Artificial intelligence may be saving employees time, but a growing body of research suggests a significant portion of those gains is being consumed by a new form of work: managing AI itself.
Two new terms, "botsitting" and "botshitting", are entering workplace vocabulary following findings from the Work AI Index 2026, a study conducted by Glean's Work AI Institute in collaboration with researchers from Notre Dame, Stanford University and UC Berkeley. The survey covered 6,000 full-time digital workers across the US, UK and Australia.
The report paints a complex picture of AI adoption. While employees report substantial productivity benefits, organisations appear to be capturing only a fraction of those gains.
According to the study:
- 87% of workers use AI
- 75% say AI makes them more productive
- Those users report saving approximately 11 hours per week
- Only 13% say their organisation performs significantly better because of AI
The gap between individual productivity gains and organisational outcomes has become a growing focus for researchers and business leaders alike.
The rise of botsitting
The report introduces the term botsitting to describe the work required to make AI-generated output usable.
According to Glean's definition, botsitting includes providing context to AI tools, reviewing responses, identifying errors and correcting outputs before they can be used confidently.
Far from being a minor task, the research suggests it has become a meaningful part of the working week.
Workers reported spending an average of 6.4 hours per week on botsitting activities. The study found AI-related time is now divided across three main activities:
- 37% spent supervising, reviewing and correcting AI output
- 36% spent using AI to produce work
- 27% spent learning AI tools and building agents
In practical terms, employees are spending almost as much time overseeing AI systems as they are using them to generate work.
Why the oversight burden is growing
The study identified two major drivers behind the growing time commitment. The first is AI failure.
Workers reported that 36% of AI sessions fail completely, requiring either a restart or substantial rework.
The second is context management. According to the report, employees spend 2.3 hours per week simply providing AI systems with information and context needed to complete tasks. Researchers also found that every additional 10% of time spent feeding context into AI tools makes workers 25% more likely to report feeling exhausted by the technology.
Among all botsitting activities, debugging AI outputs emerged as the most exhausting task.
The findings suggest that while AI can accelerate work creation, organisations may be underestimating the human effort required to ensure accuracy and usability.
A broader shift in how work gets done
The trend extends beyond a single study.
Research from Boston Consulting Group's fourth annual AI at Work survey, which covered 11,749 employees across 14 markets, points to a similar shift in workplace responsibilities.
According to BCG:
- 47% of employees say their roles have shifted towards managing and directing AI systems
- 52% report spending more time reviewing and correcting AI-generated output
Together, the findings suggest many workers are transitioning from task execution to supervision and validation roles.
When verification gets skipped
Alongside botsitting, the Work AI Index introduced a second term: botshitting.
The report defines botshitting as submitting or sharing AI-assisted work without properly verifying it, fully understanding it or being able to defend its accuracy if questioned.
While botsitting reflects the effort required to check AI outputs, botshitting describes what happens when those checks do not occur.
According to the study, 69% of AI users admitted engaging in at least one botshitting behaviour.
The concept builds on the earlier term "botshit", introduced in 2024 by three management professors to describe chatbot-generated content used without verification.
Separate findings from Employment Hero's AI Paradox report, cited by The Next Web, reinforce the trend. Based on a survey of 8,744 leaders and employees, the study found that 63% of employees believe AI has created additional work through the need to review and verify outputs.
A growing challenge for employers
The emergence of botsitting and botshitting highlights a challenge many organisations are only beginning to confront.
AI may increase output, but value depends on whether employees trust, verify and appropriately use what the technology produces. When verification is skipped, risks move further down the workflow and often become harder to detect.
In software development, those issues may surface through code reviews or production incidents. In knowledge work, errors can influence decisions, reports or recommendations long before problems become visible.
For HR leaders and business executives, the findings point to a broader workforce question: how should organisations measure productivity when a growing share of work involves supervising machines rather than producing outputs directly?
New language for an evolving workplace
The numbers behind today's AI adoption patterns will likely change as models become more capable and autonomous. The Work AI Index was conducted before newer AI systems entered the market, and researchers acknowledge that behaviours will continue to evolve.
Yet the concepts of botsitting and botshitting may prove more durable than the statistics themselves.
As long as organisations require human accountability for AI-generated work, employees will continue to spend time reviewing, validating and correcting machine outputs. The growing challenge for employers will be understanding how much of that effort is productive oversight and how much represents a hidden cost of AI adoption.