120,000 job cuts and counting. What tech's AI layoffs actually tell the rest of us

In the early 1980s, newspapers went through their own version of this. For a century, getting a paper to print meant hot metal typesetting: rows of Linotype machines casting molten lead into lines of type, one line at a time, run by operators who'd trained for years to do it fast and clean. Then computer photocomposition arrived, and within a few years most of that machinery was gone, replaced by keyboards and screens. It was one of the most disruptive shifts the industry had ever seen — entire departments restructured, whole trades made obsolete almost overnight, and, in places like Fleet Street, some of the bitterest labour disputes in British publishing history.

But the papers that came out the other side of that transition well weren't the ones that simply swapped machines for keyboards. They were the ones that figured out which skills the old system had been quietly relying on — a sharp eye for a layout that didn't read right, a nose for which story needed cutting at the last minute — and made sure those skills had a seat at the new system too. The tool changed completely. The judgment that made the tool useful didn't.

That distinction — between the tool and the judgment sitting behind it — is the one a lot of this year's layoff headlines blur together.

Since January, roughly 120,000 tech jobs have been cut, and AI has been the most-cited reason behind them. Microsoft eliminated about 4,800 roles this week alone. Amazon cut 16,000 corporate jobs in January, on top of 14,000 the previous October. Salesforce, Meta, PayPal, Cloudflare, Block, Oracle — the list runs long, and almost every announcement leans on some version of the same sentence: AI is changing how the work gets done.

We read through the details of over twenty of these announcements. A few patterns stood out that matter more than the headline number does, whatever business you're in.

"AI-driven" doesn't always mean what it sounds like. Some of these cuts are genuinely about AI absorbing repetitive work — Salesforce, for instance, said its Agentforce tool had reduced support case volume enough that it stopped backfilling support engineer roles. But plenty of the same announcements are also restructurings of teams that grew fast during the pandemic hiring boom, with AI attached as the explanation investors want to hear. It's worth asking, for any "AI made us do it" story, what workload actually shrank versus what the org chart just moved around.

The cuts are landing hardest in the functions every company has. Support, internal reporting, operations, customer service, mid-level management — these show up again and again in the announcements, across industries that otherwise have little in common. Cloudflare's CEO was unusually blunt about it, saying the roles it cut were largely "measurers": middle management, internal reporting, and oversight functions. That's a useful flag for any organization trying to work out where AI genuinely changes headcount needs versus where it just changes what people spend their time on.

Flatter is the direction, not the exception. Coinbase restructured to five layers below its CEO and COO and is experimenting with one-person teams that combine engineering, design, and product. Salesforce, Atlassian, and Cloudflare all describe some version of removing layers rather than removing output. For most organizations, that's the real shift to plan around: fewer approval steps and reporting layers, not necessarily less work getting done or less strategic thinking behind it.

Revenue is up at almost every one of these companies. Cloudflare posted its highest quarterly revenue ever the same week it cut a fifth of its staff. Oracle's obligations backlog grew over 300% even as it disclosed 21,000 job cuts over the year. That combination — record growth and shrinking headcount — is precisely why "AI replaced them" is too simple an explanation on its own. Something else is being restructured alongside the technology, and it's usually the org chart, the layers of review, and the definition of who owns an outcome.

The skill that survives every one of these reorganizations is judgment. Every company on this list still needs someone who can tell good output from mediocre output, catch a problem before it reaches a customer, or know when a number in a report doesn't quite add up. That's the layer AI hasn't automated in any of these announcements, and it's the layer that determines whether an AI-flattened team produces better work or just faster mediocre work.

None of this means every organization should treat 2026's layoff wave as a preview of its own org chart. It means the real question isn't "will AI take these roles" — it's "which of our layers of review and approval are actually adding judgment, and which are just adding time." That's a different exercise for every business, and it's the one worth doing before a restructuring forces the question.

If you want help figuring out which layers in your workflow are judgment and which are just process, that's the conversation we'd like to have.