The layoffs are the product demo

American Express is paying $336 million in severance to remove about 5% of its workforce, and the company's chief financial officer is careful about what to call it. This is not a cost cut; it is a re-organization ahead of automation. ^[1]

You don't want to automate processes that are not optimally structured, and so what we're doing here is to reduce layers and increase spans of controls, so that we can automate and take advantage of AI in the coming quarters and the coming years. — Christophe Le Caillec

The charge reads as a prepayment: money leaving the payroll ledger now so machines can take over the entries later. And the same posting keeps appearing elsewhere. Intuit cut 17% of its workforce, explicitly to fund AI investment built on models from OpenAI and Anthropic. Meta shed 8,000 people while raising its capital-spending guidance to $145 billion. ^[2] Challenger, Gray & Christmas had already counted nearly 50,000 US layoffs tied to AI by mid-year, and the labor experts in that coverage warn companies may be using the AI story to frame the cuts positively for investors. ^[2] The tally for the year already runs past 190,000 US tech jobs. ^[3] On the other side of the trade sit the sellers, and they price their agents against the wages of the people they replace. Goldman Sachs has put agentic software engineers onto production work, and its chief information officer describes the agent as a new employee even as the chief executive weighs restricting headcount. ^[4]

Devin is "like a new employee" — Marco Argenti

Workday's chief executive is explicit about the arithmetic: agents are no longer a claim on a customer's software budget but on its labor and services budget, their value defined against the pay of the new hires they replace. ^[5] In law, the entry-level associate is the business model — a few hundred equity partners supervising thousands of associates whose billable hours produce the revenue — and the agents aimed at those hours already carry private-equity valuations: Harvey at $15 billion, Legora at $5.6 billion. ^[6][7] Then there is the entry where every other posting converges. Anthropic's prospectus is public now: a $965 billion private valuation with projections up to $2 trillion, and a roadshow set for November 9 with investors trimming expectations toward $1.5 trillion. ^[8][9] The same week, the company self-published a study claiming its Claude agents now log more work hours than human employees, and that its own entry-level technical hiring fell from nearly 30% to 9% since March 2025. ^[10] Critics quoted in the coverage accuse the company of publishing the productivity claims to inflate its valuation ahead of a possible IPO. The findings are unaudited. ^[10] The window this is aimed at is visibly closing. Listed AI names have shed $150 billion since their January debuts, and OpenAI has postponed its own listing to 2027. ^[11][9] One venture investor counted the Magnificent Seven last year at roughly $560 billion in capital spending against about $35 billion in actual AI revenue. ^[12] Sam Altman has already conceded the risk on the record. ^[13]

Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes. — Sam Altman

Which brings the audit. MIT researchers found 95% of generative AI pilots produced no measurable effect on profit. ^[14] S&P Global put enterprise failure rates at 42%. ^[15] Only 2% of the enterprise leaders surveyed expect a fully autonomous procurement function. ^[16] Ford rehired hundreds of engineers after its automated inspection systems failed. ^[17]

We had been relying more and more on automated quality systems and not getting the desired results. — Kumar Galhotra

None of this makes the displacement itself fiction. Stanford's economists find the measurable hit lands exactly at the entry level — customer service and junior developers — and Pantheon Macroeconomics counts more than 10,000 US workers a month displaced, concentrated in the accounting, advertising, and legal jobs at the center of the swap. ^[17][18] The cuts are real. What they are not, yet, is evidence of capability. The sellers' own answer to that gap is a new human workforce. Postings for forward-deployed engineers — the staff who install the agents at the customer — have risen from under 200 in 2023 to more than 7,300 in the first nine months of this year, at a median advertised salary of $195,456. ^[19] The industry's fix for agents that cannot yet convert enterprise investment into results is armies of expensive humans. The layoffs are being narrated as proof of capability; on the industry's own delivery numbers, they are evidence of valuation.