Block Fired 4,000 People for AI: Here Is What Nobody Is Talking About

This Is Real. And the AI-Washing Defence Does Not Hold Up

Start with the facts. Block was financially healthy when it made these cuts. Its stock surged 24 per cent. Dorsey did not just announce layoffs — he mandated that every remaining employee use AI daily and made AI fluency part of performance reviews. This is not a company trimming costs and blaming AI for the optics. This is a company reorganising itself around a fundamentally different way of working.

The “AI washing” argument — popularised by Bloomberg and Altman — is tempting but flawed. Yes, some companies are using AI as convenient cover. But Block tripled its workforce between 2019 and 2022, growing from under 4,000 to over 12,000 employees. Some correction was inevitable. The question is whether AI changes the nature of that correction. It does.

Sam Altman’s “AI washing” warning deserves scrutiny of its own. He made it at the India AI Impact Summit on 19 February — one week before Block’s announcement. Altman has a clear interest in distancing AI from mass layoffs. If the public associates AI with job losses, it threatens the social licence that companies like OpenAI depend on. Dario Amodei, the CEO of Anthropic, has been more direct: AI will genuinely transform the nature of work, and we should prepare for that honestly rather than pretending it will not happen.

The honest reading: this is what AI-driven restructuring looks like. And it is coming.

Most Companies Will Get This Catastrophically Wrong

AI works. I say this not as a commentator but as someone who builds and deploys AI systems daily — in my own businesses and in my work with business executives through the University of Bath’s Executive Education programme.

The technology is extraordinary when deployed with the right models, the right architecture, and genuine expertise. I have seen it transform operations, compress weeks of work into hours, and enable capabilities that were simply impossible two years ago.

But that is not what most companies are doing.

The outputs from these tools routinely contain inaccurate, incomplete, or entirely fabricated information. And these outputs are feeding directly into real business decisions — market analysis, strategic planning, geographic expansion, hiring.

This is not an AI problem. It is a knowledge and leadership problem.

The numbers tell this story at scale. RAND Corporation research shows 80 per cent of AI projects fail. An Orgvue survey of UK executives found that 55 per cent of those who replaced workers with AI regretted it — a finding echoed by Davenport and Srinivasan in Harvard Business Review, who concluded that companies are laying off workers based on AI’s potential, not its demonstrated performance.

Klarna — the most instructive cautionary tale — announced in February 2024 that its OpenAI-powered assistant was “doing the work of 700 employees.” By May 2025, CEO Sebastian Siemiatkowski told Bloomberg they had gone too far. The company had over-automated complex customer interactions, rolled out at scale without proper piloting, and was now rehiring human agents.

These are real failures. But here is what almost every analysis of them misses — and it is the single most important thing to understand about AI right now.

The Technology Has Changed Beyond Recognition

Klarna’s problems happened with GPT-4 in early 2024. That was two years ago. In AI, two years is not two years. The difference between the models available in early 2024 and those available in early 2026 is not an incremental upgrade — it is a generational leap.

In terms of capability, accuracy, reliability, and the complexity of tasks these systems can handle, the progress is equivalent to what would have taken ten to twenty years in normal technology cycles before AI.

Systems that produced fabricated outputs in 2024 now deliver verifiable, high-quality results when deployed with the right architecture.

Most of the “AI does not work” narrative — the 80 per cent failure rate, the regret statistics, the cautionary tales — is based on experiences with technology that is already obsolete. Drawing conclusions about AI’s current capability from data gathered in 2023 and 2024 is like judging the commercial potential of the internet based on 1996 dial-up speeds.

This is what the debate is missing. Not all AI is created equal — and the AI available today bears little resemblance to the AI that produced the headlines everyone is still citing. Companies making decisions based on last year’s evidence are making decisions based on outdated information.

History Tells Us Exactly What Happens Next

If this feels unprecedented, it is not. Every major technological disruption — mechanisation, electrification, the internet — followed the same pattern. The technology ultimately created more opportunity than it destroyed. But the gains went disproportionately to those who adapted early and adapted well.

The key word is “well.” Early adoption without competence is worse than no adoption at all. The companies that thrived through previous transitions were not the ones that moved fastest — they were the ones that understood the technology deeply enough to deploy it where it genuinely added value.

The question is not whether AI will transform work — it will. The question is who builds real capability and who falls into the gap between AI awareness and AI competence.

UK government data confirms this divide. A DSIT survey published in January 2026 found that 56 per cent of employers already using or planning to use AI rate their own knowledge as “beginner” or “novice.” Sixty-one per cent have no staff working with AI at all. The awareness is there — 97 per cent of individuals have heard of AI — but only 17 per cent can explain it in any detail.

The divide forming is not between companies that use AI and those that do not. It is between those with genuine expertise and those operating with the dangerous illusion of it.

What This Means for You

If you are a business leader, the lesson from Block is not to panic and slash your workforce. It is also not to dismiss the signal and wait.

It is to build genuine expertise — quickly.

That means understanding the difference between a cheap chatbot and an enterprise-grade AI system. It means developing the judgement to distinguish tasks AI can genuinely transform from those it will make worse. It means building internal knowledge rather than outsourcing understanding to vendors who have their own interests.

And it means recognising that the technology changes so fast that what you tested six months ago tells you almost nothing about what is possible today.

The companies that will win this transition are the ones building real capability now — not the ones making headlines with dramatic layoffs, and not the ones still watching from the sidelines.

Block’s 4,000 layoffs are a signal. Read it correctly.