Glean just published their Work AI Index. 6,000 workers surveyed across the US, UK and Australia, and two numbers jumped out.
87% of digital workers use AI at work, but only 13% say it has significantly improved their organisation's performance. Nearly nine out of ten people are using the tools. Yet fewer than one in seven think it is actually making the business better.
The report introduces two terms that I think are going to stick.
The hidden labour of botsitting
"Botsitting" is the hidden labour of making AI usable: feeding it context, supervising outputs, debugging errors, cleaning up what it produces, switching between tools.
Workers spend 6.4 hours a week doing this. That is 37% of their AI time, more than they spend using AI to actually produce work. More evidence that having an AI training programme is equally as important as an AI strategy.
The bit that should worry leaders
"Botshitting" is the bit that should worry leaders more. 69% of AI users admit to shipping work they have not verified, do not fully understand, or could not confidently stand behind. 41% say they sometimes deliver work they could not explain if asked.
There are other findings worth sitting with. 61% say AI helps them more than their manager. 55% have sent a digital twin to a meeting on their behalf. 77% juggle multiple AI tools every week. 60% rerun the same prompt across different tools because the first one was not good enough.
Among the highest-performing AI users, 54% are using unapproved tools, 38% are downplaying how much AI helps them, and 36% are actively hiding it from their managers.
Adoption is not transformation
The companies actually seeing results are doing something different. They are not just buying tools and tracking adoption. They are redesigning how work gets done, training people properly and rewarding AI skills formally. They treat workarounds as signals that the official tools are not good enough, rather than compliance problems to police.
Adoption is not transformation, usage is not value, and the gap between the two is filled with a lot of botsitting and not nearly enough honest conversation about what is actually working.