While he hardly needs an introduction, few leaders have shaped the future of technology quite like Satya Nadella.
He stepped into Microsoft’s top job at a catalytic moment—making bold bets on the cloud, embedding AI into the fabric of computing, all while staying true to Microsoft’s vision of becoming a “software factory.”
In this conversation with Aditya Agarwal and Ruchi Sanghvi, Satya discusses:
• his approach to building early at Microsoft and why we all need to continue “refounding” • the future of AI and why critical thinking will remain at a premium • Microsoft’s partnership with OpenAI, his vision for quantum, and more
00:00 Trailer 00:57 Introduction 02:03 Joining Microsoft 06:39 Culture and relevance 11:21 Winning and not winning 17:37 Making that turnaround happen 22:30 Contrarian leadership belief 23:35 Investments in startups 27:33 Foundation models 31:07 Artificial intelligence 36:32 The future 40:39 Quantum computing 44:53 Multiverse 45:33 The next generation 50:10 Being competitive 52:34 Outro
Man, there was real daylight between Microsoft and our competition in the 90s.
There was Microsoft and there was competition, but it is not like today,
where you have whatever the seven, eight companies,
plus pretty much in any given day, anybody can come from anywhere.
How did it feel like when you were on top for a long time
and then you kind of had like a spell in between
when you were kind of wandering in the desert?
I do think success rings with it, the challenges of success,
because you start forgetting what made you successful.
And everybody who joined and rode the wave thinks they are the cause of success
versus saying, no, no, no, we rode the wave and we now need to refound.
And that, I think, is the undoing of at least large, successful companies.
You've had two great CEOs before yourself, Bill Gates, Steve Ballmer.
What is a contrarian leadership belief you hold
that you think our founders here should consider?
That's a good one. Like, so I would say...
I have not seen such an excited audience at SPC,
and we host a lot of people.
Woo-hoo!
We are honored to have you, Satya, here.
You obviously need no introductions.
But here at SPC, we explore the minus one to zero phase,
where a lot of us wander through the desert until we build conviction
on how we want to spend the next five to ten years of our lives
and what we want to work on.
So in that spirit, you've had a meteoric rise at Microsoft.
It feels, at least from afar, that at every turn, you've had to challenge...
I guess, like, just challenge everything.
So we want to visit the pivotal moments of your career
and discuss what minus one looked like in those times.
So, okay, so Satya, I'm going to start off this whirlwind minus one journey.
Hello, everyone. Good to see you all out there.
Okay.
So we're going to rewind back to 33 years ago,
when you joined Microsoft.
What was it like to interview at Microsoft in 1992?
Why did you decide to interview there instead of interviewing with startups?
Like, what was that experience like?
And I can tell you my experience interviewing with Windows NT kernel team in 2003.
Did not get the job, by the way.
I also did not get the job at Microsoft.
First of all, it's fantastic to be with both of you and with all of you folks.
I mean, this minus one to zero is a great sort of framing.
33 years ago, it's interesting.
I was actually working at Sun.
So the first job I had after grad school was in Mountain View.
And I went actually to what was a developer conference at Mascone, in fact, in 1991,
when NT was first talked about and Win32 was first talked about.
And for some sort of reason, I sort of got into my head,
which I think it's a little bit of exposed, you know, words to it,
but it was x86 was going to win.
And this is when these platform shifts sort of become sometimes inevitable.
They take a lot of time, but they become inevitable.
And that was the thing that got me to say, God, I got like, I mean, Sun,
pound for pound, was just a stunning company.
I mean, in terms of everybody else who went on to create all the great companies,
you know, and every, like, I mean, if you think about the core of the distributed systems,
we were, you know, Sun was a full stack systems company.
We made our own chips.
We had our own operating system.
We had our own.
In fact, I spent a bunch of time doing, interestingly enough,
knowledge worker tools at Sun.
But I just felt like x86 was going to win.
And that's what got me to go to Microsoft and say, I was not a Windows user.
I was not, you know, I never thought much of DOS.
But I felt like this is the company that is going to be able to capitalize on that.
But more importantly, truly be that democratizing force,
which I think is an often used word in our world.
But I really, that was what got me there and kept me there for 33 years.
But I mean, the question still is, was the interview intense?
Because, you know, I bet a lot of people in this room have maybe interviewed at Microsoft at one point.
The interview was one of those classic Microsoft.
At that time, man, I tell you, the interviewing nowadays is a little better.
At least you've got to sort of do your lead code or whatever, as opposed to
that thing was go on the blackboard or the whiteboard and then do your thing.
And I remember that.
And somehow I got through it.
But the last interview question, which I've written about, which was the worst, is,
you know, we had something called the ass appropriate.
So if you make it, then that's the one person's name.
It's never in the sheet, except you show up in their office at the very end.
And then that means that you probably are making it.
And this guy's last question to me will never forget.
Right. So he says, hey, here's I have a question for you.
You're crossing the street and a baby falls.
What are you going to do?
And I said, man, this is like some algorithm that
I didn't study.
Is it like what is this?
P equals NP, whatever.
And so I say, I'll call 911.
So he gets up and he walks me out of the door and
he says, you know what you should do when a baby falls?
You should pick them up and hug them.
I said, there's no chance I'm going to make it.
But lo and behold, you kind of learn from your life's experiences.
That's amazing. You know.
It's interesting when we Microsoft is one of those companies that I think
has just played such an important part in the story of technology
that when people bring it up today, it means many different things to many different people.
Right. But I think one of the stories that at least gets lost today
is that in the 90s, Microsoft built like much like, you know,
it kind of built a diverse set of technology that was just like
bleeding across almost like all of the diverse kind of classes.
I'm sure, you know, elements of the culture remain today as well in different forms.
But how would you describe the culture of Microsoft?
That's a great point.
In fact, we're going to turn 50
as a company on April 4th and this year.
And so we've been sort of, you know,
talking a lot about it and what exactly is it?
In fact, one of the things that
and the metaphor or the idea that Bill had
was Bill and Paul,
how was it they were building a software factory?
That's kind of the concept was, you remember, there was no software industry
before Microsoft.
And so they just wanted to be
a company, want to build a company that builds software.
And they didn't have like this is the category.
So they went in some sense.
First was, you know, let's build the basic interpreter for the Altair, right?
So that was the best place to start.
We actually had like FlightSim predates
Windows for sure.
It's one of the oldest products we have.
And so in some sense, they went to wherever
you know, you could use the software factory to create software.
So in an interesting way, I kind of think of
and that's what even led to knowledge work. Right.
And so I think the culture was, how do you create,
get the best talent to build the best software factory
that then creates the software products that are needed in the world?
That's what when I go back, I think somewhere in that framing
for all of you all as founders, I think it was a pretty powerful,
helpful thing when I look back in our 50 years, because I think
a lot of companies sometimes when you fall in love with the one product,
the one technology, the one business model,
it's sort of kind of all things run out of gas at some point.
And so I wonder, I don't I'm not definitive about it,
but I wonder whether that framing,
whether it is the cut to your point about culture,
that's when you kind of think about the talent that comes in
is trying to be, you know, accretive to the software factory.
Each time you can take a fresh look when there's a better idea
of some new widget you want to build in the factory, you go versus
just be in love with the widget you built last year.
I wonder whether that has been actually helpful in keeping
because I don't think of a culture as something.
I mean, all of us have been striving and always strive for a learning culture.
Nowadays, I use these words of growth mindset and what have you.
And I wonder whether we always had that growth mindset
because and a learning mindset because of the fundamental framing
of the the company.
Is that something that you would talk about?
Like kind of the software factories?
Was that lingo?
No, it is not like I grew up in a Microsoft.
We talked about it as a software company.
It's more in the context of our 50th year
where I'm sort of saying, like, for example, like after all,
we are now yet in another big shift.
What does it mean to be a software factory
at a time where there is intelligence
or this agentic world of ours?
What does it mean to even say you're a software factory?
I think is that how do I reinterpret this?
So I think that that's kind of a little bit of the language
I'm searching for, and I'm trying to revisit our history
to make it current, because I do I do think a lot about
I mean, longevity by itself is not sort of a goal, right?
Relevance is, and therefore,
what is it that gives you the courage to be relevant tomorrow?
That's the thing that I'm trying to kind of really study, learn,
push myself and even articulate for myself.
So on the subject of relevance, then to maybe take you through
some of Microsoft's like middle years, if you will,
maybe to ask the question in a in a most controversial way.
How did it feel like when you were on top for a long time,
maybe for like, let's say, 25 years,
and then you kind of had like a spell in between from like,
let's say, I'll just say 2000 until 2010,
when you were kind of wandering in the desert.
So what was that like?
When you're kind of at the, you know, you kind of like winning everything
and then you're kind of not winning for a while.
Like, how did the culture in the company,
like, how did you deal with it personally? I'm curious.
Two sort of couple of different reactions.
So the first is, man, there was real daylight
between Microsoft and our competition in the 90s, right?
I mean, I look at, especially in the late 90s,
there was Microsoft and there was competition.
But it is not like today, right, where you have whatever the seven,
eight companies plus pretty much in any given day,
anybody can come from anywhere, like as OpenAI has even demonstrated
to sort of create what is a generational company.
It's pretty unbelievable to see the vibrancy today.
So I do think a little bit of,
you know, sort of success brings with it the challenges of success,
which is you kind of forget what made you successful.
And and so I do think in the 2000s,
I think it's not like we didn't have ideas, nor did we start things.
I think I always say that
you miss stuff
because you're too much in love with sort of what you have today
or your business model of today and are not willing to go somewhere
which is inconvenient or just don't have a complete thought.
But one of the things that Steve Ballmer, you know, would always I mean,
one of the greatest pieces of advice I ever got was he used to have this
like you got to have a complete thought, right?
I mean, the number I mean, even myself, the number of us,
you know, we have some product idea.
I mean, the beauty of founders, successful founders is they have complete thoughts
by definition, because otherwise you're not going to be a successful founder.
And in companies, that's what you need.
You need lots of refounders who are have complete thoughts
in from time to time.
You just can't say I've fallen in love with this one piece of technology
and I've not thought about go to market or I have not.
I've thought about go to market,
but I have no idea what the product should be or what have you.
So to me, I think the culturally real learning like here's another thing.
Right. There was a new generation of leaders
who needed to go through that and become refounders
is how I think about it. Right.
So in some sense, the most dangerous times are when you feel like you're very
successful because you start forgetting what made you successful.
And everybody who joined and wrote the wave thinks they are the cause of success
was saying, no, no, no, we wrote the wave and we now need to refound.
And that, I think, is the undoing of at least large, successful companies.
Now that you kind of painted it in that way,
I think one of the things to Microsoft's great credit,
you've never been shy at taking shots on goal.
I mean, we met, I think, in 2006. Yeah.
When I was working on Facebook search and Satya was running Bing search
and we would kind of, I would fly up to Seattle once a month to see,
you know, how do we get Facebook search results to kind of show up in, you know, Bing search.
We tried. We tried so hard.
Like, how do we personalize it?
How do we kind of enable kind of Facebook data to show up in Bing?
But the thing that I always what I remember from those meetings was
like from the inside in, like Microsoft was like, you are willing to learn.
You're willing to try it out.
And it was despite being a big company,
we could actually make progress, which had surprised me at that point.
So, I mean, yeah, Microsoft took a lot of shots on goal during that time.
And I think that's to your credit.
Yeah. I mean, as a company, as I think we have had definitely
that ability to persist with things.
You know, sometimes stubbornly persist, perhaps.
And quite frankly, you have to.
Right. I think in tech in particular, you've got to take shots on goal
and and and just get like.
And I mean, the reality is the network effects of tech are unforgiving.
Right. I mean, they're great when you get them
and they're terrible when you're on the wrong side of them.
But therefore, you have to take shots on goal
as probably the way to stay relevant.
I mean, I love it.
I mean, this is what we tell founders on kind of like day minus one.
But I think it's amazing to see it apply even at the largest scale. Right.
Yeah, I mean, it is.
I mean, I think to a founder community like this,
you know, the good news for you all is, you know, people like me.
Look, I really do believe this.
I don't think there's any franchise value in tech.
It absolutely is
in that sense, a pretty scary industry, because
at least if you measure it by relevance,
you may have franchise value in terms of EPS or something,
because you can always be a thing.
But but if you sort of say, no, no, no, you want to be relevant.
You do need to actually build something.
Yeah. And that is that is not just keeping up,
but it's about sort of inventing and that then is what I think
by the fundamental bias of this industry is
there is always going to be a disruptor.
You know, I kind of time Microsoft as a comeback kid
with you becoming CEO, quite honestly.
And, you know, again, like I thought it was just going to disappear
and become one of those like, you know, old tech giants that move slowly.
And then when you became CEO, things started to turn around.
And you've talked about the culture of reinventing,
refounding, also really spoke about being relevant.
Take us back to your like minus one days of becoming CEO.
Like what was the first thing you did to make that turnaround happen?
Yeah, I mean, for a couple of things.
One is, I think, first of all, I give a lot of credit to
and in fact, all the credit for what happened in the cloud era
to not me, but to Steve, because after all, I worked for him.
I worked in that company.
He was the one, quite frankly, as in fact, a lot of what I learned
on how to be an effective CEO is actually from Steve Ballmer watching him
give me permission as a direct report to go spend the money
even when the street was not giving permission.
That's what a CEO, a real CEO does, which is it's sort of defy
conventional wisdom and to be able to empower leaders
and people inside the organization.
So in some sense, you know, you know, Steve is fine.
He doesn't care. He's he's done well with MSFT.
And but that ability
to lead
is sort of what I picked up sort of working for Steve and then observing Bill.
The thing that I'd say for us, perhaps, Richard, your fundamental question,
though, is I was very grounded in that.
I was the first non-founder, right?
Steve technically was not a founder, but he had found a status in the company.
There was no you know, it was like a company that Bill and Steve ran
that I grew up in.
So to me, the lucky break, perhaps, is
I came into that seat knowing that I just can't go in
and, you know, try and fill the shoes. Right.
So I had to, you know, as I describe it as a mere model CEO.
In some sense, now I have better language.
Thanks to Reid Hoffman, this re-founding
language, but that re-founder mindset I had, which is I said, OK,
I got to take and make first class again
things that founders take for granted, right?
Like, you know, the founder, because of their.
We call it founding moral authority. Yeah, that's right.
And so the mission is implicit, right?
Culture is implicit.
Somehow, you know, follow the founder.
There's so much that goes with it.
And so to me, re-grounding the company in its mission, right?
Going back to even Aditya's question on, hey, what was the
the software factor like that?
I came back and say when I joined the company, we used to talk about our mission
as a PC in every home and every desk.
And quite frankly, by the end of the 90s, at least in the developed world,
we achieved it.
And since then, we were a little confused.
What the heck is our mission?
Should we just shut shop, give the money back to the shareholders?
Or what do we do?
So going back and saying, no, no, no.
When we sort of what was the thing behind the basic interpreter for the Altair?
It was about putting software in the hands of others
so that they can create more software.
I said if that was relevant in 1975, in 2014, when I became CEO,
it was going to be more relevant.
And so I went back to that and said it's about empowering people
and organizations.
So the other thing that I felt Microsoft got was institutions
people want to build.
Right.
That's kind of another very unique thing, which is organizations
are also things that people want to build
and that need to, in many cases, outlast them.
And so we want to build software for those organizations as well.
So people and organizations, people ask me, are you a consumer company?
Are you a commercial company?
We're a software company that wants to serve and empower people and organization.
So in any event, the long answer to your question is regrounding on mission,
creating this language around our culture with growth mindset.
Thank God for Carol Dweck's work.
That was super helpful to me because it was not considered new dogma
from a new CEO, but it is work in child psychology that appealed to people.
Those two things is what I would say
helped me a lot galvanize the energy
to go execute on a lot of the strategies were very consistent.
If you look back, it's not like I was running one of the largest groups
in the company before I became CEO.
The strategy I've written the same stuff.
I was talking the same stuff before.
It's just nobody paid attention to it.
They said after becoming CEO, somehow things happened a lot faster.
On that note, you know, again,
you've had two great CEOs before yourself, Bill Gates, Steve Ballmer.
But what is the contrarian leadership belief you hold
that you think our founders here should consider?
That's a good one.
Like, so I would say the the one contrarian leadership,
like one of the things about sort of I don't know whether it's a founder thing,
but a lot of leaders have it, right, which is everything was broken before.
And so therefore, I rescued it.
I'm very suspect of that.
Anybody who comes in and says everything was broken and I saved the day
make make make make you sound, you know, make you feel better.
So I kind of look at it and say is if the place falls apart after you leave,
then you've built nothing.
So the contrary.
So in some sense, even as a CEO, I kind of look at it and say,
you know, it's the fourth CEO, the fifth CEO of Microsoft and their success.
That will be the true story of whether I did my job or not.
That's pretty great.
I'm going to switch tacks a little bit, just go off script a little bit.
I would say Microsoft has benefited not just from innovation,
from its research lab, but also actually its investments in startups.
I remember in 2007, the deal team came to Facebook's offices.
You know, Aditya and I were there, maybe Sol, you were there, too.
We were like hanging out on the engineering floor.
They show up at 8 p.m.
All of us were really into electronic music.
So we had these like amazing DJ decks with speakers.
And, you know, the music was bounding.
We were coding.
They all got so excited that they decided to build an all nighter,
which we were doing every night, by the way.
And then they camped out in the conference room right there
and they ironed out the deal terms.
And then lo and behold, by the end of that week,
they invested like two hundred and forty million dollars in Facebook in 2007,
which was huge.
And then more recently, you invest in OpenAI, Mistral,
like, you know, there's so many other AI companies.
I think this is particularly relevant for the audience here.
How do startups think about partnering with Microsoft?
How should they pitch your fund, which is M12?
I mean, it's such a large organization.
It's intimidating to us from the outside.
So we just don't know how to think about you.
I mean, you have a you have a crazy good track record.
Yeah, it's incredible.
It's often spoken about in terms of like Microsoft's investments.
And I think collaboration with the startup ecosystem.
Yeah, no. First of all, I think the collaboration with the startup ecosystem
and what we're trying to do, even with M12,
those are super important to us.
But we're not an investor.
So in other words, quite honestly, you know,
in Facebook, we wanted to invest.
Not that it happened, but we wanted to invest because of what we thought
could be a commercial relationship around advertising at that time. Right.
Which you were smart enough to not do with us.
And you built your own.
And then the rest is history.
But that was the strategic intent.
The intent was not an investment return, but it was about say, hey, can we,
in fact, be a sponsor at that time of Facebook in order to be able
to make some commercial advance in an area that we were interested in?
Open AI is slightly different, right?
I mean, open AI, remember, at least when we started backing them,
they were a research lab.
And then as open AI has evolved,
they become a successful product company and what have you.
We've kept up, quite frankly, with Sam and his team and his leadership team's
ambition, because we wanted to make sure
we were supportive of someone who was a partner that we were an investor in,
but more importantly, a commercial partner.
And so I would say open AI
has evolved from where we started.
But I come at it from saying, look, as one thing that I'm grounded on
is Microsoft's two things.
It's a platform company and it's a partner company.
A platform is not stable without partners.
And if you don't have partners, you can't be a platform, right?
So that type of that is when I think about startups
or even for that matter, other successful companies, I always am looking for.
I worked in my early days at Microsoft.
I even worked with on our database business
with folks like SAP, for example.
So we were able to build our relational database SQL underneath SAP.
That's the type of partnerships that are super good.
Same thing with open AI or Wintel is another one like.
So to me, I always gravitate towards
what a long term stable sort of win-win constructs
between a platform company and another partner and who themselves
can be building their own platforms.
You're obviously comfortable doubling down on your investment in open AI.
But like, you know, given, I mean, I guess the folks that you view as competitors,
whether it's like Google or like Meta, et cetera,
who have their own foundation models, like don't you feel a little,
I guess, like left out?
Yeah. Don't you want your own foundation model?
It's like, you know, the old the old quote from back in the day,
like, well, I feel like real people have fabs, you know,
I think that real companies have like foundation models nowadays.
The way I think about it is right.
I think it's it's we're a full stack systems company
and we want to have a full stack systems capability and foundation models
are an important thing.
But the way we when we initially started with open AI,
it was a sort of a bet on the conviction that
at that time, you know, obviously Sam and Ilya and Greg and others
had on basically scaling loss, right?
I mean, the first I mean, in fact, it's interesting.
I think it was Dario who wrote the paper that I first read,
which was the scaling loss paper.
And and that was the bet I placed was, man, this and especially
as it met natural language, I don't think if it was quite right now,
I have a much better appreciation of even RL,
but I don't think we would have done what we did if it was not natural language
because of being a knowledge worker company.
We were obsessed about not natural language.
That was Bill's thing all the years.
And so when somebody said, oh, I'm going to throw a compute on natural language
and maybe we could have some scaling laws that could just give
because the thing we were obsessed for thousand years
was schematized the world, right?
Which is people places things go build the greatest ontology and schema.
And that's the way intelligence will emerge.
We kind of took many, many detours on that.
And then I felt like, wow, I always felt uncomfortable
because the world is too messy for SQL to work.
And and so I said someone somewhere is going to figure out
how to get to some semantic thing in a different way.
And this was the bad.
It was unclear to me whether to work or not, but it started working.
And so I look at that and say, we have a fantastic commercial partnership.
We have all the things that we want and need the ability.
Like I've always said, like we built the system underneath it.
We built tools around it.
We're building products about it.
So to some degree, I have a long term stable relationship with open AI
and we have the IP rights.
And so therefore we will want to build like we built five.
You know, Mustafa and team have, you know, have capability
that they had in pie before.
We have it in Microsoft.
And at this point, you know, if you wanted to just to prove out,
that doesn't make sense.
I do believe the models are getting commoditized in cloud.
In fact, open AI is not a model company.
It's a product company that happens to have fantastic models at this point
and which is great for them and great for us as, you know, both partners of theirs.
And so I want to go now where I think the industry structure that's emerging
models by themselves are not sufficient, but are having a full system
stack and great successful products.
Those are the two places.
Maybe now focusing a little bit on the future.
Yeah, I kind of the set of things that we are building on top of these capabilities.
It's it's amazing that in the two and a half years
since chat GPT kind of really, I would say, entered the public consciousness.
There's just been an incredible flurry of, I mean, just a tsunami of activity
in between chips, data centers like frameworks, foundation models,
vertical AI companies, consumer hardware, consumer software,
agentic software, modern enterprise software.
There's just so much going on.
And you have a vantage point, I think, to a lot of those different aspects.
If you had to zoom out and kind of tell a lot of the aspiring founders
in this room, what might be two areas that are overlooked today
that might be like, you know, interesting kind of arcs
that are interesting places to essentially like maybe place
your seat in over for like the next five years?
I'd say there are three clear things that I think people hear
and all of us are dabbling in and they're all happening in parallel.
And it's exciting to see, right?
It's no longer just what's happening with knowledge
workers or language models themselves,
but it's also what's happening in the real world and the real world
action models and then these models for science,
whether it's chemistry or biology, all three.
There's some transfer between them, but they're also got sort of independence.
I think that that's a fascinating thing to be, quote unquote, going after.
The thing that perhaps
I've been now going back a little bit, even thinking about it
even as a hyperscaler, is what's the system architecture
that truly will support this?
If I look back at it, right,
Paul Maritz was one of the guys who used to work at Microsoft
and as and growing up in the system side for me, it was always about,
hey, can you anticipate what the workload is and then redesign
the system underneath?
I don't think we have done a first principles re-architecture of the system.
I mean, we've fallen in love with these AI accelerators.
We threw a lot of them into existing data centers and said, go forth and prosper.
And we've done quite a bit of it.
But I think time's come to say, hey, what is the next generation
hyper-converged infrastructure?
How should I think about compute storage and the AI accelerator together?
What is these distributed synchronous training jobs with sort of now
RL autograding?
How should you think of that system?
That, I think, is the overlooked part, even on the data center.
When people say data center is a computer, is that time come for V2 of it?
That's one.
And then the other one is this client.
People talk about hybrid AI, but we've not cracked.
The model architecture itself, we talked a little bit about it.
I'm surprised that there isn't a distributed model architecture at runtime,
not for distributed training, right?
At runtime, why is it that I can't split
between whatever is running on my, you know, 45 tops NPU on my co-pilot PC
and the cloud like those?
Because you can run a bunch of helper functions and so on and so forth.
But is there a way to even think about foundationally?
I'm assuming there has to be a model architecture and math breakthrough
for that.
So those are things that I'm surprised, actually,
because the open source community is gone.
I mean, what DeepSeek has done is tremendous work.
And in general, what open source is doing
in making things more efficient on existing infrastructure is great.
But I wonder why no one's cracked.
The existing infrastructure includes a lot of flops on the edge.
So that's the one overlooked place.
It's interesting.
I mean, coming back to what you said earlier, there are very few people
who are in some ways actually having a complete thought.
Yeah. Around kind of like how do these AI systems work?
And it feels as though everybody is kind of building small parts.
But there isn't that much of an overarching kind of like architecture
or a complete thought.
Yeah, that's that.
Yeah, that's a good way to say it.
And the other one that we talked a little bit about is
even in these AI systems, right, when I look at even underneath
Copilot, whether it's GitHub Copilot or the Copilot we have at KnowledgeWorker,
it's not yet a complete system, right?
So it's not like I sort of think it's not yet had the if you say
this is the new computer, it's not had the the stored program moment yet.
In other words,
there is no architecture like there's no memory system
that is robust, that is multimodal.
The tools used is a little too artisanal, at least right now.
And so somehow those would be the two other places.
But by the way, there's a lot of great work happening on all of this.
So that's why I do feel this is a little more
very early innings to building more robust AI system side.
And so at least I go there and I think this is where it was fantastic
to see the number of people who are doing not only foundation models,
but systems companies that are, you know, it's great to see that.
So when thinking about a complete system, I like to think about,
you know, where do I want to be like five years from now?
Or like, you know, what do I imagine the future as?
So if you were to define true agentic behavior,
whatever that means to you, but beyond the world of copilots,
what would that look like?
And how would it impact your day to day?
Because I like to start from there and then move down.
Yeah, it's a great one.
I've been thinking a lot about it, right?
I mean, if somebody had said to me
pre-PC that what I would be doing
is mostly in life for 33 years at Microsoft, be a typist,
you know, it would have felt like, what?
No way, right?
You know, computers are going to change my life.
They're going to be like unbelievable.
And except what I do is get up in the morning and start typing
and then hopefully get to some sleep in the night.
So I wonder whether what I'll be doing
is triaging my agent inbox,
because even in an agentic world, that's why I'm really obsessed
about even copilot as the UI for AI, because even the more I delegate,
you know, at some point, if we don't have some AI takeoff problem,
then it is aligned and it's controlled.
It's coming back to me.
It's notifying me, hopefully better quality notifications
than the ones we have today
and getting permissions, getting instructions,
helping me with what I want.
And so to some degree, there is going to be a UI layer
that allows me to then have massive, massive leverage.
Right. So I kind of think about like two metaphors in the consumer life.
I would like, hey, a friend, a coach, an advisor type of AI.
And then at work, I would love to have a chief of staff,
a researcher, a consultant,
you know, working with me on everything I do.
And so I think that that's what would be my relationship with AI
and hopefully the drudgery of knowledge work.
In fact, one of the other things I go back to is we I think we conflate
knowledge work and knowledge worker.
I think tomorrow there will be a knowledge worker,
except the knowledge work they'll be doing.
It'll be at a different level of abstraction.
And so that's what I think we're all trying to build.
Like I think about like a word Excel PowerPoint
with the tools for knowledge workers in the 90s.
What are those tools?
Copilots were sort of version ones of it.
Version two is our copilot plus agent.
What's the version three?
And I think that that's what we're trying to contemplate.
I think my eight year old and my four year old think I'm a professional
emailer because I wake up and I write emails and I go to bed
and I write emails before going to bed.
I was just going to ask, what do you think the interface?
Yeah. What does the UI look like?
I feel like the chatbot, even the copilot, they're lacking.
They're unimaginative.
They don't like really tie everything together.
I agree with you on that.
That's why this multimodal interface, like one of my favorite,
favorite things to do is this copilot voice interface for me,
which is I've even set it up to say my action button on the iPhone
with CarPlay, man, like it's like even take podcasts.
The best way for me to consume podcasts is not to actually go listen to it,
but to have a conversation with the transcript on my commute
using my copilot, like who'd have thought?
But it is more convenient because of the modality,
the fact that I can speak to it, I can interrupt it.
I mean, think about it, right?
This full duplex conversation, which was never possible.
That is a fantastic sort of new modality,
which I think there's no going back, right?
It's kind of like it's like auto suggest.
There's no going back anymore.
I think that a Bill Gates quote, Bill G.
Code that is timeless is, you know, people overestimate progress
in one year, underestimated in 10.
I think a corollary of that as a startup founder is getting to be really good
at understanding where in that one to 10 year time frame
you are for a particular piece of technology.
So when I look at a lot of the announcement, pretty cool announcement
that you have for quantum computing
that you talked about, I think just last week,
when you look at this, like, is that ready for like the startup
ecosystem to be thinking about doing something with it?
Because it feels more like in the one month in the, you know,
or like one, it still feels very early.
Or am I just misunderstanding?
There are a couple of things on quantum.
I mean, like for us, this speaks back to a little bit of our
ability to just go persist for a long time.
I'm like the third CEO who's been funding quantum
over the last 50 years.
20, at least we've been at it for 20 years, 20 plus years.
And and it's sort of a fascinating journey because we picked
it goes back a little bit to having that framing of what is it
that you want to get done right?
Because we from day one, we decided and that is, I think,
you know, Bill and at that time, a guy called Craig Mundy
and team sort of decided that, look, we want to build a real utility
scale quantum computer.
And in order to build utility scale quantum computers,
you kind of need stable qubits.
And in order to build stable qubits, you need to have like a physics breakthrough.
So they went down this thing called these Majorana particles and said,
hey, we were going to go find the existence of some guy in the 1930s
theorized that such quantum sort of phenomena exists
and it is never proven out.
And so 20 years later, we now have proven the existence of it.
And not only that, we've been able to fabricate those
atom at a time to be able to show that these Majorana zero states
can, in fact, hide quantum information in a robust way.
So error correction needs are less.
And so therefore, you can scale it up into a real chip and what have you.
It's pretty incredible to see where we are.
So therefore, today, it's a lot easier for me to find quantum
than it was even in 2014 because of this breakthrough.
And we've had a couple of false starts there.
And to be able to persist through even that has just been phenomenal
to see the team go through.
But I think the one thing that I've also come to realize is
we do need quantum, but it's not like sometimes people think of,
I think, quantum as somehow replacing classical.
I actually think classical plus quantum is a very powerful.
Like, in fact, take that atom by atom construction.
If we had a quantum computer, that would have been a lot faster.
Right, because our ability to simulate nature
goes to a different plane if you had a quantum computer.
But quantum computers are not that great.
They're great at the exploration of a data space,
but they're not a data heavy workload computer.
So I think both of these have a place.
That's one.
And even for the science models, for example,
you can use quantum to generate the label data, essentially,
to train models, which then you run as regular AI models on an HPC
or what have you. Right.
So therefore, I think there's a lot of this coexistence of these two.
And to your point, I do feel for those who are in those,
let's say that third thing that we talked about in science,
in chemistry, in biology,
I think there is real reason why you want to pay attention
to what's happening in quantum, because even with 10 logical qubits,
100 logical qubits, you could use things out of it
in order to be able to build better chemistry models,
better biology models.
And that's sort of where we are, even with Azure Quantum.
And so and that's why I think you even see the market a bit react to,
you know, like in an interesting way, like biopharma.
People are even going public.
I'm glad about that.
People are willing to invest in quantum.
It's just a little more like in that phase, which is good to see
for even leading edge deep science.
I guess we have to talk about the multiverse theory.
I mean, we just have to.
It basically suggests that quantum events indicate the existence
of many parallel universes.
Do you believe in the multiverse theory?
Yeah, the Copenhagen consensus is a little more me than
I think multiverse, but it's kind of like, man,
let the physicists duke it out.
Let me stay out of it.
It's like I'm just going to build a software factory.
What about the philosopher in you?
I exist.
So there must be something I've observed and let me leave it at that.
That is measured.
And I think that's a reasonable place to be.
Satya, you know, you're an incredible father
and you're obviously, you know, very close to your family.
One of the things that I think a lot about, I have three young children,
is that I think that by the time they're all kind of entering
the workforce, like in 10 years,
the world is going to probably look pretty different.
I don't know if you'll have AGI or something, but it's going to look
very different in terms of
what it would mean to be, I think, a productive member of society.
The work that they might have to do on a day to day basis.
I'm curious, what advice would you give me, give us to raise our children
for that future?
Yeah, I mean, it's great.
One of the things that I heard about, what does agency look like going forward?
I think is the real question.
I think we're all grappling with, does this increase human agency?
Does it somehow take away
because we are subject to someone else's decisions or what have you?
It's a fascinating thing.
So I'll come at it two ways, right?
One of the most profound moments I ever had in terms of the A.I.
demos, one was when I first saw the early versions of GitHub
copilot and said, oh, man, this thing works
because software program, believe me, it was like the toughest thing
was to convince any software developer that this thing is any good. Right.
So and but the demo that did
the most for me was I was in India.
I would say in twenty two, I think early January.
And I saw someone showed me a demo
that put together by basically daisy chaining.
I think at that time, maybe three, five GPT, three, five with sort of India
has this, you know, unbelievable sort of stack of digital public goods.
So there's text to speech.
Speech text is all available as a sort of a public good that somebody took
and put together a WhatsApp bot that a rural Indian farmer
was able to just go ask the bot to say, hey,
find out about this subsidy program I heard about on television.
And it comes back and says, go to this website, fill out these forms
and you should do it.
And then he says, I don't know what a form is and I don't know what a website is.
Do it for me.
And it started.
It's just talk about this is three, five.
And and the fact that
what dawned on me is think about the agency.
This is a rural Indian farmer
who gets something developed in the West Coast of the United States
like a few months before.
Thanks to the ingenuity of this place and some programmers in India
putting in their hands a technology that suddenly because this is a person,
you know, who is absolutely has agency except has no access
and suddenly you've changed access.
Right. That is another way to think about our children.
I think which is what would they do
if they really have with them now
the ability to code anything they want,
research anything they want,
build anything they want.
Is there like so in some sense, would we put premium
a lot more on curiosity than expertise?
I think that's kind of perhaps one of the more interesting things is
will we educate, you know, think about the confidence.
Right. You guys are you know, you actually cracked IIT.
I failed the IIT.
And so think about it.
Right. It's like, hey, it doesn't matter one exam.
Right. Go and think about like, how would you be able to
give confidence to every kid out there that they can go explore anything?
And that's that is the optimist in me.
There are a lot of other scenarios here.
But I think the core curiosity
and critical thinking are still going to be perhaps at a premium.
And somewhere coupled in it is probably confidence.
I love that answer, because I think that
I'm a pretty confident person.
So, you know, I'm going to, you know, I think get impart that to my children.
But maybe even more so kind of like that, like, hey, these computers
and this intelligence, it works for you.
Right. Like what can you make with it?
Right. Like what will you make?
And I think that's incredibly empowering.
I love that answer.
But you're clearly very competitive.
Microsoft is also a competitive and winning culture.
How do you channel that outside of work?
Maybe in your cricket team? Yeah, that's it.
I was I was happy to see today's morning match, but I'm obsessed.
I think every other South Asian is with cricket.
But I think the interesting thing that I've been thinking a lot about
is talking about competition,
at least at this stage of my life, and
is I want to make sure that I'm thinking about the game
that is being played versus the playing the game
and being a lot more deliberate about it.
I feel founders by definition are doing that
because that's how you define a company and you say,
what's the game that I want to play?
But like competition is sort of
really challenging yourself first to be in the right competitive set.
That's another unbelievable thing in the tech industry, right?
Which is if I look at it, we've been lucky enough.
Like here's my story is I came to the United States in 88.
The Berlin Wall fell in 89.
I entered the tech industry in 1990.
And man, what a ride.
It's an unbelievable time.
And so the question is and that's just luck.
Right. And so the question is, what is the
the the setup and the competitive set?
That is the most important thing is that what I've realized.
Right. Which is you want to be in the sector,
in a space with competitors who are awesome.
Yeah. In fact, that is better than being in some place
where you may be a winner, but you're not really.
And so therefore, I kind of have gotten a little more orientation
on thinking about the game, the game that is being played
or playing the game versus the game you want to play.
Just as what a time to be alive.
Yeah. Computer program.
And to be in the arena of like the greatest game,
I think of our of our time, it's a gift.
You know, we are running a bit behind.
So we have to wrap up here.
Ruchi, do you have any final things you'd like to say?
I have no final things, but I want to thank you so much for coming out.
It's been inspirational to watch you.
It's been absolutely amazing. Thank you.
Thank you so much. Thank you.
That was another episode of Minus One from the team at South Park Commons.
Make sure to subscribe to our show wherever you listen to podcasts
and find us on social at South Park Commons.
And thanks to our friends at Atomic Growth
for their support in bringing this episode to life.