Babysitting the Machine: Glean's Rebecca Hinds on the Hidden Human Labor of AI at Work
Cognitive RevolutionAI Summary
→ WHAT IT COVERS Glean's Rebecca Hinds presents findings from the Work AI Index 2026, a survey of 6,000 digital workers, revealing a paradox: 87% use AI, 73% report productivity gains averaging 13 saved hours weekly, yet only 13% say their organization performs significantly better. Two new concepts — bot sitting and bot shitting — explain where the productivity gains disappear. → KEY INSIGHTS - **The Productivity Paradox:** Survey data from 6,000 workers shows 87% use AI and report saving 13 hours weekly, but only 13% say their organization performs significantly better as a result. The gap exists because individual productivity gains fail to translate to team and organizational outcomes — a phenomenon researchers call coordination neglect, where each person looks productive while the collective output remains hollow or redundant. - **Bot Sitting Tax:** Workers spend an average of 6.4 hours per week on "bot sitting" — manually feeding AI context, debugging probabilistic outputs, and cleaning up errors — consuming roughly half of all reported AI time savings. The highest exhaustion comes from two activities: supplying context the AI should already have, and debugging LLM outputs where the probabilistic nature makes it unclear what change actually fixed the problem. - **Bot Shitting Scale:** 69% of workers admit to submitting AI-generated work they cannot explain or defend if questioned. This behavior follows a predictable cycle: pressure to adopt AI leads to more bot sitting, exhaustion from unrewarded bot sitting leads to satisficing on "good enough" outputs, and those outputs get shipped without verification. Organizations that reward token consumption metrics or tool clicks accelerate this cycle. - **Alienation and Turnover Signal:** Workers who bot sit the most are disproportionately likely to be actively job searching. The mechanism runs two ways: heavy bot sitting signals to employees that their organization lacks a coherent AI strategy, eroding confidence; while heavy bot shitting correlates with broader disengagement. A third hypothesis from Berkeley researcher Aruna suggests heavy AI users recognize their increased market value and seek better opportunities externally. - **Context Graph as Solution:** Glean's enterprise graph connects people, documents, tasks, goals, and organizational history into a unified context layer. When AI has this context, it reduces bot sitting by eliminating manual context-feeding, routes tasks to appropriate models based on complexity and cost, and enables proactive recommendations. The graph also enables dynamic project staffing — identifying the right skill combinations across a 10,000-person organization in real time rather than relying on static org charts. - **AI Detection and Retention Strategy:** Leaders should combine AI detection tools (Pangram Labs is cited as gaining credibility) with quality scoring to distinguish high-performing AI collaborators from bot shitters. Employees using AI heavily but effectively represent the highest flight risk. Effective organizations reward collective AI collaboration — not just individual productivity — through hackathons that prize best before-and-after prompts, peer feedback quality, and co-creation, not just business impact metrics. - **Mission as Organizational Infrastructure:** Organizations flattening hierarchies or cutting headcount through AI need a strong, employee-understood mission to replace the decision-making function that hierarchy previously provided. Without a clear mission that employees can connect to their daily work, AI adoption produces symbolic compliance rather than meaningful transformation. The enterprise graph can surface whether individual work actually ladders up to company mission, making misalignment visible and addressable. → NOTABLE MOMENT Host Nathan Labenz describes his own near-miss with bot shitting: his automated podcast prep agent produced a thorough research document that covered roughly 10-20% of the actual conversation topic because it missed a recently emailed report. He caught it before sending — but notes that if he hadn't, it would have been a textbook example of the exact behavior the report documents at scale. 💼 SPONSORS [{"name": "Anthropic (Claude)", "url": "https://claude.ai/tcr"}] 🏷️ Enterprise AI Adoption, Bot Sitting, Organizational Productivity, AI Change Management, Knowledge Worker Behavior, Enterprise Knowledge Graphs, AI Culture
