Workers Automating Meaningful Work They'd Prefer to Keep Due to AI Anxiety
"A fascinating study that came out of Stanford a while back that found 41% of Y Combinator AI startups are automating things that people would prefer to keep human. And again, this again boils down to the psychology of this. We can't just assume that because the technology can do something, because it can automate something, it should be automated."
About this episode
In this episode of The Cognitive Revolution, host Nathan Labenz speaks with Rebecca Hines, head of the Work AI Institute at Glean and author of the new Work AI Index 2026 report. The report surveyed 6,000 knowledge workers across the US, UK, and Australia to capture how AI is actually being used in enterprises far outside the AI bubble. The headline findings reveal a paradox: 87% of workers now use AI, 73% report productivity gains, and they claim to save an average of 13 hours per week. Yet only 13% say their organizations are performing significantly better as a result. Hines introduces two new terms to explain the gap. Bot sitting describes the hidden labor of feeding AI context, debugging outputs, and cleaning up mistakes, which consumes 6.4 hours weekly—roughly half the reported time savings. Bot shitting refers to workers shipping AI-generated work they cannot explain or defend, with 69% admitting to this behavior. The report draws on both survey data and aggregated Glean platform telemetry showing 36% of AI sessions fail outright. Hines argues the root causes include lack of organizational AI strategy, context-poor tools that don't integrate, and perverse incentives like rewarding token consumption. Workers most threatened by AI are paradoxically automating meaningful work they'd prefer to keep, driving burnout and turnover. The conversation explores solutions including enterprise context graphs that connect all organizational data, measuring and rewarding collaborative AI use rather than individual clicks, aligning work to meaningful mission, and using AI detection thoughtfully to identify both top performers worth retaining and low performers gaming the system. Hines emphasizes this is fundamentally a human change management challenge, not just a technology rollout, and that organizations must be intentional about what work should remain human to preserve meaning, ownership, and judgment.
Key takeaways
- Workers report saving 13 hours weekly with AI but only 13% say their organization performs significantly better, revealing a massive productivity-performance gap.
- Bot sitting consumes 6.4 hours per week—half of AI time savings—as workers feed context, debug outputs, and clean up AI mistakes in poorly integrated systems.
- 69% of workers admit to bot shitting, shipping AI-generated work they cannot explain or defend, signaling both AI capability and prevalence of bullshit work.
- 36% of AI sessions fail according to Glean telemetry data, requiring workers to start over or heavily revise, with context-feeding causing the highest exhaustion.
- Workers most threatened by AI paradoxically automate meaningful work they'd prefer to keep, particularly relationship-building roles like customer service, driving turnover.
- Most successful organizations measure AI impact holistically, view AI as teammate not peer, reward collaborative use over individual token clicks, and align to clear mission.
- Enterprise context graphs that connect people, tasks, documents, and mission enable AI to recommend optimal work allocation and reduce bot sitting dramatically.