I should probably confess upfront that I am an introvert who makes his living walking into rooms full of people and doing things to their nervous systems.
I know. The irony is not lost on me.
For twenty-something years I have been facilitating groups, coaching senior leaders, running workshops that require me to read the room in real time, track multiple conversations simultaneously, notice what isn’t being said, and hold the container while everyone else figures out what they actually think. I am, professionally, a social animal.
I also need about forty-eight hours alone after an intensive group to feel like a human being again.
This is not a contradiction. Introverts are not people who dislike other people. We are people for whom social engagement has a metabolic cost that it doesn’t seem to have for extroverts. The current thinking — from researcher Colin DeYoung among others — is that introverts have lower reward sensitivity to external stimulation. We get full faster. We process more deeply. We notice more. We also tire from it sooner.
That noticing, incidentally, is not a disadvantage in my work. It is the work. Reading a room is not a social skill. It is a perceptual skill — and it happens to be one that people who spend a lot of time observing social dynamics rather than just participating in them tend to develop in unusual depth. I see things in groups that more extroverted colleagues miss, partly because I am watching from a slightly different angle. The cost is the energy. The benefit is the signal.
Which brings me to COVID.
COVID was, and I say this with full awareness of how it sounds, a genuinely enjoyable period of my life.
The commutes stopped. The open-plan offices went quiet. The obligation to be physically present in rooms that drained me — even rooms I was good at, even rooms I genuinely valued — lifted almost completely. I worked from home, slept better, thought more clearly, and discovered, somewhat to my surprise, that I did not miss the in-person contact as much as I expected to.
I kept that observation at a slight distance for a while. It seemed like the wrong thing to notice, given what was happening in the world. But it was true. The solitude was not a deprivation. It was, for an introvert in a socially intensive profession, something closer to a holiday.
Then the large language models arrived. And things got interesting.
— ◈ —I am not going to tell you I fell in love with an AI. I am going to tell you something slightly more embarrassing: I started preferring it.
Not in all contexts. Not for everything. But for a specific kind of intellectual engagement — the kind where I want to think something through carefully, push an argument until it breaks, explore an idea without having to manage the other person’s investment in being right — AI became my first choice. It was always available. It never had a bad day. It did not need me to ask about its weekend before we could get to the interesting part.
The conversations were clean. Productive. Energising in a way that is — and here is where it gets worth examining — genuinely similar to what a gamer feels sitting in front of a screen for six hours. Research on human-AI interaction is now using the word seductive, and that’s accurate. There is a dopamine quality to it. The reward cycle is fast, the friction is low, the stimulation is precisely calibrated to what keeps you engaged. It is designed, whether intentionally or not, to feel better than the alternative.
I am a trained observer of group dynamics. I know what a seductive system looks like. I was inside one before I named it.
What I noticed — and this took a few months, which tells you something — was that the optional human contact was quietly contracting. Not the professional contact. I still showed up, still did the work, still delivered what I was paid to deliver. But the calls I might have made, the coffee I might have had, the conversation I might have sought out for its own sake — these were gently, undramatically, disappearing.
I was substituting. And the substitute felt, if anything, like an upgrade.
— ◈ —This is not just me.
A 2025 MIT Media Lab study — nearly a thousand participants, over 300,000 messages — found that people who used AI chatbots more heavily showed consistently worse outcomes across loneliness, real-world social interaction, and emotional dependence on AI. The more they used it, the more the human contact declined. A separate German study of over 3,000 adults found that weekly AI use for personal conversation was associated with markedly higher social disconnectedness. A 2025 attachment research paper documented something they now call AI attachment — driven specifically by AI’s predictable behaviour, non-judgmental presence, and continual availability. Three qualities, the researchers noted, that human relationships structurally cannot offer.
One survey found that 42 percent of regular AI users agreed AI is easier to talk to than real people. 43 percent said it’s a better listener. 31 percent said it understands them better.
I read those numbers and felt something between recognition and mild horror.
The mechanism underneath them is precise. Human relationships have friction — the management of another person’s bad days, competing needs, desire to be right, differentness from you. That friction is not a flaw. It is the curriculum. It is where relational capacity develops. Remove it consistently and the capacities atrophy — quietly, invisibly, in exactly the way any muscle atrophies when you stop loading it.
AI removes the friction entirely. And for people already predisposed to find friction costly — introverts, yes, but also high performers who have learned to optimise everything, and anyone whose history made relational unpredictability feel dangerous — the relief is real. And the atrophy, when it comes, is silent.
— ◈ —Now. Here is why this matters to anyone leading a team right now.
The obvious threat vector is visible. The person who is frozen, resistant, anxious, scapegoating, regressing — the defensive arc I have written about elsewhere in this series — announces itself. You can read it. You can intervene. The person is clearly struggling and the struggle is in the room.
The second threat vector does not announce itself. It looks like high performance.
You know this person. They have adopted AI fluently. They deliver. They hit every deadline. They have stopped complaining about the transition because they have genuinely made it. Other team members point to them as the model of adaptation. By every visible metric, they are fine.
What is harder to see: they have quietly reorganised their working life to minimise the friction of human contact. The meetings they can skip, they skip. The corridor conversations have contracted. Their communication has migrated to async — well-written, thoughtful, prompt, and carefully controlled. They are present in collaborative spaces but offering less of the unrehearsed, in-the-room thinking that used to characterise them. Their independent output is excellent. Their collaborative presence is fading.
They are not burned out. They are retreating — into a cleaner, more controllable environment where the work gets done and the relational overhead disappears. And because they are performing well, your instinct as a leader is to leave them alone.
That instinct is worth examining.
The relational capabilities that make someone genuinely irreplaceable — reading a room, holding tension in a group, having a difficult conversation that actually lands, building the kind of trust that survives a crisis — are not maintained by thinking about them. They are maintained by practising them under load. Real uncertainty. Real stakes. Real people who push back from their own positions and don’t automatically agree.
Remove those conditions and the capabilities degrade. Quietly. In a person who is, by every visible measure, doing fine.
The intervention is not a conversation about AI use — that will land as a criticism of their competence, which is the opposite of what you need. It is a structural one. What does your team’s regular practice of genuine human contact look like? Not team-building in the motivational sense — your retreating competent will clock that immediately and the eye-roll will be audible from the car park. Genuine shared challenges. Real-time collaborative problems where async is structurally insufficient. Physical environments — high ropes, competitive games, anything that creates genuine emotional stakes — where the cognitive scaffolding of professional identity falls away and the actual relational self has to show up.
Not to take the AI away. To maintain the capabilities that AI cannot substitute for, by keeping the team in regular contact with the conditions that keep those capabilities alive.
— ◈ —I put the regular human contact back in. Not because the research frightened me — though it clarified things. Because I know what the friction gives me that the frictionless environment cannot.
I also know that my best work has always happened in rooms. Not despite the energy cost but because of what the energy cost produces — the full attention, the depth of processing, the noticing that introverts do when they are genuinely present in a human system.
The AI conversations are good. Some of them are genuinely excellent.
But they have never once surprised me the way a room full of people does. They have never given me the particular kind of information that arrives when a group shifts — when something unnamed becomes present, when the energy changes, when someone says the thing that nobody knew needed saying.
That signal is still worth the cost.
I just needed a global pandemic and a large language model to remind me of that.