The Foundation and the Frontier

You can’t hand someone judgement. You can only draw it out of them. That makes developing other people slower and harder than it looks, and more important now that the machine is always ready to hand over the answer instead.

Occasioning, not installing

I know the exact moment, because I feel the pull of it every time I mentor someone. They’re stuck, working towards something I can already see, and everything in me wants to just tell them. It would take ten seconds. It would spare them the struggle. And it would teach them almost nothing.

You can’t hand a person their own understanding. What makes them better is the working-out, not the answer waiting at the end of it. Give them the answer and you’ve solved today’s problem and skipped the only part that would have changed them. Developing someone is, more than anything, the discipline of withholding: sitting on the answer you could give, so that they reach their own, because reaching it yourself is what makes it hold.

The last piece put a name to why. We learn through a kind of reinforcement, attempting, getting it wrong, feeling the wrongness, and adjusting. The sting is the signal. Hand someone the finished answer and there’s no attempt, and no error to learn from. So the developer’s real skill is a strange one. It’s restraint. Knowing what to leave unsaid, and how long to let a person sit in not-knowing before you step in. Generosity, in this one setting, is often the enemy.

The machine can help, or hollow out

None of that means keeping the machines out of it. Used well, AI is an extraordinary developer of people. It’s worth being precise about what “used well” means, though, because the same tool does opposite things depending on how you point it.

Point it one way and it becomes the most patient sparring partner a learner has ever had. It will run the negotiation twenty times. It will play the difficult client who refuses to be pleased. It will pose the awkward board question again, and again, and mark the same piece of reasoning against a standard until the pattern finally lands. A human developer never has time for that many repetitions. A machine has nothing but time. That is reinforcement done properly: the learner still makes every attempt, still takes the signal, and simply gets far more practice than any person could supply. The reps scale, and the struggle stays intact.

Point it the other way and it becomes the answer-giver, and it quietly undoes what it might have built. The instant it supplies the output instead of demanding the attempt, the loop collapses. Same tool, opposite effect. The whole difference is whether the machine has been set to make the person work, or set to spare them the work.

Everything we already know

But even the machine used perfectly, as the tireless sparring partner, runs into a ceiling. And the ceiling matters more than anything else I want to say here.

An AI is built from what already exists. It’s trained on the accumulated output of a field, and what it gives back is a version of that accumulation, smoothed towards the consensus sitting in the middle of it. That makes it a superb foundation-builder. It can take a newcomer and bring them, faster than any human mentor could manage, right up to the current edge of what a field knows. What it can’t do is carry them past that edge, because past the edge there is nothing yet for it to have been trained on. It pulls, always and by design, towards the average of what has already been thought.

There’s a hard line hiding in that, and it’s worth saying plainly. A machine can teach you everything we already know, which is exactly why it can’t teach you anything we don’t.

Nobody starts at the frontier

None of which makes the foundation worthless. Quite the opposite. Nobody has an original thought about a field they haven’t yet absorbed. You can’t see past an edge you have never reached. Getting to the frontier is the long, repetitive, unglamorous work of taking the known into yourself, and that is precisely the work a machine can do tirelessly while a scarce human developer can’t spare the hours. So let it. Letting the machine march people up to the edge of the known is one of the genuinely good things it can do.

The frontier itself is another matter, and it needs people. Original thought doesn’t come from more information or cleaner drilling. It comes from lived, particular, awkward experience: being wrong in a way no textbook covered, noticing what the consensus walked straight past, feeling the friction of a real situation that refuses to fit the pattern. That kind of knowing is earned, and it can’t be transmitted. It is also fragile. A half-formed original idea looks, at first, a lot like a mistake, and the reflex of any system trained on the average is to correct it back towards the average. So it needs a human beside it: someone present enough to tell a wrong answer apart from a new idea, and to shelter that idea long enough to grow.

That is the shape worth aiming for. The machine takes the known and installs the foundation. People take the new, and grow it in one another. Each is released to do the half only it can, and the result is worth far more than either alone.

The gap we are standing in

That is where this is heading, and I believe it’s where it should head. But it is not where we are, and the distance between the two is doing real damage right now.

The tidy version of the story is a clean handover: the machine takes the foundation, and people move up to the frontier. It quietly assumes two things that aren’t yet true. It assumes the tools are mature enough to be left alone with the foundational work. And it assumes the experienced people have been freed to climb. Neither holds today. The technology is good enough to take on the junior work, and nowhere near good enough to be trusted with it unwatched. So the senior people aren’t released to the frontier at all. They’re pulled sideways, into minding the machine: prompting it, checking it, catching what it got wrong, building the scaffolding that keeps it roughly honest.

And that is quietly expensive, because it is the very attention that used to go into growing the next generation. The senior hours that once went on teaching a junior now go on supervising an agent. It lands us in an awkward, specific bind. The bottom rung has already gone, taken by the machine. At the same moment, the people who would once have reached down and hauled the next generation up past it are busy holding the machine steady. The rung vanished and the helping hands got occupied, both at once. That is the gap people are falling into.

The thing to hold onto is that this is a phase, not a destination. It is the cost of an immature fit between people and a powerful new tool, disruptive enough to break the old way we grew people, not yet mature enough to deliver the new way it promises. Transitions look like this from the inside. The destination is genuinely good. The real risk is letting a generation fall through the floor while we are still crossing to it.

A choice, not a fate

None of this is fated. It’s a set of choices, and most of them are choices about where scarce human attention goes. Right now the default is to point that attention at the machine, because the machine is new and loud and visibly needs watching. The quieter, less urgent-looking work of building people slips down the list, and slips further, until one day the people simply aren’t there.

So the real task, in this middle stretch, is to protect the growing of people on purpose, against the pull of everything that looks more pressing. To ring-fence some senior attention for the next generation even while the tools are at their noisiest and neediest. That is not a technology decision. It is a leadership one, a deliberate choice about what the scarce human capacity in an organisation is really for.

And deciding what that capacity is for, then pointing it there on purpose, turns out to be the whole of the job I want to talk about next.

Next Article ->

What a Leader Is For

You delegate something that matters, and a week later it isn’t being done the way you’d have done it. Every instinct says take it back. This piece is about why that urge is so strong, and why resisting it is most of what leadership actually is.