Microsoft CPO Aparna Chennapragada on AI at work: 100M Copilot users, reasoning models shifting enterprise behavior

the final boss of product people. This is the true gatekeeper to our plan to bring back Clippy. So we will welcome uh the CPO. Welcome to the stream. How you doing? What's happening? I think we can hear you. How you doing? Okay. Yes. Fantastic. Uh thanks so much for hopping on the stream Microsoft.

So, just so you know, I have I said there's many bosses, but Clippy is the ultimate boss. Yes, Clippy is the ultimate boss. Has Clippy actually been thrown around as an idea? It feels like more and more, you know, mega companies, like the huge companies are starting to have a little bit more fun every once in a while.

It is a bit of a wild uh I've seen some like throwback Microsoft merch that's gotten out. Is there any serious discussion in internally to think about reviving Clippy? I feel like it would be so much fun. Dude, I I'm with you. I feel like um unironically Clippy was just like early, right?

So, totally totally the original convers interface interface. Um anyway, uh would you mind introducing yourself and give us a little bit of an overview of like what your day-to-day actually looks like? It's such a huge organization.

There's so much to do, but I'd love to have a little bit more context on what you oversee and what your role actually looks like. Yeah, let me give you the me. I I um I've been at Microsoft for 2 years. I'm the chief product officer for AI at work.

So focused on kind of like how do you make work great again with AI um and if I and before that I was uh at uh Robin Hood uh and before that I was at Google for many years.

But I think the steel thread through all of this for me has been like uh the way I think about this bringing near future science fiction and putting it in product and making sure that billions of people have it.

So I did that with Google search uh and then I founded this area called Google Lens with you know the where the camera turns into a search engine um and of course Robin Hood for firsttime investors and at Microsoft my focus has been and our team's focus has been we have hundreds of millions of professionals we've got to make work not just about doing more but about being better.

So how does co-pilot work? What are the agents that you get for work today? You get a badge and a PC. Should you get a team of agents? How are you thinking about the trade-off of uh put a chat box? Obviously, Microsoft has fantastic access, exclusive access to GPT5, the you know, frontier level capabilities.

Um there, but there's a trade-off from the product perspective. put the chat box next to it, which is almost just one level away from just having two tabs open next to each other and having GPT5 in chat.

com over here versus do the hard work of have the product teams go and dig into how do we add AI functionality into every subtool if you're in Excel?

How do you create, you know, smarter functions within the cells as opposed to just a sidebar which has kind of been the status quo for a lot of AI at work projects broadly? Totally. And I think this is the I mean the answer is yes is yes and right in some sense like look every shift and I've been through three of these.

I worked at Akami like 20 years ago and the internet was like a a thing was going to be a thing and then you know I was obviously mobile and cloud at Google and now at Microsoft with AI the first wave of all of these things is just add AI right like put chat next to it put bring some intelligence to it and it's no different from you know the first wave of mobile apps were like desktop form factor like shrunk into the mobile screen form factor or the first websites actually were like scanned brochures which are PDFs right um in back in the day folks uh and I think now I feel like one of the that's why you see the across the industry saying oh let's let's add AI to it but I think the yes and comes from two things one is you want to make sure that when there's a natural way because of the natural language interface you're actually asking AI to do a lot more higher order things right instead of this fine grain manipulation of like oh let's go through 2200 menus and remember the magic incantation like you want to kind of say hey do it for me right so there's a bunch of things where chat is the right answer but then there are other things where if you're an Excel bro like you want to be in the in your IDE right and get get your uh you know get your get the assistance right where you are so meeting you where where you are so if you saw last um two weeks ago we talked about and we announced the equals co-pilot but that's like is the hello world of like what we are thinking of and say how do we actually right now in teams or in your meeting how do you have assistance right along with the chatbot yeah from a from a product perspective there's something very interesting going on where I mean Microsoft kind of coined the term or the definition of like being a platform where uh the companies that built on top of Windows created more value than Microsoft captured and it was like the classic example of of a technology platform driving like incredible value.

Um, there's something potentially at least that we've been tracking where we've talked to what three or four different companies that are building AI add-ons to Excel or or trying to rebuild it or trying to rebuild it, but a lot of times it's actually integrated into Excel.

And there's one frame where okay, those companies are going to get steamrolled by Excel uh and the team there. There's the other frame which is we said it, not you. Uh well I mean that that's a risk that I'm sure every investor in those companies is discussing.

Uh but then there's another frame where Excel becomes a platform like Windows became a platform and there is a robust ecosystem of apps and services on top some of those that are complimentary to copilot functionality some of those that are competitive.

Um, so I guess the question is like is Excel becoming a platform and do you see like how do you how do you message around the openness of the Excel platform?

Are you trying to take learnings from the openness of the Microsoft ecosystem and bring that into the developer communications on top of Excel or is there a different strategy at play?

Yeah and I think you know look super sharp point and I I think in the in the fullness of time office as a whole has always been a platform right if you look at like word excel powerpoint actually outlook plugins teams has a very healthy like bot ecosystem so the platform play is always been kind of at the center of it and it's the question though is there are certain it's keeping the user in mind are there things that are much more intuitive and native to be part of the product and are there things that are kind of like let's say you're doing something very specific for accounting.

It is absolutely the right thing for that to be a plug-in and addin versus doing something horizontal that should really be a feature of the product. So that's the way we thinking about it. Um have you thought any at all about how AI can supercharge the way people use LinkedIn?

I imagine that there's a lot of I I that might be a completely separate world from what you're doing, but I it does seem like something that uh that in the future um people have been people have been out there for so long. I need a personal CRM. I need some sort of thing on top of my LinkedIn connections to remind me.

And you get those like on every social app you get the reminder, hey, it's somebody's birthday. You should write them a note. But with AI, you could imagine so much more power there to say, I'm trying to sell this particular thing.

I'm hiring for this particular thing and having GPT5 go and crawl all the profiles, put together something for you. Have you thought about anything that's happening over in the LinkedIn world or is that just kind of out of your purview?

It's you look we we Ryan and I talk all the time and he's actually like much more front and center on the office world now. So I think the the thing that's interesting there is that a lot of these ideas were just like unfundable or bad ideas before because it didn't have enough tokens.

So my favorite thing to say is like, hey, take take all these like ideas people thought were like really bad. So for example, like I'd have like a whole bunch of people pitch back in Google the whole alerting, oh wouldn't it be great if you could kind of like ongoing do like event discovery.

I'm into this favorite like my favorite band is this obscure thing and like you know wouldn't it be great if I knew something about like oh I think we have I think we're having some difficulties actually sorry I I I think we're having some technical difficulties here um can you hear us okay I I can hear you but it's fine okay sorry I think you're cutting out um le let's move on to uh I didn't say anything controversial.

No, no, it's not that. I think I think it's the uh the the backbone internet infrastructure in this company in this country that's uh that's controversial.

Um, what uh what what about Yeah, I would like I guess like my my biggest thing that I' I'd be curious to get your insight on is is are there like how many businesses in the in the Microsoft ecosystem have just totally opted out of AI that are not that are not responding to upsells that are just not interested.

Maybe they have employees that are kind of using various AI tools themselves, but I would have to imagine it's like a very very large percentage of companies within that are off contrary. I think um it's in fact we we've had like more than 90% of Fortune 500 using some form of like you know co-pilot AI products.

We have like 100 million customers that have just we've just crossed 100 million across commercial and consumer on co-pilot usage.

The thing that's actually interesting that I see is maybe a variant of what you're asking which is you know like there's one extreme uh there's a set of folks who are just leaning much the posture is much more uh proactive I would say and kind of like in the game get in the arena building stuff it's not just about oh like there's an offtheshelf thing this is a chatbot that I'll buy and see who who uses it in my company so that set of firms in fact like we've started calling them frontier firms, right?

These guys are like out there starting to kind of like invest both build and buy and partner and we are starting to see oh those folks actually get a lot more out of AI. So for example like one example for me is like these guys are using the reasoning models.

So when you if you step back and think like you know where we are in AI the first version of AI it's just been more about like next token prediction and like chatbot right? Yeah. Um but the next really real big unlock came in January and February around that time when the reasoning models came online.

Yeah, that's when we saw a very clear shift. There's certain companies that just jumped in and said, "Oh, all these things that we couldn't do 6 months ago or one year ago, let's try them now. " Has that been including Microsoft by the way? Yeah. Has that been affecting gross margins?

We saw in the Wall Street Journal yesterday that notion obviously sells a few competitors to Microsoft products saw gross margins drop from 90% to something like 80%.

I was talking to Ivan he said it actually wasn't even that bad but there was some sort of an impact obviously you can upsell against that but how is the how's the economic trade-off how are you thinking about that with uh you know selling obviously people want frontier level intelligence but uh there there it might change the economic equation.

Are you seeing anything that indicates that uh going forward the the fundamental gravity of the economic equation might be different?

I mean look look I think it's going to be proportional in the sense that like I was joking with someone saying like you know using GBD5 thinking like in your product is like taking a flying car to a grocery store right or 03 pro or whatever like so in some sense you do have to be it's no longer about like just being a rapper on the model sure yes the model is eating the product right absolutely so that's I think one thing both these statements seemingly contradictory are true and we're seeing that on one hand the model is eating the product.

What do I mean by that?

Like you know internally even us we spent a bunch of time trying to solve what I call the sixfinger problem right like remember two years ago when Deli 3 came out and like to generate humans it would be like six fingers and five like 18 toes or what have you and there's a whole bunch of startups that said oh we'll solve that right and whole bunch of product teams inside Microsoft too saying oh we'll just like figure out how to solve the six-finger problem metaphorically speaking 6 months later the model GBD 40 came out and like, you know, there's there's no longer a six-finger problem.

Yeah. I mean, hallucinations, yeah, hallucinations seem to be at an all-time low. Like, the models are getting way more. Uh, there's been a lot of reports of of consumers developing like friendlike relationships with various LLMs.

Do you have is there any evidence that that uh employees at companies are are developing real relationships with with AI in the way that some people on on Reddit have developed relationships with models like 40? Yeah.

I mean I I think we've made uh the product choices that kind of squarely put AI uh at Microsoft as tools, right? Yeah. And even even these tools, they can be agentic tools, meaning they can be working for you in background. You have a natural language interface to it.

Uh but there is this is this is not an accidental thing. We have to make intentional choices. In fact, I I I coined a term called neural software, meaning that it's not software as usual. In fact, Steve Synowski and I had a spat on Twitter about it.

uh saying uh like because for me the way I think about this is look uh yes it is software as engineers we all know that it's like stoastic models and it's like you know you you peel a few layers it's all software but it is probabilistic and there are three things that are really important going on here and we like as builders I I feel like we got to pay attention to it number one is the natural language interface right it just has this humanlike anthropomorphic interface so when poor gets replaced by five like you're like you're not like oh who moved my cheese or this menu item moved you're like who fired my you know like chef right so I think that is a point whether at work or in um in personal life but I think the second thing that we are underestimating and we certainly see that inside is the way we building products and releasing them is changing like previously we'd say oh yeah like 2.

1 to 2. 2 too, right? Like the software is updated, release notes and change log. But now it's just like you're you have to kind of slow roll out. You have to kind of do eval benchmarks on how the interaction works, not just the accuracy of things, right?

And then the third thing I'm seeing at least like and this one's a hard one to kind of like crack. Um we're still trying to figure out what that means is the composition of the team, right?

Like in the past it' be like oh you had engineers your you know your devs your test your product your UX it's all a blur right it's all a mush and like you have a bunch of generalists and you have the model whispers right and then you have go to market like that's the new triangle that that we're seeing well it's a fun puzzle to solve thank you for hopping on the stream this was great we will talk to you soon have a great rest of your day and just uh let let everybody know internally that we're we're We're we're riding riding with the XL Bros and Clippy.

Yes, we're riding with Clippy for sure that we would love to help relaunch Clippy given the opportunity. Thank you so much. Great to meet you. We'll talk to you soon. Have a great rest of your day. In the meantime, let me tell you about adquick. com. Out of home advertising made easy and measurable.

Say goodbye to the headaches of out of home advertising. Only adqu combines technology out ofome expertise and data to enable efficient seamless ad buying across the globe. Uh Nick, did you see this posed by Nick Dobos? uh OpenAI cackling right now today. Uh so OpenAI acquired Stats Sig. We talked about this yesterday.

Um but he's uh he's sharing a photo of anthropic and telemetry services for data usage. Cloud code connects from users machines to the stat service to log operational metrics as latency reliability and usage patterns.

Uh, and so is what he's is what he's saying like they are they're going to rug Anthropic by buying their their telemetry like their logging service like is this monopolistic like market? Can Anthropic not just use another company or product?

I I don't this doesn't quite like sit with me as like oh yeah it's like they totally owned Anthropic. It's like, yeah, they bought a company that Anthropic uses. Like the the Anthropic will probably keep using maybe they'll replatform. I I don't know. It doesn't seem like that big of a deal. Yeah.

I mean, I I do wonder uh curious if if Stat Sig is going to remain an independent brand or just be rolled under OpenAI. Yeah. Yeah. There's a bunch of companies that use Statsig. I'm just looking at the website. Yeah. Uh EA, Microsoft, Atlassian, and Bloomberg effective altruists use it. You said EA's use it.

[Music] Just kidding. Uh Tyler, do you think this is uh do you think that that uh what are they using tracking shrimp welfare? I frankly I don't know what stat sig really does. So I I have no like it seems like yeah anthropic can just Why don't you know what it does? It's no more complicated than palanteer.

It's equival It's equivalent to oh uh we got R for Rock dropping massive alpha on ML uh ML's closing $14 billion uh blew past the initial target of $9 billion. I assume that's market cap. Uh the rumors about Apple's acquisition were fake. Uh