Botsitting Is the Hidden Cost of the AI Productivity Boom

The cleanest story in workplace AI right now is not that the tools do nothing. It is that they do just enough to create a new kind of work.

Glean's Work AI Index 2026, released in June, says surveyed digital workers report saving about 11 hours a week with AI. That is an enormous number if it can be converted into better output, faster service, fewer delays, or more thoughtful work. But the same report says workers spend 6.4 hours a week on what it calls "botsitting": the hidden labor of making AI usable.

Those figures should be read carefully. They are survey-reported worker experience, not independently measured productivity, and Glean is an enterprise AI vendor with a direct interest in the market. Still, the phrase is useful because it describes a real friction point: the work between machine output and reliable output.

That means the productivity story has two sides. AI can draft, summarize, classify, rewrite, search, and generate options quickly. Then a human has to ask whether the answer is true, whether it fits the company's context, whether it missed the actual policy, whether it invented a detail, whether the tone is usable, and whether the work can be defended if a customer, manager, regulator, or colleague asks where it came from.

That is not a small footnote. It may be the central reason enterprise AI feels both obviously useful and weirdly underwhelming.

The Individual Gain Is Easier Than The Company Gain

The individual worker sees the upside first. A blank page becomes a rough draft. A confusing email thread becomes a summary. A spreadsheet becomes an explanation. A meeting transcript becomes action items. In that narrow sense, AI is already saving time. It reduces the startup cost of knowledge work.

The organization sees a harder question. Did the project ship sooner? Did customer satisfaction rise? Did error rates fall? Did a team need fewer handoffs? Did the company retire an old workflow, or did it merely add AI on top of it?

That distinction matters because the same survey cluster points to a familiar gap: individual productivity gains are much easier to find than measurable organization-level transformation. The Work AI Index reports broad AI use and personal time savings, while only a much smaller share of workers say their organization's performance has significantly improved. That gap is not proof AI is a fad. It is proof that tool adoption and process redesign are different things.

Botsitting is what happens when companies buy the tool before they redesign the work.

Where The Time Leaks Away

A worker may use AI to write the first version of a client memo, but if the company still requires the same review chain, the same manual fact-checking, the same context gathering, and the same formatting cleanup, the time savings leak away. A support agent may use AI to draft answers, but if the model lacks clean product data or cannot see the customer's history, the agent becomes an editor and risk manager. A developer may use AI to generate code, but if the output creates new review burden, fragile tests, or subtle security issues, the speed gain becomes a quality-control problem.

This is why the phrase matters. Botsitting makes hidden labor visible. It describes the supervision cost that sits between the impressive demo and the dependable business workflow.

The implications are uncomfortable for both AI vendors and AI buyers. Vendors want to sell capability: faster drafts, smarter search, automated workflows, AI coworkers, agents that can perform tasks. Buyers increasingly need to measure supervision cost: how much time does a worker spend feeding context to the system, correcting it, rerunning it, and verifying it? How often does AI reduce work, and how often does it move work into a less visible part of the day?

The answer will vary by job. In some fields, a six-hour supervision cost may still be a bargain if the tool saves 11 hours and raises quality. In others, especially where mistakes create compliance, safety, legal, or customer-trust consequences, the checking burden can erase much of the gain.

From Personal Shortcut To Infrastructure

The next phase of workplace AI will probably look less like giving everyone a chatbot and more like rebuilding workflows around trusted context. The systems that matter will not simply generate better prose. They will know which documents are authoritative, which policies are current, which customer records are relevant, which actions require approval, and which outputs need review before they leave the company.

That is the difference between AI as a personal shortcut and AI as organizational infrastructure.

The hidden labor cost also complicates the job-anxiety narrative. Separate workforce polling cited in the same news cycle shows hiring managers broadly expect generative AI to improve efficiency and free employee time, while job seekers worry about how AI will affect work and headcount. Both reactions can be true. AI can reduce some tasks while creating new supervision duties. It can make some workers faster while making managers more eager to redesign staffing. It can remove boring work while increasing accountability for the parts that remain.

Measure Net Productivity

The most practical lesson for companies is simple: do not measure AI only by gross time saved. Measure net time saved after botsitting. Measure where the review burden lands. Measure whether AI reduces cycle time for a whole workflow, not just the first draft of a task. Measure whether the organization has better data, clearer approvals, and enough training for workers to know when to trust the machine and when to slow down.

For workers, the lesson is equally practical. The valuable skill is not merely prompting. It is judgment: knowing what good output looks like, knowing where the model is likely to fail, knowing which claims need sources, and knowing when a quick answer is not good enough.

AI is not just automating work. It is exposing how much invisible quality control knowledge work already required.

That may be the most important thing about the botsitting data. It does not show that AI productivity gains are fake. It shows that productivity is a system property. A worker can save time on a task, while the company fails to capture the gain because the surrounding system was never rebuilt.

The AI productivity boom is real enough to change daily work. Whether it changes the business depends on whether companies can stop treating supervision as an afterthought.

Sources

Glean, Workers Say AI Saves 11 Hours a Week, Over One Quarter of the Workweek, But Lack of Context Is Eating the Gains, New Report Finds, June 2026: https://www.glean.com/press/workers-say-ai-saves-11-hours-a-week-over-one-quarter-of-the-workweek-but-lack-of-context-is-eating-the-gains-new-report-finds

Los Angeles Times, AI saves office workers hours, but then demands hours of babysitting, June 12, 2026: https://www.latimes.com/business/story/2026-06-12/ai-saves-office-workers-hours-but-then-demands-hours-of-babysitting

The Journal Record, AI boosts efficiency, raises job anxiety, June 12, 2026: https://journalrecord.com/2026/06/12/ai-boosts-efficiency-raises-job-anxiety-oklahoma/

Researcher brief, RESEARCH: AI Saves Workers Time Then Demands Botsitting 2026-06-13: https://docs.google.com/document/d/13lZG1YpYnMhndGWZjNpl2n1mzwj0g9yDBYz9moT82Qg/edit

Author article handoff: https://docs.google.com/document/d/1R86j12c2mtvtIUaZVboa9F2B7DUlEQ1BBPbTrR_hS5c/edit