If your team says AI is saving them time but your business results do not show it, there is now a rigorous explanation for why that gap exists. Glean’s Work AI Institute released the inaugural Work AI Index on June 10, 2026, and the data tells a story that every business leader needs to sit with.

The report surveyed 6,000 full-time digital workers across the US, UK, and Australia, with co-authorship from researchers at Stanford, UC Berkeley, UC Santa Barbara, Emory, Notre Dame, UCL, and UNC Charlotte. The headline findings are striking in their contradiction.

The Numbers That Should Worry Every CEO

87% of digital workers use AI at work. Those workers report saving 11 hours every single week because of it. That is over a quarter of the entire workweek.

But only 13% of employees say their organization is performing significantly better as a result.

That is not a small gap. That is AI delivering individual-level productivity gains that almost entirely disappear before they reach the business bottom line.

What “Botsitting” Actually Means

The Work AI Index introduces a term that captures exactly what is happening: botsitting.

Botsitting is the unrecognized labor of making AI work. It includes feeding AI the context it needs to give a useful answer, checking and correcting AI outputs before they can be used, debugging mistakes, reformatting results, and switching between disconnected tools to piece together a coherent output.

Workers spend an average of 6.4 hours every week doing this. That is not time spent using AI to produce work. That is time spent maintaining the AI.

When you break down how workers actually spend their AI time, the picture is sobering: 37% of AI interaction time goes to botsitting, 36% goes to actually using the tools to produce work, and 27% goes to learning tools and building agents. Workers are spending more time keeping AI functional than they are actually producing with it.

The Companion Problem: Botshitting

The report names a second phenomenon that compounds the botsitting problem. “Botshitting” is the act of shipping AI-generated work that employees have not verified, do not fully understand, or cannot confidently stand behind.

69% of respondents admitted to doing this.

This is not a moral failing. It is a structural outcome. When workers are already spending 6.4 hours a week on botsitting, the pressure to skip verification and ship faster is enormous. The problem is that botshitting quietly erodes output quality in ways that do not show up immediately, and when they do show up, the trail back to the source is often invisible.

Why This Matters for Business Leaders

The botsitting phenomenon explains something that has puzzled executives since AI tools started spreading across teams: the ROI that looks obvious on a time-savings calculator refuses to materialise in financial results.

The reason is that 11 hours of perceived time savings is not 11 hours of recovered productivity. A portion of it has already been consumed by the unglamorous work of making AI usable in the first place. Another portion disappears into fixing the output of unverified AI work. What remains is a real but much smaller gain that rarely shows up at the organisational level.

The implication for business leaders is that the AI tools your team is using are not the same as AI working for your business. Individual AI adoption and organisational AI value are two different things, and the gap between them is where botsitting lives.

What This Means for Business

Closing the gap between individual AI time savings and organisational performance gains requires addressing the botsitting problem directly. That means three things.

First, context infrastructure. Botsitting is largely a context problem. Workers spend hours feeding AI what it needs to be useful because most enterprise AI tools do not have access to company-specific knowledge, processes, and data. AI that already knows your business does not need to be fed context before every task.

Second, integration depth. A significant portion of botsitting comes from switching between disconnected tools and manually stitching outputs together. AI agents that work across systems eliminate this layer of overhead entirely.

Third, quality guardrails. Botshitting happens because there is no structural check on AI output before it leaves the building. Building review steps into AI workflows, rather than relying on individual discretion, reduces the quality erosion that happens at scale.

For organisations trying to close the gap between AI activity and AI results, this is the most honest diagnosis yet of what is actually getting in the way. The tools are not the problem. The missing infrastructure around those tools is.

If you are looking at your team’s AI usage and wondering why the business numbers have not moved, the Work AI Index is where to start.