White-collar workers are spending an average of 6.4 hours per week managing, correcting, and feeding context to AI tools rather than doing the work the tools were supposed to automate. Researchers have named this phenomenon “botsitting,” and a major study published in June 2026 found that it is destroying productivity gains, fueling job frustration, and driving high performers to look for new jobs at a rate 73 percent above those who are not botsitting.

The findings come from the Work AI Index report produced by Glean’s Work AI Institute in collaboration with researchers from Notre Dame, Stanford, and UC Berkeley. The study surveyed 6,000 full-time workers in the US, UK, and Australia between December 2025 and January 2026, all of whom primarily work on computers or digital tools.

What Botsitting Actually Involves

Botsitting is the labor of making AI usable. It includes feeding AI tools context they should already have access to, checking outputs for errors before acting on them, reformatting AI-generated content to match required standards, and debugging failures when AI tools do not integrate properly with each other or with existing systems.

According to IT Pro, workers describe spending significant portions of their day moving information between disconnected AI systems that cannot communicate with each other, effectively becoming human middleware in workflows that were supposed to be automated.

Customer service employees describe being expected to supervise AI agents instead of building customer relationships, the part of the job they enjoyed most. Analysts describe checking every AI-generated report for hallucinations before presenting to management. Writers describe editing AI drafts that require as much time as writing from scratch would have.

The Productivity Paradox

The study’s central finding is that AI adoption is producing a hidden labor transfer rather than pure productivity gains. Work that AI tools were supposed to eliminate is being replaced by new work created by AI tools. The net reduction in total work done per employee is significantly smaller than AI investment projections assumed.

The researchers describe a “botsitting tax” on productivity: companies invest in AI expecting a 30 to 40 percent efficiency gain but receive a smaller fraction of that because workers absorb the new management and correction overhead. In some cases, the overhead is large enough that teams with heavy AI adoption are less productive than teams without it.

The 6.4 hours per week figure is an average. Workers in roles with the highest AI integration, including software development, content creation, customer service, and data analysis, reported significantly higher botsitting time, with some individuals spending more than 15 hours weekly on AI oversight tasks.

The Retention Crisis

Workers who spend an above-average share of their AI time botsitting are 73 percent more likely to be actively looking for another job, the study found. The mechanism is straightforward: botsitting is unrewarding, invisible labor. Employees who absorb it receive no credit for managing AI systems. When AI succeeds, the credit goes to the technology. When it fails, the employee who supervised it is blamed.

The report states that “workers who absorb it without recognition or reward grow exhausted. Then they grow resentful. Then they start polishing their resumes.” For companies that have made large AI investments on the premise that it will allow them to do more with fewer people, the botsitting-driven retention problem represents a significant and underreported cost.

Botsitting vs. Promised AI Productivity Gains

AI Adoption Promise	Botsitting Reality
Eliminate repetitive data entry	Workers move data between disconnected AI systems manually
Automate customer service responses	Employees supervise AI agents instead of handling relationships
Speed up content production	Writers spend equal time editing AI drafts as writing originals
Generate accurate reports automatically	Analysts check every AI output for hallucinations before use
Reduce meeting prep time	Workers feed AI context from multiple systems to prepare summaries

What Companies Are Getting Wrong

According to the Glean Work AI Index, the core error most organizations are making is deploying AI tools before integrating them with each other. When an AI writing assistant cannot access the company’s CRM, and the CRM AI cannot talk to the project management AI, workers become the connective tissue. The labor savings never materialize because someone has to do what the systems cannot do for each other.

The research recommends that companies measure botsitting time explicitly, recognize and compensate employees who manage AI systems, and prioritize system integration before expanding AI tool deployment.

Frequently Asked Questions

What is botsitting?

Botsitting is the term coined by researchers at Glean’s Work AI Institute for the unrewarded labor of managing, correcting, and feeding context to AI tools. It includes checking AI outputs for errors, moving data between disconnected AI systems, reformatting AI-generated content, and supervising AI agents instead of doing the actual work. A 2026 study of 6,000 workers found employees spend an average of 6.4 hours per week on botsitting.

Is AI actually making workers more productive?

The research presents a more complicated picture than AI vendors suggest. While AI tools reduce time on some tasks, the hidden labor of botsitting absorbs a significant portion of those savings. The net productivity gain is substantially smaller than projected. In some teams with heavy AI adoption and poor system integration, the botsitting overhead makes those teams less productive than teams without AI.

Why are workers frustrated with AI at work?

Workers are frustrated because botsitting is invisible, unrewarded labor. When AI succeeds, the technology gets the credit. When it fails, the employee who supervised it is held responsible. Workers who spend large portions of their day doing AI oversight report feeling like they have more work than before AI adoption, not less. The 2026 Glean study found that heavy botsitters are 73% more likely to be job-hunting.