Good post. Hayek's knowledge problem applied to AI. The knowledge that matters is tacit, dispersed, and local. No central planner could aggregate it. No central model can either.
We test this daily at Humanity Labs. Our AI workforces run inside financial service firms. Same capability going in, different worker coming out. Shaped by each firm's processes, exceptions, and judgment calls.
The firms that win aren't buying intelligence. They're teaching it.
“Ask what the number is before making it go up.” Great line. But a few layers worth adding:
Most of the modern economy is already “tool-shaped work.” Graeber’s Bullshit Jobs thesis. Entire industries produce the sensation of output without objective production. This hasn’t stopped the economy from increasingly devoting human capacity to it. The kanna metaphor applies to most white-collar work pre-LLMs too.
AI adoption will be slower than some think and faster than others. Both can be true.
The bigger issue: this counteracts a point I haven’t actually heard anyone make. Nobody serious is saying “LLMs are NOT tools.” The real claim is “these are good general-purpose tools increasingly capable of replacing much of what humans do today.” Which says more about what humans are doing than about the tool itself.
@martyrdison Our humanity will be what defines the next generation of value. Taste. Connection. Literally stated a company about this https://t.co/TmxzJLpwxY