Microsoft's Jacob Andreou unveils Copilot agentic browser, first-party AI models, and Clippy's return as 'Miko'

Sorry. Wow. Finally, somebody that Oh, we got a short guy in the in the studio. How tall are you? 67. 67. I'm 6'8. 67. Are you getting the 67 thing? Do you have kids or anything? You know, this whole thing, the 67 thing. Come on. Yeah. Got to be in youth culture. I I don't know at all. I I found out about it.

I I literally found out about it when the Wall Street Journal published an article. Oh [ __ ] And it was like here's the here's the latest trend that's causing problems for teachers and I read it. What What are you guys even talking about? You don't know this? Okay.

So dude, so there's this meme where in middle schools if the number six comes next to seven, six, seven, yeah, it started from this meme. All the kids will go crazy and go six, seven, and then they do this because and and if you trace it back, it goes like seven layers deep into like this.

There was a kid at a basketball game who was saying 67, which was a reference to a basketball player's height and then one awful rap song that nobody had listened to before Doot Doot. Yep. Dude, dude. And that's a reference to 67th Street in Chicago, uh, which is like a popular area to sing about, basically.

I'm impressed. That's all from the article. Yeah, loosely. Yeah, Wall Street Journal knows their stuff. Knows their stuff. Anyway, the Wall Street Journal didn't have uh your backstory, so you have to give it to us directly. Yeah. Introduce yourself. Absolutely. Uh, yeah, Jacob Andreo.

Uh, I lead product uh and growth over at MAI. Uh, mostly focus on co-pilot. Uh before that I was at Snap for like nine years. Uh 2015 2023 doing the same thing basically. Uh and then in the middle venture for like a year and a half. Uh still doing a lot of Greylock. Was that Greylock? That was here in LA.

Uh no SF in the city. But I've been based in LA from 2014ish. Uh and now just spent a lot of time in Redmond and Mountain View. Uh my my two favorite places on planet Earth. It's great. Exactly. Uh co-pilot is not nearly sharp enough. Like you're gonna need to narrow that down.

That's like a million putting co-pilot and everything. Uh it's a multi- trillion dollar company. Like uh what are you focused on vertically, horizontally? How is the organization structured? Like what are you actually doing on a day-to-day basis? Yeah, I mean it's a super super cool moment for Microsoft right now.

We have uh a bunch of folks that have come together, a bunch of people that have been there. I run in people all the time. Microsoft been there for 35 years. It's [ __ ] crazy. I've never seen anyone do this. Am I allowed to swear on this show? We don't, but you can. We don't have any sense.

But feel free to if you feel like that's if you're fired up. If that's what you want the Microsoft Comm's team to really like take away from this is that you went wild on a show and they maybe shouldn't have green lit it, then yeah, we'll we'll push push the envelope.

Um, no, look, we got together an amazing group of people. Uh, a ton of people that have already been there forever. Uh and we're just trying to transform a bunch of the different things that say co-pilot across the company.

So we got a group of folks, they're super talented, working a lot on GitHub co-pilot on the VS Code stuff, uh you know, whether that's competing with cursor, all the other folks kind of in that agentic IDE kind of world. Uh and then me and my team, we spent a lot of time on like the consumer copilot app.

So the thing that's in your pocket, the thing you uh can pull out of your pocket, use voice mode, uh talk back and forth with web retrieval. And then we introduced a bunch of really cool stuff uh just in the last week. And is is is consumer copilot is that powered by open AAI models? Are you model agnostic?

Are you training your own models? Like how are you how are you describing because with some of the foundation labs they have to put a ton of their marketing behind like their model and we were having this discussion about like at some point consumers are going to stop caring about the model.

So how are you thinking about it? Yeah 100%. Yeah. So uh Microsoft AI when we're chipping co-pilot we're just building the best possible product. Uh that's it. And we're super product focused, super customer obsessed.

And at the end of the day, like whatever's going to give people the best thing in their pocket, like that's the job. And so for a long time, that was open models. And thanks to, you know, the amazing partnership that we have with them, we can put all their models directly into the product.

Uh 4 powered the mainline response for us for a really, really long time until GPD5 came out more recently. GPT5 now is honestly, it's an amazing set of models. Uh and it's been really, really popular inside the product. But there are parts of the product where we're already serving other models as well.

So we introduced a couple new modes this week. One of them is not powered by an OpenAI model. A few of our own firstparty models just started to drop. So I think six weeks ago was the first time we privately started running LM arenas on MAI fully endto-end trained models. Okay, cool.

Before that, yeah, before that we had no ability really to build our own foundation models from scratch. And in the last 6 weeks across both language and text uh and image, we've been in either the top five or the top 10 with our own first party models. And that's like a first for us. And so it's been super cool.

So does that reflect of you that essentially like the foundation model layer is commoditizing? It's high capex, but it is something that's maybe viable to get near the frontier.

Yeah, I think that a piece of it is a little bit that perceived intelligence, you know, when you try these different systems today, even when they're backed by different models, it is a little bit tough to tell, oh, this is way smarter than that. Like things are kind of getting a little bit closer together.

We saw that with uh GPT 4. 5. Yeah, totally. And then we went to five. A lot of people, even though it was a smarter model, were complaining about like the tone and the personality. And so we're spiky intelligence era. We're no longer in a race where just like smarter is just better every single time.

And so to really start to build an amazing product, it's not good enough to just have a set of models that are marching up intelligence, you have to actually build what people want in the product.

Uh and so an ability to go kind of end to end and be able to actually look at the way people are using the product, have that feedback into like model training, especially post training, and then have that then it allow you to iterate the product.

it creates this like model product user loop that is kind of the main way that you'll probably push the envelope in terms of meeting the user needs. So that's what we've really been focused on. And when when you think about the user the user of the app, uh are you thinking more business user, proumer, consumer?

Yeah, all of the above. Like there's there's everyone from like like the Xbox Call of Duty kid is a Microsoft customer to like the Fortune 100 company CEO is there's a very different archetypes. Yeah, we long term we have to serve that whole spectrum to start. Uh it's the consumer. Yeah.

Uh and it's the person that is looking for this thing. You know, they're asking us for advice on health. They're trying to learn new things with co-pilot. They're asking for a lot of help making big decisions. Those are the places where we're really playing the most.

And now you can't ship a product from Microsoft without being awesome at productivity. Uh and so in the last few months, huge push in productivity, making way easier to upload like tons of files, reason over them, create documents based on the source material you've uploaded.

And this is just like ripped like the the as soon as we put this in the hands of users completely. You know, we we kind of expected, okay, if we improve answer quality, generally this is always going to be good. Improving file upload and improving the number of of files that people could upload. Yeah.

Users just were drawn to it so quickly.

So, so I mean it's it feels like the real wheelhouse like you're just going to knock it out of the park with the customers if they're running their life on 365 like they have a they have a Microsoft email they have you know one drive going they like they set up their life in that way maybe because they use it at work maybe it's also hybrid work home but that's going to be a really great experience for them yeah so the product today doesn't actually need to be rooted in M365 at all you can just download it you can just use it and it's actually amazing for that use case longterm the system that is the highest context on you is going to be the best in both sides.

It's going to be the one that knows when your meetings are so it can schedule that dinner that you've wanted to. And so we actually think that the long view here is that these things ultimately have to come together.

But for the next 6 to9 months, we're going to ser focus on nailing the enterprise use case in M365 inside of all the work apps you already do all your work in.

And then at the same time on the other end of the spectrum for consumers, nail the co-pilot app that you just download from the app store and you just love and you use. How how should startups think about your guys's ambitions?

because I think that uh Sam had a quote it was probably a few years ago at this point maybe a couple years ago which was like don't build a company that's that's assuming models not getting better and I've seen so many companies this year we've had some of them on the show we've talked about them that are basically they may as well have said like we're building a different Microsoft Copilot for XYZ or we're building we're building Microsoft Excel and Microsoft Copilot for Excel as a new company.

And it's cool like these are big categories, right? They're extremely valuable products. It's it's worth new teams like trying to take a crack at some of this like timeless software. But I'm curious, you're I'm sure you see some of them and you're like, wait, you know that like we do that already.

Yeah, that's on our road map and we we have like you know hundreds of millions of people as opposed to like Bloomberg has an Excel plugin and they've had one for a long time. that hasn't been on Excel's road map because it's a separate thing. But yeah, how are you talking to start?

Yeah, look, I think if you get really specialized like the Bloomberg example, like we're never going to change all of Excel to go compete with like a specific vertical of a way people use Excel. We can't change like the core platform that much.

But if you're trying to build something, and I won't use the Excel example because it's too close to home, I think for people that are working on this problem, but if you built like a plugin for Word that does something that's like pretty horizontal, like it uses Word for you. Yep.

we know like this is certainly something we spend a lot of time has had templates for years and and if you think about you know prompt to template like that's a very obvious next step.

Uh do you feel like there's a there's a responsibility for the Microsoft brand to send clear messaging to startups so that you don't steamroll them? I always wonder about that. Like it feels like Sam was kind of saying like, you know, I'm former YC, you know, head. I want startups to like me, build on my platform, etc.

And I don't want someone to accidentally get crushed by the steamroller. Yeah. Yeah. I think that when you're training foundation models at the scale that they are and have kind of been doing it for as long as they have, I think that the world is still adjusting to how quickly things are changing.

And so I think the degree that the like model builders can signpost and be like hey like this is going to change a lot really quickly. Uh I do think that this is probably just the right thing to do for the ecosystem. Now that being said for us we're just focused on building an amazing product. Yeah.

And the other thing is some startups are overtly saying I'm going to kill if they're coming for you. It's not like you like gloves coming out. Maybe don't do that because we're doing that. It's like no they're trying to eat your lunch.

It's totally fair to just keep expanding your product in ways that you know are valuable for the majority of users. But the aentic stuff happening in Excel as a specific example is really cool. And so it's one of those areas where it's like totally look focus you can win a lot of things by being focused.

So who knows maybe the people doing that are still going to have an amazing product and build a great business but Excel is going to get pretty agent. Do you have anything to show us today? 100%. Let's take a lot of cool stuff. Yeah. Yeah. Show us some stuff.

Well, so just quickly the stuff your camera here so we can see your phone screen if you want to pull it off. Yeah. So the stuff that we talked about uh gosh just a little bit this morning. Um we have a huge release of AI browser. Uh and so I think we heard a lot about AI browsers this week.

Uh and the biggest difference is Edge has you know hundreds of millions of people using it and now every single one of them with a flip of a switch now has an agentic uh browser in the form of co-pilot mode. Sure. I've using it to reorder stuff on Amazon uh kickoff stuff for me. It's been awesome.

Uh we also uh get get into uh yeah get get into maybe a little bit more there just just while we're talking about it. But this is like you can chat with the pages that you're on and more content.

So sidebar is totally there uh and you can reason over multi-tab context and you can talk about the different stuff that you're doing. The the thing beyond like actions like we've seen a bunch of people talk about whether it was like comet or the Atlas stuff from earlier this week.

um you know actions are just going to get better and better as the models improve and you know cop mode and edge is you know state-of-the-art with respect to the actions that you can do.

Uh the one of the really cool differences is once you enable it it'll actually start to pick up like patterns and habits of your browsing and it'll actually create these journeys uh and do things proactively for you and you can jump back in where you left off.

And so if you're like planning a trip and doing flights in a few tabs and doing hotels and other tabs and trying to figure out stuff to go do, Edge will actually see what's going on with that and the next time you come into the browser, it'll actually have taken the next step in planning that trip for you.

And with a single tap, you just jump straight back into Flow where it's actually done the next thing. And so for me, when I'm like interviewing candidates and just doing a bunch of research and tabs is perfect for this. How how do you uh there was some conversation concern about prompt injection.

Is that like how do you think about like solving that problem? It felt like it was people were saying like, "Hey, this is a big risk like slow down. " But it felt like something that was somewhat solvable like pretty quickly. I I think that's right.

And I think that you have to look at this whole class of products right now as it sounds a little maybe too over the top, but kind of like a security preview. Like we we know there's a whole bunch of risks with allowing the browser and these models to take actions like this. You really have to supervise it.

We have a ton of checks and balances. It'll ask you a ton of steps along the way to make sure everything's still on the rails. Uh, but we're learning alongside everybody else and this area is going to evolve really, really fast. So, we're ready to just keep playing whack-a-ole as these issues come up. Yeah.

Are there any surprises? Do you have anything that that might be teed up for us? I we love hitting the gong around here. So, so we did we are now the first ever AI app to also drop groups. Okay. Are you on the most aggressive version of liquid glass by the way? I am like liquided out. I am.

It's it's all glass all the way. Okay. It looks good. H2O Max and with Pro Max you actually get the battery life so it can power liquid glass. Okay, cool. There we go. Yeah. Yeah, exactly. Um, so groups is phenomenal.

It's now the first place where if you're having a chat with co-pilot, you just add someone to the group. Use it for my wife when she's asked me questions. I have no idea the answer is all the time. I can just put her in the chat and she can just take the rest from me.

Uh, but creating a group and other people you're not bringing it to teams. It's like yeah sharing this and then and then we're both on. So when I'm in when I'm in the thread I just hit share invite with a link. I can drop it in iMessage or whatever and then she's just in the chat and they don't even install the app.

They can just use it right from mobile Safari which is super cool. And so this is the first time across all these apps that you can finally do this which how it seems like it's such a great and somewhat obvious feature. I'm somewhat I'm somewhat surprised that it hasn't. I was shocked when I when I joined Mai.

I was like no one has built anything social in any of these products. I worked on my eye at Snapchat before I left and so like no one had really done social since then. I was like I'll do it again. I'm like, "That's fine. We'll run it back. " Okay. Okay.

And then and then one of the really cool things that we did, I was Yes. Cop knows all about you guys, by the way. Oh, really? Well, we do put out 15 hours of content every day. Okay. We introduced Mo, uh, who's this new person that you can talk to, this new character, uh, really cute face.

It's not going to sex with you. So, it's completely different than the other AI appearance. It's not It's explicitly nonexplicit. That's not what it's going to be. This is a counter positioning. We love counter position. Bring the dog for that's not what's going to happen.

Please bring the dog is is not traditional market and traditional values. And so with traditional values at our core, uh as soon as we showed uh Mo to Satia, his favorite thing actually was like poking it, which I couldn't find a better way to describe that on air without it sounding a little bit weird.

Again, it's it's it's going to be strictly platonic. It's safe for work. It's totally safe for work because this is from the company that brought you Microsoft Excel and and we and of course of course and I would trust that brand. And so as as you poke Mo. Yes. And Okay.

How many times do you have to You really got to hammer it. Hammer. Clippy. Clippy is back. Let's go. We We have been We have been asking for this. Is this breaking news? Is this a scoop for us? This is This is the first time this has been on air. The train now. Clippy is back. This is reborn. Yes. Incredible.

So, Clippy walked so Mo could run. And so, we got to pay homage to our roots. And so, Sache when he saw this demanded Yes. that Clippy front. I think I I think people people have people forget we've been we've been begging for this. We've been begging for this.

Obviously, Clippy had like a bad bad rep and people were like, "It didn't do what it was promised. " But I feel like all of the nuance of like where it wasn't perfect has completely melted away and it's just lore. It's just great lore. Your parents hated it so that kids loved it and it was cool to love it.

So, everybody else got on board and so it's so ready to bring it back. So, well, I just Yeah. So many good so many good memories of being in the computer lab at school and just looking down and knowing that Clippy was there. He was riding with me.

How how are we what what does it take to update the name of the app in the iOS app store to just clippy? You got right now it's an Easter egg, but we're working here together. Actually, there's like a team of 25 people that really worked hard on Mo and they're not CMOS.

This is not, you know, bring that clippy but uh that is very exciting. Thank you so much for coming by. No, thanks for having me guys and uh breaking it down. Uh what a what a fascinating time in the industry. What a fascinating place to be. It must be thrilling to be in your seat. It's an absolute treat.

There's five places in the world you can try to build frontier Frontier AI for a billion people. And this team and this setup, it's unbelievable. So it's a really no as a as a as a as a true product guy like going somewhere where you have that scale, resources, etc.

Um, and it's actually, you know, it's good that we have just this flood of flood of talent into into early stage companies and the labs and all that stuff, but uh, it's actually incredibly important that talented people go and work at, you know, these companies that already power, you know, use cases for billions of people, millions and millions of companies.

So, I'm I'm glad you're going to do it 100%. Yeah. Anyway, so you guys fantastic.