Satya Nadella introduces “token capital”: internal evaluations, reinforcement learning on operational traces, queryable knowledge bases, agentic systems that improve over time. He presents it as a new conceptual category for the future of the enterprise.
It is not.
Defining workflows, making created knowledge usable and shareable, making institutional memory accessible and actionable: this has been the program of knowledge management since the 1990s. Nonaka, Takeuchi, Sveiby, Stewart, and Edvinsson articulated it thirty years ago. It is called knowledge capital. Replacing “knowledge” with “token” does not add explanatory power. It subtracts precision.
What has really changed are the tools. Knowledge management has always had an implementation problem, not a theoretical one. Wikis would empty out after three months; knowledge bases would turn into document graveyards. LLMs remove part of that friction, in some cases decisively enough to move from theory to implementation. Karpathy’s wiki-LLM integration is an example: it does not invent the concept of “wiki capital”; it makes the wiki, as a tool, finally practicable and effective.
But this is precisely the rhetorical move: merging a real leap in tool usability with the claim of a new category. And this is not accidental. Every component of Nadella’s “token capital” maps onto an Azure product:
- Internal evaluations → Azure AI Studio
- RL → Azure ML
- Knowledge base → Azure AI Search
- Agents → Copilot Studio
This is not a framework. It is a concept built around the solution.
There is also a deeper problem. Tacit knowledge: the way a team makes decisions under uncertainty, the reason Toyota and General Motors could work from the same manuals and still produce radically different results, is not simply knowledge that has not yet been codified. It is a conceptually different category. It cannot be tokenized. Two companies with the same model, the same internal evaluations, and the same knowledge base can produce radically different outcomes, because the value lies in the way people interact with those systems. That value is not in the token. It is in the social body of the organization.
The tool provides a real leap. The concept does not. Nadella is right about the operational urgency and wrong about the conceptual novelty, two things he strategically fuses into one.
What we are looking at is interested reductionism: the sellable part presented as the whole. “Token capital” is not a concept. It is Microsoft’s go-to-market.
Satya Nadella introduces “token capital”: internal evaluations, reinforcement learning on operational traces, queryable knowledge bases, agentic systems that improve over time. He presents it as a new conceptual category for the future of the enterprise.
It is not.
Defining workflows, making created knowledge usable and shareable, making institutional memory accessible and actionable: this has been the program of knowledge management since the 1990s. Nonaka, Takeuchi, Sveiby, Stewart, and Edvinsson articulated it thirty years ago. It is called knowledge capital. Replacing “knowledge” with “token” does not add explanatory power. It subtracts precision.
What has really changed are the tools. Knowledge management has always had an implementation problem, not a theoretical one. Wikis would empty out after three months; knowledge bases would turn into document graveyards. LLMs remove part of that friction, in some cases decisively enough to move from theory to implementation. Karpathy’s wiki-LLM integration is an example: it does not invent the concept of “wiki capital”; it makes the wiki, as a tool, finally practicable and effective.
But this is precisely the rhetorical move: merging a real leap in tool usability with the claim of a new category. And this is not accidental. Every component of Nadella’s “token capital” maps onto an Azure product:
- Internal evaluations → Azure AI Studio
- RL → Azure ML
- Knowledge base → Azure AI Search
- Agents → Copilot Studio
This is not a framework. It is a concept built around the solution.
There is also a deeper problem. Tacit knowledge: the way a team makes decisions under uncertainty, the reason Toyota and General Motors could work from the same manuals and still produce radically different results, is not simply knowledge that has not yet been codified. It is a conceptually different category. It cannot be tokenized. Two companies with the same model, the same internal evaluations, and the same knowledge base can produce radically different outcomes, because the value lies in the way people interact with those systems. That value is not in the token. It is in the social body of the organization.
The tool provides a real leap. The concept does not. Nadella is right about the operational urgency and wrong about the conceptual novelty, two things he strategically fuses into one.
What we are looking at is interested reductionism: the sellable part presented as the whole. “Token capital” is not a concept. It is Microsoft’s go-to-market.
Very interesting! Congratulations