Possible: Satya Nadella on making human and token capital compound

In this recent episode of Possible, Reid Hoffman sits down with Microsoft CEO Satya Nadella fresh off Microsoft Build 2026. The conversation goes wide: how AI is reshaping work, business, and society—and why the transformation sweeping through software development today is only a preview of what's coming for all knowledge work. Satya makes the case that human capital and "token capital" are now deeply intertwined, that companies—not just countries—must build their own AI capabilities, and that the organizations best positioned to thrive are those that can leverage their unique expertise inside intelligent systems. Reid and Satya also explore Microsoft's enterprise AI vision, Reid's work with Manas on AI-powered scientific discovery, lessons from past technological revolutions, and why demonstrating real, tangible benefits may be the most important thing the industry can do to earn—and keep—the public's trust. You can catch and subscribe to more Possible here: https://www.possible.fm/ See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.

• 0:00Speaker 00

  So what I think we have not yet conceptually gotten right and a shared understanding is what is this future of work going to look like? If you're a tech CEO, you have to be deep inside of what's the tech stack. AI is not a technology, it's the future of the firm. One of the dictums I have is don't use frontier models for non-frontier problems. I think in the AI age, that is going to be everything, right? I think I would be very surprised, Reid, if he was sitting here a year from now. If the world is not completely turned on, what is my AI supply chain look like?

• 0:37Speaker 01

  I couldn't be more delighted to introduce a special episode of Possible with Satya Nadella, the chairman and CEO of Microsoft. Satya and I have known each other a long time and part of in this is kind of AI revolution for humanity. I thought this would be a great episode to get out. I mean, we covered all kinds of important topics. Satya, as always, is elegant, is cohesive, is smart, is comprehensive, and above all, humanist. And what is our AI future? This will be an amazing episode. Actually, one of the things that's great, Satya, about filming this here is that it reminds me of the earliest days when we were talking about Microsoft and LinkedIn.

• 1:19Speaker 00

  That's right.

• 1:20Speaker 01

  Because we did one of our very, very first conversations here in the Greylock office, so it's like just awesome to be back. I want to start with something that I don't know as many people realize and appreciate about you, which is with how many books of poetry you have in your house? Can you say a little bit about your attraction to poetry, favorite poets? What the engagement is there?

• 1:41Speaker 00

  Yeah, I mean, I actually got into it in different times of my life. I remember, you know, as a middle schooler, we had this standard issue English poetry book, which I've been trying to reclaim and get all my life. And unfortunately, it's out of print, but it sort of had, you know, even getting, you know, introduced to Shelley or Wordsworth. And it had even sort of Indian authors like Siraj Nainu writing in English. And it is, I don't know, I felt maybe it was my attention span or what have you. I was naturally drawn to poetry as a thing to enjoy and love. And I've always compared it even to code, right? Which is it's sort of compression in its best form.

• 2:24Speaker 00

  And so whenever I'm bored, I get to, you know, go read. I'm not great actually at understanding deeply. I've never studied it. I'm not like so, so, so that's why my reputation of knowing about poetry is far exceeds my knowledge of poetry. But I still, you know, continue to use poetry as perhaps the best expression of the human experience, right? I mean, if you, if you sort of broadly think about literature as sort of what captures more than even history, the human experience, I think poetry is the compressed form of it.

• 3:05Speaker 01

  I completely agree. Is there a particular poet that you go back?

• 3:08Speaker 00

  Yeah, so I, the thing that I really, I would say the poetry that probably speaks to me most deeply is the Urdu poetry. That is sort of because I grew up in Hyderabad in India and Urdu is sort of the, in the air. And some of the Urdu poets, both modern and sort of, you know, people in the 17th, 18th century were just extraordinary. And in particular, you know, there's, there's this poet whose name, who goes by Galib as his name, and he's just extraordinary. Or even a modern poet like Fais. I was also very, having grown up in Hyderabad, I was very influenced by Rumi. In fact, you know, my, the high school I went to was fascinating.

• 3:52Speaker 00

  In fact, we had, you know, the number of languages which were all taught was, obviously it was English, there was Hindi, which was the national language. And then we had Sanskrit, like basically, you know, like Latin here, I guess. And then we had the local language Telugu, which is my mother tongue. But we also had Urdu and Persian. And so all of them, like in fact, when we would break for what was called second language, I, my second language was Sanskrit. But, you know, I could, I had classmates who would go to Persian, Urdu, Telugu, Hindi, and it was fascinating.

• 4:30Speaker 01

  Well, among the experiences, I'm experiencing a little bit of language envy because I, a joke that I like is, you know, what do you call a person who speaks three languages? Trilingual, two languages bilingual, one language American? And unfortunately, I resemble this joke. All right, so let's, we're just coming out of Build, which was amazing. Let's kind of actually start with kind of going back in history. So, you know, Microsoft's first product was a basic interpreter. And 50 years later, here at Build, frame the company's future on what others can build with AI. So what is the new basic interpreter in the AI era?

• 5:07Speaker 00

  That's a great one. So in fact, it's, that, that I think is the core of the DNA of the company, right? Which is what can others build? And therefore, what should we build? It's sort of the two questions that we always ask ourselves. And in the AI era, to me, the answer is the basic interpreter is the hill climbing machine, right? Because you think of, you now bring all of this AI, what is it? It is basically taking an objective and outcome and eval that you have and learning how to achieve that by learning, using data, using some reinforcement reward. And so the, the entire conference was about not, hey, here is a new frontier model.

• 5:52Speaker 00

  It was about helping every developer, every company, whether it's a AI native startup or an enterprise, building their own hill climbing machine so that they can operate at the frontier, right? So I think that that's it. The idea, in fact, you know, being very clear about the evals and the objectives that you care deeply about, knowing how to evaluate them, right? That it's in some sense is the most, and keeping, that's the new IP.

• 6:22Speaker 01

  Yes.

• 6:22Speaker 00

  Because everything else is pretty mechanical. But knowing what is the set of data that you want to train a model on and how you reward it is probably where the, the next level of IP gets created.

• 6:36Speaker 01

  Well, and one of the things, let's dig into a couple of different areas here because you and I've had a number of conversations, you know, with Microsoft strategy. And one of them is kind of this question about how enterprises keep the advantage and integrity of their own data. And Microsoft is the most natural company in the entire world to do this. So what, say a little bit about how enterprises should be thinking about like, we need our own frontier intelligence, but we also need to maintain control of our data.

• 7:04Speaker 00

  Yeah, see, that I think is the, the question, which is this economy is going to be shaped going forward by both human capital and let's call it this token capital, right? That is true for Microsoft, that is true for a new startup, that's going to be true for any bank that's been in existence for a hundred years, right? Doesn't matter. All of us will now need to sort of, in fact, the interplay between human capital and token capital and compounding the returns of it is what you need to do. So if you sort of frame it that way, then one of the most important thing is not just even thinking of data in its aggregate sense, right? See, what is the tacit knowledge of an enterprise or a firm?

• 7:47Speaker 00

  It's the unique ways that you are able to operate, pass judgment, have taste. All that's the tacit knowledge, mostly captured today in the tacit knowledge that is there with the human capital and some artifacts that are digital. So now, when it comes to AI, the, the model in some sense is able to extract, if anything, that tacit knowledge through human trajectories and encode it in a set of weights in a model. So what I claim is that every enterprise now needs to be more mindful about that interplay of humans and their digital estate working together, those trajectories, training

• 8:32Speaker 00

  essentially the models that they keep as IP versus leak it. Because if you leak it, it's a one-way door. You're done in some sense, right? Which is what is unique, what you may have spent a hundred years, right? In fact, we don't even know how to articulate it, right? Nobody sort of has a line item in their balance sheet called tacit knowledge. But we take it for granted that because we have human capital, we have it. Now, I believe that can leak. And in fact, it is leaking, right? If you look at even the number, how the model companies learn, they're essentially setting up these gyms with rewards, which are employing employees who worked at your company previously. I mean, that's sort of should tell you everything, what, what should not be happening, which

• 9:16Speaker 00

  is, so that's one of the fundamental reasons why we want this effectively, the regime to change or the paradigm to change, where you welcome the models to come in, they should hill climb inside a machine that you control. Your data is your context you feed the model. You collect, in fact, these traces or trajectories of how work gets done between humans and agents inside the enterprise. But you have a continuous loop of that and you're not letting that leak. That I think is the fundamental operation.

• 9:47Speaker 01

  Yep. And actually, the related parallel to the tacit knowledge is, and this is one of the conversations you and I had actually at the, at the Microsoft board, which was what is the future role of AI employees, right? Because part of the question is to say, hey, we're going to provision not just, you know, kind of tools to amplify human work, great, or, you know, kind of AI companies, but we're also going to provision at least specialist employees. But the challenge with that is, is that one of the things that really matters to enterprises is your employees embody a lot of, you know, tacit knowledge, how to, what to do. Do you want that in other companies and provision in other places?

• 10:28Speaker 01

  What's the, your and Microsoft view of the future of work when it comes down to how to think about AI as also maybe even specialist employees?

• 10:38Speaker 00

  Yeah, I mean, I think that that's right. So the way to think about this, you know, there are two ways I come at this, right? Which is, let's, let's take one analogy and I'll come back to it, right? If somebody in the early eighties had come to us and said, you know what, there are going to be four billion typists who are going to wake up every morning and start typing. And we were said, what for? It makes no sense, right? We have a typist pool, we have a slide pool and we're, you know, fine, you know, with that. But except we invented this complete new thing called knowledge work, where everybody was typing and creating artifacts and so on.

• 11:14Speaker 00

  So what I think we have not yet conceptually gotten right and a good understanding, and a shared understanding is what is this future of work going to look like? When you have, let's call it, you know, and let's take Microsoft, we have 20,000 employees. And we have, let's say, two million agents or 20 million agents all in a loop. What is happening? What is the tacit knowledge that gets created? What is, you know, the artifacts that are going between agents, between agents and humans and so on? All that's now going to be played out in the next, whatever, even multiple years. You can see early forms of this in coding, right?

• 11:57Speaker 00

  It's a great place to observe, in fact, the, the, even this sort of social change, right? I mean, think about, we started in a good old IDE and said, hey, AI is in, you know, in there. And it's doing code completion. That itself was useful, easy to understand. We've always had spelling correction and IntelliSense was there in VS Code, you know, from there 15 years ago. And, and it just got better. Then we said, oh, you know, instead of going out of band and going to a browser and, you know, going to Stack Overflow and searching, you can now bring all the coding knowledge to a chat session. That was also easy to understand. And you still were in the IDE, you had through the chat.

• 12:40Speaker 00

  Then we said, okay, you now have reasoning models and some, you know, prime, you know, preliminary agent loops and so on. And so you can assign tasks. So we had agent mode, right? So you not only had chat, but you can give it small tasks and you could see it complete. And then you could accept what it did and then you could insert it and what have you. Then came the big breakthrough of total autonomy and agentic loop working for long periods of time, where you could literally fire, forget, right? Where you could assign a high level intent, it would go off and do the full PR and you would accept the PR. So that transition, right, is what I think is going to happen across all work, right?

• 13:24Speaker 00

  And now we're seeing it with even in Copilot, we now have chat and we have cowork. And now, in fact, at Build, we announced something called Scout and Autopilot. So what is happening in coding will happen even in knowledge work. Interestingly enough, one of the things we launched even at Build in GitHub was a new feature called Canvas. Because what has happened is, as we've all gotten so good at using these coding agents, in fact, you know, the biggest challenge we now have is I have 100 CLI sessions open, where I'm trying to operate these 100, you know, agents. And now the cognitive load on me, managing this is so high, right? I mean, first of all, I don't, I mean, they think about it, right?

• 14:07Speaker 00

  It's a linear chat session in a command line and I have 100 of them. So guess what? We now are back to an IDE, right? So I have a new IDE, we give it a new fancy name, it's called an ADE, it's an agentic development environment. That's what the new GitHub app is. The GitHub app looks like an inbox, except it's an inbox of agents that are working across all the repos in allowing me to do the micro steering of the macro delegation I gave them. But it's a complete UI for the agents to deal with me and for me to deal with agents. And so I feel, so we introduced this new feature called Canvas in GitHub where it can like, for example,

• 14:49Speaker 00

  you know, sometimes it's easy to have a Kanban board visualization as a way to run down your PRs. Very useful, both for agents and me to interact versus some chat session. So I think that that type of innovation is how work will change. In fact, one of the other fascinating things for me is a line of even AI research would be, the models will also get much more tuned to learning how to stay the course and understand human preferences in steerability, right? Because that's what I want. And in fact, I want models that not only do instruction following, but are also really steerable.

• 15:32Speaker 00

  And that, when you have that, that's when you have confidence in it. So I think the work of the future is about tacit knowledge that gets created by this interplay, just like knowledge work got created because of digital artifacts and human capital. The new one would be this AI capital and the human capital working together, creating in some sense the digital artifacts.

• 15:56Speaker 01

  And what do you think are the things that people miss about how important it is to build out additional kind of structure for enabling enterprises? So one of them is obviously the enable and humans canvas, you know, ADs, et cetera. But what are the like, for example, notions of like, what is a security in an agentic?

• 16:17Speaker 00

  Oh, it's a great point. It's a great point. So in fact, the, to your point, one is the experience layer. Clearly that's important. And another one which we referenced earlier was, hey, we need this hill climbing machine as a concept that needs to be instantiated. But the third is the manageability of it and the security of it, starting with observability, right? One of the first things we did even, and we talked about it even yesterday, is something called Agent 365. I need to know, I need to have an inventory. I said, oh, there may be 20 million agents at Microsoft. I first need to know what are these agents, what are they doing, what are their reasoning traces? They need to be fully inspectable, fully auditable. And by the way, when the agents also have this other attribute, they are running, they can generate code and execute.

• 17:03Speaker 00

  So you need the environment in which they are executing code with maybe even file system access, network access, to be think that is governed by policy. So therefore you now need to give them identities, you need to give them sandboxes, then you need to set policies to govern them. And so we built this entire thing called Agent 365, which really has, you know, we have extended Entra so that you have an identity, we extended Defender so that you have security, we extended Purview so that you can even label the data that it's getting created automatically, so that you can have data protection. So I think security, containment, manageability, observability is the way we will have confidence around these agents.

• 17:44Speaker 00

  The one other thing I'd say, Reid, also was, you know, salient in our developer conference was, how important it is when you're building these long running agents, right? Like, if you think about, right, we have always had, in programming languages, you know, programming models for understanding the, you know, the verifiability of a program, right? And it's at runtime. So one of the attributes of these long running agents that we added to Foundry was something called asserts. So this allows us to assert what are the boundaries, right? So instead of talking about guardrails as a thing that is sort of just a classifier of some form,

• 18:29Speaker 00

  you really need to now have the ability during execution to really have that execution path not go off rails. And so there's a lot of engineering sophistication now that's emerging as we build out the platform, the runtime, the security layer, the management layer, the observability layer.

• 18:51Speaker 01

  So let's go to a little bit of the role of CEO. And I think Fortune recently described you as acting like a startup CEO inside of Microsoft's AI teams.

• 19:02Speaker 01

  But the more general question, because I think a lot of CEOs seek, you know, kind of technology, strategy, advice about what's going on in the world from you. What do you think the CEOs should be doing around AI? What, you know, kind of when they're, like, AI brings the kind of a refounding moment to lots of companies because of the nature. And so what's your advice to them for how they should engage?

• 19:29Speaker 00

  Great question, because I've been thinking about this, right? Because in some sense, you know, it clearly for, if you sort of, if you're a tech CEO, you have to be deep inside of what's the tech stack, right? There's no way you can be a tech CEO and not essentially have a fundamental worldview on where the future is going. And then to be long before its conventional wisdom, you have to pass judgment on, you know, where the company is going, right? So that's sort of the, our industry's sort of pretty binary transition. So I think there's no hope if you don't have a, the CEO leading from the front and taking the shot on goal, knowing that these are fairly harsh transitions.

• 20:12Speaker 00

  But the interesting thing that now, up to now, if you're not a tech CEO, you needed to be a great CEO, which means you need to be great at banking or healthcare or whatever it is. And you could always have great partners and technology advisors and what have you. But now I'm changing my prior on that, right? Because I think, because I'm noticing even what's happening. And I don't think the, the rest of the, the industry and the CEO community is broadly woken up to this. And I'm like perplexed at it because they're still, you know, pretty happy doing a press release with a, a tech company and saying, yeah, I got my AI strategy and, you know, pointing to eight agents they built and some outcome.

• 20:54Speaker 00

  But this is no longer, AI is not like a, sort of a technology, it's the future of the firm, right? It will be like saying, oh, I don't know what the, about the human capital in my firm, right? So I think that this is where my thing would be, that you have to now get a deep understanding of what's your token capital. Can you answer that question, right? When you concretely say, oh, last night.

• 21:21Speaker 00

  Based on all the work that happened across all of our operations, we were able to translate that into a set of knowledge that is somehow now part of my token capital. It can be some context. It can be some skill. It can be some weights in a model. It doesn't need to be any one thing, but you need to be able to specific, you know, you need to be able to identify it clearly as something that you own, you control, you created, and you put in place a system to have that compound. So I think that that is going to be probably the toughest change.

• 22:03Speaker 00

  So this change is like, unlike how this is like going to a mobile phone era or a PC era or a cloud era where all I need to do is to have an IT department that knew how to deal with a bunch of vendors and did some smart things to reduce cost or improve my efficiency. It should start there. I'm not saying that that's not the place. But this is about what what happens structurally in your industry when AI is sort of basically knows everything that it needs to know about being in your industry. And if you start there, then I think you will start understanding that this is not just a tech. It is really about a fundamental change to the firm.

• 22:45Speaker 01

  And one of the things that I think thinking nature of the firm is that you've done a great job in leading Microsoft. One of the many different genius moments and skills is going through acquisitions and partnerships, right? So it's, you know, whether it's LinkedIn, obviously something close to our hearts, you know, GitHub, OpenAI, et cetera. And part of what I think Microsoft leads the entire world on is how do you build trust in these other ecosystems? And in the world of AI, that's going to be extremely important, especially as the firm changes. So what are some of that, the kind of the the kind of lessons you've learned, the things that you are advocating for in order to build and maintain that trust?

• 23:28Speaker 00

  Yeah, it's so key because in some sense, I've always sort of thought about what's long term stable, right? Even from a Microsoft perspective, what is long term stable is for us to be a tools and a platform company where we fundamentally are defined by the amount of value that gets created on top of the platform, which should far exceed anything that is captured in the platform, right? That's the only way to have stability. If you do that, then you have trust, right? Because then the customer, the partner knows that it's not a zero sum. Especially it's not a game theoretic,

• 24:09Speaker 00

  you know, zero sum with these games where, oh, I'm going to subsidize this until two years or three years, only to then eat your lunch type of games that the tech people are really good at. And so I've always felt like, hey, look, that is all not the way. The way to think about it is to be very principled that you are a platform company, that you will really live and die by your ability to create success on top of the platform. And when they are successful, you're going to be successful. And that is the equation that builds trust and long term stability for both sides, right? I think in the AI age, that is going to be everything, right?

• 24:51Speaker 00

  I think I would be very surprised, Reid, if you were sitting here a year from now, if the world is not completely turned on what is my AI supply chain look like? Where it is helping me as a firm compound the returns of AI that I can uniquely point to as my value in a world where we know that these learning systems by definition don't have boundaries. They have to have boundaries because of mostly structures of markets and society and other mechanisms.

• 25:26Speaker 01

  Yep. So we've talked a lot about chatbots and code, which have just seen amazing explosive progress.

• 25:35Speaker 01

  What are some of the next frontiers for LLMs over the next few years?

• 25:38Speaker 00

  Well, I mean, talking about that, in fact, you know, one of the things that I know you have gotten very, very excited about, which I'm quite frankly, you know, excited for you, but it also has caused us to, you to examine where you spend your time and I'll let you speak to it. But it's sort of unbelievable to see what's happening in science. And, and in fact, the company that I, you know, you started with Sid and Manus. You know, it's just a great example, I would say, of what has happened in, let's call it coding and knowledge work. If it start happening in science, in fact, I would even claim that I wish,

• 26:22Speaker 00

  in fact, we did it in the reverse order, right? Because the social permission for AI would be so much higher if we had started with some nice discoveries in science that were having great societal benefits. Then people would have really thought of this as a thing that is really going to be more helpful. So I'm really excited for you, Reid, with what you're doing. Obviously, you've spent a lot of time, but maybe you want to talk about that.

• 26:49Speaker 01

  Well, yeah, so, I mean, it was one of the things I realized over the last month was that we're seeing such progress with Manus in, actually, in fact, we've got an internal description of move 37 for chemistry because we're seeing chemistry that that might actually in fact take shots at really interesting cancers and other things. And, you know, it's very early, right? And so, you know, these things take a while with all the ideas. But the, the fact that we are already beginning to see, like, we've got some of the best computational chemists in the world. And I looked at them and said, that's very interesting. We've never seen it before and that might work, right? And so that kind of thing. And so, you know, Sid and Ujwal and I were talking about this and I said, look, I think I need to get back to founder mode in terms of how to do this.

• 27:33Speaker 01

  And so I need to be able to kind of focus on this. And then, you know, part of the, you know, conversation you and I have had this week is to say, okay, it's been 10 years on the Microsoft board. It's been a huge honor and pleasure with, you know, and not just LinkedIn, but obviously OpenAI, GitHub, a whole bunch of things. But, you know, at the end of the year, I should really be transitioning right now to being founder mode. So like, you know, it's like, and, you know, we'll always be working together. So it's like, but, but, you know, it's time to kind of dig back into the company.

• 28:07Speaker 00

  Now, first of all, it's been such a privilege to have you on the Microsoft board, work with you, obviously through all the partnerships and LinkedIn, and we will definitely be missed on the Microsoft board. But I know we'll be very connected. But I'm also excited for you. And you know, when you say you're going back into founder mode, you know, that means I'm even more curious about what you're going to be building. Because it's, as you, I think you rightfully captured, this is the moment where I think the impact of these technologies more broadly felt in the societies than neither they are.

• 28:40Speaker 01

  Yeah, no, exactly. Let's come back to, you know, kind of questions around the patterns of thinking that happen with AI. Because one of the things that I think, like, I kind of roughly think about these things is you've got an alien intelligence, right? A different version of A that has been built to very much heavily mimic human intelligence, which creates a whole bunch of utility for us. But its patterns of reasoning are not the same as ours, right? And people frequently encounter that and they go, oh, that's scary. You know, like, well, no, actually, that's, that's, look, you have to pay attention to make sure there's alignment, but it's also wonderful because it enables new things. And so that's part of the thing, like the kinds of chemistry we're doing,

• 29:25Speaker 01

  like the humans hadn't discovered this chemistry in Manus, right? What are the other kind of thinkings about this pattern of reasoning, right? That, that, that kind of amplify human capability?

• 29:37Speaker 00

  It's a great one, like, you know, I've not thought that deeply about sort of how it improves the, I mean, if you sort of, sort of say there's, you know, induction and deduction that we do. Being in this cognitive loop with an AI, that is fascinating, right? Because my exploration space is changing, right? The, you know, like even in coding. One of the most fascinating things that one of my colleagues introduced me to is he wrote a new skill which I have in GitHub Copilot. It's called cognitive coverage, right? It's fascinating.

• 30:19Speaker 00

  What he's done is, you know, whenever an agent does some work for me, he says, just like how we had test coverage, we now have this new concept called cognitive coverage, where we as humans are going to learn from what it did. And so it just creates a quiz. So because I literally think of learning from, and it's sort of, it is matching my ability to essentially form a deductive understanding in some sense of what an agent did, right? So to your point, it is, it's a basically, it's mimicking. It's sort of an imitation game on one side.

• 31:04Speaker 00

  But that imitation game does create. So then my ability to then deductively understand it, I think is probably one of the more important human skills we have to develop. So this cognitive coverage is, I'm like fascinated by it, right? Which is, oh, I think we're going to have something like that, which is the agents are working, they have to be aligned on one side. But the other side of it is, what is new will be us knowing, oh, we have cognitively covered what AI did.

• 31:33Speaker 01

  Yeah, well, I think that point intersected with your earlier one about, you know, kind of token capital and AI capital, is that part of the skill set for the modern, you know, kind of human knowledge worker is to say, okay, how do I strategize on the use of AI, canvas orchestration?

• 31:52Speaker 00

  Yeah.

• 31:53Speaker 01

  But also within kind of capital allocation balance, right? Because part of what we've seen is people can go crazy on spending lots of tokens in ineffective ways. It's like, you know, buying, you know, go back to bad Pentagon days, toilet seats for a million dollars. And so putting your thinking together with the cognition, together with token management. And what do you think is part of the question around this skill of blending the cognitive coverage with also token intelligence? In terms of how do you amplify the outcomes?

• 32:27Speaker 00

  It's a great one. In fact, I've been thinking about like, one of the things that I think one has to study deeply is what are all the things that become more valuable in the age of, let's call it, token abundance? One is what you referenced, which is how to use tokens becomes very valuable. Whoever figures out that, oh, I can use tokens more efficiently for an outcome that matters in the world is going to get ahead, right? By definition. So then how does one build the intuition for it, right? Which is, oh, this is where, what is that outcome? How do I measure it? What's the rubric? In fact, this is where I think the evals, the, you know, if it's, it's fascinating how much time needs to be spent, especially for an RL regime.

• 33:12Speaker 00

  I think it's been the clearest, where if you really want, you want to set up the rubric and the eval dimensions or the rubrics, the scoring dimensions, such that they're really capturing the high taste that you only can define. Because if you did that, then you could be token efficient, right? In fact, the other side of the token efficiency is, in some of the examples we even showed at Build was, let's say you have, I don't know, some, you're a retailer or you're a packaged goods company. And one of the things you do is handle sort of transactions around trade promotions. Right now, you get all these claims from all the retailers

• 33:57Speaker 00

  and you have to process them and the humans are doing it. You can easily automate that using an agentic workflow. And you could say, oh, I use a frontier model for it. But then, one of the dictums I have is, don't use frontier models for non-frontier problems, right? So this is not a frontier problem. You're not trying to discover some new material science. This is a repeatable deterministic workflow. But that can benefit from all the mistakes humans can make and claims and so on. So therefore you do need, you know, intelligence in it. But you can take, you know, a model like an MAI 5B and use the traces to hill climb in your RLE

• 34:40Speaker 00

  to perform even, outperform even a frontier prompted, right? So that to me is token efficiency. So that type of human understanding of both the limits of the system, the characteristics of the system, I think, become high premium at this stage.

• 34:59Speaker 01

  Yeah, I agree. One of the other things that I think is important for people to grok, as you know, because we've been in a lot of these conversations, there's a lot of concern in other countries about, you know, kind of sovereign AI. And I think that it may be useful here, given we've just kind of gone in depth about companies and the sovereignty, like the companies, their information, their employees. What are the parallels between what we're doing for companies and how countries should think about kind of also engaging in trust in the AI future?

• 35:31Speaker 00

  Yeah, I mean, this is interesting because I, one of the reasons why I have really pivoted to companies is because companies exist all over the world, right? That's the good news here, which is because in some sense, even countries thinking abstractly about sovereignty, they can even make big mistakes where they ultimately could erode their comparative advantage that is naturally there today, embodied in the commercial activity happening in the country through its company formation and thriving. So therefore preserving that is the best form of sovereignty. Sometimes I think people think, oh, if only I had a firewall

• 36:13Speaker 00

  or all the data was resident or what have you. You know, I'm not saying those are not considerations, but this is not about even any of that, right? Which is, this is about making sure that you have an economy. That means you need to have companies. That means that those companies have to thrive in a token economy. That means they need to be able to build that IP. And so therefore, in fact, having partnerships even with companies outside that give you the ability, I think, are more important than sort of suddenly falling behind some frontier, right? And so I think that therefore, I think that this is one of those places

• 36:53Speaker 00

  where again, even the policymakers have to think about infrastructure, right? After all, you know, tokens, there are electrons on one end and tokens on the other end. So you want to be the cheapest, best environmentally good producers of those electrons to token conversion machines called data centers, right? So that one, I think, is going to be very important for every country to prioritize. Of course, the country should even prioritize, whether there's their own semiconductor production and what have you, right? That's all very valid things to do. But beyond that, I think the most important thing is to take what Ricardo was always right on, which is countries have by definition their own comparative advantage.

• 37:37Speaker 00

  And now they need to amplify that using AI. That means best thing is that whether it's the small business or whether it's the large multinational or even the public sector efficiency in the country is getting better and is operating at the frontier. The worst thing for any country to, in the name of sovereignty, if they're off frontier, then that makes no sense because you're falling behind. You have to, but at the same time, being dependent on one frontier model also makes no sense because then you're not sovereign. So what's the, the way to solve it is to be able to say, no, we will use models to hill climb on our own, one firm at a time.

• 38:18Speaker 01

  Yes.

• 38:18Speaker 00

  And at an economy and aggregate. That's, I think, the equation.

• 38:22Speaker 01

  The, the gesture at Silicon actually, I think, is a good also bring up both for companies, we're building, Maya and Cobalt, but also of course partnering well with Nvidia and AMD. So first, what's the strategic job of Microsoft's own Silicon? What might companies or countries also learn from that?

• 38:44Speaker 00

  Yeah, so it's interesting. If you think about it, right, we want, it's a great way to observe it even. Because I, I mean, if you look at what Nvidia has done, or what AMD has done, or what Intel has done, they've built general purpose technology, which, you know, like in fact, when I look at what we're doing with GPUs today, we're of course using GPUs to do model training, model inferencing. But we're also accelerating other workloads. In fact, one of the exciting announcements at Build was using GPUs to accelerate, Nvidia GPUs to accelerate our fabric data warehouse, right? In fact, one of the things that's happening with the agentic workloads is we need more performance on everything. And so that's a great, that's because of the general purpose nature, right? GPUs have, you know, Nvidia has CUDA.

• 39:28Speaker 00

  CUDA can be used as a programming model on top of those to accelerate a variety of workloads. In fact, we're using the older chips of Nvidia to accelerate. And this also works out economically for us, right? Which is it's smart for us to take a fleet and keep using that fleet over a lifetime, where we are not only using it for some cutting edge AI, but we're also using it to in fact, make an old workload even better perform. Like that's great for Nvidia, great for us. But that also speaks to what's happened in terms of the new workloads, right? The new workloads of AI are these data parallel synchronous workloads, training inference, as well as these new agent runtime workloads, which have very different call patterns,

• 40:09Speaker 00

  very different, and they're highly constantly, there's a, this didn't exist three years ago, right, at any scale, and now they exist at scale. So it behooves us to start thinking, whether it's not even before we even get to the semiconductors, I'm building my data center, my civil engineering is influenced, my cooling system, my mechanical systems. In fact, the DC to AC, that they, we're trying to make sure that the electrons are coming in in kilowatts, right, hundreds of kilowatts, straight to the silicon without any losses, right? So we're trying to minimize even the power distribution in the data center. So we can optimize to the nth degree for these workloads because they're at such scale. And that's what we're doing with Maya, right?

• 40:52Speaker 00

  Maya, for example, is being co-designed with our M A I models and the OpenAI models to get the best performance out of them. We're designing Cobalt, which is our ARM-based core for compute, and we're designing it for, for example, using all the agentic traces of GitHub, right? Coding, you know, the call pattern of a coding agent is pretty different than human apps or even asynchronous human apps. And so therefore, we're optimizing and getting massive latency gains, performance gains, and what have you. And so we're going to be a systems company that continuously optimizes for the new at-scale workloads that we have while using general-purpose technology from our partners to maximize the utility of those.

• 41:35Speaker 00

  And that flexibility, by the way, I think is what really is good. And that's where I think your question was so good, which is, countries should think of that, right? Which is, if you really think, oh, there's one thing that answers. If I said, oh, everything is Maya and everything is Cobalt, that's probably not the right thing for Microsoft. But at the same time, if you said we're open to innovation from the outside, we will innovate inside, we will in fact benchmark everything, we'll be principled about it ultimately for better economics.

• 42:02Speaker 01

  Yes, because, because creating the efficient capital token factories that create human prosperity is the goal.

• 42:09Speaker 00

  Exactly.

• 42:09Speaker 01

  And enabling many companies to do it. So a different thread. One of the things I know from various conversations with you, you're very thoughtful on, is the topic of children in the modern age, because obviously we're creating products for tutors, coworkers, et cetera, et cetera. And one of the things we're all exploring together is, what should we be doing as an industry to navigate kind of being, of helping children elevate the right way and also keeping them safe? What are some of the principles that, you know, Microsoft's thinking about, you're thinking about as ways to kind of be good as regards, you know, the next generation of humanity through children?

• 42:51Speaker 00

  Yeah, it's a great question. I, you know, the, the, there's obviously important things that we have to do around child safety, right? So when any digital technology comes out, I think we have to now think as first class, right? There are AI safety issues around cyber, bioweapons, alignment, but it's also child safety. So I think that that one thing that we want to make sure is some of the challenges of current sort of set of, let's call it chatbots, right? And their conversations with children in particular is something that we need to be

• 43:30Speaker 00

  very mindful of and make sure that we are not having children not have that agency that they need to have in order to interact with this on their terms versus be persuaded, for example, right? So those are very important things. But the other thing that you're pulling on, which I think is super important, is what does it mean to even be a child in a world like this, where there is all this abundance of tokens? And that, I think, is the question, right? How does, should, what should, how should learning happen? How should we inspire children? What's their sort of ability to go have a new pedagogical system even that they can enter?

• 44:13Speaker 00

  Because I think even, like, if you take the traditional ways, right? The anxiety that, let's say, I, I had growing up on learning. Interestingly enough, that's a sort of an artifact of the scarcity of opportunities, scarcity of good learning, and a variety of things, which may or may not be true going forward. So one of the first things I think is creating a learning environment where students and children in particular, from the earliest of ages, don't develop these phobias for math or science or what have you.

• 44:57Speaker 00

  In fact, because everyone's by definition a lot more curious, it's just that contact with the world is what, you know, erodes that curiosity and confidence that children have innately. How do we develop that even further? By giving them the ability to explore, knowing that there's no anxiety, because after all, the expertise is always going to be there in abundance. It's really your cognitive coverage of that expertise that's more at a premium, right? That's, if I had sort of, somebody had said that to me as a five-year-old, I think I may have approached life very differently. And so how do we, as a society, create the necessary conditions for that? I think it'll be probably very important.

• 45:42Speaker 01

  And I think it's one of the things, you know, I know from conversations with you intensely, is like, this is one of our responsibilities as a tech industry, right? This is like, it's a, no, no, we can't abrogate it. This is like, given the ubiquity of the AI technology, we have to be responsible for the next human generation.

• 46:00Speaker 00

  Exactly. What both sort of the unintended consequences is something we think about from day one and really build in safety guardrails and what have you. Then the great advantages of new technology, right, have to be democratized. And then I think there needs to be structural change, right? I don't think you can say, education remains exactly the same and we value the same credentials. I don't know, I don't think so. Something's got to change.

• 46:29Speaker 01

  Yeah. So one of the things we haven't had a chance to talk about yet, which is very natural in this context, is Pope Leo's Encyclical, which I thought was actually, in fact, a magnificent part of leadership on the behalf, on the behalf of the Church. I mean, Pope Francis had actually gotten me engaged in helping them talk about AI ten, eleven years ago, right? So like, the Church has been amazingly front-footed on this. And actually, part of the thing that didn't surprise me at all in the Encyclical was the humanism of it, right? Because people thought, oh, it's about religion, it's about how you, how you pray. And it's like, actually, in fact, the, the, at least the parts of the Church I've interfaced with have been humanists, right? And do you have any reflections on the Pope's Encyclical?

• 47:12Speaker 00

  To see the Pope weigh in, and from what I understand, there's historical precedent on this, right? I believe the Pope, at the time of the Industrial Revolution, also had weighed in on the condition of labor and what should it do and what have you. And so to have the Pope sort of come out in defense of what I think all of us could deep, be deeply care about, which is human dignity and human agency in the age of AI, I think is so important. And I'm glad that he has advocated what he thinks is important. The other side of this to me is, what is the society?

• 47:56Speaker 00

  Like, if I look at what is at least describing, what, what, when I think about some of the greatest ways technological advances were harnessed to create great prosperity. The story of the West is pretty unbelievable, right? There's a beautiful book, in fact, that Joel Mokier and a couple of other authors have written called, it's called, I think, Two Paths to Prosperity, which describes the thousand year history of China and the West, essentially. And, you know, I'm paraphrasing here, but fundamentally, some of the constructs, cultural, societal constructs in the West, on how to use even the scientific revolution, the

• 48:40Speaker 00

  industrial revolution, this, and they, that necessitated a real change in how the society was organized, so that it could fundamentally take advantage of this new, is sort of really massive, right? That, in fact, defined the modern world. And so one of the things I feel we now need is a similar coming together from sort of both the moral philosophy. I think the Pope basically sort of said, this needs to be the moral philosophy that guides us going forward. You combine that with even, what is the market? What is the, what is democracy? And then what is the scientific slash technological revolution?

• 49:24Speaker 00

  So if we can somehow get into a virtuous cycle where the morality and the scientific sort of breakthroughs, the political system and the markets are all reinforcing each other, then we will have abundance and we will have many stakeholders all benefiting. And if we don't, we are going to lose social permission, right? So I just don't think, I think this is what the narrow understanding of the success of the West as just a technological breakthrough, I think, is sort of not the case, right? It's an unbelievable coming together of a multitude of sort of forces in a virtuous cycle.

• 50:08Speaker 00

  It also had really bad parts to it, right? We know that. And so you have to avoid that, right? Which is, and so that, I think, was also some things that the Pope himself wrote in the Encyclical, which I thought was great, which is even to think about sort of what are the bad parts? How do we not repeat it? What is the way to be, you know, advocating for this positive cycle? I think it's beautifully captured.

• 50:31Speaker 01

  And I thought it was great that it was kind of a focus on how do you keep humans at the center? How do you elevate human dignity? How do you address these, not just of the wealthy countries, but the entire world? Is I think, you know, a great, you know, beacon in terms of how we should think about it.

• 50:48Speaker 00

  You know, I think deeply about that, right? For example, I've always, always, you know, you know, been like someone who cared deeply about the global south. In some sense, finally having their moment where there can be real catch up growth. And so I think that there's a real danger now in the age of AI for even that convergence growth to slow. And in fact, go the other way. And so what are the, again, the global structures? Because by the way, if someone sitting in the United States in, you know, in Palo Alto may think that somehow it doesn't impact them. But there's no such thing, right? You know, we share this planet and our destinies are a lot more tied than we think we are. Just because of, you know, being far away from what's happening.

• 51:31Speaker 01

  Well, um, last question before we get to rapid fire. Is you and I both know, um, you know, in the US and in Europe, there's a lot of AI backlash. Now, part of how, as you know, I've, I've been trying to address it is like with books like Super Agency, trying to say, no, no, this is an opportunity to gain agency. The transition will be difficult, but embracing the agency and the transformation, ultimately you have to. But if you do it with forethought and, and kind of like leaning into it, it can be greatly helpful. What do you think we should be trying to help people in the US and the West understand about? Why it is that actually embracing AI is more important?

• 52:16Speaker 01

  And it doesn't mean that there aren't genuine issues in the backlash. And obviously work transition, all the rest will be real issues. But how do we help people see that this could be an important part of their future?

• 52:27Speaker 00

  Yeah, it's, I think, really, at least now I've come to the conclusion that, um, there needs to be real tangible, practical, well-understood, um, outcomes that speak for themselves. Because I think what has happened, and this is one of those places where, quite frankly, our industry even, you know, the way we have talked about it. I mean, when you go out and say, hey, you know, all economic opportunity will go away for knowledge workers or, you know, white collar jobs are gone or, you know. And, and you, and then you're saying, I'm excited about building that technology, right?

• 53:11Speaker 00

  Why would anyone want you to be successful, right? I mean, I don't want you to be successful. I mean, this just makes no, you know, no social sense. So I feel now, um, that when, you know, when you have someone in a college commencement be booed because they're saying AI, is it? You know, means we have now crossed over to people don't believe us. And, and rightfully so. So therefore I think what do we need to do now is time to do the hard work. The hard work is if you're building a data center, let's make sure that that community believes that this data center is great for them.

• 53:47Speaker 00

  It's for their tax base, for their community efforts, their real estate value, their schools, uh, their water use, right? Their electricity prices. It can't be again, like, oh, I said something, no, it has to be real, right? That's kind of what the way we earn social permission. Same thing, I would say with employment. We can't abstractly even say, hey, lump of labor fallacy, always there's going to be new jobs. What are the new jobs? What are the wages of the new jobs? That I can now go apply for, train myself for? And, and how do I really start a new company or what have you, right? I think unless we really get clear, or the third thing which we unpacked a lot. Every firm needs to participate in the frontier ecosystem.

• 54:33Speaker 00

  Oh, it's not like, oh, I'm just a feeder of data to some foundation model. Like that is like, talk about sovereignty and dignity, both being lost, right? Simultaneously, right, by, for countries, communities and companies. So I think now we have to go all the way and say, okay, this is a positive sum. This is in fact, the, you know, challenges of some of the technology. We are going to really actively work it. We can articulate the tangible benefits. And this is where again, what you will do at Manus and others are also going to be very, very helpful. Because the world needs more proof points that this technology is ultimately helping human condition in our societies broadly, not narrowly.

• 55:17Speaker 01

  Yeah, no, AI for humanity. So, uh, you don't have to answer rapidly, but we ask the same questions all our guests. So the first one is, is there a movie, song or book that fills you with optimism for the future?

• 55:30Speaker 00

  It's this, uh, parallel paths to prosperity, all right? Because I, I, the reason why I like that is because it's a good call to action. For the next thousand years, what's the path to prosperity? There was a blueprint, um, at least at the part of the world got it right in the last thousand. Can we now as a, an entire planet get it, uh, right for the next thousand? I think that this is where, uh, some of our very best work has to be done.

• 55:59Speaker 01

  Agreed. What's a question that you wish people would ask you more often?

• 56:05Speaker 00

  That's a good, I would love for people to ask me, what am I not excited about? Because I'm excited about a lot of things, uh, but I'm not excited, for example, about us losing permission on AI by, you know, saying all the wrong things or doing the wrong things even, and not having a, a complete thought on how to truly have a positive sum construct.

• 56:26Speaker 01

  Yep, a hundred percent. So where do you see progress or momentum outside of your industry? And given that Microsoft powers a lot of the world's industry, um, but you know, uh, it could be robotics, AI, et cetera. But where do you see progress that inspires you?

• 56:42Speaker 00

  I mean, to some degree, you know, the, the, the work in bio, I mean, if I think about it, right? What are the most complex systems, um, that we have to have a better understanding of is human biology? So anything, any tool that can help humans take care of humans is probably the thing that will have, uh, it's, it's awe inspiring. There's this one thing, uh, we, I recently came across, right? Which is this, for immunotherapy, I believe there's a test, um, that is, uh, you know, basically a, a complex costly test that figures out whether a particular immunotherapy will work on that tumor or not.

• 57:28Speaker 00

  And so Providence and some researchers at U Dub and Microsoft research came together and built this thing called Gigatime, uh, which is a cool model, which basically is a simulation of that test and reduces the cost of what could get done only at sort of, you know, for, it took a lot of time and a lot of money, now can be done by any tertiary sort of hospital in a, you know, in any city. And that type of economic sort of availability of sort of medicine and, and, uh, and medical practice, I think is just breakthrough.

• 58:02Speaker 01

  So last question, can you leave us with a final thought on what you think is possible to achieve? That everything breaks humanity's way in the next 15 years and what's our first step?

• 58:12Speaker 00

  I think if everything breaks our way, I've always gone back, like, what's the dream? Of the world compounding at 10% GDP growth, right?

• 58:29Speaker 00

  That is what can happen, right? I mean, one thought experiment is if the industrial revolution had reached all corners of the world at the same time and every country could express their comparative advantage fully, right? That's the maximalist positive sum construct. Can we do it, right? Because we are kind of captive to this, you know, what we will describe as, hey, that's not how the world works. History is about dominance and dominant powers and, and I'm not saying we will defy all that. But the bottom line is, since you asked me to dream, I'm dreaming that if humanity can get back past their bounded rationality

• 59:13Speaker 00

  and only sort of say, oh, history repeats itself and we're only never going to get better than, you know, I know, fighting these wars and, and what have you. And say, no, no, no, let's maybe, uh, we can change the course. Then the first step would be to accept that that possibility exists is what I would say is what will get us down. Otherwise we're going to back to relitigating, oh, we are, let me look at history and then just try to match.

• 59:40Speaker 01

  Beautiful dream. Satya, always a pleasure and I'll see you next week at the board meeting.

• 59:44Speaker 00

  Thank you so much, Reid. Thank you so much.

• 59:47Speaker 01

  This is awesome. Yeah. Such a fun. Yes. Conversation. Yes, exactly. Possible is produced by Palette Media. It's hosted by Ari Finger and me, Reid Hoffman. Our showrunner is Sean Young. Possible is produced by Tenasi Delos, Katie Sanders, Spencer Strasmore, Imo Zoo, Aman Suri, Lexi Kevin, Danny Garrison, Trent Barboza and Tafadzwa Nimo Rundwe.

• 1:00:09Speaker 02

  Special thanks to Satya, Yala Manchili, Sayida Sabieva, Ian Alice, Greg Beato, Parth Patel and Ben Ralas.