Babysitting Your AI Isn't Wasted Time. It's the Job.

Last month a research team at Glean's Work AI Institute did something most of the breathless AI coverage never bothers to do. They asked 6,000 people who actually use AI at work what the experience is really like.

Not the demo. Not the keynote. The Tuesday-afternoon reality of it.

And buried in the data is a word for something you are almost certainly already doing, whether you have a name for it or not: botsitting.

It might be the most important word in the whole report. And it explains why so many smart, experienced professionals feel behind right now when they have no reason to.

What Botsitting Actually Is

Botsitting is the time you spend babysitting the machine.

It's feeding the AI the context it needs before it can do anything useful. It's checking the output to see whether it's right. It's catching the part that's confidently wrong and fixing it. It's re-running the prompt three times because the first two answers were garbage.

The researchers measured it. The average worker now spends 6.4 hours a week botsitting. That's roughly 37% of all the time they spend interacting with AI. More than a third of every "AI-powered" hour is spent managing the AI, not being magically served by it.

Most people read that number and groan. Six hours a week of cleanup? That sounds like a tool that doesn't work.

I read it and thought the opposite. That's the whole game. That's where the value is. And the people who understand that are about to separate themselves from everyone who doesn't.

The Number Nobody Wants to Talk About

Here's the statistic from the report that should stop you cold.

87% of digital workers now use AI at work. 75% of them say it makes them more productive. Those are the numbers everyone quotes. They're the numbers in the press release and the LinkedIn victory laps.

But only 13% of their companies actually perform significantly better as a result.

Read that again. Three out of four people feel more productive. Roughly one in eight organizations can point to a real improvement. The individual gains are everywhere. The organizational gains have nearly vanished.

So where does all that productivity go?

It gets eaten. The report is brutally specific about where. It gets eaten by botsitting — the hidden, unbudgeted, untracked hours of making the machine usable. And it gets eaten by something worse, something the researchers gave a name so perfect I laughed out loud when I read it.

When botsitting turns into botshitting.

Botshitting: The Failure Mode

Botsitting is the work. Botshitting is what happens when you skip the work.

The researchers define it as shipping AI-generated work that you haven't verified, don't fully understand, and couldn't confidently stand behind. Slop with your name on it.

And the numbers are staggering. 69% of AI users admit to botshitting behaviors. 41% say they have delivered work they could not explain if someone asked them to. 28% have blamed the AI for a mistake that was actually their own.

Think about that last one for a second. More than a quarter of people have looked a colleague or a boss in the eye and said "the AI got it wrong" when the truth was they didn't check. The machine became the scapegoat for human carelessness.

This is the actual risk of AI at work. It was never that the robot takes your job. It's that the robot helps you do your job badly, faster, at scale, and gives you someone else to blame.

And here is the part that matters for you: the gap between botsitting and botshitting is not a technology gap. No better model closes it. No new feature fixes it. The gap is judgment. And judgment is the one thing you have that the 28-year-old prompt jockey does not.

The Line Between the Two Is Domain Expertise

Picture two people handed the exact same AI output. A market analysis, a contract summary, a technical spec — pick your field.

The first person is three years into their career. They read the output, it sounds confident and well-organized, and they ship it. They have no internal library to check it against. They can't feel the thing that's off because they've never seen the real thing enough times to know what real looks like. So they prompt, and they pray, and they hand it in.

The second person has done this work for twenty years. They read the same output and within three seconds something itches. A number that's plausible but wrong. An assumption that doesn't survive contact with how the industry actually operates. A confident paragraph about a regulation that was repealed in 2019. They can't always articulate it instantly — but they know. And they fix it before it ever leaves their desk.

That second person is botsitting. And that supervision — that fast, almost unconscious pattern-match against decades of real experience — is the single most valuable thing in the entire AI workflow. It's the part the machine cannot do. It's the part that turns raw AI output into something a business can actually trust.

We've spent two years being told that experience is a liability in the age of AI. That the young, AI-native worker has the edge. The Work AI Index quietly demolishes that story. The scarce resource isn't the ability to operate the tool. Everyone can operate the tool — 87% of workers already do. The scarce resource is the judgment to know when the tool is lying to you.

That's not a soft skill. That's the skill. And AI just made it more valuable, not less.

Why This Is Good News, Not Bad

I work with professionals in their late 40s, 50s, and 60s every week. Many of them carry a quiet fear that they got to the party too late. That the AI wave is something happening to younger people, and their twenty or thirty years of hard-won expertise is depreciating by the day.

The data says the exact opposite.

The value of AI in any serious context is capped by the quality of the human supervising it. A powerful model in the hands of someone who can't evaluate its output is a liability — it just produces wrong answers faster. The same model in the hands of someone with deep domain judgment is a genuine force multiplier, because every output gets filtered through a brain that knows what good looks like.

You are not late. You are holding the thing that's suddenly scarce.

The report even shows what this looks like at the organizational level. Companies that build what the researchers call "context-rich" environments — where the humans are equipped to supervise the AI well — see workers who are 64% less likely to feel worn out, 52% less likely to ship work they can't explain, and dramatically less likely to botshit. The differentiator was never the AI. It was the human infrastructure around it.

That's a category of value you can sell. It's a category of work you can build a career on. It just didn't have a name until now.

Two Paths Out of This

So what do you actually do with this? It depends on where you sit.

If you're staying W-2 — make your botsitting visible.

Right now your supervision of the AI is invisible labor. You catch the errors, you fix the output, you keep the slop from shipping — and none of it shows up anywhere. That's a problem, because invisible work doesn't get rewarded and is the first thing assumed to be automatable.

Change that. Become the person on your team who is openly, vocally accountable for AI quality. Be the one who says "I reviewed this, here's what I changed, here's why." Build the review process your team doesn't have yet. In a world where 69% of people are botshitting, being the reliable human who refuses to is a genuine differentiator. It's the kind of thing that makes you the person leadership can't afford to lose — which is exactly the position you want when budgets get cut.

If you're going independent — sell the judgment, not the tool.

The companies racing to build the AI economy don't need another vendor who can run a prompt. They are drowning in AI output they can't trust. What they're short on is experienced people who can supervise the machine and be accountable for what comes out. That's a fractional offer. That's advisory work. That's a senior person bringing twenty years of judgment to a team that has plenty of AI horsepower and not enough wisdom.

You don't have to quit anything to start. You don't have to hang a shingle tomorrow or become a salesperson overnight. What you have to do is stop letting one company own all of your expertise — and start thinking about how to make that judgment available to the market on your terms. Some people do that through a new role. Some build an independent practice. The smart ones do both at once, because the window where this skill is rare and richly paid won't stay open forever.

Where the Roles Actually Are

This isn't theory. The companies building the AI buildout — the chip makers, the cloud infrastructure players, the model labs, the robotics and energy companies underneath all of it — are hiring exactly this kind of person right now. People who can bring domain judgment to AI-heavy teams.

I track hundreds of these companies in my AI Compute Funding Index — who's funded, who's hiring, and which roles map to the experience you already have. It's the map I wish someone had handed my clients two years ago, because the opportunity is concentrated in places most job seekers never think to look.

The point is this: the demand for botsitters — real ones, with real judgment — is not hypothetical. It's funded, it's growing, and it rewards exactly the kind of experience the market spent two years telling you was obsolete.

The Job Just Got a New Name

So if you've felt behind because you're "only" babysitting the AI instead of being replaced by it — stop.

You're not behind. You're doing the job. The job just got a new name.

Botsitting is not the thing slowing down the AI revolution. Botsitting is the part of the AI revolution that requires a human being with experience, judgment, and the willingness to be accountable. It's the part that can't be automated, because the moment you automate the supervision, you're back to botshitting.

The people who win the next decade won't be the ones who prompt the fastest. They'll be the ones who can look at what the machine produced and know, in three seconds, whether it's true.

You've been training for that your whole career.

Where to Aim Next

That's exactly why I built the AI Compute Funding Index — the exact companies in this piece, what they raised, and who's hiring right now — paired with the AI Compute Guide that maps the nine domains where your experience fits. See who's funded and hiring across the trillion-dollar buildout, and where a background like yours fits:

Get the AI Compute Funding Index + the AI Compute Guide →

Ready to talk through where you fit in this shift?

Schedule a Strategy Call →