502,000 AI Layoffs Planned. Zero Productivity Evidence. How to Avoid Making the Same Bet.
A survey of 750 U.S. CFOs found that 44% are planning AI-related layoffs this year - roughly 502,000 positions, nine times the number cut in 2025. The data comes from a joint study by the National Bureau of Economic Research and the Federal Reserve Banks of Atlanta and Richmond.
Around the same time, Goldman Sachs economist Ronnie Walker published a finding that reframes those numbers entirely: there is still no measurable relationship between AI adoption and productivity gains at the economy-wide level. Half a million jobs on the line, and the productivity case that would justify them has not materialized in the data.
The Pattern Behind the Numbers
In 2025, AI-attributed job cuts accounted for just 4.5% of total layoffs, about 55,000 out of roughly 1.2 million. This year, companies are projecting cuts nine times larger while the evidence base has barely moved. Block cut 40% of its workforce, Atlassian cut 10%, and Meta is reportedly planning 20% reductions, but the aggregate productivity data that would validate cuts at this scale still shows no clear signal.
The survey found that half the projected losses target white-collar positions, the same roles companies are simultaneously trying to augment with AI tools. That contradiction is worth pausing on. If you are investing in AI to make a role more productive, cutting that role before the investment has been measured means you are acting on two conflicting assumptions at the same time.
Microsoft AI chief Mustafa Suleyman predicts office jobs will “crumble in 18 months.” Anthropic CEO Dario Amodei expects entry-level AI roles to be cut “in half” within a similar timeline. Federal Reserve Chair Jerome Powell warns that AI is “quietly impacting the labor market.” The predictions are directionally consistent, but predictions and productivity data are not the same thing.
The Measurement Gap
The core issue is not whether AI will eventually replace certain roles. It probably will. The issue is that most organizations are making headcount decisions today based on that future expectation, without measuring what the roles actually do or whether AI handles those specific tasks reliably.
This is the same gap that the hiring freeze paradox exposed: 80% of organizations have deployed AI, but only 20% have redesigned how work actually gets done. Cutting headcount without redesigning the workflow creates a different kind of risk, one where the savings from cutting become costs from rebuilding when the work those people did turns out to be more complex than the AI assumption accounted for.
McKinsey’s approach was to map every role against two questions: does this role require human judgment, and does this role require human presence? They have run similar exercises across tens of thousands of positions. Most organizations planning AI-driven cuts have not done anything comparable.
What Measuring Actually Looks Like
There is a difference between cutting jobs because AI proved it can do the work and cutting jobs because you assume it will. The first is restructuring. The second is a bet.
Before any AI-driven headcount decision, the minimum viable measurement is straightforward. Break the role into its component tasks. Measure how much time goes to each. Test whether AI handles those specific tasks at the quality and reliability the business requires. If you cannot point to an internal pilot, a measured result, or a workflow redesign that accounts for the work those people actually do, the cut is based on expectation, not evidence.
The CFOs planning 502,000 cuts are not wrong that AI will change the workforce. They may be wrong about the timing, and timing is the difference between a restructuring that works and a bet that creates more problems than it solves.
Related: Zuckerberg’s AI Deputy: What Meta’s CEO Tool Reveals About Every Leader’s Information Gap looks at the other side of the workforce equation, where Meta is investing in AI tools that augment executive decision-making rather than replacing roles. And Oracle and Meta Are Cutting Tens of Thousands of Jobs examines the vendor side of the same pattern - when AI infrastructure debt drives the cuts, not AI capability.