Jack Dorsey last week cut roughly 4,000 jobs at Block — nearly half the company — and reached for AI as his justification. In a Wired interview, he pointed to what he called a step-change in AI sophistication last December, named Anthropic's Opus 4.6 and OpenAI's Codex 5.3 specifically, and described an ambition to make Block feel like "a mini AGI." Block's stock climbed on the news. Seven current and former employees who spoke to the Guardian aren't buying the story.
Several of them suggested the cuts looked less like a genuine productivity revolution and more like a company using AI hype to reframe a bruised crypto narrative for investors. Whether the tooling actually justifies the headcount reduction is a separate question — and, according to those closest to it, the answer is mostly no.
John, a current Block employee whose role involves helping staff adopt AI, told the Guardian that around 95% of code changes generated by AI tools require human intervention before they clear company standards. Customer-facing chatbots have triggered support failures. Workers say they're being tracked on AI usage — down to which tools they're using and how many tokens they're burning — and that AI proficiency has been folded into performance reviews.
Mark, a former product employee, described attending a company anniversary event and concluding that no AI system could replicate the strategic judgment his job required — then receiving a layoff notice shortly after. He and others said they came to believe, in retrospect, that company-wide AI exercises were less about genuine productivity gains and more about mapping which tasks could eventually be automated. "In hindsight, it seemed like a thinly veiled attempt to get all this input from employees on what tasks to automate," he told the Guardian.
Across engineering and product teams, workers described grinding AI fatigue alongside a bitter irony: being asked to train the systems being used to justify cutting them.
Block's experience sits inside a broader pattern. Goldman Sachs estimated last year that AI tooling was already responsible for between 5,000 and 10,000 net US job losses per month across 2025. But Block's numbers undercut the grander claims: agentic coding tools needing human sign-off on 95% of their outputs aren't replacing skilled workers at scale, whatever the investor presentation says. The distance between a compelling AI narrative and what these tools can actually do in regulated, judgment-heavy environments — financial services chief among them — remains significant, and closing it will take considerably longer than last December's memo suggested.