Block cut more than 4,000 employees on February 27, shrinking its headcount from 10,000 to under 6,000. The reason given internally was that “AI tools change everything.” Less than a month later, some of those laid-off workers were asked to return, putting the company’s AI-heavy justification under scrutiny.
Business Insider reported that rehired staff came from several teams, including engineering and recruiting. One design engineer wrote on LinkedIn that management said he had been let go because of a “paperwork error.” A human resources employee said in a deleted post that a manager pushed repeatedly to bring them back. Another former worker said Block called about a week after the layoff and asked them to return. Jack Dorsey has not publicly addressed the rehiring. Even if the number is small relative to the original cuts, the reversal suggests some roles were not as replaceable as the company’s message implied.
The cost case for AI is not always straightforward
The article argues that enterprise AI is not automatically cheaper than human labor. It cited pricing for Claude Opus4.6 at $5 per million input tokens and $25 per million output tokens. Qwen3.5 plus was listed at 0.8 yuan per million input tokens and 4.8 yuan per million output tokens. One user who relied on Claude 4.5 and 4.6 through OpenClaw for research and daily assistance said token spending reached about $6,000 in a little over a month.
That was for individual use. In a company setting, costs can rise fast once AI systems are expected to handle real workflows. Customer service is one example. Answering basic prompts is very different from replacing live support staff who deal with complex tickets, multiple knowledge bases, long conversations and production stability.
The piece also pointed to Klarna. In 2024, the Swedish payments company cut more than 1,000 jobs and said its AI customer service tools could cover the workload of 700 support agents. By May 2025, Bloomberg and other outlets reported that Klarna had started hiring customer service workers again, while its chief executive admitted the company had moved “too fast” on AI.
Layoffs remove more than headcount
The article framed the issue through the Jevons paradox: higher efficiency does not always reduce total resource use. In workplaces, that can mean AI makes employees faster, then companies respond by assigning more work instead of needing fewer tasks done. In practice, the result is often not full replacement by AI but heavier workloads for the people who remain.
There is also an organizational layer that AI does not easily absorb. Companies run on formal reporting lines, but they also depend on informal coordination, judgment and internal relationships. AI can be inserted into process flows, yet it does not naturally fit into those hidden structures. When layoffs hit, what disappears is not only labor capacity but also parts of the coordination fabric that kept teams working.
Jensen Huang criticizes AI as a layoff excuse
During NVIDIA GTC2026, Jensen Huang said in an interview that leaders who respond to AI by cutting staff are doing so because they cannot think of a better answer, even when they have strong tools available. In that view, AI should be used to expand a business and open new lines of work, not simply to package cost cutting as technological progress.
The article described AI as a broad justification for layoffs across tech. It cited another example from 2022, when Elon Musk completed the Twitter acquisition in October and then cut about half of the workforce, or more than 3,000 employees, in early November. The company later asked dozens of laid-off workers to come back after discovering some had been cut by mistake or were needed in critical roles.
For Block, the rehiring does not resolve the original decision. What it does show is that AI may reshape staffing models, but it still cannot automatically cover gaps in strategy, operations or management.