Working with these new systems is a strange experience, and I want to tell you why.
Imagine, for a moment, that you have a staff. Not a staff of ordinary people. A staff of specialists — brilliant ones, the kind of people who know more about their corner of the world than you will ever know about anything. One of them is a distributed-systems engineer. One of them is a lawyer. One of them has read every document you have ever produced and can quote it back to you. They are, individually, extraordinary.
They are also, individually, a little odd.
They are astute, but they are narrow. They are precise, but they are literal. They miss the things any human colleague would catch in a heartbeat — the joke, the subtext, the obvious thing nobody needed to say out loud. They have, in other words, no common sense. Not because they are stupid. They are the opposite of stupid. They have no common sense because common sense is the thing you build out of context, and they do not have your context. They have never had it. They cannot have it, unless you give it to them.
And here is the part that takes some getting used to. These specialists do not talk to each other.
The lawyer does not know what the engineer is doing. The engineer does not know what the lawyer is doing. There is no hallway. There is no Monday meeting. There is no quiet conversation by the kettle in which one of them mentions, in passing, the thing the other one needed to know. That whole layer of an organisation — the connective tissue, the ambient awareness, the part that makes a company a company rather than a list of departments — that layer is missing.
Which means somebody has to be it. And that somebody, it turns out, is you.
This is a peculiar thing to discover about yourself, especially if you have never managed anyone in your life. You sat down to do some work, and somewhere along the way you became a coordinating layer. You became middle management. You became the person who walks between the rooms, carrying context from one savant to another, because nobody else can.
There is a temptation, when this dawns on you, to despair. The staff are so close to being useful. And yet they keep doing the wrong thing in ways that feel almost wilful. You find yourself thinking: this technology is broken. It is not broken. You are managing it wrong.
What you learn, slowly, is that the entire game turns on a single skill. Specificity. But not specificity in the way people usually mean it — not long instructions, not exhaustive caveats, not the kind of micromanagement that produces worse work, not better. Something else. The right kind of specificity. The naming of the conceptual objects clearly enough that the savant knows which board they are playing on. Loose enough to leave room for their skill. Tight enough that they cannot wander.
The mediocre manager over-specifies. The bad manager under-specifies. The good manager finds the level at which the objects are clear and the execution is free. This is, when you think about it, what good managers of human staff have always done. The technology is new. The job is very old.
And here is the part I find genuinely hopeful.
The work is tractable. Almost always. If you bring critical reasoning, and a clear enough grasp of the conceptual objects in play, and the patience to iterate — the problems either yield, or they terminate for reasons you can name. You do not, as a rule, get stuck in some mysterious fog. You get stuck because you have not yet specified the right thing to the right savant in the right way. And once you have, the work moves.
This matters, because the dominant story about artificial intelligence is a story about power. The models are getting smarter, we are told. They will replace us, or they will not replace us, and the argument turns on how smart is smart enough. But power, it turns out, is not the bottleneck. The bottleneck is the thing a good colleague supplies for free, and these colleagues do not supply it, and somebody has to.
That somebody is going to be a person. For a while yet, it is going to be a person.
The skill of the coming decade is not technical. It is managerial. It is the old, deeply human skill of knowing your people — what they are brilliant at, what they cannot do, what they need you to supply — and arranging the work accordingly.
The staff are waiting in their rooms. Someone has to be the one who walks between them.