What happened to the companies that cut staff for AI?

Four companies are cited in almost every leadership room where the question of cutting staff for AI comes up, and the record on them is better than the folklore. One cut and rehired. One declared and retreated. One cut and reinvested. One froze and held. Read together they show that the cut was never the decision that mattered. What mattered was the decision made after it: what to do with what the machine freed, and where a human had been kept on purpose. This page sets out what each company said, in its own words where they exist, what each account does not show, and the one variable that separated the reversals from the rest.

Klarna: cut on cost, rehired on quality#

In February 2024 Klarna announced that its AI assistant had handled two-thirds of customer service chats in its first month, 2.3 million conversations, and was doing the work of 700 agents. It became the standard example in a hundred decks. In May 2025 Sebastian Siemiatkowski told Bloomberg: 'As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality. Really investing in the quality of the human support is the way of the future for us.' The company said it would recruit human agents again, keep the assistant for routine enquiries, and make sure 'there will be always a human if you want'. What this does not show: the size of the reversal or of the quality gap. No figures were published for either, and both the original claim and the retraction are the same man's framing of his own company.

Duolingo: declared, then retreated in a month#

In April 2025 Luis von Ahn told Duolingo staff the company would 'gradually stop using contractors to do work AI can handle' and add headcount only 'if a team cannot automate more of their work'. The backlash came from the app's own users. On 23 May he wrote: 'I do not see AI as replacing what our employees do (we are in fact continuing to hire)'. What this does not show: whether anything operational changed. The retreat was a statement; contractors are not employees, and the memo's contractor line was not withdrawn.

IBM: cut, then reinvested#

In the same week as Klarna's reversal, Arvind Krishna told the Wall Street Journal that IBM had replaced a few hundred human resources staff with AI agents doing spreadsheet analysis, research and drafting, and that 'our total employment has actually gone up, because what it does is it gives you more investment to put into other areas', naming software engineering, sales and marketing. What this does not show: that AI raised IBM's headcount. Total employment moves for many reasons; no figures separating them were given; 'a few hundred' is the company's round number.

Shopify: froze, and held#

Shopify did not cut. Tobi Lütke's memo of 7 April 2025 made a new hire conditional on a team showing why AI could not do the work, and asked every team what its area would look like with autonomous agents on it. The memo was published, and the company has not walked it back. What this does not show: what the rule did. Shopify has not published its effect on hiring, output or quality, so the case proves only that a published rule survives contact with the public better than an unpublished one.

The variable that separated them#

Line the four up and the technology is the same in each. What differs is whether the company decided, before the cut, what the freed time and money were for and where a human had to remain. Klarna optimised on cost, kept no stated human floor, and found the floor by hitting it. IBM stated the reinvestment and, by its own account, made it. Shopify stated a rule about the order of options and left the rest open, which has held so far because nobody has forced the open questions. Duolingo stated an intention about people without a boundary to defend, and defended nothing. The cut is the visible act. The decision is the one made before or after it, a leadership decision in every case.

The wider evidence agrees at the level of shape without confirming it at the level of numbers. McKinsey's 2025 survey found the redesign of workflows to be the attribute with the largest effect on reported profit from generative AI, held by 21 per cent of organisations; banking a saving is not a redesign. BCG's June 2026 survey of 70 executives found half already reporting deskilling in their organisations, which is what a cut without a capability floor produces over time. Both are surveys of perception and are graded as such.

What nobody has measured#

There is no study of AI-attributed redundancies and their outcomes across a population of firms. The four cases here are the four that were reported, which means they are the four with a public chief executive and a story, not a sample. Every figure on this page is a company's own. A company that cut, said nothing and did well would not appear here, and neither would one that cut, said nothing and did badly.

Before the cut#

Three questions a leadership team can answer in writing before it removes a role for AI. Where must a human remain, and how will the organisation know if that floor is breached? What is the freed money for, by name? And what does the organisation keep practising unaided so that somebody can still tell when the machine is wrong? Klarna's chief executive answered the first one in May 2025, a year late and in public. The argument for answering it first is at AI leadership, and the rule for when to stop or reverse a deployment is at deployment is not a ratchet.

Key sources

• Siemiatkowski, S. (2025), reported in CX Dive, 9 May 2025, from a Bloomberg interview of 8 May. Graded entry.
• von Ahn, L. (2025), reported by Fortune via Yahoo Finance, May 2025. Graded entry.
• Krishna, A. (2025), reported by PYMNTS, 6 May 2025, from a Wall Street Journal interview. Graded entry.
• Lütke, T. (2025), memo reported by TechCrunch, 7 April 2025. Graded entry.
• Singla, A. et al. (2025). The state of AI. McKinsey. Graded entry.
• Boston Consulting Group (2026). When Everyone Uses AI, Companies Risk Losing Critical Skills. Graded entry.

Related SuperSkills research#

On what the freed seventy per cent of a role should become, AI workforce strategy. On the bill for a cut without a capability floor, capability debt. On whether the declaration was a strategy at all, is AI-first a strategy or a slogan. On the decisions above the cut, AI leadership.

About this research#

Rahim Hirji is the author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company. He has run, grown, bought and advised businesses with AI in them. Findings are attributed to the studies and statements that produced them and kept separate from the interpretation. This is a living reference, reviewed and updated as significant new evidence appears.

How this research works  ·  Reviewed quarterly  ·  Found an error? Tell me and it is corrected on the page.