You're actually getting a little funny. Yeah. He's really posting the, uh, like, ass in his face.
Oh, ho, ho. You're not letting him put, put, put. Yo, let's actually play. You're not letting him put, put. See you in a year. Wait. Yo, work it out! Yo!
I was a streamer in 1995, and he said.
And he also said that if he brought food, he'd be like, no, wait, what did Truman do in 1945? Please take your seats. Our program will begin shortly.
I love you.
Thank you.
Please take your seats. Our program will begin shortly.
Please welcome President and CEO of Y Combinator, Gary Tan.
How are those after parties and how's the feel, how's the vibe, like who here, who here is, this is their first time in San Francisco? Okay, for you guys, I want to say, you got to be in San Francisco, you got to move here. The energy is totally different, so just, we'll start off the day with a quick sales pitch. San Francisco is the place to build the future and we hope you, but with that, we got to keep rolling.
More hard-earned lessons from builders, insights from the front lines of AI and many more chances today to meet the person that you might start a company, right here in this audience. So, let's get into it. It's my pleasure to welcome the chairman and CEO of Microsoft, Satya Nadella.
This is the home ground. Alright. Man, Santa's just there. You should move to Seattle. I started my career in Seattle. There you go. It's a fantastic place. Anyone who's successful starts at Microsoft. That's right. So, Sachin, you've emphasized before that AI is going to shape all that we do. What does this look like in practice? You know, at Microsoft, how does this actually drive your strategy and particularly thinking.
about how AI can influence ideas beyond, you know, the immediate, incredible products we make? Yeah, I mean, the way I think about this, Gary, it's... At Microsoft, I feel we're a platform company, a product company, a partner company. I think of those three dimensions. And I've kind of, in my 35 years, I've lived through client-client server, web internet, global cloud.
This is the full. How I happen. So the first thing that I think about is the platform. When I sort of look at all the folks here, the interesting thing is the compounding effects of all these platforms. So there's AI. The reason why I think the rate of difference is so, well, yeah. And so why? Because it builds on the previous generation. I think about, like, if the cloud was not there,
we wouldn't have been able to build the AI supercomputer, which then led to the models, which then led to the product. So that compounding effect, so that's why you always sort of take the previous platform and build the next platform, you want to be able to get that product. And then you've got to build the next generation products in each one of these platforms. There's a new workload, right? I mean, when I remember looking at the large-scale training job, it's kind of like a workload.
What we build, for example, the cloud, right? It's a data-fabulous, synchronous workload, which is so different than, let's say, a 100-minute job or what have you. And so the platform itself then completely changes. So to me, that's, I think, the exciting thing on the platform side is, over the age of systems software, tech, today, if I had to think about anybody who's built into the structure, not just the hyperscapers, but even the startups, obviously, there's a tremendous opportunity.
in the models and the products of products. Think of these, and then ultimately, what's it? It's for one thing and one thing both, which is to drive ultimately economic growth and GDP growth. So if I have to ask you, my benchmark for AI, is it creating surplus in the world around us? One community, one country, one industry, one company at a time. And it seems like the app level, you guys have built, you're sort of defining.
the app layer for so many decades. It feels like we're in this weird lumpy moment, where maybe the models have popped up, we're sort of punished by what's happening, but then sort of the compute and the apps need to actually catch up. The hope here is actually the people in this room will build those apps. Yeah, I'd say it's a good question, right? One of the questions is, is the model. Like SQL or is it the SaaS app itself and the model?
I think the place where does the model end and where does the product begin because if you sort of say model with some scaffolding and tool calling in some infinite loop is the product, if that is what it is, then I think that's where it gets a little confusing. But that's like saying a bunch of SQL big business logic is with SQL is what is an app. So I think it's still possible for anyone to build an app on top of a model and you.
have to sort of abstract yourself and say, yeah, the model is just like SQL was. So I think that, I mean, I always dream of a moment when AI slash machine learning will have a SQL moment because if you think about it, right? We never have a stable model. We never have a stable platform there. Because everything was vertically built and integrated. For the first time, this model layer now, we have something like a SQL engine, that we can then use to build pretty sophisticated products. And these techniques also, just the inference time compute plus tool calling.
is giving us, I think, a pretty robust harness to be able to build pretty sophisticated products. It's kind of wild how much it's the integration piece that is also happening. It's just the models sitting on their own are incredibly smart, but right now they feel there's a giant gulf between that and the data that really matters to your business. Yeah, I think that's a good observation because I think, at least my read of the situation is,
the model is an important piece. The model scaffolding and all this tool calling. So there's a real app server that you kind of have. need in order to be able to build a sophisticated application but the interesting thing is the feedback loop the data path inside the product that then is used in order to post train in order to be able to do the right tool you know selection um that seems to be the place where product is all over i know ai scaling laws are continuing to hold and the demand for intelligence appears.
to be potentially infinite uh yesterday 99 hyper intelligent beings to one human wild prognostication but seems possible given this where does the building for the future of ai truly demand for global compute infrastructure how do you anticipate these demands evolving as models don't just become larger but more intelligent and capable of complex multi-agent yeah i mean look if.
sort of really step back and say you know first if you sort of go with the compute or intelligence is one of the log of compute and then you ask the question how much energy does compute consume let's take in the United States maybe 2% today 80% drops let's say it doubles it's 6% that's like massive because then the amount of extra energy that needs to be produced in order for AI I think that's.
why we all have to sort of keep in mind that if there's one lesson history has taught us is that if you're going to use energy you better have social permission to use it so that means you've got to make sure that the output of this AI is socially useful in other words if we really are not using energy we are not going to be able to use energy in the future so that means you've got to make sure that the output of this AI is socially useful in other words if we really are not using energy we are not going to be able to use energy we are not going to be able to use energy in the future so that means you've got to make sure that the output of this AI is socially useful in other words if we really are not, creating social surplus, economic surplus, as measured by countries and communities that we just can't consume.
So that to me is the bigger thing, like everybody is today all bothered about, okay, what do I do about energy? The real question in the next five years is we've got to produce enough products that are creating great value, which I'm very confident about, by the way, in healthcare and education, in productivity. So there's many, many domains, but that's the real challenge for us as a tech industry, is to prove unequivocally that what we have created is showing up in real stats that is not just an AGI or AI benchmark.
I mean, the hope is that this will show up in sort of the real things that you sort of interact with on a daily basis. 100%. You know, you go use, you get a mortgage loan and instead of, you know, two or three months or two months. waiting around you don't know if you're going to get approved or well you know there's just so many things that are important parts of your life that you'll get drowned in paperwork bureaucracy that those things could potentially go away 100% so I think yeah even take some of the the public services right I mean.
if you don't take any country you know it's GDP or taking health care they got in the United States what is it 80 90 percent of our cost is on health care and a lot of it like everybody talks about the magical drug except all the cost is in workflow and so if you really take something like a simple thing like discharge the amount that you take it the back end of an EMR system with just an LLM and a prompt that itself is going to save so much time and money and energy.
that would sort of pay for itself kind of very direct right, There's an incredible amount of GDP on healthcare, rightly so, but every dollar that's spent on clerical work could have been spent towards some sort of treatment that would have saved someone's life. Or the simple time allocation of a physician away from paperwork to the patient is right there. So what do you see as being the biggest rate limiter for AI deployment?
You know, I think it is perhaps, see here's the interesting thing, right, and if this audience is so young, then none of my metaphors would sort of work, but nevertheless, if you sort of came in, in the early part of, let's say you were a multinational company, 3 PCs, how the heck did we do forecasts, like a simple sales forecast? The way one would do sales forecasts was you would send faxes.
People were dead. Take those fax lists and send interoffice memos, and those interoffice memos would be annotated and the forecast would come, hopefully, before the quarter end. And then suddenly people said, with email and PCs and Excel, they said, let me just send an Excel spreadsheet and email people and enter number and you have a forecast. So what happened was the work artifact and the workflow changed, right? That is what needs to happen with AI.
When someone says, I'm gonna now do my job, but with whatever, 99 agents that I am directing, on my behalf, the workflow is not gonna be constant, right? I mean, you now are really going to have to change even the scope of your job. So that change management is the real later in it, right? Because you're now taking the means of production, in an insurance company, in a financial services company, in a healthcare company,
software and saying we are going to change everything in the way we work. You might want to change what jobs they are. At LinkedIn I think they took multiple of these functions, the design function, the front-end engineer function, the product function, put them all together and said we want to have full stack builders. That's a change in scope for people. So how do you then rebuild the product team with new roles, new scopes and what have you. That's to me more the.
social great limiter, not that there's lots of other things that are not in deployment of this technology, getting it out to the world, Harvard is one, there are other issues. But I would say change management. When I look at even a lot of the AI startups, when I talk to them, everyone has now, you know, you worked at Palantir so you know this, everyone has four deployment engineers. That's like the exciting thing is the Palantir model, which I think is a fantastic thing. Why is that? That is because of all that change man. Because I think you really need to help customers, partners, really understand the benefits of any product you're creating,
but not just the technology, but even how to use the technology in the workforce. At YC, we have this funny saying that we tell a lot of people here to do, which is, you know, this is some of the smartest AI researchers, computer scientists who are just starting out in their careers. We tell them, go undercover. So go work as a medical biller and see to what degree, how many, you know, quote unquote, knowledge work jobs are actually copying and pasting from a browser into a spreadsheet,
into an email, and then click and send. And do that for a while and realize, like, actually, these are not necessarily, you know, using your prefrontal cortex and your highest mind kind of jobs. These are not, you know, can you imagine so many people, like, their lives are basically, like, you know, we used to, you know, coming up, At our age we would call it paper pushing, but they're not paper pushing anymore, but they're sending emails, they're not sending faxes anymore, but they're trying to get business done by attaching files and things.
That seems like a pretty big shift actually. I think one of the most understated things as an opportunity for anyone creating products or fundamental breakthroughs, even at the model layer, is the amount of drudgery there is in knowledge. I mean, in software engineering as well, the amount of everything, the joy of software engineering is you're out of your flow. To be able to stay in the flow, to be able to ask, that itself is a great example of what I think is going to happen to all knowledge.
You're absolutely right, the amount of cycles you spend out of band collecting information, because you think about the prefrontal cortex and the synthesis part, the amount of time you spend there is pretty normal. Now, like you said. So beyond simply adopting AI tools, what are the biggest transformational shifts you're.
seeing in the field today? I think, to me, even like, I mean, this field is changing so rapidly, right? I had not even imagined, last year, even this time, that we would get this far with RL and, with basically test and compute. And it seems pretty limited. So the way I think about it is, pre-training worked, all the post-training techniques on.
top of it were fantastic. Then, this inference time compute seems to have really added another massive scaling. I'm interested in whether there is some new algorithmic breakthrough because I always say this entire regime could be changed by one person here who comes up and says, I have an efficient thing to do or a way to do this. So you have to have an open mind that the last big breakthrough algorithmically has not yet been found. So that's one. I'm always sort of interested in that.
The other one is, what is the next step up? Because what is the pre-training to RL, the end-to-end training that's the next big sample? That I think is also what I think will happen. So I would say, if that is another scaling law for a breakthrough, it will be. If you sort of take any lab, all of us I think will be working on saying, what's a more integrated response reasoning model that we can build?
That I think is my opinion. That's what I think it is. There's something very interesting here, I think. If you think about an LLM instance as a consciousness, which I think some people are starting to say, it's sort of instantiated, you do a bunch of work with it, and then it sort of goes away, and you open a new chat box, and it's, you know. I guess I'm curious, do you think that, that might be one of the things that needs to be completed, right? Yeah, I mean, so I'm not sort of, I don't,
to me, this artificial intelligence is unfortunately the worst name we could have ever picked. So I'm not into this anthropomorphizing AI at all. I mean, I think of it more, I come at it as it's a tool. It's not trying to replicate how we think. It is, it's definitely showing signs of intelligence, but it's not the intelligence that I have. I think of human agency still will matter, will be there, and who's needs this tool. So that's kind of my position on it. That said, let's just say,
oh, yeah. The memory system is a good thing. These things do need, if I look at the next one here, I would say there are three things, right? One is memory, the other one is tools used, and then the third, which I think is perhaps the most important thing, is entitlement, which is basically, if I want to take action, what entitlement do I have to take action, right? So these three systems have to be built as first class, around the mark in order for us to build.
most sophisticated applications. That makes sense. I mean, it seems that one of the arguments people are starting to make around the future of software is, well, we have the database, and then you're gonna have, basically, middleware that is, you know, I think, what you call entitlements. Is there a lot of access control with this? What's the business logic, who gets what? And then, you know, you basically put the agent on top there. Is that sort of... That's right, so that's why I think, People say when you think about the scaffolding layer,
where you have a model called scaffolding, the scaffolding now really gets first class by thinking of these three things. Tools uses one, memory is one, and then entitlement, put that stuff together. Then you can create an agent, an agent has an ID, agent has a management mission, control on it, it has. That's the way. Do you worry with Code Gen, do you think users at some point will just prefer to make software just in time instead of using package software?
I mean, that's something that, having lots of conversations in the hallway about that, because a lot of us in this room, YC will actually fund a ton of SaaS and continue to do so, but in the background, we're starting to have that. There's some venture capitalist friends of mine who are in the audience that are actually like, I actually don't know if I can continue to fund B2B SaaS. How do you think about that? That's a great question. I mean, it's interesting. At the same time, I look at the number of people who are forking a VS Code and I say, man, you must have done something right. So therefore, there is something to be said about building a great IDE. In fact, when I think about Excel, I think of it as an IDE. So the fact that there's a great cap, you can then bring, let's call it the message.
Analyst model to this IDE and then create a loop between the canvas and the model. So I think, yes, you can generate locations just in time. You could have a prefab application that is really helping with the feedback loop to the model. And I think both of these things will exist together. I guess, do you think there's a role for design in all of this? I mean, basically a human. You know, a human being sitting in front of VS Code is sort of like the translator of the software and what the end user wants.
And I think some of this idea that software goes away presupposes that normal people walking around are going to want to create software. I don't know if that's going to work. I think that's a good point. The way I sort of say, because one of the basic questions you ask me is what happens to software engineering. I still feel, let's take the following thought experiment. If you sort of said some Martian intelligence came in the 1980s and watched how we all worked.
They would say, oh wow, these humans kind of work in the offices. They have a typist rule. They have a slide rule that people then work with. And then if they came back today, they'll say, God, man, all 8 billion people. are tightness. I mean that's what they'll survive. And so I think what will happen is all of us are going to be creating software. But there is going to be a job called a software engineer. It's going to be different. When I look at it, when you're really taking a software engineer and saying you're now a software artist, I still feel the metacognition of your company. One of my biggest things is, man, this white coding is fantastic until it does stuff that I don't know what the heck.
So that means I have a better model of my repo and exactly what happened. And I'm looking at the change logs. So when I look at my favorite feature of GitHub now is to really look at the change logs of all the agents that are working on my code. And I think that is where a lot of the software engineering will be. Like a good dev manager. I don't know which dev manager you want to find software. I really look at a dev manager. The editor's job was to make sure wills don't break and the code is got good quality.
And so to me, that is still a thing. And so there will be a level of abstraction up, even in a world of other AI. Because one thing that we don't talk about is the legal liability, by the way, until some real laws change, are going to be with humans and institutions. And as long as that is true, we're going to have to really make sure the human is in the loop at a fundamental level. And that means we need a lot of tools for humans to be in the loop in order to figure out what these things are doing.
That makes sense. Where do you think, in AI development you see so much, what do you think is underestimated and what is overhyped by the broader tech industry? Where are you sitting? I mean, it's not short of overhyping. In that way, right, we're at the... AI all the time so it's good you know for us all in this industry we live and die by our ability to get into a frenzy about something new right what is the.
Steve Jobs thing or the Bob Dylan thing which is you're either busy being born or busy dying it's better to be behaving busy being born so that's good. I think the thing that we have to most worry about and most work on as a tech community is that how do we earn that social permission is one thing that I feel to me one of the demos I saw which completely really blew me away was at the beginning of 23 when I was in.
India and I saw a local developer Daisy Jane essentially at that time either GPT-3 or GPT-5 with one of these India stacks speech-to-text, text-to-speech open sourcing and then showed a local Indian farmer who was in, able to sort of use a chatbot that was built in WhatsApp to be able to get some agricultural subsidy, right, by going to a government website. To me, it was unbelievable, like I felt like, man, how could something that was built in.
the west coast of the United States get to a real use case that fast, thanks to sort of the diffusion rate and basically people everywhere. That is the story that needs to be told, right, at scale. That is the underhyped story, because right now the overhyped thing is the model capability, and the model capability is fantastic, but man, if we can somehow get the world to recognize, that this is making a real difference in the lives of people everywhere, we are in good.
shape. If that doesn't happen, this is all about some valuations of us, our companies, and our industry. And it's the same with... I love that example. I mean, it feels like Microsoft is sort of full of examples of things that lower the floor so that a lot more people can get access to technology. I mean, you could argue that GitHub co-pilot as well. Yeah. By the way, one of the other ones, since you brought it up, there was a World Bank study.
they did, I think in Nigeria, and now they're making it to Peru or Chile in South America. We've been working at Microsoft forever on, can there be an intervention in education? That's been the dream, and we've been at it, at it for decade after decade. It's made a difference. But this study said by, actually by a co-pilot, is probably the best intervention in education, in Ecuador and Latin America.
That's been the dream, I think, that we've all had in tech, and it's right there within our graphs. I guess, are there any interesting observations? I'm curious because, you know, your co-pilot in Windows is, you know, often, you know, here in tech, like, maybe people are really obsessed with the latest frontier models, but it's easy to forget, like, Windows and the integration with Windows is actually the first interaction for EHEI, sort of AI today. Are there any observations from, like, people using that?
Yeah, no, we're very, very excited about Clippy being back as well. But seriously, I mean, like, look at, to me, the thing that I find is even in the form factor that we know and love and work in, which is a good old computer with a mouse and a keyboard, right? The dream has always been, in fact, the first research group built in Microsoft Research was Speech in 1995.
And so since, Since then, we've been saying, God, like, when will speech be first class on PC? But right now, in Copilot, the two things that are just pretty surreal to me, it's kind of like a new browser mode, where there is no vision in speech. I lean it on all the time. It can see what I see, and I can speak to it. That seems like a precision mouse move. So that is where, I think, even on existing form factors, there is a way to achieve the complete computer use.
And then there will be new form factors, right? So I think it's an exciting time to be building both hardware and modifying existing hardware for what is, I think, possible in the future. Yeah, computer use is fascinating. Computer use is actually a superset of data. Like personal data. Personal data, your work data. All your office docs, like everything is accessible, right?
Is that sort of the point? Was the movie Her correct? Literally, the operating system is going to embed itself in the most trusted agent. Yeah, I mean, I think that has been the dream, which is can these agents become your computers? And they do the computer use, and that absolutely, I think, is the direction of travel. And I think you mentioned the most operative thing, which is trust. Which is, can I trust to delegate what I want?
And that means it's about precision. It is about sort of the privacy. It's about a lot of these considerations. And I think that these all will, in time, will work out. I mean, in that respect, when you look at both EA and Apple, they sort of have to be. Yeah, I mean, so to us, you know, there are many, but it's not even sort of, there's privacy,
there is security, there's sovereignty. These are the three big, big considerations, right? Privacy, every user cares about it. Security is what every tenant or every customer will care about, you know, top of privacy. And then every country will care about sovereignty, security, and privacy. So that's the way we need to play. So you really need to build any product or any system, you need to be able to answer the questions on for the people and for organizations and for countries, how you cross all those.
three boundaries. So, Satya, you've had an absolutely extraordinary journey at Microsoft, starting as an engineer all the way up to CEO. What lessons from that path would you share for the next generation of builders? You know, look, it's not like you start any journey with sort of a specific goal of where you want to end up. But you do start with the goal of taking the first spot, sort of having the highest ambition for yourself on what you want to get done.
But I always say it's not like I was waiting to become CEO to do my best work. The first job I had, I felt, was the greatest job I could ever have. When I joined the company in 92, I felt like, wow, if I retired in that job, that would be fantastic. And that was a great mental model. When I look back at it, it's not I was not waiting for my next promotion, but using the opportunity I was given to do everything I could. And I think that that's what people who are starting out or who are founders or who are researchers or students today.
So I would say keep that alive, don't wait for the next big thing, take the thing that you have as the biggest thing and then make it expand. And then the other thing that I would say is big things are achieved by having a team around you, learn how to work in teams, making teams great. One of the things that I feel at Microsoft I learned was what it means being a project, what it means to work, in fact that's kind of the big difference between school and work.
is that you join a team and you've got to figure out how to make the team successful. The incentives are actually pretty clear, except I think the thing that is least thought, is how do you really make sure you can compose as a team and what's your role in it. Every one of us sort of looks and says somebody else's job is to align with the team, it's your job to align with the team. I'm saying... I appreciate for your own impact, how to work in a team and make a team effective.
That's magic. Here's a fun story. I actually did learn how to do product management and project management as a PM. And when I was employee number 10 at Palantir, I taught them actually how to run a project. Zero bug balance. My PM training at Microsoft turned into the thing that created how Palantir runs their product work today, which is pretty wild.
So thank you to Microsoft. I'm curious, what are the qualities that you look for in people and teams just because AI is becoming a really key part of creative work and engineering? Sort of changing the way you might interview someone and evaluate them. Yeah. Yeah, I'm always looking for three qualities in people. One is, in fact, Bill turned me on to this, which is he was describing at one point, who are good architects, who are bad architects, and he had this nice way to summarize it, which is good architects bring clarity and bad architects bring confusion.
Even if there is confusion, even if there is confusion, to people mainly to take control of it to an end, they will with no product. That's impossible. Dia?
Yeah, what's up?
God, I didn't even have time for this.
Thank you very much. Yeah, thank you very much. That was it. Reaching them was, if you look back at the history of Microsoft, how Office got built is an unbelievable thing.
The sense of thinking of these tools, right? A word process, a sheet, a slide making tool. What those tools have meant to all of us, right? I mean, that's why I always say, once you're here, somebody asks you, what's my favorite product? It's always, you know, VS Code is one and the other one is Excel. It's just pure, you feel good when you use the tool. All about the sense of power, the numbers, the analytics.
Like a spreadsheet, like what an article. It's like an unbeatable scaffolding. Columns and rows with some sort of Turing machine in the middle. It's just great. And so I would want to work on the tools. Like when I see the compiler tool, that's kind of where I feel like, you know, researcher, behavior. These are like the wordings of PowerPoint. Like every day I go to the... So to me, that's what I would love.
What are the things put in the tool that will give them that sense of empowerment? That's what I would love. I have a feeling that people who make those tools are sitting in this audience right now. Please give it up for Satya Nadella.
Please welcome, former director of AI, Tesla, Andre Goff. Hello, okay, yeah, so I'm excited to be here today to talk to you about software in the era of AI and I'm told that many of you are students, like bachelors, masters, PhD and so on, and you're about to enter a new industry. I think it's actually like...