The Short Version
Satya Nadella argues every firm will soon run on two kinds of capital: human capital and token capital, and the winners will be the ones that compound the two. I agree, with one correction. The slice he treats as a footnote, the relationships and the read of a room, is the entire game in healthcare. AI is the leverage. The human is still the point.
Satya Nadella wrote something this month that most people read as a Microsoft strategy memo. I read it as a healthcare memo.
His argument, stripped to the studs: every company will soon run on two kinds of capital. Human capital, which is the knowledge, judgment, relationships, and pattern recognition of its people. And token capital, which is the AI capability the firm builds and owns. The companies that win are not the ones that pick the best model. They are the ones that build a learning loop on top of the models, where human capital and token capital compound on each other over time.
He said one line that I have not stopped thinking about. "You can offload a task, or even a job, but you can never offload your learning."
That is correct. And in healthcare, it is the whole ballgame.
Here is what worries me about our industry right now. Healthcare leaders know they have to move on AI. They feel the pressure from every direction. But most of them do not understand how it actually works, so they are rushing in without seeing what they are giving away. That is not a strategy. That is handing over the thing that makes your organization yours and calling it progress.
So let me be precise about where I agree with Satya and where I want to push him. He is right that human capital and token capital are the two inputs. What he treats as one slice of human capital, the relationships and the read of a room, I believe is the entire game in healthcare. It is not a footnote to judgment. It is the part that never commoditizes, and building your AI strategy without protecting it is how you win the benchmark and lose the patient.
Satya Nadella sits down with Reid Hoffman fresh off Microsoft Build 2026 to lay out the human-capital and token-capital framework. The full hour is worth your time. The healthcare translation below is mine.
Why this lands differently in a pharmacy than in a software company
When Satya talks about commoditization, he is worried about a few large models absorbing the expertise of entire industries and hollowing them out, the way the first wave of globalization hollowed out manufacturing towns. The GDP numbers looked fine. The people did not.
Healthcare is the most human-capital-dense industry on earth. I spent more than twenty years behind a pharmacy counter before I ever wrote a line of code or stood up an AI agent, and the counter taught me something no model will learn from a transcript.
Some version of this happened to me more times than I can count. The refill looks routine. The interaction check comes back clean. Every number on the screen says this is a thirty-second handoff. Then the patient goes quiet half a second longer than they should, or they ask one extra question about a side effect they read about somewhere, and you realize the clinically correct answer and the human answer are not the same thing. The screen wanted me to hand over the bag. The patient needed me to pick up the phone.
A model can look up the drug interaction in a millisecond. It cannot hear the hesitation. You can encode the first kind of knowledge into token capital. You cannot encode the second.
AI plus EI equals client success
I have been saying this for a while, and Satya's framework is the cleanest macro-economic case for it I have seen.
AI is token capital. It is the leverage. It absorbs the repeatable judgment, runs the workflow, queries the institutional memory, and never gets tired at 2 a.m.
EI, emotional intelligence, is the part of human capital that does not commoditize. It is the trust, the empathy, and the read of the room that no private reinforcement-learning environment is going to capture from a transcript.
Client success is what happens when you get the balance right. Not AI instead of the human. Not the human refusing to use AI. The compounding loop between the two.
Satya frames the loop as workflows feeding evals feeding better models. I frame it the same way, with one addition. In healthcare, the signal that matters most is not whether the model improved against a benchmark. It is whether the patient felt heard and got the clinically correct outcome, together.
Before anyone tells me that feeling heard is too soft to measure, it is not, if you decide to measure it. The sentiment in a support conversation. Whether the patient came back. Whether the issue resolved without an escalation. Whether the follow-up turned into a thank-you or a complaint. We watch that experience signal on the same dashboard as the clinical one, because the day the two move in opposite directions is the day the loop is optimizing for the wrong thing.
The sovereignty test, for healthcare leaders
Here is the part of Satya's memo that should change how y'all think about your own AI strategy this year.
He proposes a test. You should be able to swap out a generalist model, the latest frontier release from whoever, without losing the "company veteran" expertise built into your system. If switching models wipes out your accumulated judgment, you never owned it. You were renting it.
For a clinic, a pharmacy, or a health system, that test is not academic. Your institutional memory, your protocols, the hard-won judgment of your best clinicians, that is your IP. If all of it lives inside someone else's model, you have handed your differentiation to a vendor. When they change their terms, you have nothing left that is yours.
For those of us in healthcare, this is not only a question of IP. It is a question of patient data. The same instinct that reaches for the easiest model tends to send your most sensitive information somewhere you do not control, to be absorbed into a system you do not own. Sovereignty over your judgment and sovereignty over your patients' data are the same fight. You either hold both or you hold neither.
The move is to build the learning loop you own. Capture your workflows. Capture the judgment of your veterans. Make your institutional memory queryable. Measure it against outcomes that actually matter to your patients, not public leaderboards. Then you can switch the engine underneath without losing the thing that makes you, you.
We built this on purpose
I am not writing this from the bleachers. At Nimbus we built our patient-facing agent, Nimy, on exactly this principle. Its intelligence does not live as a clever prompt wrapped around whatever frontier model is winning this quarter. The knowledge base is ours, written and maintained by our clinical and customer-success teams, and it re-indexes from our own system into our own vector store. The judgment of our veterans is captured as content we govern, not as weights locked inside a vendor's product. Production runs on infrastructure inside our own walls, because the second patient data leaves to improve someone else's model, we have given away the privacy and the IP in a single move.
That choice costs more up front. It is slower than calling an API and moving on. But it means I can swap the engine underneath Nimy the day something better ships, and everything that makes Nimy ours stays ours. That is the sovereignty test, passed on purpose.
Where I come out
If you lead a clinic, a pharmacy, or a health system, here is what I would do this year, in order. Capture how your best clinician actually decides, the real reasoning and not the protocol taped to the wall, while you still have them in the building. Put that institutional memory into a system you own and make it queryable. Then measure your AI against the outcomes that matter to your patients, the clinical result and whether they felt heard, instead of a public leaderboard. Do those three things and you can change models every six months without losing a thing that makes you, you.
Satya is right that the future of the firm is the ability to compound learning across people and AI. In healthcare I would say it plainer. The firms that win will not be the ones with the most token capital. They will be the ones who use token capital to free their people to do the thing only people can do. Sit with a patient. Earn trust. Make the call that requires a human conscience and not just a confidence score.
The technology is the leverage. The human is still the point.
That is the equation. AI plus EI equals client success. Balance is not a soft idea. After reading Satya, it looks like the only stable equilibrium there is.
This piece responds to Satya Nadella's June 2026 post on the future of the firm in an AI-driven economy. The frameworks of human capital and token capital are his. The healthcare translation, and the EI layer, are mine.