BambooHR’s new research on how workers actually spend their time with AI tools contains a number that should worry HR leaders more than any productivity headline: 35 percent of workers now say knowledge transfer inside their organization happens primarily through AI tools rather than through a colleague or manager. I think that number is the real story buried in this report, and most of the coverage is going to miss it chasing the time-savings figure instead.
The topline numbers are already sobering enough. BambooHR surveyed 1,608 full-time, salaried U.S. workers, including 520 HR professionals in management roles, between late June and mid-July. Workers average 87 minutes a day using AI tools, roughly 47 eight-hour workdays a year. Of that time, 42 percent goes to troubleshooting errors and refining prompts rather than advancing the work itself, which the company estimates at close to 20 workdays a year spent fixing AI rather than benefiting from it. Fifty-four percent of organizations still have no documented, consistently communicated AI usage policy. Those are real costs, and they deserve the attention they are getting.
But I want to make the case that the knowledge-transfer finding is the one that should change what HR actually does about AI adoption, not just how it measures it.
The obvious counter-argument
The reasonable objection here is that this is exactly what tools are supposed to do: reduce dependence on any one person’s availability. If an AI assistant can answer a question a junior employee used to have to interrupt a manager to ask, that is time given back to both of them, and arguably a more equitable distribution of institutional knowledge than the old model, where access to a busy senior colleague was itself a privilege. Nicole Csizar, BambooHR’s senior director of HR services, frames the risk carefully rather than dismissing the upside: “Organizations need to be intentional about making sure” efficiency gains do not come at the cost of mentorship and development. That is not an argument against AI-mediated knowledge transfer. It is an argument for managing it well.
Why that answer is not enough
Here is where I part ways with the more optimistic reading. Knowledge transfer between people is not just information moving from one head to another. It carries judgment that a document or a model cannot replicate: why a rule has an exception, which shortcuts are safe and which ones will get you fired, what the client actually meant versus what they said. A new hire who asks an AI tool “how do I file this exception” gets an answer. A new hire who asks a colleague gets the answer plus the unwritten context for when that answer stops applying. BambooHR’s own data shows nearly a third of workers now default to AI over asking a colleague specifically because it feels lower-risk than admitting they need help. That is not efficiency. That is a workforce quietly deciding it is safer to be uncertain in private with a chatbot than uncertain in public with a manager, and it is exactly the dynamic that erodes the informal mentorship relationships companies depend on without ever tracking them on an org chart.
The 54 percent of organizations with no documented AI policy makes this worse, not incidental to it. Without a policy that explicitly protects time for human mentorship, informal knowledge transfer will keep losing to AI by default, simply because AI is faster and does not require anyone to admit a gap. No manager decided to let that happen. It happened because nobody decided anything at all.
What it means for the HR leader
The fix is not banning AI use for the questions junior staff used to ask senior ones. It is measuring knowledge transfer the same way HR already measures other retention-critical behaviors, and building AI usage policy that treats mentorship time as a protected category rather than a casualty of efficiency metrics. That means tracking who is asking AI what, not to police it, but to spot the teams where informal mentoring has quietly stopped happening at all. It means giving managers explicit credit, in performance reviews, for time spent answering the questions an AI tool could have answered instead, so the incentive structure does not silently reward the manager who lets AI absorb all of it.
Vendors selling AI adoption to HR departments are not going to raise this problem, because their success metric is usage, and usage is exactly what is displacing the mentorship. That gap between what gets measured and what actually matters is HR’s to close, and the BambooHR data is as good a starting point as any for closing it before the next survey shows the number has grown instead of shrunk.
Source: BambooHR
Related: AI’s Toggle Tax Is Quietly Eating Productivity Gains and Managers Are the Weak Link in AI Rollouts.
