HBR study warns AI adoption can expand workloads and increase burnout risk

The biggest near-term risk with workplace AI may be less about layoffs and more about load. New field research suggests that when teams genuinely adopt AI copilots, the “time saved” often gets reinvested into additional tasks, extending the workday and raising burnout risk.

AI productivity gains can translate into more work, not fewer hours

Researchers from UC Berkeley observed AI use inside a roughly 200-person tech company over eight months, conducting 40-plus in-depth interviews, as reported by Harvard Business Review (hbr.org). The key finding: employees weren’t formally pushed into higher targets. Instead, the tools made more deliverables feel feasible, so workers expanded their own to-do lists. “Freed” time got absorbed by extra tickets, revisions, and faster turnaround expectations, with work creeping into lunch breaks and evenings.

For B2B marketers and e-commerce operators, the pattern is familiar: faster drafting, summarization, and reporting can lead to more campaigns, more variants, and more stakeholder pings per week. The danger is a silent shift in baseline output—where “normal” becomes whatever a team can produce with copilots.

Evidence is mixed on measured efficiency, but expectations still rise

The HBR piece also points to broader uncertainty on whether AI tools reliably compress task time. An experiment from METR found experienced open-source developers took 19 percent longer on tasks while believing they were 20 percent faster (metr.org). Separately, an NBER paper tracking adoption across thousands of workplaces estimated average time savings around 3 percent, with no meaningful changes in hours worked or earnings by occupation (nber.org).

Even if gains are real in your org, the management implication is the same: without explicit guardrails—capacity planning, SLA boundaries for “responsiveness,” and incentives tied to outcomes instead of throughput—AI can turn into a work-intensification layer.

The practical takeaway: treat AI rollout as an operating-model change, not a software install, and define what “done” and “enough” look like before output expectations reset upward.