The AI productivity revolution has a dirty secret: humans are working an extra day each week just to make the bots functional.
The Summary
- White-collar workers spend 6.4 hours weekly "botsitting" AI, feeding context, debugging errors, and cleaning up outputs, according to research from Glean's Work AI Institute
- 87% report individual productivity gains, but companies aren't seeing companywide performance improvements, creating a productivity paradox
- The hidden labor of making AI work is pushing frustrated workers to look for new roles, revealing the gap between AI's promise and its current reality
The Signal
Glean's Work AI Institute surveyed 6,000 full-time workers across the US, UK, and Australia between December 2025 and January 2026. All worked primarily on computers or digital tools. The researchers partnered with Notre Dame, Stanford, and UC Berkeley to document what they're calling "botsitting," a term they coined for the invisible maintenance work AI tools create.
6.4 hours per week. That's nearly a full working day spent not on the creative, strategic work AI was supposed to enable, but on the grunt work of making the bot useful. The pattern holds across roles and geographies: feed the model context it should already have, check outputs for hallucinations, debug when it misunderstands instructions, clean up formatting errors.
"Workers now burn an average of 6.4 hours a week botsitting — most of a full working day, every week."
Here's the paradox that Business Insider's Juliana Kaplan and Jacob Zinkula identified: individual workers report real productivity wins. 87% see gains in their own output. But their companies aren't capturing those gains at scale. The botsitting tax explains why.
The work itself is "often tedious" and "exhausting," according to the report. It's not the high-value work executives imagined when they mandated AI adoption. It's digital janitorial work:
- Providing context the AI can't access or remember
- Verifying outputs against reality because the model confidently invents facts
- Reformatting results to match actual business requirements
- Explaining to the bot what it got wrong, again
This creates a second-order problem beyond lost productivity. Workers are looking for other roles because of botsitting frustration. Companies pushed AI tools to cut costs and boost output. Instead, they're creating retention problems among the exact knowledge workers they can't afford to lose.
The discussion on Hacker News drew 255 points and 204 comments, suggesting this resonates beyond the survey sample. When 6,000 workers independently report the same pattern, and thousands more pile into the comment thread, you're looking at a structural issue, not an adoption curve problem.
The Implication
If you're mandating AI tools at your company, measure botsitting hours alongside productivity gains. The net benefit might be smaller than you think, or negative. If you're a worker drowning in bot maintenance, document the hours. When your manager asks why projects take longer despite "AI assistance," you'll have receipts.
The agent companies building Web4 need to solve for memory, context persistence, and autonomous error correction. Until agents can actually remember what you told them yesterday and verify their own outputs, humans will keep working that extra day per week. The productivity revolution starts when botsitting hours hit zero.