What CEOs should believe when a peer says AI made the team 100x more productive

The myth

Cloudflare reported a record quarter on May 8 and announced its first mass layoff in 16 years on the same earnings call. CEO Matthew Prince told TechCrunch the cuts were a structural response to AI, with team members becoming “two, 10, even 100 times more productive than they had been before.” Internal AI usage at the company was up 600% in three months. Revenue hit $639.8 million, also a record.

The myth that travels out of an announcement like that, in board decks and CEO group chats this week, is simple. AI deployment produces workforce productivity. Deploy the agents, watch the headcount need fall, book the savings, tell the analysts.

It is a clean story. It is also wrong in the way that confident stories often are.

Why it sounds right

AI productivity gains are real. I have watched them happen inside engineering teams shipping twice as fast, finance teams closing the books with two analysts instead of five, support orgs absorbing triple the ticket volume on the same headcount. Prince is not lying. The Cloudflare R&D team is using AI coding tools across the board, every line of deployed code is reviewed by an autonomous AI agent, and the HR, finance, and marketing teams are running thousands of agent sessions a day. None of that is theatre.

CEOs are also under enormous pressure to tell investors a productivity story. Spotify, Coinbase, Freshworks, Meta, and now Cloudflare have all turned that pressure into a layoff with an AI label. The narrative compounds. When a peer CEO says his team is 100x faster, the natural assumption is that the same number is sitting somewhere in your own org, waiting to be unlocked by the right procurement decision.

That assumption is where the myth gets dangerous.

What the evidence says

Look at the April jobs report that Fortune published the same morning as the Cloudflare announcement. The information sector lost 13,000 jobs in April. Financial activities lost 11,000. Information payrolls fell to their lowest level since March 2021. White-collar industries are now in their 16th consecutive month of net job loss while the headline unemployment rate sits at 4.3%. None of those workers are showing up in a productivity statistic. They are showing up in a missing line on someone’s payroll.

KPMG’s Canadian leaders survey, published this week, supplied the second number that matters. 70% of leaders say AI delivers meaningful business value. 3% have achieved measurable ROI. That is not a rounding error. That is a 67-point gap between what executives believe and what they have proven.

The 100x quote is doing a lot of work in that gap. So is the framing across the industry that AI is “rebuilding” companies. Brian Armstrong at Coinbase replaced “pure managers” with “player-coaches” the week before, capped his hierarchy at five layers, and raised the employee-to-manager ratio to 15-to-1. That is org redesign disguised as a productivity claim. The productivity gain is the layoff is the org chart. Three things, one announcement.

The verifiable fact: white-collar payrolls are contracting at scale. The verifiable claim about AI productivity at the org level: still mostly missing. Individual workers report gains. Organizations do not.

The reframe

Here is the better mental model. AI productivity is not a property of deployment. It is a property of org redesign.

The companies converting AI into measurable lift are redrawing the work, redrawing the manager layer, and only then sizing the team. Most companies in the headlines are doing it backwards.

The companies converting AI into measurable lift are doing three things in this order. They redraw the work. They redraw the manager layer. Then they size the team. The companies in the headlines are mostly going in the opposite direction. They cut first, and then decide what work the remaining team will own.

That is why the Cloudflare and Coinbase announcements are interesting for a different reason than the productivity claim itself. Both CEOs spent the announcement talking about how the team is structured. Prince said his R&D team is organized around the AI agents that review the code. Armstrong said his pods are capped at small numbers because that is what works with agentic tools. The structure changed first. The headcount changed because the structure changed. The productivity claim came after both.

If a CEO peer tells you his team is 100x more productive and the only visible evidence is a headcount cut, the productivity has not been proven. What has been proven is that the company can run with fewer people for one quarter. Whether the gain is sustainable, repeatable, and visible on the P&L next year is a separate question, and it is the only one investors and boards will care about by Q4.

So what

There is a short test that takes about ten minutes to run before believing the next AI productivity claim, including the next one Cerevisor is going to make about its own work. Ask three things.

What work did the remaining team take on that was not happening before the cut. What is the new manager span, and how was that number set. What financial outcome (revenue per employee, cost per transaction, gross margin, time to ship) moved in the same quarter as the headcount change. If the answer to all three is fuzzy, the productivity claim is a story, not a number.

The thing for a CEO to do this week is not to match what Cloudflare announced. It is to write a one-page version of that story for the company in question before a press release writes it. Name the work AI is going to take on. Name the manager layer that is going to change. Name the metric that proves it. Then decide on headcount.

The order matters. Most CEOs right now are doing it in the wrong sequence and dropping the productivity multiplier into the announcement to cover the gap. That is the trade no leadership team should be the last one in its peer group to notice.