the six where software postings are clearly rising, up 21.5% on a year ago. Australia is the only one above its pre-pandemic level. Germany's index fell 18% in the past year to a new low of 48, Britain's is stuck at 60, and France's is 52. In Germany and Britain the weakness is not confined to tech: all postings are down 12% and 10% respectively.

Two things are going on. The first is that Europe's economy is weak, so there is no AI rebound to see. The second is that the AI rebound in the US is a capital story. The companies posting for AI roles, and the data-centre construction showing up in the US construction numbers, are where the investment is going. Where the investment is not going, the correction simply continues.

India is the clearest case of a market that is growing and shifting at the same time. Nasscom reports the tech industry still adding people, around 135,000 in the financial year to March 2026, a 2.3% increase to 5.95 million. But the growth is in global capability centres, where hiring is up 4%, and in AI roles, while the IT-services business that built the industry is hiring 4% fewer people than a year ago. China has an AI hiring boom and a youth unemployment problem at the same time, which is only a paradox if you assume the boom is hiring the young.

I want to be careful about what this evidence does and does not establish.

It does not establish causation. Everything above is correlation across time and across countries. Rates, the over-hiring of 2021, the collapse of venture-funded start-ups, the long decline of legacy media and telecoms, and offshoring all push in the same direction as AI, and the information sector the BLS tracks includes film studios and phone companies as well as software firms. The most honest statement is this: AI is the explanation most consistent with the timing of the decoupling, with the composition of new postings, and with the age profile of who is losing out, in three countries at once.

It is not the whole picture. CompTIA's figures show 23 of the 30 largest tech occupations gaining postings in September. The BLS's own ten-year projections, published in August 2026, expect software developers to grow 15.8% by 2034 (267,700 jobs) and data scientists 33.5%. The occupation the BLS projects to shrink because of AI is customer service representatives, down 153,700. The official view is that AI removes clerical and support work and adds technical work. That is consistent with the sector data, and less dramatic than the headlines.

The data has holes. The BLS revised its 2025 employment estimates down by 898,000 in February, so the figures for the past year are less firm than they look. No household survey was run in October 2025 because of the federal shutdown. China changed its youth-unemployment method in 2024 after suspending publication, and reports in 2026 say companies there cannot cite AI as a reason for layoffs, so Chinese cuts are probably under-reported. The layoff totals come from a crowd-sourced tracker. Postings measure what employers want, not what they get.

For companies. The market is offering you senior, AI-fluent engineers at a moment when they are not changing jobs. Hire them. But the same market has stopped producing the next generation, and in five years the mid-level engineer you need will be the junior nobody hired in 2026. Companies that build deliberate apprenticeship pipelines now will own a scarce asset later. The other conclusion is for anyone selling headcount: IT services is shrinking in the US (1.3% fewer jobs in a year) and in India (hiring down 4%), and nothing in the data suggests that reverses.

For engineers. The premium is real and it is specific: AI skills applied to a domain, at a senior level. A wage premium of around 56% is the largest skills premium in any of this data. The flip side is that the quits data says the market will not reward moving right now; the leverage is inside the firm, in being the person who makes a team more productive.

For graduates, and for the people who teach them. Three countries, three definitions of youth, one ratio. Entry-level hiring is the part of the tech labour market that AI has already changed, and it changed before any policy response began. The practical advice follows from the data: the market is paying for judgment and AI fluency, so the fastest route in is to build demonstrable work with AI tools in a real domain, rather than wait for the junior roles of 2021 to return. They are not returning.

Interest rates ended the tech hiring boom in 2022. AI is the reason the bust has not ended in 2026. The sector is shrinking through attrition rather than layoffs, the roles being posted are senior and AI-specific, and the people locked out are the young, in Bengaluru and Shanghai as much as in San Francisco. None of this shows up in the headline unemployment rate, which is why it has been easy to miss. It shows up in who gets hired.