As a researcher-developer and a former educator, I spend a lot of time thinking about what AI can and can't do– as well as what it ought not to do. Lately, the questions that hold my attention and preoccupy my mind are not about its capability. They are about us — about how this technology and field of study are shaping the way we think about work, our relationship to it, and the boundaries we take for granted.
A recent Harvard Business Review experiment made the stakes tangible and quantifiable. Researchers gave managers the same flawed documents to review, with one differentiator: whether the draft was attributed to "an AI tool" or to "an AI employee" on the team. That single relabeling undermined the managers' sense of personal accountability by about nine percentage points — they quietly handed the responsibility and judgment over to the machine. They escalated 44% more work up the chain, rather than standing behind their own human review. They caught 18% fewer of the errors that they were specifically tasked to catch. As a result, people reported more uncertainty about their own professional identity and less trust in how AI was being used around them.
No one in that experiment decided to be less careful. No one chose to feel less responsible or less certain of their role. A simple rewording produced the dramatic effect — quietly, below the level of deliberate thought. These findings alarm me: AI is reshaping our sense of responsibility and even our identities, while bypassing our awareness.
We might adopt AI under social pressure — because we feel that we cannot afford to look like we're falling behind — and we do so, before we actually understand the thing that we're adopting. The study itself measures what drives adoption, not understanding; the gap between our adoption rate and our actual learning curve is my own concerns over their data. Nonetheless, I see, hear, and notice the gap everywhere. It is a gap that is increasingly hard to ignore. We are scaling something faster than we are comprehending it, and letting it steer the engine while we're still building the controls.
Real AI literacy means understanding that these systems can't be held accountable, so a human must always be. It means knowing what limits a model has and what problems could accumulate as a result, when to trust a generated output and when to challenge it, and how to stay intellectually engaged with content that something else drafted first. It is as much about responsibility, as it is about capability.
And the more I surface these questions, the more I realize that what is most at stake isn't productivity. It's trust. That's the quiet cost that remains mostly invisible — and it lands on the desks of the people in HR and learning and development, who are carrying the heaviest and least-acknowledged reality of this steep learning curve. The cost isn't only individual. It's the slow, collective erosion of the judgment, accountability, and care that good work truly depends on — compromised across thousands of small moments where someone quietly assumed the machine had it handled.
Consider the job search. AI now sits on both sides of the table: candidates draft their applications using AI, and employers increasingly screen — sometimes even interview — with AI. Each side starts to suspect that the other side isn't fully human. When the first handshake of a working relationship is mediated by systems neither party fully understands, what is the relationship built on? An HR team can process more applications than ever, faster than ever — but it may be purchasing that speed at the cost of the very trust the relationship will later depend on. The most pressing endeavor should not be about processing more applications faster; it should be about keeping the human moments visible and honest, and being candid about when a human ought to step in.
Consider an onboarding scenario: a new hire joins a team that may list AI among its members, alongside human colleagues who have been handing varying degrees of their own judgment to AI. If the seasoned employees aren't sure what's still theirs to own — research says many aren't — a new hire has no map at all. Who do you learn the job from, when part of "the team" is software? Onboarding becomes the space to make the implicit explicit: here is what we own, here is what we delegate, and here is what we always check ourselves as humans. If all of the moving pieces are communicated clearly, you signal real trust into someone's first week at the workplace. If each piece does not land clearly in its place, no one is sure who is accountable.
Promoting AI products and packages to people and organizations is easy. Teaching them when to trust it, when to push back, and when to remain skeptical is difficult– often thankless– yet more valuable work. This highlights another layer of trust: employees need to believe that being trained to work with AI is an investment in their development, not a primer to replace them. People can tell the difference between genuine literacy and a pressure to adopt — the distinction has the potential to determine whether AI training builds or erodes trust.
If there is a hopeful outlook ahead, it might be one where we begin to understand trust as something we design and build towards, rather than assume. I don't think that any single vantage point — such as Silicon Valley — could design such a process well, alone. What is perceived as responsibility and accountability, what is deemed fair, and what constitutes good work and good management vary significantly across cultures, value systems, and aspirations. The teams who are committed to building processes including AI to benefit human flourishing— that people could choose to trust— ought to be: ethically grounded, multicultural, multigenerational, interdisciplinary, and cross-functional. They are more likely to be the ones who not only see the blind spots of more homogenous groups, but also continue to preserve, steward, and iterate with more wisdom and creativity.
This is not an argument against AI. It is an appeal to close the distance between how expediently we adopt it and how well we truly understand its impact. It is a wake up call to recognize that workplace trust— from the first day to the hundredth day— is what that distance costs us, when we leave it unchecked. How we divide this work between people and machines, and what we choose to protect through this cesarean task, is still a choice.