There's a new word for the thing you've been feeling.
It's called botsitting.
Researchers at Glean's Work AI Institute - a collaboration that includes experts from Stanford, UC Berkeley, Emory, and Notre Dame - surveyed 6,000 full-time workers across the US, UK, and Australia this year. The LA Times covered it today. The findings should be required reading for anyone running a business with AI tools.
Here's the summary:
AI now automates more than a quarter of digital work. It saves the average worker about 11 hours per week.
But workers also spend 6.4 hours per week doing something the researchers call botsitting: feeding context to AI tools that don't know your business, supervising outputs that look finished but aren't, debugging errors that crept in somewhere between prompt and output, cleaning up AI-generated work that landed wrong, and switching between a stack of tools that weren't built to talk to each other.
Eleven hours saved. Six hours consumed. Net: 4.6 hours.
And only 13% of organizations say AI has significantly improved their company's overall performance.
This Is Not an Argument Against AI
Let's be clear about what this story isn't.
It's not "AI is a scam." It's not "go back to spreadsheets." The 11 hours are real. The gains in certain tasks - drafting, summarizing, generating first versions of things that used to take a full afternoon - are real.
But there's a hidden tax that's eating the dividend.
And here's why it hits small business owners harder than the enterprise workers in this study: you're doing all of it yourself.
When a large company's AI stack generates 6.4 hours of weekly overhead, that overhead gets distributed across a team. Someone handles the context-feeding. Someone else catches the errors. There's probably a person whose whole job is managing the AI tools that manage the other AI tools.
When your company is you - or you plus two people - that 6.4 hours comes straight out of your time, your cognitive bandwidth, and your ability to do the thing you're actually trying to do.
You opened your business to serve customers, make something, or build something. Not to babysit bots.
Why Botsitting Happens
The researchers identified several specific behaviors that generate the overhead.
Context-feeding. Large language models are trained on the internet, not your business. Every time you start a task, you're re-explaining your brand, your customers, your policies, your tone. If you're doing this across four different tools, you're doing it four times. Over a week, those explanations accumulate into a part-time job.
Supervising outputs. AI outputs look finished. They're often not. The email subject line is good but the body contradicts your pricing. The blog post structure is clean but the facts are confidently wrong. The customer follow-up has the right tone but the wrong name in paragraph two - and you only caught it because you read it slowly before hitting send. That supervision takes longer than people expect, because the outputs look so polished.
Debugging errors. When something goes wrong with an AI output, diagnosing it is often harder than fixing it. You're working backward through a process you didn't fully control. This is especially exhausting when the error doesn't happen at the start of a workflow - it happens at step 7 of 8, and corrupts everything that came before it.
Tool-switching overhead. The average AI-enabled worker is running multiple tools that don't share memory or context. Every switch costs time in re-orientation and re-prompting. The tools save time individually; the switching costs time collectively. At five or six tools, the switching overhead can neutralize a significant portion of the individual gains.
The Other Word: Botshitting
The researchers named a second behavior that's worth understanding.
Botshitting is what happens when botsitting becomes overwhelming: you stop verifying the AI's output and just ship it.
Sixty-nine percent of AI users admitted to doing this. 41% have delivered work they couldn't explain if someone asked. Twenty-eight percent have blamed AI for mistakes they themselves caused.
The study puts it plainly: botshitting is "offloading your critical human thinking, judgment, and understanding" to an AI and then publishing the result without owning it.
For a solo business owner, the stakes here are different from an enterprise employee. When an enterprise worker botshits a deliverable, it's a department error. When you botshit something that goes to a customer, it's your name on it. Your reputation. Your relationship.
The customer getting an email with the wrong name, the wrong price, or the confidently wrong information doesn't think "their AI made a mistake." They think "this person made a mistake."
The Actual Measurement Problem
One more piece of context.
The QuickBooks AI Impact Report surveyed 34,000 small businesses this year. 77% of US small and mid-size businesses now regularly use AI. Forty-one percent report increased revenue. Seventy-four percent say their productivity improved.
Here's the catch: more than half of them are measuring those improvements based on a "general feeling" - not specific, tracked metrics.
This matters in a specific way.
If you feel more productive because you're doing more tasks per day, but a significant portion of those tasks are AI-generated work you haven't fully verified, and 6 of your 11 saved hours are being consumed by botsitting - your "general feeling" might be measuring throughput, not outcomes.
Faster isn't always better. More isn't always value. The question isn't whether your AI tools are busy. It's whether the work they're producing is actually moving your business forward.
Most small business owners don't have a great answer to that question right now. That's not a character flaw - it's a measurement gap.
What To Do With This
This is where the research is actually optimistic.
The businesses getting real, compounding value from AI - not just a "general feeling" of being busier - have a few things in common:
They use fewer tools. The botsitting overhead is not linear. Going from two tools to five doesn't double your overhead - it multiplies it, because each tool requires its own context, its own supervision, and creates its own cross-tool integration debt. The businesses with the lowest overhead are using AI in a narrow, targeted way: one or two tools, for clearly defined purposes, with clear "done" criteria.
They keep humans on the verification layer. The 69% who are botshitting are doing it because verification feels like extra work on top of AI work. The businesses avoiding botshitting have made verification normal - they've built it into the workflow rather than treating it as an optional add-on. They're not checking the AI's work because they don't trust AI. They're checking it because they trust their own standards.
They've audited their stack honestly. If you're paying for 5–8 AI subscriptions and could articulate what each one specifically does for your business revenue, you're in the minority. Most small business owners adopted tools in 2024 and 2025 under FOMO pressure and haven't stopped to ask: which of these is actually earning its seat?
The Question Worth Sitting With This Weekend
If you took an honest accounting of last week - how many hours did you spend feeding context, supervising outputs, debugging errors, and switching between tools that don't talk to each other?
That number is your botsitting overhead.
Now compare it to the time AI actually saved you on core, valuable work.
That math is what's actually happening in your business. Not the "general feeling." Not the headline about 11 hours saved. The actual arithmetic, in your specific situation, with your specific tools.
If the math is good, great. Keep going.
If the math doesn't feel right - if the botsitting is eating more than you realized, if the botshitting risk is higher than you're comfortable with - that's useful information. Not a failure. Information.
The problem isn't that AI exists. The problem is that the way most small business owners adopted AI was optimized for FOMO, not for outcomes.
The good news: that's solvable. Not by abandoning the tools. By getting ruthlessly clear on which ones are earning their place.
Sources: Glean Work AI Institute, Work AI Index 2026 (6,000 full-time digital workers, US/UK/Australia; co-authored by researchers from Emory University, Stanford University, UC Berkeley, UC Santa Barbara, UNC Charlotte, University College London, and Notre Dame). Coverage: LA Times, June 12, 2026. QuickBooks AI Impact Report, 2026 (Intuit, 34,000 SMBs).