If working intelligence is becoming abundant, the obvious question is the alarming one: what is left for people to do?
It is asked in a worried register, and the worry is understandable, but the question is shaped wrongly. It assumes that the value of a person was their intelligence, and that as intelligence gets cheap, the person gets cheap with it. That is not what happens. Abundance does not destroy value. It relocates it.
When something that used to be scarce becomes plentiful, value moves to whatever is still scarce and still needed. The interesting work is to be precise about where it has moved.
What does not automate
Strip the cheap things away and look at what is left holding weight.
There is judgement: not the production of an answer, but the prior act of knowing which question is the one that matters, and which of ten sound answers is the one to act on given everything this company is and intends. A model will give you ten good options. It cannot tell you which one you should want.
There is taste: the developed sense of what good looks like in a particular domain; the ability to feel that something is not right before you can articulate why. Taste is compressed experience. It is what is left of a thousand prior judgements, and it does its most important work precisely when it cannot yet be put into words.
There is integrity: holding a standard, or a promise, or a line, when it is expensive to hold and easy to drop, and when a fluent, confident, plausible case for dropping it is available on demand. Integrity is not knowing the right thing. It is paying for it.
There is sensing: noticing what no one thought to ask. A model answers the question put to it, superbly. It does not feel the unease that a question is the wrong one, or register the weak signal at the edge that does not fit the frame. Someone has to be looking at what the loop is not looking at.
And underneath all of it there is responsibility: the fact that someone, a person, has to mean it. Has to stand behind the decision, carry the consequence, be answerable. A model has no stake. It bears nothing. It cannot be the one who is accountable, because accountability is not a computation. It is a person putting something of their own on the line.
These do not stay human because models are too weak to take them. They stay human because they were never information problems. They are what it means to have a stake.
The company that concentrates its people
This reframes what an AI-native company actually is.
It is not the company with the fewest people. That is the lazy reading, and a company run on it will hollow out exactly the capabilities that were keeping it honest and alive. The replacement story treats people as cost. A company that believes it will, with great efficiency, automate away its own judgement, its own taste, and its own capacity to notice that something is wrong.
The AI-native company is the one that concentrates its people. It is ruthless about not spending them on what is now abundant: the gathering, the drafting, the first-pass analysis, the routine production of competent output. And it is equally ruthless about spending them, deliberately and well, on what does not automate. Its people are positioned where judgement is exercised, where taste is applied, where integrity is tested, where sensing happens, and where responsibility is held. That is the whole design intent: get the human contribution out of the abundant work and into the scarce work.
A company that does this does not feel emptier. It feels sharper. The same people, no longer spending most of their week on what a model now does for nothing, are spending it on the part of the work that was always the reason they were valuable.
Designing for the scarce thing
If judgement, taste, integrity, sensing, and responsibility are now the scarce and decisive inputs, then a serious company does not leave them to chance.
For most organisations, these qualities are hoped for rather than designed. They are assumed to arrive with good hiring and good culture, and then left alone, unexamined, unsupported, structurally invisible. That was affordable when they were a small part of the value and routine competence was most of it. It stops being affordable when they are most of the value.
Designed-for means built into the structure. Sensing becomes a real capability, with people, attention and standing, rather than a lucky by-product. Integrity gets protected by how decisions are made, so that holding the line is supported rather than left to private courage. Judgement gets developed on purpose, because taste is compounded experience, and a company that stops generating real experience stops compounding it. These become things the company architects, maintains and is willing to spend on. First-class concerns, not residue.
This is the foundation the rest of the rebuild stands on. A company that becomes legible to its model, and rebuilds its decisions into loops, and runs on a substrate intelligence can act on, has done essential work, and has done it so that its people can be moved to where they are irreplaceable. The point of the rebuild was never to remove the humans. It was to find out, precisely and honestly, what only they can do, and then to build the whole company around making sure they are the ones doing it.