Skip to main content

Why Companies Regret Firing Everyone for AI

Pooja Dutt's walk through the firing-and-rehiring whiplash of the AI boom, and the uncomfortable reason the layoffs reversed: the benchmarks were testing the wrong thing.

This page summarizes the video essay Why Tech Companies Regret Firing Everyone (for AI) by Pooja Dutt. It is cited here because it is one of the clearest plain-English accounts of what happens when leaders treat AI as a one-to-one replacement for people before the work has proven it can be. Watch it on YouTube.

The whiplash​

Between 2022 and 2026 the largest technology companies cut staff in waves on the theory that AI would replace them, with public predictions that AI would do ninety percent of developers' work by 2030. Meta announced an AI buildout in the tens of billions the same week it posted thousands of the engineering roles it had spent two years cutting. Then the reversal: a 2026 Forrester study found that fifty-five percent of employers regretted AI-driven layoffs, and a survey found that sixty-six percent of companies that made AI-driven cuts had already begun rehiring, many within six months. Klarna replaced roughly seven hundred service staff with AI, watched quality fall and customers revolt, and went back to hiring humans.

Why the benchmarks misled everyone​

The cuts were justified by benchmark scores that looked decisive. AI's score on HumanEval rose from thirteen percent in 2021 to over ninety-five percent by 2025, which read like obsolescence for human engineers. The catch is that HumanEval is only 164 Python problems, and frontier models had largely seen those exact problems in training. On a harder, more realistic benchmark, BigCodeBench, the best models scored about thirty-five percent against ninety-seven percent for humans. The benchmarks were not lying so much as testing the wrong thing. The models were faster and stronger at narrow, seen problems, and still hallucinated, still made mistakes, and still could not be left alone to produce what an experienced engineer produces.

The lesson for owners​

The deeper point is that headcount reduction is not the same as ROI. Companies cut first, discovered the demo did not survive contact with real work, and paid twice: once in severance and once in rehiring. The takeaway for any owner weighing transformation is that capability on a benchmark is not capability in your business, and a layoff justified by a demo is a decision made before the evidence existed.

Further reading​

Source: Dutt, Pooja. Why Tech Companies Regret Firing Everyone (for AI). YouTube. Watch.