Satya Nadella wrote something interesting, and worth interrogating.
Here is the link: snscratchpad.com/posts/frontier-ecosystem
He argues that firms need to build two types of capital: human capital (judgment, relationships, ingenuity) and token capital (owned AI capability). The loop between them, he says, is the new IP of the firm.
He's right about the loop. But he doesn't follow the logic to its conclusion.
I've written along similar lines, but I've maintained the models currently aren't good enough to take the human out of the loop, and it's possible that they never will be.
Let's assume for argument's sake that we have a viable path to get them there.
Humans contribute expertise. It gets encoded into the firm's AI systems. The system compounds. Firm IP grows. Repeat. What happens to the human in cycle three? Cycle ten? The expertise has been extracted. The firm owns the distilled version. The original is now optional.
This isn't a new story. Frederick Taylor did it to physical labor a century ago. Time-motion studies, standardized workflows, expertise lifted from workers and encoded into processes. The worker became interchangeable. Satya's token capital is Taylorism for cognitive work.
Human capital in his framework isn't a permanent complement to token capital. It's the input that gets consumed to produce it. The loop systematically converts one into the other.
There's a collective action problem he doesn't resolve. Every firm optimizing for token capital accumulation, individually rational, produces exactly the hollowed-out world he says society won't tolerate.
Let's not forget - corporations produce goods and services for people. If LLMs and robotics replace people wholesale as firms cut headcount to juice short term earnings; who will buy said goods and services in the long run? What is even the point of making things that aren't for human consumption?
Business leaders, especially in tech, are surprised at the AI backlash. They shouldn't be. Their marketing pitch:
a) AI is powerful and dangerous and will change everything in ways we don't know yet
b) It will take your jobs
c) Therefore, give us trillions of dollars in private and public funding, let us build datacenters to make (a) and (b) a reality.
What?
I think there's a good chance that we never get to wrestle with these types of outcomes. But we shouldn't leave it to chance. We shouldn't forget that individuals and firms have agency even in an agentic AI world. The primacy of human labor needs to be preserved, understanding that the shape of that labor will change as technology advances.
We can and should build toward a future that serves human goals.