In May, Cloudflare reported the best quarter in its history. Revenue up 34%. Record numbers across the board.

The same week, it cut 20% of its workforce.

I'll be honest, when I first saw those two headlines side by side I assumed the usual story — a stock-price move dressed up in AI language. Then I read what CEO Matthew Prince actually wrote about who got cut, and it changed how I think about this whole year of layoffs. His words: "the vast majority of those we laid off last week were measurers." Middle managers. Operations specialists. Chunks of finance, legal, compliance, internal audit.

Not the engineers. Not support. Not sales. The people whose job was to watch, count, and report on the people doing those jobs.

Once you see that pattern, you can't stop seeing it in this year's cuts.

Amazon cut 30,000 corporate roles across three months and described it as reducing layers and removing bureaucracy. Coinbase flattened its entire company to five layers below the CEO. Google, without ever making a single announcement, has cut roughly a third of the managers who oversee small teams. Jack Dorsey took Block from over ten thousand people to under six thousand and talked about smaller, flatter teams changing what it means to run a company at all.

Different companies, different words, same target. The layer in the middle.

Why that layer existed

Here's the thing nobody says about middle management, because it sounds rude: most of it was never really management. It was measurement.

Think about why the layer formed in the first place. A VP cannot personally watch two hundred people work. So a structure grows whose entire function is making work legible to people above it — collecting status, chasing updates, normalizing numbers from six different systems into one deck, translating "the migration is a mess" into "we're tracking amber on workstream 3." The old rule of thumb that a manager can handle about seven direct reports was never a law of nature. It was a limit on human attention. Information was expensive to move up an org chart, so companies hired people to move it.

The measurer's product is legibility. That's it. That's the whole product.

And now legibility is becoming free.

The ticket summarizes itself. The codebase explains its own changes. The support queue writes its own weekly report, and it's more accurate than the one the team lead used to spend Friday afternoons assembling. When a director can ask a system "what happened this week and what's stuck" and get a straight answer in eight seconds, the four people whose job was producing that answer over four days have a problem — and it arrived before anyone laid a finger on the people doing the underlying work.

That's the story of 2026's layoffs that the "AI replaces workers" headline keeps missing. AI didn't come for the work first. It came for the apparatus of observing the work.

I should confess something here, because this newsletter runs on declared bias. One of the products my company builds is conversation intelligence — software that watches every support call, scores it, and tells an operations director what happened without anyone compiling anything. I sell legibility. I am, in the most literal sense, part of why the measurement layer is in trouble. So when I describe this shift, I'm not describing something being done to the economy by mysterious forces. I'm describing something I do to it, on purpose, for money. Factor that in.

The rhyme from the 1990s

Now — before anyone flattens their whole org chart off the back of one newsletter — this has happened before, and the ending is instructive.

In the nineties, ERP systems and the "re-engineering" craze gutted middle management once already. Same logic: the software makes information flow, so the human routers are redundant. Companies cut deep.

And then, quietly, over about a decade, a lot of it grew back.

Why? Because it turned out the layer had been doing invisible second jobs the whole time. The manager who compiled the Friday report was also the person who noticed the new hire was struggling before it became a resignation. The ops specialist who chased status updates was also the one who resolved the turf war between two teams before it reached a director. Judgment, teaching, conflict resolution, morale — none of it was on the job description, all of it was real, and none of it was replaced by SAP.

The people who only measured never came back. The people who measured and decided got rehired, eventually, at greater expense, by companies that discovered the difference the hard way.

Forrester is already predicting the same correction this cycle — their forecast is that half of the AI-attributed layoffs get quietly rehired, often offshore and at lower salaries. I think that's directionally right. Some of these companies are about to learn, on a two-quarter delay, exactly which of their measurers were secretly coaches.

The test I'd run on your own org chart

So here's the practical bit, because you're not reading this for the history lesson.

Take any role in your company that sits between the people doing the work and the people accountable for it. Ask one question: is this role's primary output a report about other people's output, or a decision that sticks?

If it's the report — the status deck, the compiled metrics, the rolled-up summary — the economic clock on that role is already running, whether you act on it or not, because your competitors' tooling is making that output free. Be honest about it now, kindly, while there's time to move the person toward the judgment work.

If it's the decision — the call on what ships, who's struggling, which fire is real — protect that role, even if the person also happens to keep score. Do not cut the coach because he carried the clipboard.

And whichever way you cut, watch the two-quarter lag. When companies remove the measurement layer, nothing breaks on day one. Quality decays slowly. Onboarding gets worse invisibly. The new person flounders for a month before anyone notices, because the person who used to notice is gone. By the time it shows up in a number, you're two quarters past the cause. Every rehiring wave in that Forrester prediction will be some CFO discovering this lag in real time.

Flat orgs aren't free. The coordination the measurers did still has to happen somewhere. The bet — and it's a real bet, not a certainty — is that it moves into the tooling. Where the tooling carries it, the flat structure holds. Where it doesn't, the layer grows back, and it grows back expensive.

One more thing before I go. The man running the biggest flattening experiment in corporate history — 8,000 people out at Meta this spring — published a long essay last week arguing that none of this needs to cost jobs at all, and that AI's real purpose is empowering individuals. It's a genuinely serious piece of writing, it's the best case anyone has made against the doom position, and it has one broken sentence right at its load-bearing center. Next issue, we read it properly.

For now: go run the test. Report or decision. Tell me what you find in your own org — I suspect you already know, and have known for a while.

Sandeep Bansal is the founder of GoZupees, which builds conversation intelligence and AI agents — the tooling this essay argues is dissolving the measurement layer.

Next issue: Zuckerberg's superintelligence essay, taken seriously — and where it breaks.

Sources worth your time:

  • Matthew Prince on the Cloudflare cuts (via Fortune, May 2026)
  • TechCrunch running tracker of AI-cited layoffs — Amazon, Coinbase, Google, Block entries
  • Forrester, Predictions 2026: The Future of Work (via HR Executive)
  • Challenger, Gray & Christmas, July 2026 Job Cut Report