Liam, a composite of clients I have worked with, is a quantitative superstar at a financial services firm where he has worked for five years. At least once a quarter, he brings forward an idea or a sharply framed argument that catches senior management’s attention and often ends up shaping strategy.
His ideas used to carry brushstrokes of insight from colleagues across his global organization and across disciplines. Liam talked with people constantly, and it showed.
Then his firm, like many, encouraged employees to push the limits of their AI use. Liam went full throttle. Priding himself on rigor, he instructed his AI tools and agents to be hard on him: no sycophancy, no easy consensus, respond the way top engineers, quants, and traders would. His ideas got sharper, and a few more of them.
Yet the pushback got worse. Colleagues were more cautious, less enthusiastic, sometimes indifferent. Frustrated, he raised it with his manager, who referred him to me.
I had plenty of advice ready. Then, in a flash of metacognitive brilliance, I realized my advice was exactly what his AI would have given him: fast, confident, and blind to what was actually happening–and wrong. So instead of prescribing, we built theories together.
AI makes answers cheap. Co-authorship becomes scarce.
Liam cited efficiency as a reason he talked with colleagues less. Why endure the messiness of human inquiry when a tool can stress-test your analysis in seconds? The tools are convincing. Randazzo et al. (2025) describe GenAI as a “power persuader.” In their study of BCG consultants, the more professionals fact-checked or pushed back on an AI’s analysis, the more intensely the AI tried to persuade them, a pattern the authors call “persuasion bombing.” Liam had built himself a tireless sparring partner. Yet after a few intellectual rounds in the ring, it kept telling him he was right.
What he didn’t account for was the downstream friction.
Nobel Laureate and AI pioneer Herbert Simon warned that “a wealth of information creates a poverty of attention.” Liam used to earn his colleagues’ attention by asking for their input. Now he arrived with freight cars of polished analysis from his silicon-based colleagues, and the path to agreement grew longer, not shorter.
Before, colleagues recognized their own fingerprints on his ideas. They felt some sense of co-authorship. Now they were asked to evaluate work they had no part in creating. Pushback is often what people do when they weren’t part of building something. Liam was getting resistance where he used to get contribution.
The hidden cost: AI triggers anti-mattering
Mattering has been studied for more than 40 years, yet it is far less known in organizations than belonging. Isaac Prilleltensky, a pioneer in the field, describes it as feeling valued and adding value. One of its oldest signals is simple: being depended on.
Liam had stopped depending on his colleagues. Without intending to, he told them their thinking was optional. Psychologist Gordon Flett calls what follows anti-mattering: the sense that you are insignificant to others. It shows up as withdrawal, thinner collaboration, and less shared learning.
The signal travels in both directions. In a Microsoft Research and Carnegie Mellon survey, knowledge workers who trusted AI more to do a task applied less critical thinking to it, while those more confident in their own abilities applied more. People who start to believe their thinking is less valuable may simply stop offering it.
AI doesn’t cause anti-mattering on its own. People choose how it is used. That is why the remedy is human.
Care is the lever
Many leaders treat care as a nice-to-have: an emotional perk rather than a business essential. That misreads and flattens what care is and how it works.
In The Thin Book of Trust, Charles Feltman describes care as the belief that someone holds your interests in mind and wants good for you. People may trust your competence and reliability, but if they believe you look out only for yourself, they limit their trust to specific transactions. Care is judged by the receiver, not declared by the giver. Liam never intended indifference. But when he stopped asking and started presenting, colleagues could reasonably conclude he valued the machines’ input over theirs.
Georg von Krogh, drawing on the philosopher Milton Mayeroff, shows why this matters for performance. He calls knowledge creation a “fragile process fraught with uncertainty and conflict of interest.” In high-care organizations, people give their knowledge freely. In low-care organizations, they hoard it, and exchange becomes quid pro quo. Tacit knowledge, the kind LLMs cannot capture, moves through caring relationships or not at all.
What care produces: sense-making that gets stronger under stress
Karl Weick, who studied how groups make sense of crises, argued that resilience depends on respectful interaction: trusting others’ observations while respecting your own enough to voice them. Unexamined AI use threatens that balance. When the machine’s answer is fluent and confident and a colleague presents it as their own, only those who believe their perspective matters will offer a different one.
LLMs will keep getting better at sounding intuitive and judicious. But intuition grounded in experience, perception rooted in being present, and judgment that someone is accountable for are not outputs. They demonstrate skin-in-the-game. They are human capacities. They atrophy or grow depending on whether organizations create the conditions for people to use them.
What your organization can do
Ikujiro Nonaka and his colleagues call those conditions ba: shared spaces, physical, virtual, or mental, where knowledge is created.
| Type of ba | What happens there | AI’s role |
| Originating | Face-to-face sharing of feelings and experience. Where care, trust, and commitment first emerge | None. This is the human core |
| Dialoguing | Colleagues turn tacit knowledge into shared concepts through dialogue | Can supply material; the dialogue must be human |
| Systemizing | Explicit knowledge is combined and organized | Strong. This is AI’s home ground |
| Exercising | People learn by applying knowledge in practice | Can coach; people must do the reps |
Source: Adapted from Nonaka & Konno (1998), California Management Review, 40(3), and Nonaka, Toyama & Konno (2000), Long Range Planning, 33(1). “AI’s role” column added by the author.
AI dominates one kind of ba and is absent from the one where care begins. Three practices follow.
Bring people in before the machine polishes. Share rough ideas with colleagues first, then use AI to stress-test what you built together. Leaders presenting AI analysis should frame it as input, invite challenge before stating their view, and speak last.
Protect ‘originating ba’. As AI removes routine interactions, deliberately keep the ones that build relationships and judgment: apprenticeship and mentorship, working through problems together, debriefs after hard moments.
Make contribution visible. Name who shaped the work, especially where human judgment improved on the machine’s. It tells people they are depended on.
The scarcest resource
Liam’s problem was never the quality of his ideas. It was that he had stopped actively building them with others.
So that is where Liam and I started. He now brings rough ideas to a few colleagues, in person, before any AI sees them. They question, sketch, and shape the idea together, and only then does he take it to the machine to stress-test. It’s early days, but the work is being made together again.
The scarcest resource in the AI enterprise isn’t compute. It is people who believe their perspective is worth offering, and the relationships that make offering it feel safe and worthwhile. Leaders can’t buy that. They can only create it through care.
Belonging gets you in. Mattering keeps you relevant. As machines do more of the work, the same is true for organizations: the ones that stay relevant will be those whose people still believe their thinking counts, feel valued, and consistently add value.
