我们需要一个 AI 生态系统，每家公司都能在其中赢得一席之地

The world is going to be very skeptical of tech and tech companies that say, “Trust us, we’ve got it. The future is going to be glorious.” You kind of have to deliver tangible benefits because it’s too important this time around. It’s too much of the economy for it not to be the case.

True ambition is about making the impossible possible. I take great inspiration from the people who are managing the Azure network. We built, in the last 15 months, more Azure capacity than we built in the first 15 years. I mean, it’s crazy—wild. Our job is not to do Azure networking. Our job is to build the agentic system that does Azure networking.

The way to get to information, the way to educate yourself, the way to continuously keep yourself updated has changed so much. Maybe the next big startup could be someone who builds a new university, a new pedagogy even, of how to get someone to go through a curriculum and find economic opportunity that’s highly valuable.

I’m so excited to be here.

No, thank you so much. It’s great to be with both of you. I listen to both of your podcasts all the time. It’s great to be on it.

Thank you so much. So, you were talking about these amazing announcements from across the Microsoft estate all morning for, I think, 3 hours. What’s the most important reflection or takeaway you have?

I’d say perhaps the biggest one for me is let’s conceptualize this more as an ecosystem play, as opposed to a single model or even a single platform. At least for me, having grown up at Microsoft and having seen 4 major platform shifts, I sort of fall into the camp where a platform is defined fundamentally by its ability to create more value around the platform versus what’s captured in the platform.

And so, if you view what’s happening right now, I think this morning’s keynote was about how any company, whether it’s an AI-native company or a traditional enterprise company, can participate as a first-class participant where they can point to AI they create. It’s not that they don’t use other people’s AI. Of course they will. But to me, what’s the path? What’s the recipe? How do I do it? What does the stack look like? What does the tooling look like? What is valuable? How do you do that? That’s our job to do.

Ecosystem strategy is very complicated, right? Because you end up building certain components, partnering for certain components, supporting them. You just announced this big suite of models. Tell us a little bit about the training strategy for Microsoft.

The thing that we wanted to do with the MAI models was to build, as Mustafa talked about, first of all, a great lineage. Starting with pretraining with very good data quality, doing all the ablations, making sure of the quality, because, in some sense, it’s become even harder to build a clean-lineage model. There’s so much stuff out there that you truly need to ablate out to be able to have a fantastic pretrained model.

In fact, that’s one of the challenges of a lot of the open-weight models: They look great on 1 benchmark or 2, but they’re not great in practice. That’s why, in fact, even in the RFDs, they’re pretty good. I’m really excited about these MAI models, because how the heck can a small 5B model hill-climb?

It goes back a little bit to what I think is ultimately the key thing to do, which is to pursue finding that cognitive core. To me, it starts with a clean lineage, then creating the ability for companies to be able to use this—not just as a generalist, but to create their own specialist by building this hill-climbing scaffold around it. It’s not just the model; you have a hill-climbing scaffold around it.

Then you will start building your RL. You will start collecting the traces. Most importantly, you’ll have private evals, because we know all the evals out there are good and interesting, but they’re not really that critical at this point because they all can be maxed. The point is that each company will have its own private eval. That end-to-end platform story around our models is what I think is interesting.

The one other thing, Sarah, since you brought that up, is I do feel there’s a new frontier. People talk about the frontier, and you’re operating at the frontier. Interestingly enough, if you add a little temporality to it, you can use—for example, the Land O’Lakes demo we showed was pretty cool. We used GPT-55, right? Then you collected a bunch of traces, and then you took a 5B reasoning model and achieved higher. So, that’s another aspect of what it means to operate at the frontier.

I think, first of all, I have to congratulate you on basically building a frontier neural lab inside of Microsoft in 2 years. I’m wondering, you have all this AI strategy that you’re rolling out. What do you know now that you wish you would have told yourself 2 years ago—or 3 years ago? 3 years for the Jensen partnership, 2 years for MAI.

I think the thing that I reflect on quite a bit is that I got into all this when I got excited by the scaling laws paper. The OpenAI partnership came about when those folks said, “Hey, we’re going to really throw a lot of compute at transformers.” And they’ve helped.

The thing that I always look back and say is, “Wow, these things do have capability that they’re climbing up with.” This crude way of saying it is, “Intelligence is the log of compute.” It kind of works. Now, what I think we underestimated, perhaps, is the real-world complexity of deploying these so that they actually deliver value in the real world.

The outcomes as measured by any benchmark are interesting and important, but the true eval is when people out there are able to do unique things that only they can value. And it’s very measurable. I wish we had even had more of that in our consciousness, because, as an industry, when people say, “Wow, I don’t want a token max,” it’s an artifact of us not having thought of ourselves as an industry that we are using tokens to create value every step of the way. I think that’s what I wish we had gotten there, but I’m glad we are here.

What are some other use cases that you’ve seen that have created the most value for your customers? I know that people talk a lot about code, and I think it’s pretty clear that that’s something that’s having a very large-scale impact. Are there other areas that you find in common where your customers are really benefiting?

To your point, obviously coding is now good, but it’s interesting, by the way—you love to even talk about the coding. Coding is working so well that we now have to rebuild the IDE. It’s kind of nuts to see what we saw. It’s like, “Oh my God, I have these 100 agent sessions. The cognitive load it transfers back to me as a human is so excessive that now I need a new UI.” By the way, chat as the only artifact is also impossible. That’s why we need a canvas.

One of the things that we are starting to see—we started seeing it with Cowork, but even in some of the work we showed with Copilot, and what you see with Claude is a good example—is that a lot of human capital is doing the glue work. If you can now augment that with tokens and agents that are long-running and durable, then your ability to scale even what is still judgment and glue work gets amplified, just like coding does.

I’m positive that 6 months from now we’ll all be saying, “Oh, wow. All through the night, there was a bunch of stuff that all these autopilots that I have working on my behalf, with my delegated authority, so to speak, and with even my identity, did a bunch of work.” Then, of course, I’ll need my new IDE to say, “What did you do? Did I do this work?” I think that’s where the compression of workflows and completion of tasks is where a lot of the value gets created.

You raise a really interesting point, which is that there’s the actual agent doing the code, and then there’s a harness around it. That’s the environment, the context, and everything you’re setting up as a developer around an actual coding agent. What is the harness for the enterprise? Is there an equivalent concept for broader productivity work, or how do you think about that concept generally?

That’s right. In some sense, you kind of want the harness to define the models, the data, and the tools, so that you have a loop across those 3. What we are trying to do, first of all, is make sure that each of our products that we build—whether it’s GitHub Copilot, Security Copilot, the stuff we showed with M365, or even Discovery for Science—are multimodal harnesses with tool access. That way, you can do this progressive disclosure of tools, even so that they’re token-efficient.

So, in our case, we have the GitHub harness, which essentially we're using across all our products. It's available in Foundry. We're open: you can use your Llama harness, whatever, or you can use any open harness or any harness of your own, and train with your tools, multiple models, and your context.

That's the pitch, because right now a lot of the dialogue is, “Hey, if I train the harness, tools, and the model together, you get evals.” What we are proving—and the best example of that is what we did with M365—is that when it launched, it found bugs or vulnerabilities that were not found by Mythos [?]. So there is an existence proof, I would claim, that you can have a multimodal harness that can, in fact, be more performant in the real world.

The premise behind training at the independent frontier labs is really, “We're going to have these models, and we'll have an API business, and we'll support enterprises and startups, but a first-party product, be it productivity, code, or search, drives the majority of revenue.” That's a different value equation than you're describing. I think with the Microsoft ecosystem, if that's the case—tell me if it's the case—because obviously you have first-party products and enablement products, what is the role of the developer? What's going to be hard, what is the set of skills, and what is the value capture for the developer in that world?

Yeah, so I think there's always going to be a case that someone who's super successful and is a platform builder can also have first-party products. It was true with Windows; it was true with the SaaS side and the cloud side as well, with us and others, and so on. But the thing is, it should not be a limiter to other people achieving that same success, right? That, I think, is the core difference: the network effects this time around, around intelligence, are such because they learn from data, and not really lots of data—it's just a few samples that you have to see to understand what's novel about something.

That's why I would say every company having private evals may be the biggest IP. I think about it like: what's that private eval that you can then use even a frontier model to hill-climb on without leaking the traces? Maybe that's one of the biggest drivers of IP. Another acid test is: you have a private eval, you're using Model A; can you switch it to Model B and climb up? If you can, then you're in control. If you can't, you're not in control. And that's where even the harness decision becomes super important, right? Therefore, having an open harness, letting all models come in, and having your evals, your context, and your tools help you hill-climb—I think those are the skills that an AI-native startup needs, a SaaS company needs, or every enterprise needs.

Yeah, I think in a very real way, you're Microsoft—historically as an operating systems company and then becoming a cloud company—maybe the third act is that you're a harness or evals company, whatever the conglomerate of concepts that you want to put together. I think enabling every company to have frontier intelligence—or whatever the exact term was that you used—is the mission, right? That is the platform promise: if you build with us, you will get your intelligence for your data.

That's it. To me, if there was one tagline for this entire developer conference, it's, “Can everybody operate at the frontier with their frontier intelligence?” To me, that is so important, because otherwise I don't know how you achieve stable equilibrium, right? How do I then go and say, “Wow, my company is going to have a terminal value because I now know how to continuously compound on top of a platform that gets better”?

When Windows came out, Adobe built on it, Autodesk built on it. Or even take what Jensen said: “We built DGX.” And he built CUDA on top of it. I always say to Jensen, “God, I got the short end of that.” I wish we had recognized it. Nevertheless, that idea that you can build a platform layer that someone else can then extend and build their own intelligence layer, in this case, I think is everything, right? Without it, why have a developer conference? I can just come and have you all worship at the altar of one model. But that's not a developer conference.

Backstage, we had a discussion about what is IP, or what the value in a company is. It used to be the length of human experience at a company, and now it's this other thing, which is the evals—the experience in applying agents to the company. I just want you to flesh that out a bit more, because—

Yeah, it's a great way to frame it, right? At the end of the day, every company is going to have both human capital, which is still going to be super valuable, because humans and their ability to find the gaps that exist at all times are going to be the way we all create value. I'm definitely in the camp that this is going to be about expressing new forms of human agency and ambition, even as token capital goes up. Let's say any corporation has lots of tokens and a lot of human capital. The question is, how do you compound the two?

If you take teams, I have a bunch of agents doing work and a bunch of humans doing work, and the traces between those are really important context for how that enterprise is creating value. Then that goes back to training not a generalist model, but the company-veteran agent. That is super valuable, again, because when a company goes and says it should, in fact, go onto the balance sheet, that's how I think about it.

Human capital was never possible to put on a balance sheet because you didn't know how to capture the tacit knowledge. Whereas now, I think you can, with the agents that have learned through time, through all the traces. So that's what, at least, we think will happen.

I think the SEC is going to have to have accounting standards for token expertise.

You're talking about the equilibrium state and a stable equilibrium where companies have this compounding value and can see terminal value for themselves. Another challenge to that equilibrium is that there are applications and workflows that are common to a vertical or a horizontal. This was the generation of SaaS companies, and Microsoft has lots of SaaS properties as well. Then there are things that are very specific to every enterprise, which they differentiate against.

I'm sure you have heard and participated in much of the debate about the end of software, because all these workflows are cheap to generate now. Do you think the equilibrium looks different between what agents get built in enterprises versus in their vendors in the future?

Yeah, so I think what's happening there is that we had a particular way we captured, I would say, workflow in apps, right? We built up a data model, right? We schematized some part of a business process.

Mhm.

We then built a bunch of business logic—

Yep.

—and then we put a bunch of UI on top of it. Right? So that's kind of what every SaaS company—

A little configuration.

For 20 years, that was—

And that was it.

So, interestingly enough, now you kind of get to relitigate that vertical stacking, right? I still think, for example, that the data model you build underneath every SaaS application is super good. Why reinvent it? My general ledger better be a general ledger. I don't need new schema creation. In fact, that entity relationship is actually a pretty good, robust thing that I want to feed.

And you want that to be stable.

That's right. Then same thing with business logic. If you look at it, we have this product called Power BI. It is dashboards galore that people created. The beauty underneath those dashboards is a very rich semantic model. Someone took the pain to create a dashboard and do all the measures, and you want that—that's business logic—to be available to you.

So I think the challenge of the SaaS business model is that we packaged things one way. We now have to learn how to unbundle these things and rebundle them in new ways and discover new business models. If you look at what's happening today with Microsoft 365, it's a great example.

We have this thing called Work IQ. In fact, what we are realizing is, oh my God, if you look at it, there's a historical parallel too. We sold first Exchange and SharePoint, and before Teams we had a thing called Lync Server and what have you. We thought, oh, that's all going to move to the cloud, but little did we realize that the number of people who would use servers in the cloud would be 10× or 100×, because people were not buying servers; they were just buying a subscription.

The same thing is now happening with Microsoft 365 because with Work IQ we have exposed what was perhaps the most important database in a company that never got used as a database because it was only captive to our apps, right? It is where all email operated on it, Teams operated on it, Word, Excel, PowerPoint, and SharePoint. But now, this is one of the coolest things I get to do with Work IQ: I go to a GitHub repo and say, “Hey, I attended a bunch of design meetings last week related to this repo. Can you capture all that and tell me what changes I should make?” Think about that. It literally can go look at all those transcripts and come back with a plan to change a codebase.

Previously, you could never have thought of using Microsoft 365 for something like that. So the value-creation opportunity now in the agent world is in fact 10x more. But it does require us to, for example, rearchitect, because there's going to be usage around Microsoft 365, which is going to be perhaps more than even the end users. What I used to serve an inbox or a mailbox cannot be used to serve an agent. That's sort of what we're doing.

I don't believe in permanent business models for any of these domains, but in the near term, do you have a prediction between outcomes-based pricing, token-based pricing, and enterprise bundles?

Yeah. The way I think about this is, we've always had—let's even take per-user pricing. Per-user pricing is really an artifact of someone creating a budget and needing certainty. Somebody wants a budget, and they need a per-user price.

Mhm.

Per-user pricing is just a set of entitlements to usage. That's kind of what it is. So, the way I see it, the first bundling will be to take some usage, bundle it into per-user stacks, and then sell subscriptions. Subscriptions are going to be there, and per-user pricing is going to be there.

Then the next big thing will be consumption. People will say, "I want consumption." It's also possible that people will say, "I don't even want to pay for any of the subscriptions or the consumption. I want an outcome." But remember, most people love outcomes until they have an outcome, because once you have an outcome, it's like giving away royalty.

I've talked to customers who love outcome-based pricing, and I say, "I'm all in," until they say, "Oh my God, what are you talking about? You're sharing in my outcome. No, no, no. I want you to go back to per-user pricing, and I want you to use consumption pricing." So I think that debate will go on.

All of these business models have a particular time and a place, versus one to rule them all. If you're a SaaS vendor or a platform vendor, having that flexibility is important. Quite frankly, we face this with GitHub. We recently announced per-user pricing on GitHub because, little did we know, GitHub Copilot was constructed at a per-user level before we understood even the intensity of agent usage.

It was an interactive way for a developer to use code completion, maybe tasks. It wasn't like, "I launched 10,000 agents that are going on all day." So that's what the adjustment is about. There will always be a per-user model, but there will have to be a consumption meter.

How do you think about the durability of SaaS more generally? One thing I've observed is that, in a lot of enterprises internally, there will be teams that almost have agent euphoria. They're so excited about the explosion of things they can build that they're trying to rebuild a lot of applications, or going to their SaaS vendors and saying, "We're not going to work with you anymore," or, "We're considering an internal project."

It seems like in 6 to 9 months, maybe some of those people will come back and say, "Actually, we can't rebuild everything." How do you think about what's durable in this world and what isn't?

I think we have to go through 1 full budget cycle on this to really see the emergence of equilibrium. At the end of the day, there's a marginal cost to even generating the app. A simple way to say it is that you should always acquire something if the marginal cost of building and maintaining it on your own is higher. That's a quantifiable thing.

The maintenance part is important. You've got to remember that all the security issues that AI finds, you better fix them too. Of course, there's a coding agent to help you with that, but then that burns tokens. So whose responsibility is it? It's kind of a cycle that you've got to think through.

I think we've gone through the excitement that I can generate a lot of software. The next thing will be: What software do I really want to generate? What software do I want to use from others? How do I compose these 2 into some agentic workflow that I have agency over?

There'll be very little tolerance for anybody who is inflexible at the vendor level. But at the same time, anyone who has that flexibility, shows up, and delivers the value will be back at it again. We're selling software, but we're just using different business models, in fact.

Speaking about building software, one of my favorite moments from a previous Build, maybe 1 or 2 years ago, was a section where you were building your own software. I'm curious if you're building anything now.

Yeah. First of all, let's face it: building software has made it possible for even the incompetent CEO of a company like ours to build. So thank God.

That said, I do feel that something like GitHub Copilot—to me, especially the new Sessions app, or the new app—has made it so much more possible for you to have agency over artifacts that you felt you couldn't touch before. For me, as a CEO, even being able to go to a codebase and learn about it is powerful.

I remember joining Microsoft a long time ago. Everybody had to go in and look at whatever Cutler's, Malik's, or what have you, to learn how to write good C/C++ code. Now, the ability to be more full-stack, up and down, is so good. But that doesn't mean every one of us should be doing the same thing. The question is, how do you then have the ability to inspect things, learn things, and see things? I think it's just so much more.

What I'm building a lot of is these long-running Foundry agents. There are autopilots. The easiest thing is, I think I just built one even last week where the idea was: Can I have an agent that is continuously monitoring—essentially my own chief-of-staff autopilot? We're going to have that, obviously, in Scout. That's what we showed.

But it is so easy and trivial to build. I took Work IQ and said, "Take Work IQ, go and build a Foundry long-running agent. Store all the memory using Raven, basically as my backend as a service." And lo and behold, it built it. Not only did it build it, I could say, "Publish to Teams," and it published the damn thing to Teams. The ability to have some end-to-end project like this completed is pretty miraculous.

Do you think that impacts the different types of engineering roles that exist in the future? Right now, I think there are a dozen different types of engineers you can be, from QA to front-end engineering. There's a big swath. I've heard some people argue that in 4 or 5 years, we'll basically end up with 4 engineering roles: people who are managing agents; forward-deployed engineers, or FDEs; security engineers; and people working on large-scale infrastructure for a small number of services. Everything else just collapses into the agentic world.

Yeah, you think that's a correct view of the world?

Yeah. I think we'll have to experiment our way through it. But what you said is that there are some things that are very much at scale. At LinkedIn, they did structurally change and basically built up a new discipline called full-stack builder.

They said, "Let's bring people from design, product management, and front-end engineering and put them all together." But they also have an edge. It's not like the design person no longer has the design edge, or the front-end person no longer has the front-end edge. But you can give yourself bigger scope in the role so that you're not confined to one role.

Equally, infrastructure has become very critical. One thing we've realized is that even for the Excel team, for example, building the RL environment in which a reward can be learned is actually one of the hardest infrastructure problems. So you need new talent—distributed-systems people—even in what was considered an end-user app team, because it's a different skill set.

So, yes, infrastructure science is the other one, obviously. We'll see how these evolve. The world will always have a bunch of specialists.

I think the generalist role is going to be the most exciting, because the leverage of a generalist is where we're going to see the maximum returns. When you said, "Hey, I code," I'm now a generalist. What I basically translated was knowledge work, which I did when I created a Word document or a spreadsheet, and now I can build an app. It's in the same sentence.

That idea that, "Oh, wow, my generalist skills have gotten higher leverage," is what I think we're going to see across the board.

Music to the ears of CEOs and VCs who are a little dangerous and a lot of fun.

Golden age for idea people.

Idea people with a lot of agency. If you take that idea of personal agency and just zoom it out to the organizational context, my partner Mike Vernal, who actually started his career at Microsoft, just wrote an essay where one of the big takeaways is that it's an age where you can be much more ambitious, and you need to be, given the pace of the environment and how quickly users and companies are actually open to adopting new technologies.

How do you think about—I feel silly asking this of somebody running a trillion dollar plus company already—how do you think about how Microsoft can be more ambitious now?

It's a great question.

I think the thing in these types of transitions is to have a conceptual model of how work can change to go after outcomes that you could hardly imagine previously. In fact, Kevin Scott has this nice line: when you're making hard things easier, that's sort of one point of leverage, but true ambition is about making the impossible possible. So now, the thing that is missing a little bit in all of our organizations is: What is that new conceptual model of what we can build? What was impossible, and what can we build?

I'll give you one example of this. I take great inspiration from the people who are managing the Azure network. This was even last year. We were scaling. You saw that I talked about how we built, in the last 15 months, more Azure capacity than we built in the first 15 years. It's crazy. It's pretty wild. And it's the same team.

They saw that and said, “Bob, this just ain't going to work if we don't reconceptualize our work.” They essentially said, “Our job is not to do Azure networking. Our job is to build the agentic system that does Azure networking.” These are the folks managing the 500-plus fiber operators managing the WAN all over.

Fiber operations ultimately is a physical operation. Things get cut, things have to be repaired. We have fancy words called DevOps and so on, but basically, emails are coming in and you have to respond to them and take care of it. So they built this agentic system. They even have a character for it. It's called Miles, and it does all this stuff.

They started screaming for more tokens and saying, “Look, we don't need headcount. We need tokens in order to be able to manage our operation.” That reconceptualization of what their work is—they basically took their work and made it meta. That meta-work is now their new work.

Mm-hmm.

Right. In the ’80s, if somebody had come to us and said, “4 billion people are going to get up in the morning and start typing,” my model would have been, “We need 4 billion typists.” But we're not doing typing. We're doing knowledge work.

So that, to me, is it. Whether it's Microsoft or any organization, we have to give ourselves permission to do new types of metacognition and meta-work using these new tools to change the outputs that matter, and then really make the impossible possible. Completing that connective tissue across those things, I think, is where a lot of the enterprise value will get created.

So, you talked about the data centers?

Yeah, please ask.

Oh, okay. Well, this leads nicely into the data center buildout. I'm always impressed at the sheer scale of the buildout from Microsoft, but also everyone else. This is redefining what it means to be a hyperscaler. It feels like unprecedented scale in terms of the finances, the way you run the company, and also the communities that are impacted. Talk a little bit more about what you're seeing on the ground.

I think there are 2 aspects of it. Obviously, the buildout is extraordinary. Nothing like this has happened, and it's great to be one of the participants in it. But you brought up the other part. At this point, it's clear that unless we as an industry are very principled about ensuring that the benefits of all the stuff we're talking about are felt in real ways at the community level, this is not just a campaign. It has to be real.

People have to be able to say, “Look, this is not changing energy prices for me. In fact, if anything, it's bringing down prices because, long term, there's going to be a better grid. There is going to be more energy.” Water consumption is, in fact, not an issue; water is being replenished. You have to really educate folks on what's truly happening in the closed-loop systems we're building.

We have to invest in the training, the jobs, and the tax base. In fact, the least-talked-about things are the amount of jobs that get created during construction and after construction, and what tax base is there in the community. All this has to be real. If that is the case, then we will have permission. If it is not, we won't have permission. It's as simple as that.

We have to take it as an industry pretty seriously. I think it's good for communities to be skeptical and ask the hard questions, for us to do the hard work, and earn that. But at the end of the day, if we can really be the—I've always felt like, in human history, if you use a lot of energy but also create a lot of value for society, the story has been fantastic. If you don't do that, it's not been that great.

This time around, I'm a firm believer that, ultimately, if you do have an token economy that drives productivity, drives economic growth, and drives widespread participation and better health outcomes, then I think we will be in a great place. That's at least what we all have to be focused on.

Yeah. It makes me think, actually, that with all these initiatives that you're doing, it might be easier to see ROI in the communities first before in enterprise.

I think both sides. In fact, it comes back together. The people in the communities are going to be employed and are going to be participants in the real economy. That's the question.

If the broad economy is doing well and the communities are doing well, the dots get connected. The market forces are such that we will connect the dots. That's it. You have to be able to see the evidence. It can't be about any one company, but it has to be broad economic growth and broad community permission.

What have you most updated your thinking about regarding the societal impact of AI?

I think the most critical thing is the first question we even started with: We need to tell the story and make it real that everybody has a real shot to participate as a first-class participant in this new economy.

In the next 12 to 18 months, we need a way for people to say, “Oh, wow, I get it.” There’s going to be tremendous capability and a tremendous amount of infrastructure, but I can see what is going to happen, whether it's the benefits, like health outcomes, my ability to create a startup, or my ability to run my local store more efficiently. It's just happening, and I see that benefit myself.

To me, that's earning permission in a path-dependent way. We can't wait. The one thing I've now learned is that I think the world is going to be way more skeptical of tech and tech companies that say, “Trust us. We've got it. The future is going to be glorious.” You have to deliver tangible benefits. Frankly, politicians winning elections because they have advocated for that—that will be at least my adjustment, because without it, thinking that somehow it's too important to summarize—it's too much of the economy for it not to be the case.

One very simple framework I have for the broad benefits of AI, beyond the communities just working in technology, is wealth creation. It's going to happen in a ton of different companies, startups, and large companies. Then you have health care. You had amazing demos today. There are companies like OpenEvidence, and I think that is happening.

Education seems like another obvious good where we haven't seen as much impact as I would expect. Do you have a hypothesis on why that might be, or if it'll come?

I think this is where we have to rethink how we think about education. Recently, I met with the founders of Alpha School and learned a lot about what they were doing. It is fascinating to listen to how they rethink what education really looks like, because I think it's actually very important.

I'm not saying anything traditionally being done is less important. I was even looking at the assignment guidelines for a Stanford CS class. It's fascinating to see. They were making sure people were learning how to apply softmax appropriately versus saying, “Hey, fix my training run.”

I think learning concepts is important. It's going to be critical. But the way we create the incentives, what the credentials are, how we value those credentials, and what the employment opportunities are for those credentials—that has to completely change, given that the way we get information, the way we educate ourselves, and the way we continuously keep ourselves updated have changed so much.

Interestingly enough, maybe the next big startup and success story could be someone who builds a new university, or even a new pedagogy for how to get someone to go through a curriculum and find economic opportunity. That could be highly valuable.

Yeah. That has felt perhaps impossible for a long time, but it's a great note to end on and something that might be possible. Thank you, Satya.

Thank you so much. Thank you. I appreciate it. Thank you all.