I counted fourteen write-ups of the Uber layoffs in one afternoon, and thirteen of them used the same three-word explanation: replaced by AI. The fourteenth quoted the company's own framing, which said something quieter — that the organization had grown too complex and siloed, and needed restructuring before AI could be layered onto it. Those are not the same claim. One says a model closed your tickets. The other says a reorganization happened, and AI is the reason offered for it.
If you answer support tickets for a living — at Wayfair, at Capital One, at a Series C company nobody outside your city has heard of — the distance between those two sentences is the distance between a skill problem and a strategy problem. You are being handed the first one.
Where the belief got built
The belief has a birthday, and it is more recent than it feels.
In 2022, tech shed roughly 165,000 jobs by layoffs.fyi's count. In 2023, about 260,000. Those cuts were priced in interest rates and pandemic over-hiring, and everyone knew it at the time. But a layoff announcement is a communications product, and "we hired too many people during a bubble" is a bad one. It makes the executives who did the hiring look like the problem.
Then, in May 2023, IBM's chief executive told Bloomberg the company would pause hiring for roles it thought AI could do — around 7,800 back-office positions over five years. Not fired. Not replaced. Paused. It was a forecast about attrition, and it was reported as a plan.
The load-bearing moment came in February 2024, when Klarna announced its AI assistant had handled two-thirds of customer service chats in its first month and was doing the work of 700 full-time agents. That number — 700 — is the single most-cited data point in the entire AI-replaces-support genre. Teleperformance, the outsourcing giant, saw its shares fall hard on the news. The market believed it immediately.
Here is the part that did not travel. By 2025, Klarna's CEO had told reporters the cost focus had gone too far, that quality suffered, and that the company was recruiting human agents again. The reversal got a fraction of the coverage the original claim did. The 700 kept circulating anyway, cited in analyst notes and trend pieces and, eventually, in stories about job cuts at companies that had nothing to do with Klarna.
That is how a field comes to know something. One press release, one number, one un-audited claim, repeated until repetition looks like evidence.
Is AI actually causing these layoffs?
Sometimes partly, rarely alone, and almost never in the way the headline says. When employers name a reason for announced job cuts in the monthly tallies from Challenger, Gray & Christmas — the outplacement firm that has tracked US layoffs since the 1990s — the reasons that dominate are cost-cutting, restructuring, market and economic conditions, and closings. AI appears in those tallies and has grown, but as of writing it remains a small share of the reasons companies give for themselves. The press attributes AI far more often than employers do.
This matters because companies have every incentive to say AI. "We are cutting 10% of support because we built something smarter" reads to investors as a margin story. "We are cutting 10% of support because the unit was expensive" reads as a cost story. Same headcount, two very different stock reactions.
The quieter mechanism
Support work has been migrating for twenty years, and not to models. It moves to business process outsourcing vendors — Teleperformance, Concentrix, TaskUs — and to Manila, Bogotá, Hyderabad, Cairo. A role eliminated in Chicago frequently reappears as a vendor seat somewhere the fully loaded cost is a third as high.
AI is genuinely part of this now, but usually as a tool the vendor operates, not a replacement for the vendor. The bot handles the password reset and the where-is-my-order. A human handles what the bot escalates, and that human increasingly works for a company whose name is not on your paycheck.
Which makes AI a very convenient explanation. It is forward-looking, it is flattering, and it does not require anyone to say the word offshoring on an earnings call.
What believing the wrong story costs you
If you think a model took your job, you conclude you must out-compete a model. You go learn to write prompts. You add "AI-assisted" to your LinkedIn headline alongside four hundred thousand other people who read the same advice.
If you think your function was restructured and repriced, you draw a different conclusion. You go find the work that does not survive a handoff: the escalations, the refund exceptions, the account that churns if the conversation goes badly, the ambiguous claim nobody has written a policy for yet. Deflection rates — the share of contacts a bot closes without a human — plateau, and the tail that escapes them is harder, angrier, and worth more.
| What the headline says | What company language usually says | What you can check yourself |
|---|---|---|
| "Replaced by AI" | "Restructuring," "org design," "efficiency" | The earnings call transcript, free on the investor relations page |
| "The bot closed the tickets" | A deflection or containment rate | Whether escalation volume rose after the bot launched |
| "Those roles are gone" | The roles moved to a vendor or a cheaper region | Open reqs posted by BPO firms naming your product |
What to try this week
Spend one hour, not more. Open your employer's most recent earnings call transcript or quarterly filing on the investor relations page. Search six words: deflection, containment, outsourc, vendor, restructuring, headcount. Write down which ones appear and what number is attached to each.
Then open your own queue and count, for one week, how many tickets reached you after an automated first touch failed. That count is not a metric your company reports. It is the shape of the work that is left, and it is the most honest forecast of your role you will get from anyone.
You may find AI mentioned twenty times and deflection mentioned zero. That tells you something. You may find a containment rate of 55% and a vendor transition announced in the same paragraph. That tells you more.
The work that survives is not the work a model cannot do. It is the work a company cannot afford to get wrong.
