Months ago, I built a crawler. Newsfeeds, YouTube and TikTok stats, SEC filings, earnings calls — the whole pile. I pointed it at one question: when does the AI labor story stop being a thesis and become something the market has to price in?
I called it the Canary.
I'll confess: I built it partly because I didn't trust myself to see it in time. Running an operating company can blind you to the thing happening at your feet. Quarterly numbers, board prep, whatever's on fire — the immediate eats the structural. So I built a light. If a sitting CEO of a publicly traded company stood up and said we're laying people off because of AI — no dialect, no "structural optimization," no PR fog — the light would come on.
Today the light came on.
Matthew Prince, Cloudflare. He posted the rationale and it's a clean read, so I won't repeat it here. The short version: the company is going to hire Builders and Sellers at rates it has never seen, and needs a lot fewer Measurers — the people in the middle whose job is to watch the work and report on it. He said the quiet part. He said it on the podium. He signed his name to it.
That's the canary. That's what I built the tool to detect.
Roughly 200,000 jobs were eliminated across fewer than 25 large public companies over the last 18 months. Microsoft. Meta. Amazon. UPS. Intuit. Nissan. Boeing. Chevron. BP. Not penny stocks — name brands.
Every one of those announcements spoke a careful dialect. Structural optimization. Operational efficiency. Margin expansion. AI-enabled productivity initiatives. If you ran the operation you knew exactly what those words meant. If you sat on the board, you knew. The investors who'd been pricing in margin expansion knew. The employees being walked to the door knew.
Nobody said it on a podium. Until today.
Most analysts will read Prince's post and write think pieces about Cloudflare's roadmap, about the labor market, about productivity gains. That's the surface. The structural truth is underneath.
The 200,000 jobs were the layoff. Prince's post was the permission. Permission is what cascades — not productivity.
The cut wasn't new. Every public company has been cutting. What Prince did was say why. That's the move that compounds.
For 18 months, every CEO has been watching every other CEO, waiting for somebody to drop the dialect. The first one to do it gets the heat. The second gets the cover of the first. By the fifth, it's policy language. By the tenth, it's how the entire industry talks. Permission travels faster than productivity does, because permission is a social signal and productivity is an engineering one. Boardrooms can move on social signals overnight. Engineering takes quarters.
The next 18 months of layoff announcements will read nothing like the last 18. Watch for it.
Prince handed the org chart a clean map. It deserves to be taken seriously.
Builders make the thing. Sellers move the thing. Measurers watch the thing get made and the thing get moved, and they report on it. In a pre-AI world, the Measurer layer was a rational expense — companies got too big to coordinate themselves without it, so you hired humans to be the connective tissue between systems and people. Project managers. Middle managers. Analysts. Coordinators. Reporting specialists. The plumbing of a modern company.
AI eats that layer. Not at the edges. At the foundation. Software that can read every system, summarize every meeting, surface every anomaly, and route every decision doesn't need a person sitting between two other people to translate. The economics of the coordination layer were unwritten the moment GPT-4 shipped, and Prince just said it out loud.
I've been writing about this for two years. AI collapses the human coordination layer. That's the whole game. It's the thesis underneath everything I publish here. I didn't need a sitting CEO to say it to know it was true — I built the Canary because I knew one eventually would, and I didn't want to miss the day it happened. Today it happened. The cover for every other CEO not to act on it is gone.
If you run a company, you have Measurers. Some are titled "Project Manager." Some are titled "VP of Operations." Some have $250,000 salaries and an MBA. Some have $50,000 salaries and a knack for spreadsheets. The category isn't about the level. It's about the function: are they making, are they selling, or are they watching the people who make and sell?
Count yours. Honest count. I'll wait.
Now zoom out. The role-level cut Prince just made is the inside view. The outside view is where most operators won't want to look.
There are only three kinds of company in the world Prince just described, and the gravity well between them is closing fast.
Type 1: Native AI Lean. Started post-2022, built around AI from the foundation up. No legacy headcount, no legacy tooling, no legacy process. These companies operate the way a Navy SEAL element operates — no wasted movement, no inappropriate toolset, the team is small because the work doesn't need more. The product reflects the operating model: simple UI, ruthless feature surface, near-zero human labor in COGS, gross margins that don't look real to anyone running a pre-AI P&L.
The other day I spoke with a new voice AI agent company. One of 50 that I've spoken to in the last 24 months. All 50 are slowly chipping away at traditional voice and telephony models. But this one had receipts. Their approach was different: they were going to own as much of the stack as they could, and because all of their developers had Vibe Coding Motherfucker wallets, they were actually pulling it off. They weren't giving up telephony margin to Twilio or Plivo. Weren't giving up STT/TTS margin to Deepgram — they built the AI voice stack. Eight employees. Set to do $5M ARR in their second year of selling. A brand and name I'd never heard of. They just snuck up on a market with a superior product, radical execution, and a sales motion that doesn't waste a call. And they won't be the last to do it.
Type 2: AI Reformed. Pre-2022 companies that found religion. Old enough to have built real distribution and customer trust, brave enough to gut the operating model and rebuild it. Most won't make it — the gravitational pull of the legacy org is brutal, and reform is harder than starting from scratch. The ones that pull it off become juggernauts. A distribution moat plus a Type 1 operating model is the hardest combination to beat in any market.
Type 3: Dead, or dead man walking. Won't or can't pivot. The reasons vary — leadership refuses, product is too complex to rewire, cash is too tight, board is committed to the old plan, leadership doesn't understand (or believe in) the shift. The diagnosis doesn't matter. The outcome does. Type 3 companies get death by a thousand cuts from 100 Type 1 competitors below them, and get out-massed by the 1 or 2 Type 2 juggernauts above them. The squeeze closes the same way every time.
There is no Type 4. I've looked. Hybrid is a transition state, not a destination — you're either moving toward Type 2 or sliding into Type 3, and the slide accelerates as your Type 1 competitors compound.
Prince's announcement is the kind of moment that makes the slide visible from outside the building. The next ones will make it visible to your customers.
Three orders, before the permission reaches your industry — because it will, and you don't get to opt out of timing.
One: ask the right question about every Measurer. Not "could AI do this job today?" — that's the wrong question, because today's AI isn't tomorrow's. The right question is what is this role's compounding curve over the next 24 months, against the compounding curve of the tools I could buy or build to do it? If the answer is "the tool wins, and the gap widens," you already know what the next year looks like, whether you act on it or not.
Two: name your type, out loud, in a room with your leadership team. Type 1, Type 2, or Type 3. No fourth option. No "we're a hybrid." No "we're getting there." If the team can't agree, that disagreement is the most important data you'll get all year. Run toward it, not away from it.
Three: if you're Type 3, decide whether to reform or sell. Those are the moves. Reform is brutal, expensive, and most leadership teams aren't up to it. Selling is honest. Pretending is the third option, and pretending is what kills companies between now and 2028.
The companies that figure this out will run leaner than the ones that don't. The ones that don't will get taken apart by four-person competitors they can't see on the board deck yet.
I've written that line before. I'll write it again. It's still true.
I built the Canary because somebody should be watching.
Today the light came on. The watching is over.
The deciding is on you.