There are now several theories of how work will change in the AI age, and Microsoft CEO Satya Nadella has come out with a new one — and it’s centered on human capital and something he calls token capital.
In a long-form post on X, Nadella argued that the real question facing companies today is not which AI model to adopt, but whether they are building a compounding learning loop between their people and their AI systems. “The real opportunity is not in picking the best model but instead in building a learning loop on top of models where human capital and token capital compound,” he wrote.
He defines human capital as the knowledge, judgment, relationships, ingenuity, and pattern recognition of a company’s people, while token capital is the firm’s AI capability it builds and owns. The two, in his view, are not in competition. Human expertise grows more valuable as AI capability grows, and the engine that drives token capital growth is human agency — people setting goals, connecting dots across domains, and recognizing the patterns that matter. “Without human direction, you have compute running in circles,” he said.
This framing is consistent with views Nadella has expressed before. He has previously talked about macro delegation and micro steering as the new model for how professionals will work alongside AI — delegating broadly to agents, then steering them at key decision points. The new post takes that idea further and asks what it means at the level of the entire firm.
The answer Nadella lands on is architectural. Companies need to build agentic systems that improve with each use, while retaining ownership and control over the intelligence they accumulate. He names three specific mechanisms: private evals that measure model improvement against real business outcomes rather than external benchmarks; private reinforcement learning environments that let models grow stronger on actual traces from inside the organization; and a knowledge base that makes institutional memory queryable and keeps token usage efficient. Together, these form what he calls a “hill climbing machine” — a loop that compounds the more it runs.
The sovereignty question sits at the center of his argument. “A company should be able to switch out a ‘generalist’ model without losing the ‘company veteran’ expertise built into their learning system,” he wrote. That portability is the test of whether a firm actually controls its AI capabilities, or whether it has just rented intelligence from a larger platform. Nadella has long believed that foundational AI models are getting commoditized, which makes this point especially pointed — if the models are interchangeable, the learning loop built on top of them is where the durable value lives.
The broader concern animating the piece is concentration. Nadella draws an explicit parallel to the first wave of globalization, where outsourcing hollowed out industrial economies while aggregate GDP numbers looked fine on the surface. He wants to avoid a similar dynamic in AI, where a small number of models absorb the expertise of entire industries and the economic returns flow to very few players. “The last thing any of us want is a world where every company across every sector is ceding value to a few models that eat everything they see,” he wrote, adding that there is simply no societal permission for an AI future that works that way.
His proposed alternative is a frontier ecosystem rather than a frontier model — one where every organization can own the loop that encodes its institutional knowledge, and where the platforms enable more value to be created on top than is captured inside. That is, notably, the ethos Microsoft has operated under through several previous platform cycles, and Nadella is explicitly positioning it as the right frame for this one too.
For enterprises thinking about their AI strategy, the post is essentially an argument against treating AI adoption as a procurement decision. Picking the best available model and deploying it is not the same as building an asset. The learning loop is the asset, and building it early — while the institutional knowledge is still fresh and the competitive advantage is still capturable — is the argument Nadella is making.