Botsitting, botshitting, and the hidden human labor of AI at work

Botsitting, Botshitting & the Hidden Human Labor of AI at Work

01

Executive Summary

The botsitting-botshitting cycle

Footnotes

The Work AI Index 2026: Global

Botsitting, botshitting, and the hidden human labor of AI at work

Download the PDF

Yi Zhu

Jacob Ewing

Designers

Abhijith Nair

Mohammed Hafeez

Niranjay Bhosale

Developers

SECTION 01

Executive Summary

AI has arrived in the workplace. The organizational impact has not.

87% of digital workers now use AI at work. 75% say it makes them more productive, saving them roughly 11 hours each per week through automation alone. Yet only 13% say their organization is performing significantly better as a result.

So where are the gains going?

They’re being swallowed by a new, largely invisible form of labor. We call it botsitting: the work required to make AI usable, including feeding it missing context, checking its outputs, debugging its mistakes, rerunning prompts, and cleaning up the confident-but-wrong answers AI leaves behind. Workers now burn an average of 6.4 hours a week botsitting — most of a full working day, every week.

When that labor is untracked, unbudgeted, and unrewarded, workers start cutting corners. They stop checking outputs and deliver work they can’t fully explain or defend. That’s when botsitting turns into something more dangerous: botshitting — shipping AI-generated work that workers haven’t reviewed, don’t fully understand, or couldn’t defend if asked. Today, 69% of AI users admit to botshitting at work.

The organizations pulling ahead aren’t simply using more AI. They’re building what we call the human infrastructure of AI. And they’re doing it at three levels.

Individual

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.

At the individual level

High AI achievers (people who report both productivity and quality gains from using AI) don’t just prompt and pray. They use their judgment. They spend more of their time botsitting (40% vs. 33% for low AI achievers) and are 18% more likely to deliberately refrain from using AI on certain tasks. But they’re also more likely to bend or break the rules to get value from it: 54% use unapproved tools or approved tools in noncompliant ways, and 36% hide how much AI is helping them — often because they’re working around an official system that is too slow, too narrow, or too disconnected from how the work actually gets done.

Team

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.

At the team level

High-achieving AI teams treat AI as a teammate rather than a tool (75% of high AI achievers trust AI as a teammate vs. 32% for low AI achievers). 64% of high AI achievers say AI is easier to collaborate with than their human colleagues and 74% say AI helps more with daily work than their manager does. Additionally, 44% say it is more fair than their boss — a number that climbs when managers have too many direct reports and too little time for any of them. This, however, does not mean human managers are becoming obsolete. Managers who are high AI achievers are offloading 32% more of the coordination work to AI, reclaiming time for coaching, mentoring, and helping their people build new AI skills.

Organization

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.

At the organizational level

Leading AI organizations resist AI addition sickness: the reflex to solve every problem by buying more AI, adding more tools, or pushing people to use AI whether or not it helps. They start with the work, selecting tools and platforms that fit the job instead of letting vendor contracts dictate their AI strategy. And they understand that giving AI access to data is not the same as giving it context. More than half (53%) of workers say critical information they need to do their jobs is not accessible through their AI systems. By contrast, workers in “context-rich” AI organizations are 64% less likely to feel worn out by AI, 52% less likely to ship work they can’t explain, spend 9% less of their AI time botsitting, and 31% less likely to botshit.

There’s plenty we still don’t know about how AI will transform work, but this much is clear: Organizations must build the human infrastructure (not just the technology infrastructure) that makes AI worth using, or they’ll keep paying the bill — in botsitting, in botshitting, and in the exodus of people who got fed up cleaning up after the bots.

SECTION 02

Introduction

Robin

It’s 11 p.m., and Robin, a junior software engineer, pastes a thousand lines of AI-generated code into a pull request and goes to bed. By morning, the build has broken. A senior engineer, already behind on her own deadlines, spends half the morning untangling code that no one on the team can explain — including Robin.

Robin is one of the 41% of workers who now ship AI outputs they can’t explain.

Evelyn

It’s Tuesday afternoon. Evelyn, a product marketer, runs the same prompt through three different AI tools because the first one didn’t sound right. Neither did the second. The third isn’t right either, but it’s “good enough,” and the deadline is 4 p.m.

Evelyn is one of the 60% of US workers who rerun the same prompt through multiple tools because the first output wasn’t good enough.

Michael

It’s 4:47 p.m. on a Friday. Michael, a financial analyst, uploads last quarter’s numbers to an AI assistant, skims the summary, and fires it off to his CFO. At Monday’s quarterly business review, three of the figures don’t match the spreadsheet they came from. The discrepancy derails the discussion, and Michael — who never opened the source file — blames the tool.

Michael is one of the 28% of workers who now blame their own mistakes on AI.

This is what AI at work looks like in 2026. ‍ The Work AI Index from the Work AI Institute is an effort to understand the hidden human labor AI has added to the workday. We surveyed 6,000 full-time digital workers1 across the United States, the United Kingdom, and Australia, spoke with dozens of AI leaders, and analyzed anonymized, aggregated workplace AI interactions from the Glean Work AI platform. What we found is a workforce that has embraced AI — along with a thick, mostly invisible layer of human labor holding the whole thing together.

AI is everywhere. The gains are not.

1

AI adoption is near-universal.

87% of digital workers use AI at work. 75% say it makes them more productive. 77% juggle multiple AI tools every week at work, with 33% using four or more.

AI adoption is near-universal.

87% of digital workers use AI at work. 75% say it makes them more productive. 77% juggle multiple AI tools every week at work, with 33% using four or more.

1

Workers are handing over bigger parts of their jobs to AI and want to hand over even more.

AI now automates 27% of their work output. Within a year, they expect that number to climb to 35% — a 30% jump in twelve months. And they want it higher still: 57% say they want AI to automate more of their job than they think it actually will.

Workers are handing over bigger parts of their jobs to AI and want to hand over even more.

AI now automates 27% of their work output. Within a year, they expect that number to climb to 35% — a 30% jump in twelve months. And they want it higher still: 57% say they want AI to automate more of their job than they think it actually will.

1

Workers are turning to AI first, sometimes before they turn to their colleagues, their managers, or even their own judgment.

48% reach for AI before they try to solve a problem themselves. 52% find it easier to collaborate with AI than with their human coworkers. 61% say AI helps them more with their day-to-day work than their own manager does.

Workers are turning to AI first, sometimes before they turn to their colleagues, their managers, or even their own judgment.

48% reach for AI before they try to solve a problem themselves. 52% find it easier to collaborate with AI than with their human coworkers. 61% say AI helps them more with their day-to-day work than their own manager does.

And yet the gains keep evaporating somewhere between the worker’s desk and the board deck. Workers say AI automation alone saves them roughly 11 hours a week (just under a third of their work week). But only 13% say their organization has significantly improved performance and outcomes because of it.

The productivity paradox of AI at work

11 hrs

Workers say AI automation saves them 11 hours a week

87%

of digital workers use AI at work

75%

say AI makes them more productive

13%

say their organization is performing significantly better because of it

So, where are those 11 hours going?

As it turns out, not to the higher-level work leaders promised AI would free people up to do. The hours are going into the work nobody planned for — the human labor of making AI itself usable. We call this work botsitting2.

DEFINITION

Botsitting (n.)

The largely unrecognized, unbudgeted, and untracked labor of making AI usable — feeding it context, supervising its output, debugging its mistakes, and cleaning up after it.

Methodology and caveats

The Work AI Index draws on a survey of 6,000 full-time (30+ hours per week) digital workers across the United States (n=3,000), the United Kingdom (n=1,500), and Australia (n=1,500), conducted between December 2025 and January 2026. The sample is nationally representative by age, gender, and income. “Digital workers” are those who report doing most of their work on a computer or digital tools. We focused on this group because AI is currently most embedded in digitally mediated work. Workers in other roles (frontline, manual, hands-on) use AI differently, and those experiences deserve their own study. ‍ Our sample skews higher on AI adoption, seniority, and digitally intensive sectors (especially the tech sector) than the broader working population. The main findings, however, hold after adjusting for role, industry, demographics, employment status, organization size, and AI usage intensity. Driver analyses use logistic regression to control for these factors. We report odds ratios for binary outcomes and percentage-point differences for continuous ones. ‍ The survey data are self-reported, which means they are subject to social desirability bias and recall bias. We screened out inattentive and “speeder” respondents using standard attention checks. Where possible, we triangulated survey findings with interviews, case studies, third-party research, and anonymous, aggregated telemetry data from the Glean Work AI platform. ‍ Despite these caveats, the findings point to a shift in how work actually gets done, and to a widening gap between what leaders think AI is accomplishing and what their employees are doing to make it work. Most “state of AI” analyses (and most AI strategies inside organizations) treat AI as if it lives apart from the messy reality of work. They focus on model performance, speed, and which tasks and jobs are theoretically at risk. They pay far less attention to what happens when the technology meets real workflows — what workers actually do with it, and how organizations deploy, manage, or mismanage

[truncated for AI cost control]