Every week another CEO writes a letter that goes something like this: “Intelligence tools have changed what it means to build and run a company. We are reducing our workforce by [X]% to operate leaner and faster.” Markets cheer. Stock pops. The narrative writes itself: AI is replacing workers en masse, the future is here, get out of the way.
The data doesn’t back this up. Across the entire U.S. economy in 2025, jobs explicitly cut for AI reasons totaled 54,836 — that’s 5% of total layoff plans for the year, according to Challenger, Gray & Christmas, the firm that actually tracks layoff reasons. Across three years (2023–2025), the cumulative number cited as AI-driven is 99,470, which is 3.5% of all layoff plans in that window.
Meanwhile, U.S. productivity grew roughly 2.7% in 2025, nearly double the prior decade’s average. Business applications hit close to 500,000 filings per month in late 2025. 60% of new business founders used AI to launch in 2025, double the rate of two years earlier. New companies are forming with smaller teams from day one, and the existing workforce is producing more per head.
So which is it? AI is replacing 40% of a workforce, or it’s making everyone more productive while new businesses sprout up faster than ever?
Both stories can’t be true. One of them is.
The mass-layoff narrative is mostly press-release engineering#
Let me lay out the pattern, then go through the receipts.
The pattern: Company X tripled headcount during the pandemic on the assumption that 2020–2021 demand was the new normal. It wasn’t. By late 2023 they’re sitting on bloated cost structures and slowing growth. They’ve also bet on a bunch of “next platform” initiatives that didn’t pan out. They need to cut. They could say “we overhired and bet wrong on the next platform” — which makes the CEO look bad. Or they could say “AI has fundamentally changed how we build a company” — which makes the CEO look prescient and pops the stock.
Guess which one they pick.
Block is the loudest recent example. Dorsey just laid off 4,000 people (40% of the company) and the shareholder letter opens with the AI thesis. Stock rallied 16%. What the letter doesn’t open with: Dorsey acknowledged internally in November 2023 — well before this AI narrative — that “the growth of our company has far outpaced the growth of our business and revenue,” and imposed a 12,000 headcount cap. Then came five rounds of cuts totaling more than 7,000 positions while gross profit roughly doubled. The company tripled headcount during the pandemic. The math was always going to need to work itself out. The Web5 initiative was killed entirely. TIDAL is being scaled back after a $300M acquisition produced negligible quarterly revenue. Bitkey is shipping but immaterial. Out of every “next big thing” Dorsey has announced since 2021, exactly one (Proto Bitcoin mining) is a real new product line.
That’s not an AI revolution. That’s a CEO using the moment to clean up a balance sheet that needed cleaning regardless.
Klarna is the cleanest case study because it ran the full loop. CEO Sebastian Siemiatkowski boasted in 2024 that AI was “doing the work of 700 employees” and that the company had stopped hiring “largely due to AI.” For a while the numbers looked great. By May 2025, Siemiatkowski admitted Klarna had “gone too far” and “focused too much on cost.” Customer satisfaction had cratered. Complaints surged. Users called the AI responses “generic, repetitive, and insufficiently nuanced.” Klarna started rehiring humans. They’re now running a hybrid model where AI handles tier-one volume and humans handle anything that requires judgment, empathy, or escalation. The savings from the original AI substitution turned out to be partially offset by the cost of unwinding it.
IBM is the most over-hyped case. The viral story is “IBM laid off 8,000 HR employees and replaced them with a chatbot.” The actual story, per CEO Arvind Krishna in the Wall Street Journal: the real number of HR roles replaced was in the hundreds. IBM’s total headcount grew during this period. They reinvested savings into engineering, sales, and AI roles. AskHR (the chatbot) handles 94% of routine HR questions well. The other 6% — sensitive issues, judgment calls, complex situations — still need humans, and the gap caused real morale and operational problems internally.
Amazon cut 14,000 corporate roles in October 2025, the largest single cut in its history. The framing referenced AI and “biggest bets.” Amazon also overhired massively during 2020–2021. Andy Jassy has been steadily reversing that overhiring since 2022, well before generative AI was anyone’s stated reason for anything.
Salesforce — Marc Benioff has cited AI repeatedly for various cuts. Salesforce also did a 10% workforce cut in 2023 that they explicitly attributed to COVID overhiring at the time. The 2023 framing matches what’s actually happening; the 2025–2026 framing has shifted to AI because AI is the more flattering story.
This pattern keeps reappearing. The Oxford Internet Institute’s Fabian Stephany puts it directly:
Many firms overhired during the pandemic, and current layoffs may reflect a market correction rather than actual AI-driven displacement… It’s to some extent firing people for whom there had not been a sustainable long term perspective, and instead of saying “we miscalculated this two, three years ago,” they can now come to the scapegoating, and that is saying “it’s because of AI.”
Wharton’s Peter Cappelli is even more blunt:
The companies that are laying off are not struggling. The reason companies — especially the rich tech companies — are cutting jobs is not because they are in financial trouble or that AI has taken jobs or that they know already which jobs will be taken over.
Forrester’s J.P. Gownder:
We know that AI is not yet doing the work to justify the level of layoffs that have been happening. When you try to identify the very specific jobs and roles impacted by AI, the picture starts to get cloudy.
And the most telling admission, from someone with every commercial incentive to claim otherwise — OpenAI CEO Sam Altman at the India AI Impact Summit:
There’s some AI washing where people are blaming AI for layoffs that they would otherwise do.
When Sam Altman is telling you AI-washing is real, AI-washing is real.
A 2025 IBM survey found that only 1 in 4 AI projects delivers the return it promised and even fewer get scaled up. Orgvue/Forrester research found that 55% of companies that rushed to replace workers with AI now regret the decision. The aggregate picture is not “AI is silently doing everyone’s job.” It’s “AI is doing parts of some jobs, often less well than expected, while companies use the narrative to do something they were going to do anyway.”
What the AI economy actually looks like#
If you turn away from press releases and look at the data, the actual shape of what’s happening is clearer than the headlines suggest.
Productivity per worker is rising. Erik Brynjolfsson at Stanford — who has been studying technology and productivity since before ChatGPT existed — estimates U.S. productivity growth in 2025 at roughly 2.7%, nearly double the 1.4% decade average. The 2025 jobs print got revised down to 181,000 from an initial 584,000, while GDP grew 3.7%. Fewer workers, more output. That’s the literal definition of productivity growth.
Brynjolfsson frames AI as a classic J-curve technology: companies invest, restructure, and absorb the disruption first. Measurable output comes later, in what he calls the “harvest phase.” Every general-purpose technology has worked this way — electricity, the internal combustion engine, the PC. A decade or so of restructuring, then output surges.
New business formation is hitting historic highs. U.S. Census Bureau data shows business applications climbed to roughly 500,000 filings per month in late 2025. Per Gusto’s 2026 report, 60% of new business owners used AI to help launch in 2025 — double the rate from two years earlier. Businesses started in 2024 in AI-enabled industries are operating with roughly 6% lower headcount at 12 months than 2023 cohorts. In the SF Bay Area specifically, 16% lower.
Smaller new firms, more of them, more productivity per worker. That’s not “mass exodus of jobs.” That’s a restructuring of how work gets done.
The labor market shift is real but narrow. Brynjolfsson’s own research found AI is hitting entry-level workers disproportionately, especially those aged 22–25 in highly AI-exposed occupations. UK graduate job postings are down 67% since 2022; US graduate postings down 43%. Whether that’s AI, economic uncertainty, post-COVID normalization, or offshoring is genuinely contested, but the directional concentration on entry-level white-collar work is real.
So the honest summary of where we actually are:
- Aggregate AI-attributed layoffs are small — single-digit percent of total layoffs, even with employers eager to claim the label.
- The pandemic overhire correction is large — most of what’s getting labeled “AI” is this.
- Productivity per worker is genuinely rising — measurable in the macro data.
- New firms are forming faster, smaller, and more AI-native — near-historic application rates.
- Entry-level white-collar work is taking real damage — narrow but real displacement.
- The bigger labor effect so far is “no-hire, no-fire” — companies aren’t replacing people who leave, which doesn’t show up as layoffs but shows up as fewer openings.
None of that matches the “AI replaced 40% of the workforce in a year” headline. The headline is wrong by roughly two orders of magnitude.
What this means for engineers, founders, and operators#
A few things follow from the actual data rather than the narrative.
Be skeptical when a company cites AI for layoffs without showing the work. Specifically: which workflows were automated, what productivity number changed, and what the rehire rate looks like 12 months later. If a CEO can’t tell you those things, “AI” is the press release version of “we overhired.”
Watch the rehire patterns. Klarna and IBM are the early cases. Forrester’s data suggests this will be widespread — “half of all AI-related layoffs will be quietly rehired” is the prediction, and the pattern is already visible. If the savings story were real, you wouldn’t see the rehires.
The real AI signal is in productivity and new-business numbers, not layoffs. 2.7% productivity growth and 500k monthly business applications are the actual indicators of a productivity-driven economic shift. Layoff announcements are noise on top of an overhiring correction.
The companies actually being transformed by AI are the new ones. If 60% of new founders are using AI to launch and the new-cohort firms are running 6–16% leaner from day one, the real AI story is happening at the formation end of the economy, not the layoff end. The interesting question isn’t “which incumbents will cut 40%?” It’s “what new categories of business become possible when a three-person team can do what a thirty-person team used to do?”
Entry-level hiring is the actual risk zone. If you’re managing a team, the AI productivity gain is letting you skip the bottom rung. That’s also where you used to develop the next senior engineer. Skipping the bottom rung saves money today and creates a talent crater in three to five years. Worth thinking about before you defer hiring junior people indefinitely.
Don’t confuse a stock pop for a strategic insight. Block’s stock rose 16% on the layoff announcement. That’s the market rewarding the framing, not validating the strategy. Klarna got a $19.65B IPO valuation partly on the AI-first narrative — months later, they were rehiring humans. The market is responding to a story; the story is downstream of business reality, not upstream of it.
The story that’s actually happening#
There’s a real AI transformation underway. It’s not the one in the headlines.
The headline story is: AI is replacing humans en masse. Companies are getting leaner. Workers are getting displaced at unprecedented rates. The future is here.
The actual story is: companies overhired during the pandemic and are correcting, using AI as the press release because it sounds better than admitting the miscalculation. Meanwhile, AI is genuinely raising productivity per worker, new firms are forming at near-record rates with smaller native teams, and the long-run economic shift looks more like the women-entering-the-workforce era or the post-electrification productivity boom than a mass-displacement event. Entry-level work is taking real but narrow damage. Most of the laid-off workers will get jobs again, often at different companies, sometimes at companies that don’t exist yet because the founders haven’t started them.
These are very different stories. One is a horror movie. The other is a slow, awkward transition with real winners and real losers but no apocalypse.
Tell the difference, and the next five years get a lot easier to navigate.