Satya Nadella published a sprawling post on X this week that reads less like a product roadmap and more like a philosophical position paper on the future of the firm. The Microsoft CEO introduced a framework he calls "human capital" and "token capital," arguing that companies will soon need both: the judgment, relationships, and pattern recognition of their people, plus the AI capabilities they build and own.
The post touches on familiar themes about platform ecosystems and value creation, but its core concern is more specific. Nadella is worried about who captures the learning when employees interact with AI systems. As he put it, companies should be able to "switch out a 'generalist' model without losing the 'company veteran' expertise built into their learning system."
That sentence deserves more attention than it received. Nadella is describing a scenario where the accumulated wisdom of an organization, the tacit knowledge that veteran employees carry in their heads, gets encoded into AI systems that the company can own and control. He frames this as sovereignty. Others might frame it as extraction.
Tacit Knowledge Has Become Legible
For decades, management theorists have drawn a distinction between explicit knowledge (documented processes, written procedures) and tacit knowledge (the unwritten heuristics, judgment calls, and contextual awareness that experts develop over years). Tacit knowledge was always valuable precisely because it was hard to transfer. It walked out the door when employees retired.
AI changes that equation. As a recent piece in the California Management Review observed, the reasoning patterns and interpretive skills that experts develop "rarely appear in manuals or dashboards" but increasingly can be captured by AI systems that observe how work actually gets done. Nokia's enterprise blog made the point even more directly: once fine-tuned, models themselves "become carriers of tacit knowledge, a form of institutional memory that persists even when employees leave."
This is exactly the compounding loop Nadella describes. Every improved workflow generates better training signal, which accelerates the accumulation of tacit knowledge unique to the firm. The companies that build this early, he argues, will have an advantage that is hard to replicate.
The Ownership Question
But whose tacit knowledge is it?
Under most employment contracts, work product belongs to the employer. If you write code on company time, the company owns the code. The legal framework was designed for artifacts you can point to: documents, inventions, creative works. Tacit knowledge was never explicitly addressed because it couldn't be explicitly captured.
Now it can. When an employee's judgment, reasoning patterns, and decision-making heuristics get absorbed into a company's AI system, that system becomes more valuable. The employee's expertise has been converted into what Nadella calls token capital. The employee receives their salary. The company receives an asset that compounds.
IBM noted in a recent analysis that employees using company AI tools could potentially hold copyright over content they create, though "these rights would likely be transferred to the company under employment contracts." The same logic extends to the training signal itself. If your daily workflow improves a model, that improvement belongs to your employer.
This is not necessarily unjust. Companies have always captured value from employee expertise. What's different is the permanence. Previously, when a skilled employee left, their tacit knowledge left with them. The company had to rebuild institutional memory with new hires. Now that knowledge can persist indefinitely in AI systems, and the former employee has no ongoing claim to it.
The Globalization Parallel
Nadella draws an explicit comparison to the first phase of globalization, where "entire industrial economies were hollowed out by outsourcing." He warns against repeating that dynamic, "with a small number of AI systems capturing all the economic returns, while entire industries find their knowledge commoditized right out from underneath them."
The irony is that his own framework could produce a version of the same outcome at the individual level. If companies successfully encode employee expertise into systems they own, workers become more replaceable even as they become more productive. Their judgment gets amplified, yes. But it also gets replicated and scaled in ways that diminish their bargaining power.
Nadella seems aware of this tension. He emphasizes that human capital "does not become less valuable as token capital grows" and that "human agency will be the driver of token capital growth." Without human direction, he writes, "you have compute running in circles."
That may be true at the frontier, where novel problems require novel judgment. But most work is not at the frontier. Most work is pattern-matching against situations that resemble past situations. That is precisely the kind of work AI systems excel at once they've absorbed enough expert examples.
What Comes Next
The legal and policy frameworks have not caught up to this reality. IP law was built for artifacts, not for the diffuse expertise that gets encoded into model weights. Employment law assumes a transactional relationship: labor for wages. It does not account for the ongoing extraction of judgment that AI training enables.
At Microsoft Build 2026, the company announced Microsoft IQ, an enterprise intelligence layer designed to make institutional memory queryable. The pitch is compelling: AI systems that understand your organization's specific context, not just generic capabilities. But the same infrastructure that makes institutional memory queryable also makes it extractable, transferable, and ownable in ways it never was before.
Nadella's post is best read as an opening argument in a debate that will intensify over the next several years. He wants companies to own their learning loops. He wants value to flow broadly across the economy rather than concentrating in a few foundation model providers. Those are reasonable goals. But the question of what workers are owed when their expertise becomes company property remains unresolved.
The California Management Review framed the opportunity for companies: "Your next competitive moat will come from designing new systems that capture tacit knowledge and make it explicit." The same sentence describes the challenge for workers. Their competitive moat is eroding.
