AI Strategist Fires Half of Her Agents

Practitioners evaluating autonomous-agent tooling should treat claims of full automation with caution. Early deployments frequently shift work from task execution to context management, monitoring, and error handling, reducing net time savings and increasing ops friction. Industry-wide survey data from Glean's Work AI Index 2026 (6,000 workers across the US, UK, and Australia) puts numbers on what is often treated as anecdote: workers spend an average of 6.4 hours per week on botsitting tasks, and more than one-third (36%) of AI sessions fail outright, requiring a restart or substantial rework.

What happened - Business Insider reports that AI strategist Sol Rashidi, chief strategy officer at Cyera and a senior fellow at Harvard Kennedy School, told the outlet, "I just fired half my agents because they were unreliable." Rashidi had four agents running and deactivated two, and told Business Insider she was "spending more time babysitting them" than doing useful work.

Research context - Glean's Work AI Index 2026 finds that for every hour workers get useful output from AI, they spend roughly another hour making it usable. Only one in five companies has a mature governance model for autonomous AI agents, and workforce readiness remains the top barrier to scaling AI, the report finds. These structural gaps go beyond individual agent quality: most organisations lack the observability tooling and operational processes needed to manage agents at scale.

Technical context - The Rashidi anecdote illustrates three recurring operational gaps practitioners encounter when scaling agents: context continuity (keeping prompt state and retrieval reliable across sessions), exception handling (agents failing silently or producing incorrect outputs), and observability (debugging autonomous chains at scale). These are implementation- and integration-level problems that platforms must address before agents deliver consistent out-of-the-box ROI.

For practitioners: Track time spent on agent maintenance separately from task completion rates. Instrument chat and agent workflows for traceability. Pilot agents on narrowly scoped tasks where failure modes are low-cost, and apply a governance framework before expanding to multi-agent or production-critical workflows.