The AI productivity promise is cutting your jobs. What can be done?
502,000 jobs. Not one company has published the productivity data that would justify them. Here is what you can do before your company is next.
TL;DR
Howard Yu and Jason Averbook named the pattern last week: companies are cutting jobs based on AI productivity gains that have not arrived yet. The NBER data confirms it. It runs deeper than a leadership mistake. The stock jumps when the announcement is made, not when the results come in. I dig one layer deeper and give you the instrument to change what the board sees before it takes a decision to cut jobs.
The paradox
February 26, 2026. Jack Dorsey posts a memo on X. Block is cutting 4,000 jobs - nearly half the workforce. The reason he gives: AI had fundamentally changed what it means to run a company.
XYZ jumped twenty percent that day. By March 27 it had given most of it back. The stock closed at $55.98, down 15% over the past year.
On March 24, the working paper was published. 750 CFOs. Duke University. Federal Reserve Banks of Atlanta and Richmond. Their finding is that CFOs plan to cut 502,000 jobs in 2026 citing AI. Nine times more than the 55,000 cut last year.
Peter Girnus, one of those 750 CFOs, posted on X the next day: “500,000 people will lose their jobs because of a technology whose economic benefit Goldman Sachs cannot detect.”
Howard Yu and Jason Averbook have been tracking this pattern closely. Jason unpacked the NBER data. Howard named the Circuit City echo:
Companies cut the people who carry the institutional knowledge that makes the tool actually work.
Their question is whether the organization is ready.
Mine is different:
Why do boards keep authorizing cuts when the evidence is not there - and what can one person do before the decision is made?
The system is working
A CFO who waits for internal evidence before announcing cuts misses the premium.
A CFO who announces cuts citing AI - even with zero internal data - collects it immediately.
As Averbook notes, AI-related stocks have driven roughly 75% of S&P 500 returns since ChatGPT launched. That creates a powerful incentive to frame any cost-cutting as AI-driven, whether the evidence exists or not.
The market is working exactly as designed.
Some skip the productivity claim entirely. Microsoft spent $80 billion on AI infrastructure. They need the margin back. They found it in letting go of their people.
Yu named the consequence - Circuit City cut 3,400 experienced salespeople in a single morning because the spreadsheet said it made sense. Less than two years later, bankruptcy. The knowledge walked out. The revenue followed.
The spreadsheet missed it. It always does.
This is a story about a system, rather than bad CFOs.
The 750 who took that survey are rational people. The system pays for the announcement. The evidence was never a requirement.
Again, this is not new. Last week it was compliance certificates nobody examined. This week it is productivity gains nobody measured. The same system. Different outcome.
Fixing this does not mean finding better executives. Change what the board sees before it decides to lay its people off.
What you can actually do
I know what the spreadsheet misses. I have been ON it.
As CIO of a precision manufacturing company with decades of history, my job was to keep IT running, build new solutions, explore what IoT could do for the factory floor. A bit more than usual CIO scope.
On the side of all that, I built an Academy.
I handpicked the team of technical writers and trainers. Gave them one task: capture the knowledge before it walks out the door. The experienced engineers heading toward retirement - the grey hair that knew why the third exception in the approval workflow existed, why a specific machine behaved differently on a cold morning, why that client needed a particular configuration nobody had written down.
We captured it for internal reuse. We converted it into external manuals and trainings for clients purchasing complex end-to-end manufacturing lines.
Do you know many CIOs building Academies? I did not wait for someone to ask. I could see what was coming.
Then leadership changed. New conditions. Quite incompatible with my vision, so I left.
Individual initiative without structural mandate is fragile. I learned that the hard way. The instrument only works if it is treated as a requirement by the board - not a side project built quietly by one person who suspects what is coming.
Now. You can bring three things to the board's attention before it's too late. None of these require board authority to propose.
Map what will be lost before it is gone
Before any restructuring citing AI, build a workflow dependency map of the affected roles. A capability map instead of skills matrices that simply count certificates.
A map that shows which decisions require knowledge that systems have yet to capture, which client relationships require someone who knows the history, which process exceptions exist because someone learned the hard way why they matter.
Two to four weeks. Less cost than one severance package. The board sees what it is about to destroy - before the decision.
Measure before you cut
Most restructuring proposals arrive with a financial model and a headcount list. What they never include is a productivity baseline - what output, at what quality, at what speed, the affected roles are generating today.
Without that number, there is no “after”. Just a “before” and a “hope”.
Build it now. Task level, team level, process level. Two to three weeks of measurement before any announcement. The leader who walks into that room with that baseline owns the conversation.
Write the risk brief nobody asked for
Take the capability map and the productivity baseline. Put them on one page. Frame it simply: here is what we are about to lose, here is what we currently produce, here is what AI cannot yet replace.
Nobody asked for this. Still, put it on the table.
A board that cuts without it has half the data. A board that has it - and cuts anyway - at least owns the choice with eyes open. That is a different kind of accountability.
The one test that matters
No company citing AI to justify cuts has published the productivity data that would prove the cuts were right.
Klarna replaced 700 people with AI. Then hired them back.
That is the test. Find the productivity data. If it is not there, you know what the layoff announcement is worth.
If you want more of this kind of work
I write these Signals between working with teams who are trying to build AI governance that holds under examination. Each one takes me 6–8 hours to research, cross-check and write.
If this helped you see something more clearly, the best way to support is to:
or
Key Sources
Artificial Intelligence, Productivity, and the Workforce: Evidence from Corporate Executives, Peter Girnus – Response to CFO survey, Jason Averbook – What the March 2026 AI jobs data actually says, Howard Yu – How organizations lose their minds, Challenger, Gray & Christmas – Job Cuts Report, Klarna – AI workforce reduction and rehiring, The AI readiness cult by Andrei Savine
The 9x jump in AI-cited cuts to 502,000 across the NBER CFO sample maps onto the next 12 months of earnings calls. Block's +20% pop followed by -15% slide indicts the market's pattern-matching on the announcement. The Klarna walkback is the case study every board should review before signing off, and I plan to write through it at theaifounder.substack.com. What's the realistic path for a board member to demand productivity baselines before approving an AI-justified RIF?