A wave of tech layoffs is sweeping through the industry, driven not by poor performance but by executive confidence in AI agents, “confidence that may be running well ahead of reality”.
The most striking example came on May 21, when ClickUp founder Zeb Evans announced a 22% workforce cut, roughly 290 of 1,300 employees. The reason wasn’t financial pressure. Evans framed it as a structural bet: the company now runs around 3,000 internal AI agents, a three-to-one bot-to-human ratio. Remaining employees are being offered salaries up to $1 million for those who generate “100x impact” by managing these agents.
Evans said the quiet part aloud, these roles, in his view, are simply obsolete.
He’s not alone. The same week, Meta cut thousands of roles despite strong revenues, Oracle reduced headcount, and GitLab restructured around what it calls “the agentic era.” Layoff trackers estimate over 100,000 tech workers have lost jobs across roughly 250 companies in 2026 so far, nearly matching all of 2025, with seven months still to go.
The problem? The evidence that AI agents can carry this load doesn’t yet exist.
A 2026 NBER survey of nearly 750 executives found real AI productivity gains, but also a troubling “productivity paradox,” perceived gains running well ahead of measured ones. Separately, METR found that workers overestimated AI’s effect on their task time by 40 percentage points on average. Executives, who mostly experience AI through polished demos rather than messy real-world implementation, sit at the furthest end of this distortion.
PwC’s 2026 research adds more nuance: three-quarters of AI’s measurable economic gains are going to just 20% of companies, and those leaders are winning through slow, deliberate reinvention with heavy investment in training people, not just deploying tools. A majority of enterprises using generative AI still report no measurable impact on profitability.
Even optimistic forecasts don’t justify the urgency. MIT researchers concluded that AI agents will handle most text-based work at minimally sufficient quality by 2029, with agents consistently outperforming humans a few years after that. That’s a reasonable planning horizon, not a basis for cutting a fifth of your workforce today.
The core mistake executives are making is confusing the demo for the destination. The demo always works. Real implementation is where hallucinations slip through, edge cases pile up, and human judgment quietly holds things together.
Companies acting on honest, measured assessments of AI capability will build real advantages. Those mistaking confidence for evidence may find the cost arrives a quarter or two after the headcount is already gone.
Previous
