The US labor market added 178,000 jobs in March, according to the Bureau of Labor Statistics. The monthly change was limited even as companies touted AI-led growth and productivity gains, leaving questions about whether AI is translating into broader hiring or output increases.
Most hiring came from non-tech sectors. Healthcare added 76,000 jobs, construction 26,000, and transportation and warehousing 21,000. By contrast, tech-linked areas showed little strength: computing infrastructure and web search portals were largely unchanged, while computer systems design and related services lost 13,000 jobs.
That pattern contrasts with public claims of a tech hiring recovery. Marc Andreessen pointed to rising job openings at tech firms, but openings have not consistently translated into hires. March’s strongest gains were outside core tech, while many digital services stayed flat or declined.
Entry-level roles are under particular pressure. A Goldman Sachs analysis, cited in recent reporting, estimated AI displaced about 16,000 jobs per month over the past year. A 2025 SignalFire study found new graduate hiring in tech down about 50% from pre-pandemic levels, attributing the shift to smaller funding rounds, leaner teams, fewer graduate programs, and increased AI use. SignalFire summarized the change: the once-wide door for new grads is now barely cracked.
Goldman Sachs warned that displaced workers often move into more routine roles, which can erode the value of their skills and weaken long-term labor outcomes. Those shifts have intensified debate over who bears the costs of technological change even as some leaders predict long-term gains.
Executives report broad AI adoption and early returns. Harvard Business Review said 80% of leaders use AI weekly and 74% report positive returns from early deployments. Yet many workers report a different reality. Mercer found 43% of workers say their jobs are more frustrating since AI adoption. Workday reported that for every 10 hours of claimed efficiency, nearly four hours are spent fixing AI output. HBR highlighted “workslop”—polished-looking but shallow AI-generated content—seen by 41% of workers, with each instance adding roughly two hours of rework. Workday also found only 14% of respondents consistently achieve net-positive outcomes from AI use, suggesting widespread errors, extra review, and low trust.
Part of the divide may stem from how tools are used. Senior leaders often apply AI to strategy, drafting, and synthesis—tasks where models perform relatively well—while routine operations that require steady accuracy see less reliable results. Brian Solis of ServiceNow labeled the extra burden an “AI tax”: more checking, more rework, and more anxiety.
OpenAI has acknowledged AI’s disruptive effects and proposed policy ideas—expanded healthcare coverage, retirement savings support, and an industrial agenda—to begin addressing transitions. The company warned that if policy does not keep pace with technological change, institutions and safety nets needed to navigate the shift could lag behind.