Last week John and I argued about what happens to expertise when AI can replicate it. We both made our cases. We were both right. We were also arguing about the wrong thing.
The real question isn't what AI can do. It's what that reveals about us.
Most people frame this as a replacement story. Machines get more capable, humans get less necessary. It's a tidy story… and maybe also wrong. What's actually happening is a reallocation. AI is absorbing the hygiene work. Repetitive tasks, first drafts, pattern synthesizing at scale. Not meaningless work. But it's work that was always compressible, even when we dressed it up as something more important than it was.
When that compresses, something else has to expand. And here's where most organisations trip. They treat the efficiency gain as the point. Use the freed capacity to do more of the same, faster. The numbers look good and quarterly slides look great. And is anyone asking what we lost?
But the race may never have been what we thought it was.
When intelligence was scarce, value came from knowing more, doing faster, solving efficiently. When intelligence is abundant, those become table stakes. The centre of gravity moves, not toward more thinking, but toward better judgment. What is worth doing? What should we stop? What does good actually mean here? AI can generate infinite options. It can't decide what matters to humans. That responsibility will return to humans, whether we want it back or not.
And when output is cheap, ideally discernment gets expensive. Taste, real taste, not aesthetic taste but the pulling together of experience in the form of values, context, and sensible restraint, actually becomes the work. The editor who reads a perfectly structured AI draft and says this is technically correct but doesn't feel like us. Also the designer who looks at an optimised layout and feels something is off (and they are right). That instinct is the actual deliverable.
In a world where anything can be generated, the question isn't "can this be made?" anymore. It becomes "who stands behind this?". A consultant who follows through when it's inconvenient for everyone. A leader who says the difficult thing when the room wants reassurance. Trust compounds through friction, pressure and resolution. It's not an output but it's a pattern of being human. You have it or you're still building it, and it's ok if everyone can tell which it is.
As cognition speeds up, the real bottlenecks move to the body via the nervous system. Who stays grounded when everything feels uncertain? Who can hold a room when tension intensifies? Picture a leadership team in crisis. The smartest person is spiralling through scenarios. But maybe the most valuable person slows their breathing, names what's actually happening, keeps the group from fragmenting. That person isn't smarter they are just more stable. In thinking about high-change environments, the person who doesn't destabilise the system under stress is likely worth more than the person with the best ideas. We just haven't learned to pay for that yet.
Then there is meaning. AI can generate infinite stories. But meaning isn't created by volume of information. It's created by skin in the game, accumulated loss, shared history, and the people who were there. A handwritten note will always land harder than a perfect email and a concert means more than a recording. The restaurant your grandmother took you to matters in a way a Michelin-starred meal probably doesn't. Meaning requires friction. AI removes friction. Humans have to figure out what means something to them.
Other institutions might call what remains a competitive advantage. We call it stewardship. Attention to people and protection of long-term outcomes. The refusal to extract at the expense of the system. It's visible because it isn't optimised to death. The manager who remembers what you said three months ago and the company that absorbs a loss rather than pass it to a customer. Or the team that ships slower because they won't cut corners. That visibility builds trust. Trust builds loyalty. And loyalty builds something that doesn't show up on a quarterly report until suddenly it does, and by then everyone wants to know "the secret" (I think we all knew the secret before we designed the system).
So orgs face a real choice right now. Use the efficiency AI creates to keep extracting more, or use it to become even more human. The first path is easier, more legible, and produces numbers that look good on slides. It creates oxytocin (mostly the harmful kind). The second requires the thoughtful designing of trust, not just straight output. Developing people, not just deploying all the tools.
Most organisations will choose the first path. Eventually the second will feel inevitable.
AI doesn't reduce the importance of humans. It strips away the illusion that intelligence was the most important thing about us. What's left is harder to define, scale, and fake. Harder to generate because it lives in the part of us that is as old as time, and we just have to remember again.