Botsitting is becoming normal work
Teams adopted AI tools because they promised speed. But many workers discovered a second job inside the first one: supervising the machine. They write prompts, wait for output, check facts, rewrite sections, remove risky claims, and make sure the final answer matches the business context.
That hidden supervision work is often called botsitting. It is not wasted time. In many workflows, it is the quality-control layer that makes AI usable. The problem is that most productivity dashboards do not know how to see it.
Where botsitting shows up
- Customer support: Agents review suggested replies before sending them to customers.
- Marketing: Writers turn AI drafts into brand-safe, accurate, original content.
- Operations: Managers check AI-generated summaries against real dashboards and tickets.
- Finance: Analysts verify formulas, assumptions, and source data before sharing reports.
- HR: People teams review AI-generated policy drafts for tone, fairness, and legal risk.
The productivity paradox
AI can reduce first-draft time while increasing review complexity. A task may finish faster, but the worker may carry more cognitive load because they are responsible for errors they did not manually create. This is why a simple "AI saved us 60%" story can be misleading. The real calculation must include supervision and rework.
Mini case study: the support team
A 20-person support team adds AI-assisted ticket replies. First-response time improves. But managers notice that experienced agents now spend more time checking generated replies for policy mistakes. Junior agents send drafts faster but need more QA review. The company only sees the benefit clearly after separating three metrics: AI draft time, human review time, and reopened-ticket rate.
The result is a better workflow: AI handles common first drafts, senior agents review edge cases, and the QA team audits risky categories. The team keeps the speed benefit without pretending that supervision is free.
How to measure botsitting
- Track time spent in AI tools separately from final-output tools.
- Record review blocks as real work, not idle time.
- Measure rework caused by inaccurate, incomplete, or unsafe AI output.
- Compare output quality before and after AI adoption.
- Ask employees where AI reduces work and where it merely moves work.
The best-practice policy
Do not punish people for slower AI-assisted sessions when the work requires careful review. A legal policy draft, client billing summary, or security report should not be optimized for speed alone. The right policy is risk-tiered: fast review for low-risk drafts, deeper review for client-facing or compliance-sensitive work.
For more on the evidence layer, see verified time tracking and AI-agent productivity measurement.
The bottom line
Botsitting is not a failure of AI. It is the human control layer that makes AI work useful. The companies that measure it honestly will make better AI decisions than companies that count only automation speed.