AI is causing productivity declines | Brian Merchant

Summary

The discussion argues that many companies adopted AI too quickly, often before testing whether it could reliably perform the work being cut. According to Brian Merchant, the result has sometimes been lower-quality output, additional review work, lost institutional knowledge, and employees being expected to do more with fewer resources—rather than clear productivity gains.

Business takeaways and practices

• Evaluate AI against real work before restructuring. Identify which tasks AI can perform reliably, where it fits into existing workflows, and where human expertise remains essential. Merchant criticizes top-down mandates driven by executive enthusiasm or fear of missing out rather than operational evidence.
• Let employees test tools in context. Provide access and allow teams to experiment with appropriate uses—or decide a tool is not useful for their work—instead of imposing a single company-wide method.
• Measure the full cost and output, not just headcount savings. AI-generated work may require substantial human checking and correction. Assess quality, rework, workflow disruption, subscription costs, and productivity alongside labor savings to determine actual return on investment.
• Watch for labor intensification. Automation may not reduce workload: employers can use it to raise output expectations, speed up work, or assign additional tasks to remaining staff. Merchant describes this as a recurring pattern in which productivity gains accrue primarily to owners rather than workers.
• Account for burnout and work quality. Mandatory AI use can make work less satisfying, disrupt established processes, and leave employees responsible for maintaining quality while meeting higher targets. The discussion warns that these pressures can increase burnout.
• Recognize the risks of deskilling. Merchant argues that automation can shift work away from skilled employees toward cheaper, more easily monitored labor. In creative fields, AI-generated assets may increase competition for illustrators and artists while lowering the perceived value of their skills.
• Treat employee knowledge as an operational asset. Some companies reportedly found that cuts removed people with critical skills or institutional knowledge, leading them to rehire. Workers with scarce, business-critical expertise may have more leverage to negotiate pay, job-security commitments, or other protections.
• Use employee feedback and collaboration. Merchant recommends that workers share information and build solidarity to seek safeguards against poorly considered AI adoption. Managers should include frontline employees, who understand day-to-day work, in adoption decisions.

Metrics and examples cited

• AI-related rehires: An SHL survey of workers in the US and UK found that about 20% of people replaced by AI were later rehired—roughly one in five cuts reversed.
• Returning workers’ expectations: About half of surveyed returning workers said they would need a significant pay increase; 43% wanted job-security guarantees; and about 30% wanted an apology from their employer.
• Workload and burnout: A Hubstaff report cited in the interview found that 46% of computer professionals using AI said their expected workload had increased, and 7 in 10 had experienced burnout in the prior year.
• Company examples: Merchant refers to Klarna as a case where a company reportedly concluded it had gone too far and needed to rehire. The host also mentions Ford and Amazon hiring again after reporting large losses; the subtitles provide no further details.
• Productivity context: The host cites Apollo data claiming that revenue or productivity growth outside the top seven S&P 500 companies was absent. This is presented as context for the AI-adoption debate, not as a detailed analysis in the interview.
• Adoption targets and timelines: No specific targets or timelines are proposed. The central recommendation is to make AI use more selective and judge it by demonstrated business results.

Presenters and sources

• Isaac Pound, host of The Tech Report
• Brian Merchant, author of Blood in the Machine and its newsletter
• Sources cited: SHL, Hubstaff, and Apollo data; Merchant also references interviews from his “AI Killed My Job” series.