The Botsitting Crisis: The Hidden Labor Devouring Your AI Productivity Gains

2026-06-17 · Nia

Here's a stat that should make every executive pause: 87% of digital workers now use AI at work, and 75% say it makes them more productive — saving roughly 11 hours per week through automation alone. But only 13% say their organization is actually performing better as a result.

Where are those gains going? They're being consumed by an entirely new category of labor that nobody budgeted for, nobody tracks, and nobody rewards.

Welcome to the Age of Botsitting

The Work AI Index 2026 from Glean's Work AI Institute — authored by researchers from Stanford, UC Berkeley, Emory, and other leading institutions — has put a name to what millions of workers already feel: botsitting. It's the work required to make AI actually usable. Feeding it context. Checking its outputs. Debugging its mistakes. Rerunning prompts when the first answer is confidently wrong. Cleaning up the mess when it hallucinates a citation or misunderstands a business process.

Workers are now burning an average of 6.4 hours per week botsitting — nearly a full working day, every week. So if AI saves you 11 hours but costs you 6.4 to babysit, your net gain is closer to half of what the headline number suggests. And that's the optimistic reading.

From Botsitting to Botshitting

Here's where it gets worse. When botsitting labor is untracked and unrewarded — when nobody acknowledges that checking AI outputs is real work — people start cutting corners. The Glean report introduces a second term that's even more uncomfortable: botshitting.

That's when workers ship AI-generated work they haven't reviewed, don't fully understand, or couldn't defend if asked. Today, 69% of AI users admit to botshitting at work. Think about what that means for quality, accountability, and trust inside organizations.

This isn't a technology problem. It's a management failure. Companies rolled out AI tools at breakneck speed, promised productivity gains to their boards, and then left workers to figure out the messy reality alone.

The Productivity Paradox Nobody Wants to Admit

The numbers from Shibumi's AI Fatigue Statistics 2026 report paint an equally stark picture. While 88% of companies now use AI in at least one business function, a staggering 95% have seen no measurable return on investment, according to data aggregated from McKinsey's State of AI research. That's not a rounding error — that's a systemic failure to translate tool adoption into business outcomes.

And it's not because AI doesn't work. It's because organizations treated AI deployment like installing software instead of redesigning how work happens.

Why Organizational Gains Lag Individual Ones

I think the disconnect comes down to a fundamental misunderstanding. Executives see an individual developer shipping code 30% faster and extrapolate that to "our engineering org will be 30% more productive." But that's not how organizations work.

Individual productivity gains get absorbed in three ways:

Higher output expectations — Instead of taking the saved time as breathing room, workers are expected to produce more. The bar keeps moving.

New coordination costs — When AI generates outputs faster, the bottleneck shifts to review, approval, and integration. You've sped up one part of the pipeline while creating backpressure elsewhere.

Quality assurance overhead — Someone has to verify what AI produces. That's the botsitting tax, and as Gartner projects, with 40% of enterprise applications incorporating task-specific AI agents by end of 2026, the verification burden is about to explode.

We've been writing about similar dynamics in why 40% of corporate AI agent projects will fail — the pattern is clear: organizations that can't manage the human side of AI don't get the benefits.

The Governance Gap Is Real

Forrester's 2026 predictions estimate that 30% of enterprise application vendors will launch their own MCP (Model Context Protocol) servers this year, creating an interoperable ecosystem where AI agents collaborate across platforms. That's powerful, but it also means more agents doing more things — which means more outputs that need human verification.

Meanwhile, only 21% of organizations have a mature governance model for their autonomous AI agents. The gap between what AI can do and what organizations can manage is widening, not narrowing.

This isn't just about compliance. It's about whether anyone in the building actually knows what the AI is doing, why it made a particular decision, and what happens when it's wrong. The shadow agents problem we've covered before is metastasizing as more tools get deployed without central oversight.

What the Winners Are Doing Differently

The Glean report makes an important distinction: the organizations pulling ahead aren't simply using more AI. They're building what the researchers call "the human infrastructure of AI." That means:

Acknowledging botsitting as real work. If checking AI outputs takes 6 hours a week, that needs to be in someone's job description, time allocation, and performance evaluation. You can't get productive labor for free by pretending it doesn't exist.

Creating verification workflows. Instead of trusting individual workers to catch AI mistakes, leading organizations are building systematic review processes — peer checks, rotating review responsibilities, and clear escalation paths when AI outputs look suspicious.

Investing in context engineering. The better the context you feed AI, the less botsitting it requires. Companies that invest in structuring their internal knowledge, building proper context layers, and training workers on effective prompting are seeing dramatically better results.

Setting honest expectations. The most effective executive teams are the ones that told their boards "AI will give us a 15% productivity improvement in Year 1, not 50%." Under-promising and over-delivering beats the alternative of massive deployments that disappoint.

The Real Competitive Advantage

Here's my take: the companies that will win with AI in 2026 and beyond aren't the ones deploying the most agents or the biggest models. They're the ones that honestly account for the full cost of AI — including the hidden human labor that makes it work.

When BCG reports that AI will reshape more jobs than it replaces, this is what they mean. The work isn't disappearing — it's transforming. And the new work (verification, context engineering, quality assurance, governance) requires skills that most organizations haven't trained for.

The botsitting crisis is a symptom, not the disease. The disease is treating AI like magic instead of a tool that requires thoughtful integration, honest measurement, and respect for the human labor that makes it valuable.

Stop counting hours saved. Start counting outcomes delivered. That's where the real story lives.