A new survey of 750 CFOs reveals the contradiction at the heart of AI-driven layoffs: executives plan to cut nine times more jobs for AI this year than last year, even as most admit AI hasn’t delivered measurable productivity gains.
The Duke University CFO Survey, released March 24, found that 44% of surveyed companies plan AI-related job cuts in 2026. If these numbers hold, approximately 502,000 roles would be eliminated—compared to 55,000 AI-attributed layoffs in 2025.
“What executives are seeing with productivity today is actually more of a wish than a realized fact,” said John Graham, the survey’s director. “Companies see the potential of AI without the financial results to match.”
The Productivity Paradox Returns
Robert Solow won the Nobel Prize partly for observing, in 1987, that “you can see the computer age everywhere but in the productivity statistics.” Nearly 40 years later, economists are making the same observation about AI.
The PwC 2026 Global CEO Survey found that 56% of CEOs report getting “nothing out of” their AI investments. Only 12% reported that AI both grew revenues and reduced costs.
Deloitte’s analysis puts the failure rate higher: MIT research found a 95% failure rate for enterprise generative AI projects, defined as showing no measurable financial returns within six months.
Meanwhile, global AI investment crossed $250 billion in 2025 and continues climbing.
The math doesn’t add up. Companies are spending more, cutting more jobs, and getting less measurable return.
Who’s Getting Cut
The Duke survey is clearer on who faces the axe: workers in “routine, clerical and administrative roles.” Highly skilled positions—architects, engineers, specialized analysts—face lower risk, particularly if they can use AI to augment their work.
Larger firms are moving faster. Among companies with 500+ employees, 78% invested in AI in 2025, and 76% of large automating firms plan to use AI for labor replacement. Smaller companies are catching up—80% are expected to invest in 2026, up from 48% in 2025.
The cuts are concentrated. Block eliminated 4,000 jobs (40% of its workforce) in early March. Atlassian cut 1,600 (10%) on March 11. Both explicitly cited AI. Neither has demonstrated that AI actually performs the work those employees did.
The Quiet Admission
The most revealing finding from Duke’s survey: “AI investment is not expected to have much effect on the number of employees or produce measurable cost savings in 2026.”
Read that again. CFOs are cutting jobs for AI while simultaneously admitting AI won’t produce measurable savings this year.
This suggests three possibilities:
-
Belief in future returns — Executives are betting that AI will eventually deliver, and early cuts position them to capture those gains.
-
Market pressure — Investors reward AI-driven restructuring. Block’s stock jumped 20% after announcing its 40% workforce reduction.
-
Convenient cover — Some cuts would have happened anyway. AI provides better framing than “we overhired during the pandemic.”
The truth is probably all three. AI offers a narrative that justifies cuts that might otherwise face more scrutiny.
Where AI Actually Works
Stanford’s 2025 AI Index narrows the productivity gains to exactly two domains: software development and customer support.
Generative AI users save roughly 5-6% of weekly work hours on average. But aggregate productivity impact remains 0-1%, with gains dissipating through rework, substitution, or non-economic use of time saved.
The organizations seeing returns share common traits: they embedded AI extensively across products rather than treating it as a cost-cutting tool, and they built structured workforce capability programs. The 12% of CEOs reporting both cost and revenue gains had done the organizational work—not just deployed the technology.
The Historical Pattern
Solow’s original paradox took a decade to resolve. From 1973 to 1995, despite massive IT investment, productivity growth remained depressed. It wasn’t until businesses restructured around the technology—not just deployed it—that gains materialized.
The current pattern may repeat. Companies buying AI expecting immediate returns are making the same mistake as those buying mainframes in the 1970s expecting instant transformation.
The difference this time: executives are cutting headcount before the productivity shows up. They’re betting on a future that hasn’t arrived.
What This Means
For workers, the message is grim. You can lose your job to AI whether or not AI can actually do your job. The belief that AI will eventually work is sufficient justification for current cuts.
For investors, the AI rally continues regardless of fundamentals. Companies announcing AI-driven layoffs see stock bumps. Whether AI delivers measurable returns matters less than whether markets believe it will.
For executives, the pressure to demonstrate “AI strategy” creates incentive to announce cuts attributed to AI even when cuts would have happened regardless. The narrative has become more important than the numbers.
The Duke survey captures this perfectly: a 9x increase in AI-attributed layoffs coinciding with CFOs admitting AI hasn’t produced measurable results. The layoffs are certain. The productivity gains remain theoretical.
What You Can Do
If you’re in a target role (routine, clerical, administrative):
- Document everything you do that requires judgment, context, or human interaction
- Develop AI literacy—being able to use AI tools makes you harder to replace with them
- Watch for signs: task automation pilots, “efficiency initiatives,” reorganization discussions
If you’re hiring:
- Be skeptical of AI-attributed cuts at other companies
- The talent being let go may be doing work AI cannot actually perform
- Companies cutting now may face capability gaps when they realize their mistake
If you’re an investor:
- The productivity paradox resolved before. It may resolve again.
- But “eventually” is doing a lot of work in current valuations
- Companies that build organizational capability around AI—not just deploy it—have better track records
The 9x increase in AI layoffs tells us more about executive belief than AI capability. When the productivity data arrives, we’ll know whether that belief was justified.