The Oldest Pitch in Computing

Karpathy is out here talking about ACI — Agent-Computer Interface — as the next layer of the stack, and buried inside it is the same idea that W. Ross Ashby wrote down in 1956 and Engelbart built a lab around in 1962.

Intelligence amplification. IA, not AI. The machine doesn't think for you — it makes your thinking go further. You stay in the loop. The tool sharpens the mind that holds it.

The reason this framing keeps coming back is that the pitch is basically perfect. "We make your smarts smarter" is the kind of sentence that slips past every objection — past the fear that the machine replaces you, past the sci-fi dread, past the union hall — and lands somewhere that feels true. You are still the person doing the thing. The thing just costs you less.

Engelbart knew this. His whole deal was augmenting human intellect, not simulating it. He built the mouse. He built hypertext. He ran a research program for decades that nobody funded properly because it turned out "make humans better at thinking" is a harder sell to the Pentagon than "build a machine that thinks instead." And yet here we are.

The ACI framing is doing the same move. Positioning the LLM not as the agent but as the interface — the new shell, the new OS, the layer through which a human reaches into a computer. Karpathy draws it as an OS diagram and the thing that jumps out is how much it looks like Engelbart's oN-Line System, just with a transformer where the text editor used to be.

None of this is a criticism. It's just funny, in the particular way that the history of computing is funny — which is to say, it is the same three ideas cycling through different substrates forever, each time discovered by someone who has never heard of the last person who discovered it.

The idea is unkillable because the idea is correct. Amplify the human. Everything else is a distraction.