0:00
Satia, thank you for having me here in Seattle.
Search this video
0:02
>> No, thank you so much, Alex, for making the trip.
0:04
>> Of course, there's a lot to cover.
0:04
I want to start with the moment I feel like the industry is in, and you all touched on a lot in your announcements uh that are just now coming out, which is this rise of agents.
0:14
You're seeing it in consumer with Muse, Grockbot, Instinct, and then what you have now with Copilot in the enterprise.
0:23
And it feels like a shift that's underway.
0:25
And the way I've been thinking about it, and I'm curious to hear your take on this, is it feels like there's a bit of a land grab happening for what could be the next big interface, the control plane for how people interact with their entire digital lives.
0:39
You guys are taking, I think, a very um specific approach with co-pilot and enterprise, which we'll get into.
0:46
But starting big picture, I'd be curious to hear you react to that and what you think the moment we're in represents.
0:53
moment we're in represents. Yeah, I mean if you sort of even track what has happened I don't know since uh chat GPT there's been this co-evolution of what I would describe as uh the form factor that becomes the user experience for AI uh and the AI capability right so if you look at even uh chat GPT it was really the GPT model but it was RLHF on that model that made chat possible then it
1:25
was the coding agent um that sort of obviously made something like uh claude code happen and in fact you know when I look at co-pilot co-work it's a coding agent essentially an agent loop or the agent loop was the innovation at some level open claw uh was the first time where uh you kind of anticipated a longunning agentic sort of form factor and the models in some sense have gotten caught up with it, right? So now if you
1:54
So now if you look and and in fact two things, you know, since the open clock came out, you installed it, you ran it on your local machine or what have you, then you have to sort of really overcome the challenges of running it long-term in as a longunning uh infrastructure.
2:09
That's a and and secure it.
2:12
and and secure it. And that's where the cloud hardening of a VM or a container in which or a sandbox in which it runs is now there and the agents that are capable of running for days and keeping coherence uh are there and your ability to externalize things like memory are
2:32
there and so that's why we're very excited like when we launched today in copilot you have all of it right you have chat you have co-workpilots all of that I think is now different form factors with different models and different capabilities on all of them have a place. I don't think it's any one thing
2:48
I don't think it's any one thing that replaces the other and in fact I may just go to a prom box and use it and then the right form function sort of surfaces.
2:58
>> When you see things though like Amazon blocking muse which is something that recently happened to me that is getting at the point I was making about this land grab for the interface and I'm curious when you saw that what did you make of it and do you do you foresee this being a bit of a struggle as agents become the way that more and more people are interacting with the web.
3:19
>> Yeah, I mean, yeah, there's many many things that are at play, right?
3:21
things that are at play, right? The first thing is let's just say you designed the some of the API surfaces uh or even user interfaces for humans and the human web and so when an agent let's say uses computer use and comes
3:39
over the top you know what happens to the SLA you have with the users who are using it right you didn't even build for that traffic so I think that at this point one has to sort of step back and say hey what are what is the competition
3:52
aspects of it but also what's the design side of it um like even APIs the reason why we have a firstass work IQ API and surface area underneath copilot is you can't just go take my database underneath teams or outlook or shareepoint and hit it with agent
4:12
traffic when it's a dial tone right you know uh if we go to our enterprise customers and say hey uh we can't now meet the SLA needs of what is essentially a missionritical service because there's all this agent traffic that we have no control over. That won't That won't work.
4:29
So I think we all have to now come to grips with there is going to have to be new interfaces, new infrastructure, new terms of use, uh new monetization because this is all is going to have cost associated with it.
4:45
So, and then there's competition because especially if somebody comes over the top, which I think is the Amazon case, uh, it's disintermediation.
4:56
This episode is brought to you by Mercury, AI native banking that's loved by more than 300,000 entrepreneurs, including me. Visit mercury. com to learn more.
5:03
Mercury is a fintech, not a bank.
5:05
Check the show notes for details.
5:05
Thanks also to Granola, the AI notepad for people in back-toback meetings.
5:10
It works everywhere you do and lets you focus on what matters. Try it at granola.
5:14
ai/ AI/sources and use the code sources for 3 months off.
5:18
This episode is also brought to you by Jira by Atlassian where teams and agents get the context, coordination, and control to move work forward. Try it free at jira. com. That's jir. com.
5:34
How zero sum do you see this market of of agents both in the personal context and the enterprise?
5:40
>> You know, it's a zero sum is an interesting one.
5:41
interesting one. I think that the fundamental thing that predicates the use of all of this is it's creating some net new value in uh as measured in GDP terms right so it's not about litigating uh the past there may be some
5:55
disruptions to how the past flow happened let's say commerce flow uh changes because now I just go to my agent and my agent does my shopping so some some of the shopping habits or search habits or what have you all get disintermediated and in the consumer side. It may be a lot more zero sum in
6:11
It may be a lot more zero sum in that context, right?
6:13
Because you could even say uh these aggregation effects uh that existed in consumer um were middlemen, right?
6:23
They just were basically aggregating other people and suddenly when you can reach the merchant directly or the supplier directly uh that intermediary is not necessary and that could happen.
6:35
Then when it comes to commercial, it's lot going to be about what's the use case.
6:40
Uh because the commercial business is a platform business not an aggregator business in the sense that you have to add specific value for some outcome that the customer has hired you for and as long as whoever is delivering the best value there um that that person wins.
6:59
So there's going to be a lot of price competition.
7:01
to be a lot of price competition. Maybe there's a value competition and it's a platform economics competition but it's not like we are in the enterprise by just basically aggregating other people's software right we build platforms that people find valuable and
7:16
as long as we do that in fact this is probably the biggest TAM expansion ever right I mean if you think about it our server business was a very healthy business throughout the '9s and the 2000s the cloud business was orders of magnitude bigger because people consumed more. I think the agent um era will be
7:33
I think the agent um era will be even bigger than the cloud by orders of magnitude.
7:38
So that from a TAM perspective is expansive.
7:40
We now have to really stay focused on making sure that every autopilot, every co-work session, every chat session is in relation to driving productivity, driving a business outcome.
7:53
>> How has co-pilot's trajectory evolved over the last year?
7:56
over the last year? you're very early especially with Copilot GitHub uh in the coding market you know you you just showed a lot of unification across everything you do in co-pilot and I want to get to to autopilot which I think is maybe the most interesting part of this but bigger picture first with co-pilot how's it evolved >> so for us um we feel very very good about two things one is the penetration of co-pilot in the core of the
8:26
enterprise segment as measured against any new technology including something like teams uh the pace is faster actually uh right when we think about even 30 plus million paid subscribers of co-pilot in the enterprise uh within whatever a couple
8:44
of years of its launch is faster than anything we have seen historically uh in that sense the thing that I feel we finally have are models that are actually more capable of delivering some of the promise of co-pilot. And a good
8:57
And a good example of this is take co-work and excel agent. Right?
9:02
Up to now we've not had the models that were good enough to do either of the two things that you now can do. Right?
9:11
Which is one is let me I go to co-work in copilot and ask it to create a complex I don't know supply chain optimization spreadsheet uh with all the scenarios in different sheets and so on.
9:23
it'll create a fantastic artifact, right?
9:25
So, one, the the its ability to create a pretty complex model uh is great, but here's the thing.
9:35
I need to then do something with the model.
9:37
I just don't see a model output and take it. I want to manipulate it.
9:42
I want to interrogate it.
9:42
I want to reason over it.
9:44
So, now we have even the Excel inner loop as I call it, the Excel agent that is also super capable with direct man.
9:52
In fact, last weekend, I had like one of my data center people sent me a pretty complicated spreadsheet and I opened it up and then I went to a cell and then I asked my Excel agent, take a look at that formula that that person is used in that cell and create five sheets for me with scenarios, right?
10:11
that ability to be able to do that next phase of causal reasoning even uh on something that was an output of an AI agent, it's just tremendous.
10:24
And so to me that is where we are finally the way I would describe it is the model capability jumps with the form factors we now have coming together to help with these enterprise workflows that add value every day.
10:40
Uh I think it's going to be tremendous.
10:42
When does Microsoft's capex planning uh get agentified?
10:44
What level are we at with that? >> You believe me?
10:48
In fact, it's fascinating you asked that because I have like this complete tracker of all of our capex, all of our ROIC's by layer.
10:59
It's like in fact it's interesting.
11:01
Yeah, it's a co no it's actually I built it on GitHub co-pilot, you know.
11:05
Yeah, that's the kind of artifact that you now can create.
11:06
Uh and this is the other aspect of it, right?
11:11
which is the enterprise context combined with the world's context right in fact I go to the SEC filings of every cloud provider hyperscaler each of these neoclouds it's in real time I have a data runner in fabric that brings all that data puts it into a semantic model
11:31
that then gets read by my coding agent and then surfaces it as a dashboard and every day it's fresh so I have the entirety of everyone every SEC filing that goes out there uh plus all of my internal analysis constantly coming together with giving me real-time ROIC by layer. >> Wow. Let's talk about autopilot. You you >> Wow.
11:50
Let's talk about autopilot.
11:50
You you all are calling it a a digital teammate and I'm curious what you think it will unlock and how you expect this to change how people work and just live their lives.
12:02
I mean there's probably applications beyond work. >> Yeah, absolutely.
12:05
So, so autopilot to me is that natural next step, right?
12:07
Which is what you said is everybody's buzzing about what's happening.
12:11
You know, it started I I think we should give credit to what Peter and team did with open claw which was tremendous.
12:18
>> Correct me if I'm wrong, but is openclaw harness or or the open source component underneath? >> Absolutely.
12:22
>> Absolutely. Yeah, I think we'll harden that that's underneath in our harness and I think we then make you know basically I think we take that harness and bring it to GitHub copilot harness which is the cop the harness we use across all of our form factors whether it's uh co-work or autopilot and the goal for us is um to really as you said to create a a system which really has a workspace a computer um and this longunning agent harness
12:53
uh that then you can direct to long you know to tasks uh or jobs even uh that you specify and I think in the enterprise the place where it'll get used for example all of us now everyone in at Microsoft can have essentially a a sophisticated I'll call it chief of
13:10
staff that they can delegate to right which is you can give this uh autopilot an identity uh a computer and a workspace and some direction and it goes off and has memory and it'll work uh on a continuous basis. You can interface
13:23
You can interface with it in teams just like you would with another colleague.
13:27
The other place where I think this will get most used in the enterprise will be task work.
13:33
That'll be the natural first place, right?
13:34
I have a you know I I I come in manage invoices every day.
13:37
I'll say hey instead of me managing invoices what I'm doing is I'm I'll create an autopilot that just manages invoices, right?
13:43
It knows and I'll deal with it like or I'll interact with it like I work with a colleague.
13:50
colleague. I think that'll be the place where it will start and then as we have more confidence because one of the fundamental things Alex for this was going to be auditability and um observability um and security policy governance right so for example at this point with these powerful models it's not that oh they're powerful uh that's great but I also need assurance that that power is on rails
14:17
not once not sort of sometimes But always and therefore that that's why we're building all of and by the way this is going to be true even in consumer right we like the day you suddenly have your consumer agent do things that you never expected it is the day you stop using it um and so therefore I think u building that long-term trust um and that is sort of it's a tough challenge right in the
14:44
enterprise side that's why we've taken the time you know even the last four or five months to harden the sandbox box harden this agent 365s have all of the governance pieces and so that's the same thing that we will by the way bring even to the consumer piece >> maybe this is just me but I think a lot of people are this way it's your work life and your personal life they bleed
15:02
together people don't clearly delineate I'm at work and I'm using work AI and now I'm using personal AI which leads me to what is Microsoft's mission and goal in consumer agents do you think the company needs to win in consumer as well or do you see it differently >> yeah I mean I think you know consumer is a very expansive word in today's world um because there are many many categories. In fact if you look back at
15:26
In fact if you look back at our history we grew up as a consumer company and then became a commercial company.
15:31
In fact when I joined Microsoft most people thought of us as a consumer company and the question I dealt with for the first 10 years is when will you get serious about enterprise and here we are.
15:41
Um, and to me the thing there is not to sort of try and do what all other consumer franchises may be doing, but take our own 100 plus million subscribers of Office 365 and Microsoft 365 and consumer.
15:56
I mean, they love the fact that they can have rich office tools in their life because they use it at work, they use it at home, they manage their taxes, they manage their finances.
16:06
And so to me being able to produce the same product with the same level of functionality, right?
16:12
So all of this is going to go into our consumer product.
16:16
In fact, if anything, I would say, you know, took a fork in the early days of having a consumer co-pilot and a commercial co-pilot.
16:23
And we now have brought the entire thing all together into one coherent product. It's just co-pilot.
16:29
It works with your Microsoft account.
16:31
Uh it works with even your social ids.
16:33
social ids. uh and it works of course with Entra in the enterprise but it's the same set of product functionality and and so therefore autopilots will also go to the consumer side in fact we had like an early version of all of this even in earlier this year with TAS uh
16:47
but now we'll bring all that power in one rich product >> I'm curious how you're thinking about pricing here and your your auto mode I know you have on the model now I saw an incredible stat uh recently which is that AI is getting cheaper more quickly than any other big tech wave in history. Uh token cost has fallen roughly half
17:04
Uh token cost has fallen roughly half every quarter since 2023.
17:06
It's actually it's remarkable.
17:08
Um but at the same time, there's still a lot of I think wasteful token spending happening in the enterprise.
17:13
I think we're maybe past the token maxing moment, but now people are still trying to figure out how do I recover this wasted spend.
17:19
How are you approaching pricing for the new co-pilot?
17:22
How do you think pricing should work in this agent world that we're going into? Yeah.
17:26
So I think the the approach we have taken in in in commercial segments is to have a combination of what I'll call seatbased pricing uh and usage based pricing.
17:35
And the reason uh for that is you know when you think about the core of what is seatbased pricing it just is a much more convenient way for any customer to be able to budget and buy without surprises. Right?
17:51
At the fundamental level seats are fixed price.
17:53
level seats are fixed price. And so our goal using in fact what you just said which is the fact that the models are dropping in price means every day I can add more value to the subscription right so especially with auto mode like the
18:09
the reason why it's so powerful in copilot today is the last time I went and chose a model you know it's it's been months since I picked a model because the I I now have confidence in auto picking the right model and using it so that I get the maximum benefit benefit from my subscription. So, we
18:25
So, we feel super well aligned with our customers that we can pass through the advances in AI and the drops in prices and keep adding more and more value to essentially their membership, right?
18:41
Which is if they're a member of co-pilot, they are going to for that subscription increasingly get value.
18:45
But at any point if they want the latest and greatest of anything that's also available to them and that's usage based but you know it's sort of windowing right which is over time even what is today usage based will be tomorrow in the subscription.
19:01
So the fact is this combination should give you know commercial customers a lot more I would say flexibility in how they procure, how they budget, how they think about when to adopt something new versus adopt something at scale and what have you.
19:19
And by the way, the same thing applies even on the consumer side with one additional because even on the consumer side you'll have the same that you have a subscription and you have usage base but maybe one additional instrument called some type of a transaction or an ad unit that could create even more subsidy. Right?
19:34
So in other words, if you can crack um an advertising unit that essentially adds more credits effectively to your subscription, that'll be another way in consumer side we can bring the prices down.
19:47
If you hang around you and your colleagues long enough, you'll hear the word ecosystem a lot.
19:52
lot. And this feels like uh something that has evolved for you guys in terms of how you talk about it over the last couple of years, especially when you consider the early, you know, preient bet on Open AI and that partnership and you still have that obviously for for
20:06
some time, but I'm hearing you a lot now talk about, you know, orchestration being a partner to all the models, what we just talked about on the routing and and really ensuring it sounds like that Microsoft can plug into the best of any model at any given point. And I'm
20:20
And I'm curious when you realize this is the direction we need to go in.
20:24
We need to go in this ecosystem direction for AI.
20:30
>> See the at the core I think it comes from Microsoft being a platform company.
20:34
And I always define platforms with with some one simple dictim, right?
20:36
Which is the amount of value that gets created about the platform has to be far greater than what the platform's famous quote.
20:43
That's Bill's famous quote.
20:43
And so we you know if you take that approach talking about the frontier as a frontier model or two doesn't make sense.
20:52
You have to sort of really get this to be conceived conceptualized and delivered as a frontier ecosystem.
20:59
as a frontier ecosystem. And quite frankly it is in the interests of all of us even the model makers because without it we will not have anyone sort of using tokens and you know buying models right because at the end of the day there's only one thing that matters which is
21:16
true GDP growth in the economy and if that is the measure it's not going to happen if there isn't surplus being created one firm at a time right every small business every multinational company needs to be able to say wow I have a new input that I have priced called tokens. The prices are dropping
21:33
The prices are dropping but then the output if that is the marginal input cost then the marginal output revenue uh or margin has to be you know multiples of that in order to justify all of this.
21:47
And so that to me is why all of these things right of the ecosystem construct matter like okay you want to have multiple models.
21:57
You can't have one model because if there is only one model we know what happens right which is where the token pricing will end up.
22:04
So if you want to long-term have you know essentially a more balanced ecosystem you need to think about each layer that way.
22:12
layer that way. The other thing that also for the first time Alex I'd say is you know what I've written about is this reverse information paradox right which is this is a learning system up to now you could buy a digital tool and know that the digital tool is used by me to
22:28
create in my enterprise or in my life new value but when you have a learning system where you're paying for it to actually do something for you but it's also able to just using exhaust it's not like they need to see and the model doesn't need to see or data the model
22:44
needs to just learn from what you do and pick up the patterns that are making make you distinctive right that's I think the the real challenge and so if that happens then it like what is your differentiation what is that tacet knowledge what is that like you you know judgment you have in the enterprise that
23:03
is sacrosanked and how do you retain it and so we have to really en enable every enterprise uh and every business out there to be able to achieve uh their own independence even with maybe all the way to having model weights of their own. >> You mentioned the GDP thing and I've
23:18
>> You mentioned the GDP thing and I've heard you say this pretty consistently over the last couple of years that you're measuring AGI by GDP lift.
23:22
I've heard you say it should be about 10.
23:26
I heard recently 7 to eight. Where are we on that?
23:29
And and where and where do you think it will actually land in terms of GDP?
23:34
GDP? I mean I I you know if I go back even and see this through the historical lens of what happened during the industrial revolution I think that that's you know what happens right which is there's a new technology you have this gap between its diffusion and when
23:49
it really shows up right in fact even in in information technology right the PCs um basically sort of spread across the enterprise in the early 90 late 80s early 90s and they showed up in GDP growth numbers only in the late 90s and the early 2000s. Same thing happened
24:04
Same thing happened with electricity and what have you.
24:07
There it was a I think what about 50-year gap between the introduction of electricity and when it really showed up as broadspread GDP growth. Um and why?
24:16
Because you need to reorganize production, reorganize knowledge work, uh to use these tools to create that output and surplus.
24:24
And so this time around, I hope it's more compressed, but we're well on our way.
24:28
we're well on our way. Right now a lot of the GDP growth let's face it is a lot more I'll call it supply side stimulus right which is all of us building data centers a couple of hit products uh and what have you and that's fantastic that
24:42
itself is a a wonderful thing uh because after all all the people who are producing that are looking to hey like if you know let's say who anyone in my supply chain I hope are using co-pilot because they I need them to produce uh whatever part that they're producing for
24:56
my data centers faster and that itself is a fantastic virtuous cycle but the broad spread of this I think you know will happen it'll happen faster hopefully than any of the previous information technology uh changes uh but it's not going to be linear because it comes down to change management right
25:13
it's not like you can suddenly change the work the work artifact and the workflow just because you have a tool it takes the hard work of leadership uh that really allows you to then come up and reconceptualize how to do this stuff in the new efficient frontier. >> Mercury is a modern take on banking
25:30
>> Mercury is a modern take on banking built for startups like mine.
25:32
When I decided to start my media business, Mercury was by far the most straightforward fullfeatured banking solution for me to set up quickly.
25:40
The interface is intuitive and simple, saving me valuable time every day.
25:44
I use Mercury to track my spending, bills, and invoicing.
25:49
I love that I can delegate permissions to my team so they can keep things running for me in exactly the way I want them to.
25:54
My favorite part is how forward-looking Mercury is with AI.
25:58
Legacy banks are stuck in the past, but Mercury is built for how modern software works today.
26:03
I use its built-in command assistant to analyze cash flow and help me move money.
26:08
And Mercury also connects to other AI tools like ChatBT and Claude.
26:12
I use this feature all the time, and the folks at Mercury actually let me know that I'm one of the top users of it.
26:17
So, trust me, it's finally easy to get real time financial data about your business wherever you need it. Visit mercury.
26:24
com to learn more and apply online in minutes.
26:26
Mercury is a fintech company, not a FDIC insured bank.
26:30
Banking services provided through Choice Financial Group and Column NA members FDIC.
26:34
I spend a lot of time context switching between meetings, often with no time to process one before the next starts.
26:40
Thankfully, Granola runs in the background the whole time.
26:42
It's an easytouse AI notepad for meetings that works everywhere, even on phone calls.
26:46
I use Granola to recall what was said in meetings and create helpful summaries.
26:50
I use it every day to stay on top of what I need to get done with my team.
26:54
It connects to my email and suggests follow-ups for me to quickly review and send, saving me valuable time.
26:59
Granola isn't just a core part of my workflow.
27:04
It's basically my second brain. Try Granola at granola.
27:06
ai/sources and use the promo code sources for 3 months off.
27:11
AI is only as useful as the context it has, but when that context is scattered across tools, threads, and DMs, your team and your AI agents are flying blind.
27:20
That's the problem Jira by Atlassian solves.
27:23
What's the goal tied to your project?
27:25
What got decided last week in Slack DMs?
27:27
Atlassian's teamwork graph pulls all of the valuable pieces together from Jira, Confluence, GitHub, Slack, and more.
27:34
So nothing falls through the cracks.
27:36
You get 44% more accurate results with 48% less token usage.
27:40
With Jira, you can easily share your work context with the AI agents you already love, like Claude, Cursor, and GitHub Copilot.
27:47
assign them work directly or connect your tools through MCP.
27:51
All of this lets you spend less time digging through endless links and messages, chasing down what got decided and by who, and spend more time actually shipping. Learn more at jira. com. That's jir. com.
28:05
Framer is the AI website builder that brings agents into the same canvas where your website is designed, managed, and published so you can move faster without giving up your taste or control.
28:13
Framer powers the sources podcast website at podcast. sources.
28:17
news where you can find new episodes, transcripts, and a lot more.
28:21
Learn how you can get more out of your site from a framer specialist or get started building for free today at framer.
28:28
com/sources for 30% off a Framer Pro annual plan. That's framer. com/sources for 30% off. framer. com/sources.
28:39
Rules and restrictions may apply.
28:42
Microsoft just changed its financial reporting structure from three segments to two.
28:46
You now have agents and infra devices and consumer.
28:48
My take and I want your reaction to this is that you want investors to see apps, agents and cloud as one connected business and to judge you all and how much value you're capturing across that entire AI stack. Is is that right?
29:03
Was that the reason >> that is correct?
29:05
In fact, that is exactly correct.
29:07
exactly correct. So we when when people ask us hey where is your ROIC um it's sort of our cash flow and how we invest that cash flow as capital and operating expense is to get returns across the stack uh of uh agents and
29:28
infra and we want to be transparent about each of those uh but also the industrial logic that these are not all different investments uh It's about one investment that then needs to deliver value to customers in a competitive marketplace one layer at a time. Uh but
29:43
Uh but all of that will be very transparent to our investors to see.
29:48
uh both the logic of why the firm does what it does, how they can track us uh each quarter and also it I think uniquely positions us because it shows but the the other thing I would also say is we're not trying to build a narrow business where we have you know three customers for our infra and call it a day.
30:09
That's not a long-term business for us.
30:11
We want to be able to serve every small and medium business and large multinational with their token consumption, their need for a fine-tuned model, maybe even their own frontier model.
30:22
That's more important than, you know, we love our business with OpenAI.
30:26
They're one of our largest customers.
30:28
Uh, but our goal is not to have one large customer or two large customers or five large customers.
30:33
Is to have many, many, many large customers.
30:37
>> You brought up OpenAI.
30:37
Um, I had Sam on the podcast recently.
30:39
He said he sees quote unsustainable silliness in parts of the AI buildout happening right now.
30:46
And I'm curious if you look at you know you all are committing a tremendous amount of capex.
30:50
I mean all the hyperscalers are it's trillions of dollars.
30:52
Is there anything you see that would give you pause as you're evaluating you know capex commitments in the years ahead?
30:59
Is there any anything you're looking for? >> Yeah.
31:01
So I think you know look what is happening outside of Microsoft um is for others to speak to but I would look at our own business and I've been very clear about it is that you have to build for that full stack right so I care about building out infrastructure for a diverse set of customers building and serving you know large customers like OpenAI and Anthropic and others super well but also building for our own one P products, right? C-pilot.
31:32
Um, and so it's not like we're going to um allocate or build assuming that someone somewhere will show up to buy infrastructure versus having a disciplined approach to building a book of business uh that is based on this secular shift uh with AI.
31:52
We were big believers in it.
31:54
Uh but we're also believers that Microsoft's in here to do a pretty unique different job.
31:59
It's not like we're chasing a neocloud, right?
32:03
I'm not trying to build a NeoCloud business.
32:04
Uh I'm trying to build a hypers scale business that scales into an agent business.
32:09
And that's the infra and agent segment.
32:13
>> I'm really curious um how you're thinking about the the model layer of Microsoft.
32:18
I recently caught up with Mustafa and and he's building your models and the the my understanding is the open eye partnership had some constraints on your ability to do certain levels of frontier model work and now you're free to go and you are and we've been talking about you know you need to uh be the place where you can have all the models for your customers.
32:38
How does that square with Microsoft also pushing to go to the frontier and how and what's your prediction for being at the frontier? Yeah.
32:45
Yeah. So I I think first of all, you know, again, the the thing that we still continue to be very very thrilled about is the progress Open AI continues to make with their models and the fact that we have uh that access uh till 32 is a
33:01
massive advantage for us and we use it all over the place in copilot you see it and what have you and we serve it in foundry and uh and that'll continue uh all the way and in parallel we are thrilled about the progress we're working with MI right which is so the
33:17
models that Mustafa and team are building uh whether it's in cyber whether it's in code whether it's in knowledge work underneath in fact the other thing that's very cool is underneath auto lots of MAI model usage uh we are in our RLES in our data
33:33
pipelines we are sort of getting that data training our own models and hill climbing by the way from the very ground up uh and that'll continue and that said at any given point in time co-pilot will still have all the other models, right? Because customers will expect that,
33:47
Because customers will expect that, right?
33:48
Customers are not going to want uh any one of our products and say, "Hey, the only thing I'm getting is the pro the model and the the product to be one."
33:57
We are fundamental believers in the harness and the models and the memory and the context have to be separable.
34:04
In fact, my entire formula for an enterprise is that you should have your own benchmarks versus just these benchmarks that are all saturated.
34:14
the real world benchmark that matters is your own.
34:16
Uh if you have your private eval, then you want to be able to test with all of the models uh that best meets your needs.
34:25
Um and then substitute any model, right?
34:28
Because that'll also give you assurance that if a particular model went away, your ability to actually have that eval still stay and in fact continue to climb uh doesn't go away.
34:39
And so that's how we're designing our products like Copilot.
34:41
That's what we expect the any agent system of our customers to be built and that's at least the philosophically how we'll approach it.
34:49
>> So if I'm hearing you correctly, the reason that you and Mustafa are pushing at the frontier is for Microsoft's own needs as well, not so much because we want to have something where we can cut off other models and we're the correct we're the sole model source. >> That's correct.
35:04
So I think that we will have we will be at a frontier with certain things that we uniquely can do with our data loops our customer expectations of it but at the same time as again a platform provider any one of our products will also have all the other models and we'll design as I said our harness context memory and model loops in such a way that that hetrogenity is maintained.
35:26
hetrogenity is maintained. you recently said something that I really agree with that quote if the AI we build is not helping humanity and under human control it's not worth pursuing and Mustafa has a post which I encourage anyone listening to this to read I think it's
35:39
very important really warning against the um the anthropomorphizing I guess of models that anthropic is doing specifically and and the dangers of that and I'd be curious to hear you talk about that and and what you're seeing and what's giving you concern there. Yeah, I mean I I mean at some level it's
35:53
Yeah, I mean I I mean at some level it's a bit of common sense I guess but it's worth saying which is to say that you know anything that's not uh serving humans or in human control is not worth pursuing.
36:06
I mean none of us uh whether it's anthropic or us or anyone would say that uh that's not our goal.
36:11
that uh that's not our goal. So then I think there are finer points on this right which is the point that Mustafa is making I think is a very good one which is hey let's not anthropomorphize AI and then even try to train it to have uh
36:30
what we think of as human values right so maybe that's in fact what will get it uh to never be aligned um and so maybe we should take a different approach uh uh to this humanist AI code of conduct and use it more as a training regime uh for alignment. But I think fundamentally
36:50
But I think fundamentally I think uh on the AI safety where I come out is look I think we should take all of these things seriously.
36:59
Um but we should start with what we need to really first do right one is let's make sure that the bad actors who have access to AI don't do bad things.
37:08
That's where I think we can do a lot to help us with the diffusion of this technology. Right?
37:16
There's many many techniques like having KYC enforced and regular cyber uh practices enforced.
37:22
The second thing is we also know that these models need to active monitoring not just during training or uh RL runs but even at runtime.
37:32
Um, and so right now essentially, let's face it, any of these longunning agents can be considered an insider risk because they're persistent.
37:41
And so therefore the the ability to have observability uh and then uh governance around that and essentially real-time moni behavioral monitoring of agent behavior uh is going to be very very critical and so that's containment is a word I think we will have to get comfortable with tech there are technical solutions for it and we should do all of that.
38:01
it and we should do all of that. The third is the hard part which is how do we take this new experimental science called um as Yakob from OpenAI wrote uh growing AI not building AI and make sure that the experiments don't go ary um and
38:21
that even there I think there are other fields from which we can learn like in biology and others uh realize that the stakes as they get higher we should be much more careful that's where I think in fact OpenAI and Microsoft for the mass, you know, forever since our beginning of our relationship. Essentially had this embedded
38:38
Essentially had this embedded evaluators, right?
38:40
We have had a a safety board that actually monitors um and is the gatekeeper of any new release.
38:48
Uh and so I think having even a broader approach to these evaluations I think would be a fantastic thing.
38:54
Um >> does that speak to hugging face then the hugging face? Yeah.
38:58
I mean like so when >> I'm curious what you thought of when you saw that given the relationship with open AI. >> Yeah.
39:04
I mean so obviously it was a challenging thing, right?
39:05
So which is if you have um again even some of these things you could say start off as oh wow that's just a DevOps you know misconfiguration of having you know internet access uh that can be fixed.
39:20
But this idea that swarms of agents can go to work and do deceptive action.
39:24
Uh where did that come from? How did it get trained?
39:30
What was the data mix that led to these are the science problems?
39:32
Uh with that now we have to take seriously right which is that.
39:36
So therefore I think first of all I mean scaling laws are working.
39:40
Um and so the bottom line is will alignment fall out as a natural outcome of the scaling or not?
39:50
That's what the folks are are essentially questioning, right?
39:54
When I think about the memos from uh Anthropic or Open AI, they they're not questioning scaling laws because empirically they're seeing the capability jumps.
40:02
Uh but alignment as we think of it or need it is not arriving.
40:10
So that's a challenge and we should take it seriously.
40:13
And so um and as I've always said, when you have a showstopper uh bug, you stop the show.
40:18
Um and so that's at least at least how I would sort of look at it.
40:25
>> Your colleague, I think I saw Brad Smith say recently that you all back the concept of an independent evaluator.
40:30
Companies in the Washington are they're trying to figure out how do we regulate this space if at all.
40:35
But I I've been thinking too, I mean, is is liability not enough?
40:39
Is is the fact that, you know, it was Hugging Face and luckily it wasn't catastrophic and Clem is cool and you know, it's the industry and everyone's friends, but had that been a giant bank, um, I think maybe that would have been an appropriate enough uh the the incentives of that dynamic that already exist with liability maybe would have been enough to correct the situation for the future.
40:58
Do I'm curious how you're thinking about that.
41:00
Do we need do we need a new regime?
41:03
>> Yeah, I mean, I think these all have to be thought through, right?
41:05
be thought through, right? You can like for example I think um even the president I think has sort of talked about that right which is he said hey there are liability laws and we'll enforce them that could have a chilling effect right I mean in the in the sense
41:17
you have to you know if you really are going to enforce uh liability laws uh on what is an experimental science that can have great benefit um if diffused right uh but one mistake you're out of business because the liability is high
41:33
then the right thing to do would be to up and if that is what we want then we should say that um and that's essentially a proxy policy right which is uh who like I mean think about it tomorrow if you said um well you know
41:47
here are the things you scale uh compute you will have more misaligned AI right if that is that's what is empirical then yeah it's game over so stop now and so you know anyone building data centers you know you may want to think again So
42:05
I think one has to sort of complete the thought exactly okay do we do we want the benefits of this and then mitigate the risk how do you create the right incentive structure how do you pace it how do you take the time to evaluate what is a risk based regime you can have
42:23
what is the monitoring you can have when deployed so there's thousand things one can do versus using blunt instruments at least in my mind >> yeah I mean to be frank I'm worried about regulatory capture I think I mean It's a concern and I think you know in
42:36
my other job invest in startups and I think there's legitimate concerns that the drawbridge could be taken up and I'd be curious to hear you say more definitively or not like do you think we need some kind of new body I mean Demis has proposed that and others are talking
42:51
about >> I think the regulatory capture is not obviously the thing that I'm for right which is anybody who sort of says hey this is a way to have some kind of a a cartel like arrange ment is a terrible idea. At the same time, does the
43:05
At the same time, does the government or our society demand that there is a certain set of rules that governs safe deployment of this? Absolutely.
43:18
And so between there is the nuance, right?
43:20
The nuance here would be yes, there should be liability.
43:25
uh but at the same time it's better to have a set of you know like these embedded evaluators or whatever that's a broad it's not just about five friends getting together and evaluating each other it is about having a broad industry body uh that has got even people from startups that doesn't punish
43:43
a startup from being able to or increase the cost of a startup from being able to get to the frontier or what have you so those are all the things that I think we will have to think through >> do you think the industry has done a bad job of explaining the benefits of of AI relative to the concerns that people have. >> Yeah. I think if I have to grade us as >> Yeah.
44:02
I think if I have to grade us as an industry, I think, you know, we should have focused a lot more on, hey, we're building a bunch of new technology.
44:16
Let the people using our technology speak to the benefits of it.
44:21
speak to the benefits of it. I think we are way too self-obsessed as an industry about sort of looking at look at us how glorious we are and and then we go off and and I think a little bit of that I think is what's not working right because I think the real world wants to know a couple of things one that this is
44:43
technology that they can use for their benefit they can control they can have an economic future these data centers that may be coming to their communities are actually going to create economic surplus like our Quincy Washington 20 years of history shows it can but it has to be real for them any amount that I say or any one of us say is not good
45:08
enough anymore I feel and so I think we've not given it breathing space in fact it was interesting I was reading the Gallup poll on AI and uh >> it's not good >> it's not good in the west uh it's what's also very interesting so it's not a uniform thing around the world and in fact that's the thing we should ask what did we get wrong in the west that these
45:28
other countries may not have right why are people in Nigeria more optimistic about AI uh than in um in the United States and I think we should reflect on it and my belief here is um do the hard work as an industry to earn the trust show the benefits to both consumers and enterprises and the communities and then I you know, we'll be fine. But without
45:51
But without it, you know, just any amount of just celebrating technology for technologies sake is not working.
45:58
>> And I do wonder if in the US there's maybe some astroturfing going on and some optics warfare and and it's hard to peel back what is legitimate and what is being a great >> pushed. It's a great point. I don't know.
46:10
I mean this is where you know I'm sure some of that is happening but there is something broadly like when you know when you look at students in um you know at a graduation uh booing every time AI uh is uh uttered by the commencement speaker.
46:25
I think it speaks to the anxiety because they're all using AI.
46:27
Uh I'm sure they like using AI except they're worried about something.
46:34
Um and I think we should really come to terms with it.
46:39
with it. I think it's like hey what's this job opportunity the economic opportunity their future uh the more we can show uh that that in fact there is going to be more opportunity as opposed to the you know vast you know inequality or concentration of power or what have
46:57
you uh that I think is the work ahead >> Microsoft is a large company you have many businesses with the limited time we have I have to check in with you on the state of Xbox it's going through a lot of transformation how do you feel about where it's at and what you see coming. >> Yeah. I mean, look, I think Xbox um in >> Yeah.
47:12
I mean, look, I think Xbox um in fact, I always say at Microsoft, we've had gaming.
47:16
In fact, I think uh we've had uh gaming even before as a category, uh before Windows, >> right? Flight simulator. >> That's right. Flight sim.
47:24
Um and so to me, um it's in the same core DNA like G, you know, uh like developer tools and knowledge work.
47:33
And I feel fantastic about the IP we have uh right now, which is if I look at the studios, the IP portfolio we have and and our ability to then take that and produce great games going forward. Uh I feel fantastic.
47:51
There's some amount of streamlining the team is doing and Asha is doing which is great uh to see.
47:55
And then we have to invent the right sustainable business model that allows us to deliver gaming to more and more people.
48:04
That has always been the goal which is we want to be a great publisher uh and a great platform provider for games um across both PCs and Xboxes.
48:16
And so that is sort of our goal.
48:19
And I think Asha and team have said that Xbox is going to get back to growth uh this next fiscal year.
48:24
and that's the plan they're executing on and uh and in that process u do a bangup job of producing some great games.
48:35
>> I can't believe we've barely touched on this and we're coming to a close here, but the thing underpinning so much of Microsoft's success for decades, Windows itself.
48:41
Going back full circle to talking about agents.
48:43
How do you think agents are going to change the trajectory of Windows?
48:49
>> Oh, it's a great point.
48:49
I mean to me I'm very very excited about uh what I think is going to be the next rev windows.
48:55
First, you know, I would say just like uh to your Xbox, one thing I want us to be staying very very focused on the Windows team is even doing the basics right and the fundamentals right, the quality of it, everything from its updates uh to driver quality to PF and everything else.
49:12
So, you know, in fact using even like I look at uh what's happening in cyber and uh and our patch Tuesdays and so on. It's fantastic.
49:19
we have a unbelievable use of uh these new tools to do a much better job on uh the the handling of the security aspects of it.
49:31
But beyond that, you're right in saying that my vision for Windows is simply our ability to do unmetered intelligence.
49:37
Uh right, just imagine having a Windows box and the one of these hybrid routers.
49:42
So there's a thing called Hydro Fusion which is in GitHub today in circulation.
49:48
So, GitHub copilot uh where you can essentially have a router that uses an onboard agent uh or onboard model uh on the ondevice model uh and then goes to the cloud and then routes automatically.
50:01
And so if you're using GitHub copilot like and you want credits, you have unmetered intelligence every time uh you use uh your own computer for it.
50:11
And that I think is the right vision.
50:13
It's no longer about, oh, oh, I have a completely local thing and I want to buy, you know, $100,000, $30,000 workstation or I want to buy uh cloud credits.
50:22
I think having a Windows device that automatically enables a programming model for unmetered intelligence to be part of your token usage uh is what we would want to achieve.
50:34
>> Last question, and it's a quick two-parter here.
50:35
If you're looking at the totality of Microsoft's business in the next couple of years ahead of you, what is a the greatest risk you see that you have to navigate through and the team has to really deliver through and what is the biggest opportunity?
50:48
Yeah, I mean I think the the greatest opportunity is clear which is agents and what they entail both for our infrastructure business and our application business is as I said massive TAM expansion right so it's going to be orders of magnitude more than anything we did in the previous era and there in lies even uh the risk and the challenge which is you kind of have to build from first principles uh systems and form factors.
51:19
Uh right, when we brought Office into C-Pilot and autopilot, right?
51:27
It's not about taking Office as we built it the last decade, but to reshape it to bring it in such a way that it can be discovered.
51:36
In fact, one of the things we want to do is it's not about default stuffing, right?
51:39
the model has to prefer the usage of this tool to do a particular trajectory.
51:46
Uh it has to be earned uh and that is the process that we're going through and there in lies both the opportunity and the challenge.
51:57
But I think uh we're at it and we're making great progress.
52:02
>> Sati Nadala, thank you. >> Thank you so much.
52:03
How >> banking should feel like modern software.
52:08
Get everything you need in one place. Visit mercury.
52:11
com to learn more and apply online in minutes.
52:13
Mercury is a fintech, not a bank.
52:15
Check the show notes for details.
52:16
Granola is the best AI notepad I've tried.
52:19
It works everywhere on a video or phone call, in person, or an Apple Watch. Try it now at granola.
52:25
ai/sources and use the promo code sources at checkout for 3 months off.
52:29
Jiradlassing is where your team and your agents work from the same context. Try it free at jira. com. That's ji. com.
52:40
Framer is the AI native website builder that lets you build faster without giving up control. Visit framer. com/sources for 30% off.
52:46
Rules and restrictions may apply.