LaunchPod - Raj Singh
===

[00:00:00] 

Jeff: What's up Ra? How you doing, man? Good to see you again.

Raj: Good to see you too, 

Jeff: we got to meet the other day in San Francisco at dinner. I'm back in Boston though now, but really still we got to do that. Thanks for coming on the show. We're gonna talk a little bit about what Egen browsers Mozilla and where it fits into the whole browser ecosystem building with ai.

All sorts of fun entrepreneurial stuff you've done on your end. ~But you know, just for the audience, you've what founded like four companies, exited them all fairly successfully. You did a stint at Salesforce as a product leader. ~You're at Mozilla now as a VP of product. Can you just give us like , the quick TLDR on, how did you get here and, you know, what's the background , on Raj Singh that we need to know?

Raj: Sure. Yeah. I would say my career. You know, has been all startups.

And for the most part has been all my own startups beginning from masters not finishing it and doing file sharing company to my most recent company, which was in the space of meeting summarization that was acquired by Mozilla. I would say thematically it's been all consumer.

I tend to focus on my own pain points. I like to describe my career as a map. I kind of jump around to my own passion projects. That certainly is not probably the most economical or financially rewarding path. You can certainly take more of a ladder career and probably do have a much better outcome possibly.

But it's been great and I've no reservations , [00:01:00] and I've been able to just build a lot of great products.

Jeff: Yeah. But would you have had more fun? You may have maybe. Maybe you could do better financially, but would you have had so much fun?

Raj: Hundred percent. I mean, I think, I think one of the things that's unique is like, you know, sometimes I tell myself like, am unemployable right?

Jeff: No, I, I know what you mean. Like, you start to get the skillset where it's a very narrow set of things you do very well, to found a company almost, you have to be wide and, and able to do a lot of things, but at the same time, there's a lot of jobs that, that might be difficult to, to do full-time.

So,

Raj: Absolutely. Yeah, absolutely. No, and it's different. And, you know, and just to be clear, you know, I'm, I'm leading obviously a new products team within Mozilla. And you know, I say that half jokingly, but there is, there is some very interesting contrast between sort of new product development and operator kind of work.

Jeff: Yeah. I mean, clearly you're employable 'cause you're employed. So, cool. You, you said you've, you know, you pretty much have constantly just been building your own companies. Yet you are, you are at, you know, Mozilla now, which, while, you know, very technology focused is, is one of the older kind of stalwarts of modern tech, right?

I remember when Firefox came out, it [00:02:00] was mind blowing , how cool it was to have a new browser option and all the, you know, cool nerd kids I knew were all jumping to it. The uncool nerd kids like me who are jumping into it too. but 

Like you said, you didn't start out there. , You started a company and you got acquired. But Mozilla, so maybe jump into that because I think this is a message everyone needs to hear of, like, the whole idea of pivoting , and you're never gonna have the right idea or rallying, have the idea right away.

So you started out it was lentil and loop. And then the last one was Pulse all. Same company. You ended on, on meeting summarization. But let's jump into that. Maybe. Like how'd that start and what was the genesis of that?

Raj: I tell people I've not been lucky enough to work on something that just worked on day one, and I do tell people. Those things do come around, when there's something just working on day one, it's a rocket ship. It doesn't matter what you're doing in that company, get on board. You could be the worst employee in that product, but you're gonna see hypergrowth and and it's just an experience in itself.

my last startup came together was really inspired from some of the stuff I was doing in my startup before that. So, which interestingly was also based on my previous experience. if we go back to like 2000. 9, 8, 9 timeframe. I was doing a lot of outbound partnership type [00:03:00] stuff with the stuff I was building, and as part of that, I was meeting operators and handset manufacturers in different countries.

And there'd also often be like many people in the room. There'd be a lot of different locations and a lot of different places you wouldn't know who these people were, right? Like you'd be in a meeting and like, where are all these people?

Where am I going next? Et cetera, et cetera, right? And so, I was like, can I build something to help me better prepare for meetings, and I built a calendar company that Salesforce had acquired that was all focused around helping you prepare for that meeting.

So, fast forward, if we're gonna talk about the last startup. One of the interesting things when I was looking at calendars was we had this top slide, we'd called the lifecycle of a meeting, ? There was like the two weeks before the meeting, just before the meeting, the morning of few minutes before the meeting.

The meeting itself. Just after the meeting, maybe a week after the meeting. So, we had built something called lentil, which would let you summarize your meeting.

It was basically would listen into the meeting. It would transcribe and it would summarize it. And this was really hard in 2018, 19. pre l, LM era for the most part, right? So we had to basically gather our own data set of meeting transcripts. We had to annotate them using a data [00:04:00] annotation team.

We were training on a language model, not a large language model that was called Bert. We would fine tune against it and then we would basically there's a whole process like identify, is this an action item? Is this a question? different sorts of annotations to try to figure out what was relevant, you know, out of, out of a particular meeting.

Transcript, transcription at that time was also not very good. It was crazy. Like you had to add punctuation. You couldn't figure out names. Right? There was removing pauses and ums and all these sorts of things, and I can't even recall all the different. Things that we had to do in terms of tactics and strategies and sort of cutting edge research to figure these things out.

There's something called diarization, which is like, we take it for granted now, but can you actually break up a transcript by the speaker?

Now it's much easier to do. ? They're multimodal with it, right? They're looking at the voice as well beyond just sort of the transcript.

So we built this and the end result was the summaries weren't very good, you know,

And this is the problem. There was two sides of this issue. One was the summaries were not that good and two.

There was a A two by two that we built, which is small meetings, like one-on-one meetings versus multi-party meetings, like many to many. And then [00:05:00] internal meetings and external meetings. And if you think about that, those sort of four quadrants. The highest value meeting summary quadrant turned out to be external meetings, one-on-one.

And why? Because one-on-one meetings tend to be very high value because there's a lot of information being exchanged, versus a meeting with 10 people in there. You know, people describe them as just meeting hell or meeting waste, right? And then external. And why external? Because one-on-one, obviously it's hard to take notes, but external, 

There's a longer lifetime value of that meeting. I'm collecting a bunch of data and I might not look at that data until three months later. Internal meetings, it turns out one-on-one. Sure. But internal, many to many. There was often a dedicated notetaker, and the meetings were often recurring meetings

So the meetings were transient, meaning what was useful this week, by the time we were like two weeks from now, it wasn't particularly interesting anymore. Like we've moved on to the next sort of topic. And so we went after the wrong quadrant, ? This was what I would describe as a segmentation mismatch.

We went after the worst of the four quadrants, which was many to many internal, and I think the mismatch was we were trying to build a tool to replace meeting notes. And really the way we should have [00:06:00] positioned it is build a tool to help the person taking notes.

In the many to many internal meeting, So that was kind of a mismatch.

Jeff: If you're enjoying Launch Pod, the best way you can show support is simple. Follow the show so you never miss an episode. Leave us a quick review to help more product leaders find it, and share this episode with a friend or colleague who'd get value from it. Every follow, review and share really helps us grow.

Raj: So fast forward, we built this thing and we're testing it, and I'm not seeing the kind of results that I would like to see. So the nature of all products, and this is what I always tell people, and I've been through so many startups, and as I said, I've never had a startup that worked from day one, is you bet that the team's product intuition can figure it out, right?

That's ultimately the bet. Can you pivot? And pivoting is not easy. You know, and I think there's like two terms here. I don't know if you've heard the term pivot, and then there's also the term hop, which is sort of like pivot is like change your go to market or change your segmentation. But hop is like, do something completely different, right?

These were not hops, these were pivots, but it's not as easy as it sounds. ? As the team gets bigger, it's like steering a bigger ship. So it's hard to, so anyway, so. COVID comes as we're testing this thing and we're like thinking like, okay, how can we sort of pivot? And suddenly video communication, collaboration is like [00:07:00] the hottest topic, right?

Because everybody's working from home. And frankly, at that time, zoom was great, but not that great. And Hangouts was still terrible, ? Which later or later became meets. You know, and Blue Jeans was kind of internal, like conference room only kind of experience. It really wasn't meant for hyper remote 

And so all of a sudden we're looking at this and we're like, sometimes the macro conditions enable a certain opportunity, right? So we're like, Hey. Asynchronous meetings and asynchronous notes is gonna be more important than it's ever been now.

We have fully distributed teams. People are hiring everywhere. All of a sudden people are saying remote was the future. It's funny how fast that pendulum has swung. Not saying it's not the future, but definitely everybody's r ting back to the office. anyway, so we're like, what if we build video communication as well?

So we have video communication and we also enable the note taking.

Jeff: you built your own video meeting software.

Raj: Yeah. And in that stack, you know, it's interesting, and this is a very common product startup mistake, and this is speaking as somebody who's built multiple startups, and I've had multiple, you know, eight digit exits. Like going through this kind of mistake, right?

Like you, you make the same mistakes over and over again. And the issue is what I [00:08:00] call people think of technical debt, but I call it like product debt or sunk cost, We had built all this note taking stuff and now we're like pivoting to like, let's do video communication with presence, which is what the virtual office world wanted or the remote workers wanted, like to kind of be able to basically what later became Slack huddles.

but we wanted to retain all this work we did around meeting summarization, notes. It could be useful like bolted on, but it wasn't core to the new job to be done. The new job to be done was less about note taking. It was more about make it easy to connect and make it easy to see who's online and around, but we kind of kept it there and it impacts the overall experience. But now there's a maintenance cost and there's engineers thinking about it and there's the designers thinking about it, it's interesting in that path to find PMF, sometimes you have to be willing to cut off an arm to kind of figure it out and then we can reattach it later, right?

And so like, you I think about that and how that added to the complexity of the overall experience when we could have sort of streamlined it. And so eventually, we built this video communication experience. With presence. So it kind of could tell you who's around at any given time, which is like, you know, kind of slack and statuses and teams and statuses were there, but it wasn't fully flushed out and [00:09:00] developed like it's today.

And so you can kind of sense like who's online and then I can quickly drop in and have a chat. And so it was really kind of the way people work when they're physically in the office. They, they kind of like, Hey dude, you got a minute. And they go talk, right? we built this experience and it was well received.

I mean, during COVID everybody was playing with all of these tools, but. We fell for another sort of trap, right? And this is something we couldn't have predicted, which is normally when I'm building a startup, I like to think about insertion points. So we're looking for an angle of attack, something that's being ignored by the incumbents.

The incumbents being sort of your top 10, hyperscalers your fortune. You know, the largest tech companies in the world, they, their priority lists are so long, they're ignoring tons of things. And you take that one little thing they're ignoring. It could be a feature of their product, it could be just a space or a segment, or an audience or a geography, who knows.

You make that your whole product and usually you continue to be ignored and they might even see that you exist. But now you get to 1% market share, now you get to two, 3% market share and all of a sudden they're like, Hey, we should be doing that. There's something there. And they go copy it. Right?

And at that point, when they start investing in it, counterintuitively actually boost your growth because they educate the market and now you're that [00:10:00] standalone player and you're growing. so what happened with COVID was video communication, collaboration. Single handedly became the top 10 priority of every company in tech,

right? 

Apple went all in on like, oh, we gotta make FaceTime better. And all multi-platform. Facebook even launched the portal device. they made video like a core piece. Like they were like 

adding video to everything. Right?

Uh, 

Jeff: got way better

Raj: zoom substantially improved. Google meet.

Honestly, like within a year was a radically different product. Like it was no longer dropping. I mean, it was just insane, the whole thing, right? So what's crazy and fascinating about all of this is it it, you know, it sort of became their top 10 priority. And as a startup, that's really early, right? So it depends what stage you're at.

If you're kind of already past the threshold it's kind of a different game. Like you're kind of like in scaling mode, but you're sort of in the still pre PMF, early seed sort of phase. That's a very difficult position, ? To be building into a category where you're no longer an insertion point, but you're a frontal assault.

And I always tell people frontal assaults rarely work. It can work from a marketing perspective. Like if I want to go and just go heads on and say, we're, [00:11:00] you know better than Google Analytics and we're gonna replace it, right? But fine. But from a product perspective, it has to have some sort of insertion point.

And so this became very challenging because what ultimately happened, and if you look at the category holistically. The incumbents just got bigger, right? Slack launched huddles. Zoom got better. Google meet got better, and the whole sway of startups in this category ultimately got acquired. Or didn't make it right?

And so we saw this coming and so now I had to execute pivot number two. sounds easy, but I always tell people when you fundraise, you're typically fundraising for 18 to 24 months. You only have enough time for one pivot. . To do two pivots is very hard, right.

Usually you just don't have the timeline, but you have to have enough conviction to try to pull it off. And it's also hard because when you hire a team or you pull a team together, they're typically hired to build a certain kind of idea and then when you pivot.

It's like, well, that's not really what I joined this for. The mission has changed, or like what we're building is different, right? And so keeping everyone motivated, keeping everyone together, it sounds easy. It is easy when you're two, three people, but it gets harder when you're like 8, 10, 15, 20, 25,

30 people, 

You're dealing with emotions. You know, as somebody had mentioned that the best way to sort of evaluate product leaders is to have them talk through [00:12:00] pivots and have them talk through shutting down a product because it is so hard. so anyway, so we're there and we said, okay.

We're not gonna win on video communication. What does our product look like if we drop video communication? And by this point, we had dropped our meeting summarization. Right? We 

wanted to streamline and make the product simple.

So 

yeah, now we're like, okay. You know, and, and so much of product evolution and figuring out product is frankly.

, The negative list, right? The reduction, right? , Like how can I reduce Right. People don't

think 

about recap. Exactly, right. It's all about removal. Right? Removal as a product, removal as a feature, . So we said, well, people are using our product because they get presence, but then they're saying the video on Zoom and the video on Hangouts is way better.

? And we can't build out all those features. We'll just be doing what I call table stakes features and it'd eat up the whole team and we simply can't compete. So he said, what would our product look like if it was just presence? It was just status. It would just tell you who's around, who's available right now, et cetera.

We said, well, we probably wouldn't go to market as a standalone product. We'd probably build it as an add-on. So we built this add-on for Slack, and at this point we renamed the company Now Pulse. So we, or renamed the product, we renamed the company. And we launched Pulse as a [00:13:00] add-on for Slack that we just auto update your status.

There's already examples of this, like Google Calendar now is a plugin. It'll tell you when you're in a meeting. That's like an example of what it would do, but it would do additional things. It would tell you if it was a holiday in your country, it could tell you what time zone it is.

It would share that as part of your status. It could tell you if you're in this Google Doc, so other people can see if you're in the same Google Doc, it could tell you if you're coding right now or you're in focus or whatever. Like we made it super intelligent. It could tell you the weather where you live.

It could tell you if it was your birthday, whatever you're into, ? , Like you could control it. The music you're listening to from Pandora, ? It almost became a cultural building sort of thing. 'cause you could learn about your teammates, actually see when they're around and not disturb 'em at the right or wrong time.

And this worked right? It started growing on a $3 a month product Art MMR just completely grew and it was growing for the most fascinating reason. Companies themselves, and I, and I hate to say it because this is kind of double edged, and we weren't playing on either side.

Companies themselves wanted a better sense of what their teams were doing.

Jeff: Yeah. I remember that discussion of like, are people actually working? That was the big worry. All, all the leaders had at that time was, people are remote and they say they're working, but, but how do we know?

Raj: [00:14:00] And a lot of that's anxiety driven, right? And I

don't completely that, but, you know, it didn't help that there was all this press about people being over employed and having three side jobs and , all these Reddit forums about how to not get caught anyway, so teams or managers or leaders or companies wanted a better sense of what their teams were doing on the other side. Employees wanted more control around their status, right? So for example, I'm in Slack.

If I step away for five minutes, it has me as missing. Like I'm not there. I'm like, no, I'm, I'm doing stuff, but I just wanted to go to the bathroom 

I work on the weekends. Let me, you know, take an hour to go cut my hair. I don't have any hair. Not a good example, but you get the idea, right?

Uh, what we did is we had features for both sides. We had enterprise policies, so the enterprise could set up things, but we had things for consumers. Consumers could turn on something like make me always look green. So they're always online, even when they weren't there, right? Like, so like it just went viral, ?

And it was growing and it was great. And then what happened was, mozilla turns out was looking for a company to do article summarization. And they came across us and we started chatting and they're like, Hey, can you bring this into the browser? 'cause we wanna summarize articles.

And it was kind of different to what we were doing [00:15:00] now and we were tired. Right? COVID was a very exhausting period. We were all parents. We all had young kids most of our team. And so that whole combination of thing going through two pivots and so it was an attractive offer and we're like, we gotta do this.

So we ended up joining Mozilla to bring article summarization into the browser. So we kind of wound down Pulse. We took the stuff that we built previously. Our team was like a machine learning team. And so we started building this out. And then the real irony here, and I always tell people everything's about timing within three months.

Chat, GPT or GPT three became publicly available, released whatnot, and everyone's like, dude, this is better. I'm like, yeah, we should use that. And fast forward, now that stuff is all being done using LLMs.

Jeff: right. But you're still at Mozilla and, and I mean, that's the fun thing, you know, I say fun with, you know, air quotes around it. Fun thing about acquisitions is they, they're requiring you for, you know, for something for tech or for people and usually for tech, but there's always the risk of or I guess the risk to them, not to you of.

Some huge function level change in, in technology that that none of us saw. It's funny 'cause chat, CPT two had been released, people knew this existed. Like this was not a secret thing. then, 

Raj: yeah, so [00:16:00] three different comments there. So one, technology as a disruption doesn't happen very often, but 

when that happens, it like radically 

changes 

everybody. I had a board member once that said something along those lines that I thought was really interesting. They said something like you know, I, you know, I haven't been lucky to in like, invest on day one into something that was taking off like that same sort of theme I talked about earlier.

But when market shift or technology shift, I've had teams that can rapidly react to that and recognize like they're now way ahead if they just change their go to market. Right. So that's sort of really sort of interesting with respect to acquisitions. Yes, there's I like to say there's, teams only.

That's very common. Now in those cases, they're not even company acquisitions. They're probably basically like offer letters out to the teams plus assets. There's team plus tech, right? And then there's sort of like, you know, a strategic acquisition, right? 

And, So , definitely all sorts of camps.

Jeff: but there's no way, I mean, it's funny 'cause because GP two existed. It was, it was 

fairly 

Raj: Oh. 

Jeff: no one just, it wasn't that, it wasn't that interesting. It just, but three came out was just like mind blowing and, and like you said, change the game there. 

Raj: So interesting about that too is they just threw more money at the problem. It wasn't a [00:17:00] fundamental technical architectural improvement. They just simply threw more compute. And actually in some ways, that is what's fueling the race today, because people don't know if there's gonna be suddenly another step function or breakout if we just throw more compute.

Jeff: Well, it's been this whole post. I mean, we're getting a little off topic, but this is fine. Like the whole AI post AI world, I feel like one of the most interesting things has been this kind of just surprise breakout. function increase that happens , almost seemingly randomly at times where like, you know, GT three came out and that was just overnight.

The world was different , but even within companies, I've seen companies, we, I mean we built something internally that was using GPT for like a chat layer in the product. , And I was against the idea We Do an analytics tool and session replay and all that kinda stuff, and we'd been using AI to , watch the sessions.

But, you know, people had in their mind that, need like a chat layer. And I was like, do not build another freaking chat to chart, analytics tool that's a what? A solution without a problem. And so. We'd been working on this chat, and the chat was just kind of wasn't great.

It was internal only. And I went on vacation with my family I took four days off and I was still on Slack almost every day. Came back and it worked in this incredibly different [00:18:00] way where it triggered the session watching agent to watch sessions for you and explained what was going on.

And it was incredibly powerful. I watched 24 hours of sessions in five seconds, the findings, it was spitting out when I asked it to like, look at our, like AdWords landing pages were. Often better than like senior level people I had employed to run that function for me. And that was like a four day difference.

, But like you said, it was just like, in their case, they threw money at, at GT two for a GT three, but it is just sometimes these like, unexpected jumps that do that and, you know, that happened. GPT three, they no longer needed summarization at Mozilla.

Raj: Yeah, I, I always say when it comes to dashboards. Like they're not actionable. Like we need action. Right. Make actions. And I, so I think the key thing you're really highlighting, and this was like a big issue in the quantified, quantified self universe, right? All these things telling your heart rate over time or telling you whatever some interesting stat or whatnot.

It's not telling me the action, like, what's the action I need to take? So

I think, 

Jeff: Do I need to exercise more or do I need to

Do right? But I think that's what you cracked here, and I think that's great.

So curious, like,~ like, you're at Mozilla now and we had this whole outline prepared for, you know, conversation today. And then we met in person. For most of the world who wasn't there, Raj spoke in San Francisco and I did not expect it to go as deep into age agentic browsers and a bunch of stuff that it did, but makes sense given you're at Mozilla and you have a great background in ai. but that's one of the areas I do wanna dive into.~

~'cause I feel like it's something people are interested in, like Right. They, you know, ~fairly recently, open AI released their browser. There, you know, there's strawberry. There's perplexity has one . These are coming out left and right. And [00:19:00] just kind of curious to get your take do you think this is the interface AI is going to?

Like, we've had these like chat windows for a long time and now it's API based is the browser going to be a, I dunno. I, I feel like a lot of people haven't found great uses for it yet. Aside from, it's just cool. But what, what's your take on like age agent browses as, as a totality and, and what's going on here

Raj: chat was always a modality, but open-ended chat in this way. I mean, this is incredible, right? Like for the first time it's good enough that people feel that they can. Express whatever creative ideas they have and get the result they want, and have communication, collaborate, whatnot.

Right? this is the new interface for the web. And if you think about a lot of AI applications in the past or anything that had suggestions or recommendations, they might have an open-ended interface, but they would often suggest things to ask, like they'd have a button click here, or they build wizard workflows for those common tasks.

But we're hitting a point now where chat is so good. I think early when Chat G Petit was released, there was some questions on like, would this become the standard UI to everything? But it's very plausible. It could become the standard UI to everything. And I'm already seeing that sort of behavior where you enter a travel site, do I really need to navigate all these things [00:20:00] on the left between the hundreds of options and airlines and times, or can I just chat with it and have it like show me recommendations as I continue to chat.

Right. And so I think all of that's very sort of interesting. Chat as a modality now A agentic browser. Yes. It's a little bit of a wild, wild west right now. Like people don't know what's gonna take off and so there's a lot of experimentation. There's a lot of AB testing happening, people are going down different tracks.

We're seeing a lot of money being spent. . . I mean, this whole industry if you just follow the money. Fascinating, right? Nobody's really. Contribution margin positive here, or EO positive, like everybody is like massively underwater from a gross margin perspective, I think the reason is, is because of all this experimentation and I think people trying to figure out what's gonna work.

I think the reason it's also important is the browser itself. is probably going to be the interface of the next generation of software. I mean, it's always been that way, right? The web increasingly we spend more and more time on the web and the computers themselves are basically just a browser the whole premise of the Chromebook.

But it's also a source of incredible intent data. It's where you consume information, it's where you spend your time, it's where you go to different destinations, whatnot, and all of [00:21:00] that data is building a increasingly personalized understanding of you. if you look at what is lockin and retention and why these different AI chat apps are showing smile curves, which are so rare from a retention curve perspective is in part because it's learning.

Not learning in the sense that it's like getting smarter, but it's learning because it has memory and it knows who you are. And so it can then tailor it's response. Better to you and in a more personalized way. And I think that's all very sort of incredible and powerful. So we're seeing a lot of activity you know, whether it's OpenAI with the Atlas browser, whether it's perplexity with Comet obviously we're doing stuff there as Mozilla Chrome, you know, edge whatever.

I like to say. Browsers has been a relatively boring space for a long time. I mean, outside of the Chrome, most people would not have been able to tell the difference between most browsers, but suddenly it's like one of the hottest categories. And I think that's for due reason because this is a new surface everyone's competing on.

And one of the fascinating things, I should also call out this whole notion of agent mode or operator mode. We've seen since web two where we basically opened up all APIs, the internet become increasingly closed. . Obviously it's accelerated now with [00:22:00] LMS indexing the whole internet, but APIs have been closed, right?

Because it's not in LinkedIn's best interest to open up APIs. It's not in Facebook's best interest to open up his APIs. People are building competitors. People are, you know, working against you. But now with agent mode. We are basically enabling a new API layer of the internet because the AI can basically browse the web as if it's human.

And so this is incredible because this is completely changing the interaction pattern. And this has actually even led to things like Amazon coming out and saying, we're sending a season desist perplexity because we don't want comment to run , agent mode on our, our.

Jeff: I'm still curious how that's gonna work. Like I get they could have a judgment against it, but, maybe there's like behavior patterns or something that are, are telltale signs, but AI is just gonna keep getting better on, on web action and stuff like that. How do you eventually even police that?

Raj: I don't know how they're gonna, I don't think they have a way to really block it outside of sort of this legal perspective. 'cause it is the end user's browser.

Right.

You know, know, it's like the consumer is choosing to do this. I think the issue is, is how, how fundamentally it's disrupting the business model of the internet, which is advertising.

Right. And advertising's [00:23:00] completely based on people seeing things. But if an agent is now seeing things, which is an AI and it's navigating Amazon to do a product comparison across three products or looking at a range of things and like. Gathering a bunch of reviews for me, that's a dramatic shift in how I'm interacting with Amazon.

And that means there's a whole bunch of ads that are not being seen, and there's a whole bunch of user analytics that are measuring the wrong way, right? And so you can imagine the next generation of telemetry even segmenting between AI traffic and user traffic.

Jeff: or the other piece is, is data scraping, right? I work in marketing, so I, there's a lot of these pieces that, that I keep up with, even if I'm not, you know, doing some of the, the stuff you're not supposed to.

You see it and you read about it and, you know, there was a bunch of companies on LinkedIn that basically made their entire operating principles scraping data from LinkedIn, right. And then, and reselling it. And then they basically got caught and LinkedIn cracked down on that 

and put the kibosh on them.

And you can see this over time, where like you couldn't use, Chrome extensions to, to, you know, pull data and export it out of the window from Chrome and all that stuff wouldn't work on LinkedIn. But then people started to find ways around it and do that and dah, dah, dah.

But like, how do you stop it when it's just the agent just running in the, in the browser and it looks like a [00:24:00] human doing it.

Raj: Yeah, historically, the way a lot of these companies were doing it was through headless browsers in the cloud, and so it's much 

easier to stop. Right. It was clearly a violation of the term, right

It's like through a, yeah, it's exactly, but now it's running locally on the desktop, I think this is kind of fighting a trend.

This is like blockbuster fighting Netflix or 

whatnot, or Kodak fighting. Digital, right? Like, it's not gonna happen. So I think this is the way things are going. I think the real issue is a lot of these companies particularly web publishers, have to rethink like what this new model looks like.

is it is it an ad supported business model? Because from a publisher perspective, how do I react? Like, do I just become a provider of data and the aggregation layer, like, you know, AI chats the aggregation layer and you view and consume through that. Or am I still a destination?

Right? And what's that business model gonna look like? So it's all very scary. And then you have things like cps, which are basically unstructured APIs as a way to think about it, right? I like to call it that because I think it's just easier to understand. And so now I can basically ask a site for something and it will figure out what APIs to call to give me that, right?

And so I think this whole universe is all very fascinating. It's all very transformational. I don't think anybody can predict it, but I think this is why [00:25:00] everybody's playing on every surface. And I also think this is why so much money is being spent. Unclear on , how to cover the cost of all this operational spend, the hundreds of million that are being spent in inferences for the browser.

Jeff: Right. And then it is kinda an interesting thing because , it brings a world where, you know, for a long time, like you said, browser was kind of an uninteresting area of, of technological innovation, and now it brings browser back to the front. But at heart there's, there's only so many. Rendering engines that exist.

There's only like so many flavors of, of a lot of these things. I don't think people realize , how probably much more like influence Mozilla has had in some of these things than the people realize because there's only a few players who are really operating some of these levels.

It gives you guys, you know, really good insight probably into what's on. Or, or I guess does it, 

Raj: the ecosystem has been consolidating for good or for bad, right? So, if you look at like Edge you look at brave, you look at opera these are all based on Chrome chromium, right?

So Chromium is a one of the major rendering engines. Then you have WebKit, which is from Apple which is Safari, right? And then you have Gecko, which is Firefox. And those, that's it. I mean, those, those are really the three main rendering engines of the internet. And because two of them are [00:26:00] basically.

You know, hyperscalers, super large companies we are kind of the check and balance, right? We're the ones who are thinking about the user thinking, privacy centric, thinking about what's best. We kind of play that third party, and because we are one of the major rendering engines, we're one of the three.

even though our market share might be at four or 5%, , we have a seat at the table when there comes to web standards meetings and JavaScript and the future of CSS and things like that, we're there. Because we are one of the major rendering engines, right? And so, what we don't want is a world where everyone consolidates to Chrome and then they start building their own proprietary markup and language and whatnot, whatever you wanna call it, which gives them them advantages and effectively monopolizes their control of the future of the web.

And so, in many ways, I consider Mozilla and Firefox as an existential, reason that it needs to exist.

Jeff: Anyone who's done marketing, you have like 30 different standards you have to pay attention to for email. If you're gonna do any HTML email, there's like 30 different things you have to test it on first to make sure it's gonna work. Which is, which is good in, in, not in a good way, or it's bad in that it's really complex, but it's good in that there's no one player who can [00:27:00] really.

Make a huge change and, and change the whole paradigm. It's gonna be, you know, pretty open. And that's a piece I didn't realize about Mozilla and makes a good point about like it being central to you know, so use Mozilla everyone and keep, the internet open. 

Raj: We obviously, as you can imagine, as we're thinking about AI and browsers. We're thinking about how we can do this in the most privacy sensitive way 

Jeff: Right. 

Raj: most user friendly way and things like that.

Jeff: And. We're not trying to train and collect on all your data.

Yeah, exactly. Exactly. Whether or not Mozilla is thinking about, you know, an agentic browser, I know you're working on solo now over there, which is kind of , a set of tools for solopreneurs and, that end of it. But no matter what you're thinking about, you know, what you're building, if you're working with ai, it starts to get into you know, I think everyone talks about like, what's the quality of token?

What is the best token? And that's the AI you want to use. But from a product billing standpoint, that's, that's really not true. And, and I know we don't have a ton of time to go through this, but there's, there's a lot more think about when you're building AI into your actual product, whether it be a browser or whether it be software or, or whatever you're doing, than just who has the best token, right?

Raj: A hundred percent. Yeah. So I don't, I don't look at it like that. So I [00:28:00] sort of say like, best token is my, euphemism or analog for basically who's the best ai, right? But there's two other parameters that are critical, which is one speed, right? So the speed of inference certain distilled models run a lot faster.

Or if you're running 'em on LPU inference focused clouds like rock or ceris, you're getting, you know, more tokens per second or whatever the benchmark is, right? So there's speed and then there's cost. 

Some LMS are just more compute intensive, right? Like they're a lot of parameters or whatever it is, right?

Or whatever, API interface, right? And then there's longer tail things too that might matter, particularly as you think about browser and the amount of context like size of context window and things like that. But for the most part, those are all at a level that they're, they're sufficient. So I really think of those three parameters, cost, speed, and the quality of the token as what really influences.

How you select and how you decide what AI to use, and that's why we're seeing so many of these AI coding tools like Cursor or whatnot. You know, they might be training their own vertical models or using lightweight things here, right? They're doing some level of orchestration so they can optimize on either speed or optimize on [00:29:00] cost, or they can optimize on output.

Jeff: I mean, was it Cursor just, I think released their own model actually.

Raj: Yeah. 

Jeff: and, and, but you have to be in cursor to use this. I think that was a big pushback people had was they wanna use the API and, but even right, if you look at applications it's gonna start making sense for, for certain things that you're gonna wanna.

have the application actually push and, and host a slimmed down version of, of one of these models, maybe on the client side or something. So you're not spending the API cost to like ping it constantly. There's a lot of different things you can do here that people need to think about. Unfortunately this is, this is probably a whole topic un itself but speaking in Mozilla can't steal your entire day as much as I would like to.

But appreciate you coming on. This is super, you know, interesting. Super good insight. Love to have you back on again soon. Hopefully we can go down more of these rabbit holes, but appreciate the time, man. It was great catching up again, and hopefully see you soon.

Raj: Yeah. Thank you Jeff. Yeah, lots of deep topics. This is such an exciting space.

Jeff: There's, like you said, there's a lot going on in this space and it changes every day. So, all right, . Have a good weekend. Good to see you and hopefully talk to you again soon.

Raj: Okay, bye.

Jeff: But.