AI Didn't Cut Those Jobs. Your CFO Did.

On Friday, American Express told analysts it will eliminate 5% of its global workforce, roughly 3,500 people, and book a $336 million pretax restructuring charge. On the same earnings call, the company raised its 2026 profit forecast above Wall Street expectations. Chief executive Stephen Squeri framed the cuts as making the firm leaner and more agile; the CFO's version was blunter in press interviews: fewer operational layers, more automation.

A day earlier, Barclays said it would cut 5,000 roles in what the wires described as an AI-powered cost-cutting drive. Axel Springer is trimming about 100 jobs in a newsroom restructure it attributes to AI. In Beijing, Reuters reports, the government is now instructing its own agencies to soften AI-related displacement before it lands.

Put those four items side by side and the media read is obvious: AI has started eating white-collar work. I think that read is lazy, and it is becoming the most convenient excuse in corporate communications.

The $336 Million Charge Is the Tell

Notice what Amex actually did. It did not say that a model replaced a team. It took a restructuring charge, cut 5% of staff, and pointed at automation as the direction of travel while revenue and profit guidance went up. That is not a technology story. That is a cost program with a press-friendly label.

The label matters because companies choose it strategically. Layoffs justified by efficiency invite accusations of greed and trigger uncomfortable questions about buybacks. Layoffs justified by AI arrive pre-authorised, because the story is that the technology makes the decision, not the executive. Nobody has to be the villain when the model takes the blame.

Goldman Sachs economist Joseph Briggs made the analytical version of this point on Thursday. "Concerns of imminent AI-related labor displacement have cooled in recent months after hitting a fever pitch this spring," he wrote, before laying out why the quiet is temporary. Digging through decades of technology shocks, he found that "automation-related labor disruption is limited when the overall economy is healthy but accelerates during cyclical slowdowns." The share of Americans in routine jobs, he notes, has been stable during expansions and has fallen in steps at each downturn since the early 1980s, never climbing back.

In other words, the AI layoffs are queued, not caused. They are waiting for a recession-shaped permission slip. Under Goldman's modelling, up to 9% of workers, about one in eleven, could be displaced through the AI transition, with slower hiring dragging employment down another three points. That is a real number, and it deserves attention. But it is a forecast about timing and the business cycle, not evidence that Friday's cuts were technically inevitable.

Pantheon Macroeconomics, tracking AI-exposed sectors month by month, reckons the net US job loss is running above 10,000 roles a month. Read the composition and it gets stranger: construction employment is up, because data centres are being poured as fast as concrete allows, while accounting, advertising and back-office functions shrink. A technology that is simultaneously hiring builders and retiring bookkeepers is not a simple job-killer. It is a reallocation with a body count.

Where AI Actually Adds Headcount

The jobs growth story is real too, and it is almost invisible because it happens four people at a time. New US Chamber of Commerce research finds small businesses are nearly eight times more likely to report hiring caused by AI than cuts caused by it.

• 47% of US small businesses report direct job growth driven by AI, against 6% that say they let people go because of it.
• About two-thirds now use AI somewhere in the business.
• More than 80% of AI-using small businesses grew sales over the past year.
• In Oklahoma specifically, 94% of small businesses using AI reported workforce growth over the last twelve months.

Jordan Crenshaw of the Chamber's Technology Engagement Center offered the least glamorous and most believable example in the whole debate: a coffee shop owner who cut the time spent building his own marketing and website copy by roughly 70%, then used ChatGPT to design a bagging rig that halved his packing time during the Christmas rush. The result was more product out the door and more hours on customers. That is a headcount story nobody will ever put on a front page, because a five-person shop hiring a sixth does not generate a news cycle or a monthly statistic.

I am not arguing that AI displacement is a myth. Goldman's 9% is plausible precisely because it is modest per year and brutal cumulatively. My argument is narrower and, I think, more useful: the AI jobs debate has become unfalsifiable. When a company raises guidance and cuts staff in the same breath, and we file it under AI, we lose the ability to tell a genuine capability shift from ordinary margin management dressed in a robot costume.

Three things worth watching instead of the headlines:

• Whether a company discloses which roles it eliminated, not just a percentage, and whether it hires back into AI functions at the same site.
• Whether the restructuring charge is presented as one-off cost repair or as an ongoing reallocation, since the second is a technology claim and the first is not.
• Whether governments design mobility programmes, as New York's AI Jobs Institute and China's agency guidance attempt, or simply absorb bad quarterly numbers as evidence of the future.

The next time a company announces thousands of layoffs and blames artificial intelligence, read the earnings call first. If the profit forecast went up in the same hour, the model did not make that call. A human did, and a human chose what to call it.