Skip to content

By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.

Product SchoolProduct School

Snapchat SVP of Engineering on How Everyone Ships Code Without Breaking Quality | Saral Jain | E304

In this episode of The Product Podcast by Product School, Carlos González de Villaumbrosia sits down with Saral Jain, SVP of Engineering at Snapchat, the last independent social platform operating at global scale, with 956 million monthly…

Saral JainguestCarlos González de Villaumbrosiahost
Jul 15, 202632mWatch on YouTube

Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.

Transcript

  1. 0:00 – 0:34

    🎬 Prototypes Beat Mockups

    1. SJ

      The bottleneck is not Figma mockups. The bottleneck is we just build the prototype and then show the prototype and the best idea wins. It reviews more than 90% of the code within five minutes of the code being published. Why don't you go ahead and build it? And then Casper goes ahead and builds it. It's a super wild time to be an engineer. Adoption wasn't the problem. Translating that adoption to business impact was the problem. The boundaries between these roles will continue to kind of become lesser and lesser clear, and I don't think it matters what your title is. It really matters what you're spending your time doing.

  2. 0:34 – 2:55

    🎙️ Meet Snap Engineering Lead

    1. CV

      Hey, this is Carlos, CEO at Product School and your host on The Product Podcast. Today's guest is Saral Jain, Senior Vice President of Engineering at Snap. Snapchat is the last independent social platform operating at global scale, with nine hundred and fifty-six million monthly active users closing in on the one billion mark and a community that opens the app more than thirty times a day. The company generated close to six billion dollars in revenue in twenty twenty-five and grew twelve percent year over year last quarter while returning to daily user growth. Saral joined Snap nine years ago, right around the IPO, after nearly a decade leading engineering teams at Amazon Web Services. Today, he leads all engineering for Snapchat, spanning product experiences, multi-cloud infrastructure, and the company's machine learning and generative AI platforms. In our conversation, we cover letting designers and product managers ship production code with an AI agent running the first review pass on ninety percent of code within five minutes. Casper, the AI teammate any teammate Snap can invoke from Slack or Jira to build a working prototype from a conversation, turning company-wide AI adoption into business impact after early prototypes were being built and thrown away. Running lean startup squads that makes engineers, designers, and data scientists to launch zero-to-one bets inside a mature platform. Let's get into it. Welcome to The Product Podcast, Saral.

    2. SJ

      Yeah, thanks. Thanks for having me. It should be fun.

    3. CV

      It's been fun to kind of dig a little deeper into Snapchat and fir-first of all, realize that the company's almost fifteen years old.

    4. SJ

      That's true. Yeah, I mean, most people don't realize we have been around for a while. It's almost fifteen years old, and, uh, uh, we have thrived to almost a billion, uh, Snapchatters, uh, in our community right now. So it's been a fun ride.

    5. CV

      You mentioned-- You said over a billion, uh, users?

    6. SJ

      Yeah, almost a billion users. So we have around nine hundred and fifty-three million monthly active users right now using our platform very, very en- in engaged way. So almost a billion people use Snapchat.

    7. CV

      That's crazy, and that makes me feel a little old because I was an early user, but I have to confess.

    8. SJ

      You and me are both, uh, in the same boat there. [chuckles]

    9. CV

      And so you've also been at the company for what? Like nine years?

    10. SJ

      That is correct. I joined right around the IPO, so it's been around nine years I've been here. I played various different roles at the company as well.

    11. CV

      And now you're currently SVP of, uh, Engineering, right?

    12. SJ

      That is correct. I lead most of engineering for Snapchat.

  3. 2:55 – 4:47

    🏢 How Snap Is Organized

    1. CV

      So how does... Help me understand the org design. So I understand you, you, you report directly to CEO, and then how is your team structured?

    2. SJ

      Yeah. So I think, uh, we have a very functional organizational structure where all of the functional leads, whether it's the head of product, whether it's engineering, whether it's sales, legal, marketing, comms, all of those functional leaders report to our CEO. Uh, and so we don't have, like, the typical general manager model where there's, like, a single accountable owner. All major initiatives, we have to all work across all of our, our teams to, uh, get stuff done, which, uh, which has pros and cons, but it really works very well for a company like Snapchat to have functional leadership report to our CEO.

    3. CV

      He's effectively the CPO as well, I guess.

    4. SJ

      He is. He's a designer by heart, by training, and he's our chief product officer and design officer as well. Although we do have a head of design and a head of product that reports to him.

    5. CV

      So let's talk about it. Like, what is the relationship between specific product management and you as the engineering leader?

    6. SJ

      Yeah. It's a very, very tight partnership. Like, almost everything we do, we do in very, very close coordination, essentially. It's a very, uh, design, uh, focused company. And so a lot of the time, the ideas come from our design team essentially, and the product team essentially, who really pay very close attention to our community and the pain points of the community and what we want to create for our community essentially. And so we have very clear long-term thinking of what we are trying to build. So the product team comes up with some of the ideas. The design team comes up with ideas. The engineering team also kind of own some of the ideas essentially, and work very, very closely together to get them executed.

    7. CV

      So engineering and product are peers, and now you get-

    8. SJ

      That is correct

    9. CV

      ... design. Is, is design also a peer to the other two?

    10. SJ

      Our design team reports to our product manag- head of product essentially. And so, uh, think of it as three in a pod essentially. So the design, product, and engineering work very closely together, but design and product report to the same, uh, leader essentially.

    11. CV

      Got it.

  4. 4:47 – 6:11

    💻 Designers Shipping Code

    1. CV

      Well, one of the, the things I heard your CEO mention, uh, recently is that AI is writing over fifty percent of the code and that, um, designers are even shipping code. Yeah, so for you as an engineer, [chuckles] curious to know, how do you feel about non-engineers shipping code and how you're managing that?

    2. SJ

      You know, I have actually gotten very comfortable with that idea because it is a reality in the industry right now that the boundaries between designers, product managers, and engineers are actually changing in front of our eyes right now. We have, uh, uh, product-minded engineers kind of working through the product process essentially. We have designers, design engineers, product managers shipping, uh, code, uh, in production. And, uh, that is the reality we have to live in because the, the boundaries between these roles will continue to kind of become lesser and lesser, uh, uh, clear. And I don't think it matters what your title is. It really matters what you're spending your time doing essentially. And so that's really how we are approaching it, and we definitely have some designers who are shipping production code. Uh, but we have over a decade of in- of investments in our code base that makes, uh, it, uh, easy for people who have non-traditional engineering, like non-engineering backgrounds be able to contribute code without increasing the blast radius of outages or performance regressions too much essentially because we have invested in a very robust platform under the covers.

    3. CV

      I'm very curious always

  5. 6:11 – 7:33

    🚀 Startup Squads Explained

    1. CV

      about this topic. I'm seeing a lot of the AI native companies thinking about smaller pods. Anthropic is an example. I hosted their head of product recently. And in those pods, as you mentioned, regardless of the title, everybody's shipping. Uh, so how does that work in practice for your org?

    2. SJ

      Yeah, I think we have a similar concept. We call them startup squads, and this is something Evan published in his letter last year as well. Uh, he typically writes an annual letter to our, uh, community, to our, uh, employees, to our stakeholders, essentially. And in that, he mentioned this concept of stop-- startup qua- squads. Uh, and these are typically very, very lean cross-functional teams of engineers, designers, product managers, data scientists, all working together on a problem and almost treating it as a large startup, a zero-to-one initiative or a net new revenue initiative that they're working very closely on. Um, the role, uh, distinctions in within that startup squads are not as clear, essentially, because we are all working together as part of a startup to kind of make something successful, essentially. And so, yes, there are designers, uh, who ship code. There are product mind-minded engineers who, uh, who have a knack of asking not just the how but also the why, essentially, who are actually kind of coming up with the product intuition as well. And, uh, the teams are working really well. Some of these we have, uh, net-- we, I think we invested in nine startup squads essentially, and most of them have generated incredible results in a very short amount of time.

  6. 7:33 – 9:53

    🛡️ AI Code Review Guardrails

    1. CV

      And so from the quality assurance perspective, like especially as now more and more code is being generated by AI, as more and more people have access to shipping into production, like how do you create the mechanisms to ensure that you maintain high quality as you increase speed?

    2. SJ

      Yeah, 100%. That is like one of my top focuses right now because, as you know, like we ship to almost a billion people in across the world essentially, and quality is the utmost important thing because we cannot afford that to be degraded. The way we as- approach it is foundational platform investment. So we have a very, very good code review agent, for example. It's called Code Pal. So no matter who ships the code, the first pass of code review is done by Code Pal essentially. It reviews more than 90% of the code within five minutes of the code being published essentially, and it does a very, very good pass at identifying all of the major issues that can happen. It understands the, not just the code that is being changed, but literally the tens of millions of, uh, lines of code across all of our repositories all at once. So it can really identify patterns that humans might not be able to in many cases essentially. And so that's one good example of how we have invested in underlying platforms that can help us ship code faster across different, uh, job profiles without reducing the quality. Uh, we have many such investments like, uh, uh, in our client platform, in our server, uh, uh, server architecture essentially, where we are solving all of the hard problems once so that the feature engineering teams can focus on their feature logic essentially. Does that, that make sense?

    3. CV

      Yeah. So you're saying there is an co-agent that does the initial pass and then you still have an, a human pass, like the engineers ultimately kind of vouch for the code that goes live or is there anything else?

    4. SJ

      I think that is where we are at today, which is, uh, ultimately every single piece of code needs to be owned by a human, uh, because, uh, uh, that ownership is extremely important for us. So the first pass is done by a, uh, an AI agent, but then a human does review the code manually. Uh, uh, although in the future we are getting increasingly more comfortable with depending on the service or the kind of code having the AI pass be the authoritative pass, although we are not completely there yet. And AI adoption will be uneven across a large company like ours anyway. But, uh, today the humans are absolutely, uh, responsible for the ultimate code that gets shipped.

    5. CV

      Got it.

  7. 9:53 – 12:01

    🏗️ Why Spin Off New Companies

    1. CV

      And so on the, on the desi- org design front, I'm also curious about how you are now structuring teams across multiple companies even, right? So you recently, well earlier in 2026, spun off Specs, your AR division. Then most recently you also, uh, kind of spun out another company, the Gen AI video group, Dotmo. So we're curious to know what is the rationale behind spinning off, uh, teams or companies instead of just keeping it as part of the same umbrella?

    2. SJ

      Yeah, that's a very good question. I think that's more business rationale than like, uh, anything to do with product development or AI essentially. But like we, uh, Specs is a great, great example that is, uh, is such an exciting moment for our company right now that we are in-- we have invested in our Specs for over a decade. We have invested in AR for over a decade, and things are converging where AR, AI capabilities are strong enough and the hardware is strong enough now that we are close to our consumer launch. Uh, we announced our Specs, uh, uh, this year and it's a different product. Like Snapchat is a mature platform used by a billion people. Specs is a net new, uh, computing platform that is, uh, being done for the first time in the world. It's almost like a startup, right? And so those two require different le- different kinds of thinking in terms of how to operate a business, different kinds of investments essentially, and that's the rationale of spinning it off as a separate, uh, entity essentially. Similarly, the Dotmo which is the new, uh, gen AI, uh, startup that we spun off is a very, very, uh, uh, creative, uh, idea because we have, uh, almost a decade of investments in our own AR technologies and like we have, we have been developing our own image and video generation models. Um, we think, uh, we can, uh, continue to kind of double down on those investments which are very capital inve- intensive investments, uh, as a separate entity versus part of the same umbrella. But they share a lot of technology together, right? Like, uh, I mean, we-- as I mentioned, the foundations we have built over many, many, uh, years, uh, are, are shared across all of these. And so it's, uh, it's, uh, it's very exciting for me to see that we are able to kind of, uh, support the, not just Snapchat but multiple different products using the same foundation.

  8. 12:01 – 13:29

    💪 Competing With the Giants

    1. CV

      Yeah. It's to me also very interesting because outside Meta or Google, Snap is, is kind of the, the independent social platform, right? Like the, the, the David fighting against the giants and somehow being allies. So some of these moves that you are making, like spinning off companies Uh, I- I- I've seen them for much larger platforms, right? So curious to know how you're able to s- to, to really build your own infrastructure and manage your budget, assuming you have less budget than those giants, to still be able to stay relevant and, you know, competitive in such a, in such a market.

    2. SJ

      You know, I think about that so much. Like, I believe, uh, Snapchat is one of the only independent platforms that have the kind of reach that we do, essentially, and that has been true for the entire history of Snapchat existence. We have always competed against companies that have 100X more resources, 100X more budget, 100X more people to do it, and the only way we are able to do it is it is such a creative company. It is... We move so fast, and we understand our community so well. And, uh, those are part of the DNA of the company, essentially, and that, that has served us really well for the last decade, and I believe they'll continue serving us really well in the next decade. And especially with AI, uh, I feel AI is the best thing that could have happened to a company like ours, which is just so creative, so many ideas, and often the bottleneck is the execution. But with AI, the execution bottleneck is kind of going away, so it's, it's a great time to be, uh, in, in our shoes right now.

  9. 13:29 – 14:58

    🔄 AI After Restructuring

    1. CV

      Yeah, I want to double-click on that and, and kind of bring an elephant also in the, in the room, right? Like, so recently also Snapchat or Snap a- announced around a layoff of 1,000 people, yet... And they kind of put AI as one of the reasons because you're finding more efficiency with that. So I'm really curious how you are able to still, you know, even reduce your team and yet maintain that crazy pace, yet spin off companies. Like, what is kind of the secret sauce behind leveraging AI in ways that other companies are not able to leverage yet?

    2. SJ

      Yeah, I mean, uh, look, I think that restructuring, uh, was a reality of the business, essentially, and, uh, AI is, uh, one of the things that helps us feel confident that, uh, we are well positioned going forward. And we are using AI across many different aspects of the business. One is AI products for our community, for our Snapchatters, essentially, who use Snapchat. Second is, uh, AI for our employees, uh, like how we are rethinking the way work is done within Snapchat and being more efficient using AI. And then the last thing is AI for our partners and advertisers, essentially, how can we rethink our business, uh, using AI? And we are making tremendous progress on all three of these verticals, essentially, and that gives us the confidence that AI can, uh, help us, uh, continue moving really fast. But, uh, I will say that, uh, in general, I believe that s- smaller teams are more flexible, more nimble, and are able to move fast, and that is something that has proven really well for us over many years, and we'll continue doubling down on that.

  10. 14:58 – 17:09

    ✨ AI Features for Snapchatters

    1. CV

      So you mentioned there are three, uh, specific applications. One, like, external application for the user in terms of AI, the internal application for your team, and then also for your, for your ad partners. So want to kind of go a little deeper into each of those. So if you think about your external users, what are some of those AI features that might not even be re- maybe users don't even realize it's, it's AI, but it's kind of powering the experiences that people have on, on Snapchat and other companies.

    2. SJ

      Yeah. Look, I mean, I think outside of ChatGPT and maybe NanoBanana, there haven't been a lot of breakout AI features where there are, that are, you know, like, massively adopted, essentially. And so for us, we actually don't think of features as AI features or not AI features. We think of features as are they adding value to our community and our core product, uh, value, right? And I think, uh, in that we have launched, uh, recently a couple of very, very interesting features that are gen AI features. One is My AI, which is our chatbot, essentially. It's an AI assistant, uh, that, uh, people on our community use. Uh, but it's, uh, it, it is very unique. It has a personality, very different kinds of use cases that people use My AI for. Uh, and similarly we have Gen AI Lenses, generative AI lenses. AI Lenses are a core part of Snapchat. Snapchat opens to the camera, so the first thing people play around with are those filters and lenses that, like, are unique to Snapchat, essentially, and many of them are powered by generative AI right now. Um, last December, I believe we launched a lens called Imagine Lens, which is an open prompt lens. You can actually look at anything in the world or your face on the camera and give it an open text prompt saying, "Transform my world in this particular way," and it'll do it. And, and I think, uh, uh, it has been such a massive hit with our community. I believe, like, over 700 million people have interacted with our Gen AI Lenses since they have launched, so, uh, definitely a big viral hit.

    3. CV

      You're right. Like, a lot of these features, they do not need to be called AI features.

    4. SJ

      Sure.

    5. CV

      At the end of the day, if it's a good experience for the user, it's a good experience. Doesn't really matter what's underneath it. And you ma- are making me realize, I mean, uh, Snapchat has been the pioneer with far, for a lot of these new features, right? Since the days of the Reels to now Lenses, to the location

  11. 17:09 – 18:26

    🏰 Defending the Product Moat

    1. CV

      of your friends.

    2. SJ

      Sure.

    3. CV

      A lot of the things that now are becoming commoditized, right? So where is the moat these days for those type of new experiences, especially as people are able to copy that faster?

    4. SJ

      Yeah, that's a great question. Look, I think the moat is the platform and the network effect, essentially. Like, we have always pioneered some of the most widely used features in the industry, essentially, whether it's, uh, ephemeral messaging to stories, to maps, to Gen AI Lenses, to just filters. Uh, Snapchat's DNA is being creative, essentially, and, and I think, uh, our design team and product team is some, some of the best in the world, essentially. Uh, but, uh, we have a habit of doing it again and again. But the most interesting thing is the reason our community loves these features, they are all interacting together with each other. Ultimately, people come to Snapchat to talk to their close friends and family in a way that mimics the real world. It's fun. It's casual. Uh, it's, like, not, not for, you know, like, uh, likes, uh, essentially, and I think that is why people love Snapchat. And we build features that are all tying to this, uh, uh, DNA of being able to talk to your close friends and family in a, in a fun, unique way, essentially. And, uh, that's just how we think about product development, so that's our moat.

    5. CV

      That's true. The, the filters, I don't know, they're- ... feature that you kind of pioneer-

    6. SJ

      Right

    7. CV

      ... that now it's assumed that everybody should have features-

    8. SJ

      Right

    9. CV

      ... on their camera.

  12. 18:26 – 21:00

    🌍 Who Snapchat Is For

    1. CV

      Uh, so let's talk about, you talk about close friends, you talk about family. I want to talk about user segmentation, because over the last 15 years, of course, there's a lot of new users that are getting to the platform. You mentioned you have over a billion, um, monthly active users. So who is Snap, so who is Snapchat for, and how are you making sure that you continue to be relevant to some of those new users versus maybe other alternatives?

    2. SJ

      Yeah, that is such a great question. Snapchat is for everybody. I think, uh, in the past, like, there have been like, uh, like, you know, uh, conversations about a particular demographic is the right demographic for Snapchat. But when you're reaching a billion people, you can safely assume that everybody is your user essentially, and that is true across geographical boundaries, across demographics, across age, gender, and everything. Uh, and um, uh, our community's so engaged with our platform that, like, it's one of the most often opened app in the world. So not only are the people coming every day, they're opening it, like, on an average 30 times a day, right? And so, uh, it is a very, very kind of, uh, uh, uh, repetitive behavior essentially, and we, we love, uh, uh, uh, that, that about our platform. And so no, it's-it's for everybody. And, uh, we understand our user base every, really well, and our features kind of resonate with a large u- user audience as well.

    3. CV

      I was having a similar conversation with the VP of product at Robinhood, right? Like, they started as a easy trading app for first time users that were getting into finance, and now obviously as they evolve, they have the sophisticated financial investors all the way down to Gen Z-ers that are also trying to get into the financial markets, right? So how do you go about segmenting customers or providing different experiences to make sure that you can really cover such a broad spectrum age-wise?

    4. SJ

      Yeah. I mean, I think it, uh, it kind of all starts with each feature, what is the value proposition? I think to give you an example, we are, we have a feature we are, uh, recently launched called Topic Chats. Topic Chats is a place where community, our, our entire community can get together, talk about often, uh, uh, talked about topics. And most recently we have realized that, uh, some of these games, whether it's, uh, the NBA, whether it's the World Cup that is going on right now, has like a massive, massive use of a Topic Chats functionality. And so for us it's about having like a really, really good perspective of what we are launching and for who we are launching that, and then doubling down on building creative features to make it, uh, useful. Uh, but, uh, uh, that segmentation we try not to do ahead of time. We kind of are thoughtful about which features will resonate with who in the community, but sometimes our community surprises us, and so we kind of then are, uh, creative and, uh, build features that can have the largest impact.

  13. 21:00 – 23:19

    📈 AI for Advertisers ROI

    1. CV

      So shifting gears a little bit, I also want to talk about the kind of AI use cases for your ad partners. I guess that's still the main part of your business model, right? Like partners that want to advertise on, on Snapchat. So what are the specific AI-powered experiences for them?

    2. SJ

      Yeah. That is such a, a important thing in our business. Like, Snapchat's ultimately an ads platform, essentially from a revenue perspective, although we do have a very thriving subscription product right now. But, uh, ads is the real way we kind of make money, at least right now. Uh, uh, we have-- Again, I think the way we approach AI for our community is the same we, uh, way we apply, uh, apply AI for our advertisers. AI for AI's sake is not something advertisers are interested in. They are interested in performance of our ads. And so ultimately, AI is only valuable for them when the ROI of every dollar they spend on Snapchat increases essentially, and that's how we are using AI. So we have built, uh, solutions for smart targeting of customers, smart measurements of, like, how ads are performing, smart creatives. How can we, how can advertisers create, create the creatives automatically without kind of, you know, too much investment on their side? So it's a pretty big, uh, investment in all parts of the advertiser journey using AI, but ultimately not for AI's sake, but, uh, just for improving the ROI for our advertisers.

    3. CV

      So as I think about this, you are in the advertising business, you are competing for attention ultimately. And, um, what, what is the kind of the u- unique value prop, or how is AI helping you accelerate that value prop so you can capture a bigger part of that, that, that wallet?

    4. SJ

      Yeah, 100%. Look, I think, uh, we have to be the, uh, we have to get advertisers to what is the core moat of our product, which is, for example, our messaging product, right? And so we have recently launched AI Sponsored Snaps, which is, uh, a Sponsored Snaps from our advertisers on the chat feed, but it's an AI-enabled product where the users can interact with their favorite brands using AI conversations essentially. And, uh, uh, we recently piloted it with, I believe, Experian, for example. Uh, and, uh, we are seeing great lifts and great results. Advertisers are seeing great results. And so, uh, when we think about advertisers, we also have to tie it to our community and our usage of our app essentially, and then AI is a, a value add on top of it.

  14. 23:19 – 26:32

    🎯 Jobs To Be Done for Impact

    1. CV

      And then finally, and we already touched on internal use of AI, as you mentioned, smaller pods or squads, uh, everybody shipping, kind of creating the constraints for people to stay creative and, like, use AI in ways that are helping you ship faster. But one of the challenges that I'm seeing with organizations is that as they build that system for everybody to be enabled to leverage AI, uh, not everybody's on the same page. So you have this group of people who are, like, super users who are already creating their own context and, like, really going for it. And then you have people who are curious, but they, they still need to, to kind of brought up to speed, right? So, and as we get into a more multiplayer platform with the idea of, hey, everybody should be able to leverage AI without having to set up the entire system or context themselves. Like, how are you building that internal infrastructure?

    2. SJ

      Yeah, that is such a great question. I'll talk about engineering, and I'll also talk about the larger company, because we are seeing different things in different use cases. Uh, uh, first talking about the company, uh, when we started first rolling out some of these AI solutions, uh, like Gemini and ChatGPT Enterprise, we quickly realized that almost everybody's using it on a daily basis. So adoption wasn't the problem. Translating that adoption to business impact was the problem. A lot of prototypes were being created, and many of them were being thrown away essentially, and that's not necessarily, uh, useful for, for the business essentially. And so we quickly pivoted to what we call the Jobs To Be Done framework. We actually sat as a leadership group Talked about every function, whether it's sales or engineering or marketing or legal or any other function in the company. What are the actual jobs do they do for their stakeholders or, or users essentially, right? Whether it's internal employees, external Snapchatters, or anybody else. Uh, and after we identified the jobs, we then started talking about how AI can help us rethink the job, accelerate the job, or maybe not touch that particular job. I think once we created an operating system that is oriented towards our jobs that we need to do for our community, then AI became a massive accelerant essentially. And of course, even there we had to go through a change management process of training and enablement and making sure that we are building the agents that can help us accelerate the jobs. But crafting and grounding stuff into the jobs to be done framework was extremely helpful for us.

    3. CV

      How do people use that framework in practice?

    4. SJ

      In practice, we actually, uh, created a list of our jobs to be done per functional area, and then when we create, like, our roadmaps for the startup squads or for generally the entire team, we align the roadmaps to the jobs to be done. So at any given point in time, you can actually see what the jobs to be done across the business are and what are the roadmap items that any individual are working on, which jobs to be done are they tied on. And so when they actually start using AI, they then, instead of working on some throwaway prototype, actually use AI for one of the actual roadmap items that are tied to the jobs to be done, and that's how business impact happens essentially.

    5. CV

      Got it. So it helps you kind of put the guardrails around what the company's prioritizing, and I guess you also have the infrastructure in a way that allows people to start building with AI without having to do all the, all the back end work.

  15. 26:32 – 32:56

    🤖 Casper Agents and the Future

    1. CV

      Um, I was so curious, you mentioned at the beginning you have a coding agent that is helping do the initial pass for code and... But you also mentioned there is non-coding, uh, other teams outside engineering that is also leveraging AI. So how are you thinking about those agents? Is this like a you have a generic agent that is kind of overseeing everything, or is it more of a very specific agents on a, uh, divided by function?

    2. SJ

      I think that's such a great question. Uh, so we have two kinds of agents. Some are foundational agents. So a code review agent or a coding agent are foundational agents that anybody in the company can use essentially. But then some are vertical agents as well. We have an agent that does a great job at A/B tests. It kind of looks at all of our A/B ta- test data, experimentation data, and, like, able to understand what's going on. We have a data science agent. We have a cost agent, which finds cost inefficiencies in our infrastructure. So we have vertical agents and horizontal agents. Uh, uh, one of the best examples of a horizontal agent is called Casper. Casper is our remote coding agent, essentially. So what it does is it, uh, we almost call it an AI teammate. It is not an engineering agent. It is any team in the company can use Casper. Uh, and, uh, Casper is available where the people already are. So it's in Slack, it's in Jira, it's in Coda, oth- other tools that we use essentially. And at any point, you're having a conversation with your teammate, and you're just thinking about a product idea or some iteration, Casper is listening into that conversation. And at one point, you can just invoke Casper saying, "Okay, you have all the context. You know everything of the history of the company essentially, and all of our knowledge base and, uh, literally all of our tens of millions of lines of code across all repos. Why don't you go ahead and build it?" And then Casper goes, goes ahe- goes ahead and builds it. Uh, and then the code review agent, which is CodePal, reviews the code, and they have a back and forth, and that's how engineering is, uh, working right now. It's, it's a super wild time to be an engineer.

    3. CV

      That's so cool.

    4. SJ

      Yeah.

    5. CV

      Um, in the same way we talked about designers now shipping code, I also want to ask you about engineers prototyping and kind of getting into the designer territory. Like, how do you go about it now?

    6. SJ

      I mean, that's actually, like, such a refreshing, uh, change from my perspective as well. Uh, not that, uh, we didn't have that in the past, but we always look for engineers, uh, that are product minded because it's such a... Ultimately, it's a product we are shipping to our consumers, right? And, and so we always look for traits in an engineer where they are not just asking the how to build something, but why should something be built essentially, and that trait is, uh, goes a long way in building prototypes as well as, like, building actual, uh, products for our community. But, uh, we are seeing so many examples of engineers coming up with, like, uh, insane ideas, and now the bottleneck is not Figma mockups. The bottleneck is we just build the prototype and then show the prototype and the best idea wins.

    7. CV

      So as, as you build those prototypes, are you still relying on third-party tools like Figma, or do you go straight into Cloud Code and, and prototype?

    8. SJ

      Yeah, most of the prototyping is being done in Cloud Code right now, but we do have our own design system, uh, that Cloud Code understands essentially, and so it still looks and feels like Snapchat. Uh, uh, and of course, Figma does play a big role, uh, in, in terms of ultimately shipping the features that we ship essentially. But prototyping is being done by feature, by tools like Cloud Code, Codex. Uh, we have rolled out every AI tool you can possibly imagine. And so [chuckles] I wouldn't necessarily single out Cloud Code, but, uh, certainly Cloud Code is a big part of it.

    9. CV

      All right. No, and, and I, I find that really cool because you're literally setting up the conditions for, in this case, an engineer to be able to prototype, and the d- the design system is there. If the tooling is there, you can go straight to the actual experience you want to create, um, and vice versa, right? When a, a, a non-engineer is trying to ship code, if the conditions are there, they can go straight and request and try to do what they want without having to worry too much about, like, all these other things that might be more, more technical.

    10. SJ

      Yeah, 100%. I think, like, we want to get to a world, and we are in this world where in a, in a scenario when building is so cheap, what to build becomes the most interesting thing, right? And that's where the best ideas win, and the best way to find the best ideas is to have lots of them. I think Evan has talked about this quite a lot as well, and that's where prototypes come in pretty handy.

    11. CV

      Yeah. So as you, as you continue to make these big bets, spinning off companies, uh, leveraging AI in a crazy way to build faster and, and better Then there's also the reality of the market, right? Like, kind of like the weather. You check the ticker, and it's like, "Oh my God." Why isn't the market, in your opinion, kind of rewarding all of those investments and improvements that you're making, even though you're objectively actually beating expectations across the board?

    12. SJ

      Yeah, look, uh, market is sometimes a lagging indicator of, of, uh, execution essentially, and so we are heads down focused on execution, executing our strategy. That's something Snapchat has done really well for over 15 years now, and so we'll continue doing that. Uh, uh, uh, we know for a fact that our community loves us, uh, that's why they use... All, almost a billion people use us every day. We know our, uh, advertising performance keeps improving, and we, uh, we, we know that we are focused on being disciplined about the gross margins. And so, uh, we know the investors will eventually kind of catch on to that, but that's not something that you as an engineering leader wants to distract your team on a day-to-day basis on.

    13. CV

      And for, for you from a planning perspective, like, how far do you try to plan knowing that there's so many changes?

    14. SJ

      I think that's a great question. I think in, in the past, we used to do these year-long plans, and we still have a sense of what our strategy is for the next year. But we bre- try to break down planning on a quarterly level right now so that we have clarity on what we want to achieve every quarter tied to our yearly plan. And then finally, we are extremely flexible, and things change on a daily basis. We sometimes actually do daily check-ins on key initiatives, and, uh, as things change, we're adaptable because that is in our DNA, to be able to move fast and be adaptable and flexible on things. So I wouldn't recommend, especially with AI, to, to kind of do these long horizon plans and stick to them because that is just not the reality of the world at this point.

    15. CV

      Right. I mean, and it's so part of your DNA. You're not... You're the ultimate survivor, like literally 15 years fighting against giants and, like, innovating and, like, finding creative ways to, to continue the, the, the good fight. Um, Saral, it's been a pleasure to learn from the inside, you know, how you guys are building and thinking about AI. Thank you so much for going deep into the details with me.

    16. SJ

      Thank you. That was such a fun conversation. [outro music]

Episode duration: 32:56

In the uListen app, the transcript lights up word by word as you listen. Get uListen

Transcript of episode B4Ore8A2RA0

Get more out of YouTube videos.

Read along on your phone, search it beside the video, or ask your favorite AI app what was said.

iOSAndroidClaudeChatGPTChrome