How to Design Your Stack: An Interview With Sherry Jiang, Peek Founder
It was how do you actually create the best growth levers in the product? Your legibility of traction and monetization just has a longer time horizon, right? It's like a lot harder to get to your first 100,000 $10,000. I still feel like Convex is a very modern solution, right? I I would say um I'm loyal to what's good. It just makes it so much easier. I think there's some magic that comes with that. Click on something and immediately when I open up the Convex dashboard, I just see it there. So, I think that's just like a testament to like how seamless it is to set up something that used to feel so arduous. Yeah. Some people are just loyal to a brand even when the brand does not deserve it sometimes, right? But it was really it was like a very like Gen Z take on like productivity, which is like we don't want these boring productivity tools. You need an AI that just bullies and roasts you. Hi, I'm Wayne Sutton, head of community at Convets. I'm here with Sher Junk, founder of Peak, and we're here to talk about her Convict experience. Hi. Hey. How you doing, Sher? I'm doing great. Okay. So, should we do the whole spill? Who are you? Where you from? Why you here? What's going on? Wow. Okay. Well, uh, I like to build stuff. Um, but, um, I live in Singapore right now. Um, I have a startup called Peak, uh, in the consumer AI space. Uh on the side I actually also do uh vibe coding education. So I've taught over a thousand people how to go from zero to one building with things like cursor v0ero. Um and yeah convex fan as well. Awesome. Convvis fan customer. Yeah. Awesome. With peak. Yes. With peak and my own projects as well cuz you know you got to build outside of your own stuff sometimes. Is that how it started doing your side projects and then convets? What was that user journey like? Yeah. Well, I actually learned about Convex from my co-founder Jeff. Um, and then um I think the first time I got like a lot more exposure to it was actually during the Cursor hackathon cuz then I got to see people build with it. Yes. And I saw someone basically creating ASKI art that was like a flip book with a database. That was just a crazy idea. It was wild. And I was like, huh, okay. I think I'm going to I'm going to give this a try, right? Um, so I started to build my I mean, I've been building my own like personal software. like I have my own to-do list, my own compon board, and I use convex for all of it because it's just so it's so simple to set up. Um you don't need like MCPS or any of that to like make sure it's talking to whatever Postgress solution you use. Um and uh yeah, I mean I think uh the simplicity was the first thing that got to me, but then um I think in terms of our decision with peak was actually more around this whole like real time UI bet that we we want to take. Awesome. Awesome. Great. Well, thank you for using comments. Thanks for building with I say thanks for building with us because we still Thank you. Yeah. Yeah. Yeah. All right. So, Peak, what's the story? I heard you vibe coded it, got into accelerator, raised some money. What is Peak? How where did how did it come from? Yeah, for sure. Well, um I have been building consumer fintech for almost a decade now. Um uh I worked at uh Google actually on payments. So a big part of what I was doing wasn't just like okay um what's a transaction how how do you send money here and there it was actually understanding human behavior around money because um how we spend what we think about prioritizing what we spend is a huge part of it um so I was really interested in how people um even though they want to be better about saving and spending money and their priorities why don't they do it right and so that's a whole bet around peak so we want to help bring a layer of behavioral intelligence to help people have a better sense of self-awareness on their own spending patterns. Um, and AI is a really cool way to ex kind of bring that together because what what are some of the best ways to learn um about how people think about things? It's through conversation. Yes. Right. Um but it was so hard to articulate what this could be and abstract. Yes. And when we were raising money, I was describing this just as you. So, you're probably like, "What exactly does that mean?" So, I was like, "You know what? I think I'm just going to have to build a first version, get a bunch of feedback from our customers, which are uh normally, I would say, early career folks, Gen Z's." I just put it in front of 20 people. Um, they really liked how chill it was, talking about things that are more emotionally charged, like, you know, impulse shopping. And then I used that to basically, uh, get into an accelerator. So, um, I had recordings, I had transcripts of people using the product. So, it's just so much easier in this day and age to show not tell what you do. And I felt like that was the best way to like bring the vision of Peak uh alive, I guess, or to share it that way. Awesome. Awesome. I'm I'm going get we going get back to Peak, but you you touched on so much right there in terms of like your behavior uh behavioral experience um working in fintech, working in the financial industry at tech companies. Um what what was your early career like in tech? I heard you was at Google and all the things. Well, you say you was at Google. So, what was that like? Yeah. Um, Google was actually my second job. Um, but I was there like I guess a year after graduating, so it's pretty much where I grew up professionally. Um, and uh, I I got to work on some cool stuff, right? I think um, often times people think, okay, when you're at Google, you work on a product with a billion plus users already, like maps or search. And that never really interested me. I was like I want to go for the most like you know maverick out there bet that maybe won't work out but it's like more fun and um at the time um building digital payments for India was uh it was a bet from the company um because it was still a cash driven economy. Um so that's actually what um then moved me out to Singapore to work on that business. Um, but what was really really fun during that time was I um as the growth product marketing manager, I got to run a lot of behavioral science experiments cuz a big part of our growth strategy wasn't just like running ads and all that, but it was how do you actually create the best growth levers in the product. So I got to design game mechanics. Nice. Uh like uh scratch cards cuz one of the you know principles in behavioral science is uh variable rewards. So I mean basically if you uh if you never win something it's not very motivating but if you win all the time then you don't get the anticipation that is the reward. So we kind of tested that 35 to 65% win rates were like just enough that people were really excited if they used their product and then got a scratch card and you know got some kind of reward. Nice. Nice. So I got to really understand the psychology of how people think about money. Um and you know think about how to use it in a positive way for people to develop a you know a more I would say like uh healthy relationship with money less fear around new um foreign factors like digital money etc. Yeah. You have all this amazing experience at Google and you start building peak. At what point do people like who are you? Why do you build this app? and you be like, "Yeah, I used to work at Google. I used to build this thing, so I know what I'm talking about." Does that ever come into the conversation? Um, I think yes and no. Um, I do think uh I would say broadly in this current time building in consumer for AI is uh I wouldn't say like not understood, but it's definitely not the popular choice. I would say like people are like, "Oh, why aren't you doing something in B2B?" I mean I think something like 90% of startups built everywhere including here are B2B. Um why why is consumer the bet right? But I think when I explain my background around like why like you know I've always been interested in like understanding how tech can improve people's everyday people's daily lives right it starts to make sense a little bit more. Um I think the other part that I do try to articulate is there are actually systematic ways to understand human behavior. And I think that's also why um when people don't know that they think okay consumer is like a black box right I want to understand things that are a bit more codified a bit more like you know you can break down like how maybe a certain company's workflow works right okay even if it's like a legal company maybe there's a set process of how I don't know this contract goes through right the same thing actually happens for the way humans make uh decisions as well but these are more um internal things that are not as externally expressed unless you spend a lot of time, right? And so that's why I've kind of brought that background because I would say I've had an unfair advantage of just talking to so many people around how they think about money and like where they've, you know, fallen for like a 50% off deal and end up buying stuff that they don't want and how they feel about the evaluation of that purchase decision. So all that has been in my head for a long time and um I'm able to in a way externalize and create what that knowledge base looks like, right? It is still like a you know it is still break downable into like a step-by-step workflow if you think about it. Yeah. You touched on something that I saw a tweet you posted and it was about I think it was um the Airbnb founder talk about how like you focus on consumer like everybody's forgetting this bigger market and and you was like yes finally someone said it. Yeah. Yeah. So what I mean basically validate what you just said it's like yeah consumers a bigger market but everybody's doing B2B. Uh what what was your kind of vibes when you saw that post? Yeah, I mean definitely very validating. Um cuz uh I obviously like um agree with Chesy as well, but I honestly do al do understand um why there's a lot of hesitance to go into B2 B TOC cuz it's freaking hard, right? It's so hard. Um your like your legibility of traction and monetization just has a longer time horizon, right? it's like a lot harder to get to your first hundred thousand $10,000 than like oh I you know signed a contract right um and then I think also there's been a bit of this like um I would say pendulum swing since the zerp era where there was so many consumer companies and money was cheap and it was like grow up cost like who cares about revenue right and now the pendulum swung in the other direction where I think there is a bit more conservatism in the whole capital allocation space VCs basically And look, if you're a founder and you're trying to build a team and you know not starve, like sometimes you need a bit of money. So I think there is a bit more I would say alignment that people would feel like they need to have with what the thesis thesis are out there, right? Um, but I think one thing that um, I'm really like I guess lucky to have is I've spent so much time in consumer where like I know it feels like there's a lot of these different challenges, but I've broke down and solve those challenges before. I'm like, oh, like I'm more comfortable with trying to figure out how to market to millions of people than I am like to knowing how to, you know, do manage a sales pipeline, you know? I just haven't done it as much, right? um that's not really my thing. Um and then I think I also like you know earning stuff on the side and everything so I can you know keep working on what I think is a longer term vision. So that's uh I hope more people come into the space and build consumer. I also understand that it's hard. Um, but hopefully a few people were inspired by Chesky and are like, you know what, like why is it that we spent three years um or has it been three? Yeah, about three years since Chhat GPT blew up. Why is it that when my mom or my cousins are using AI? They really feel like it hasn't changed that much, right? Like that that to me is a it's one of those itching questions that I I seek to answer. Yeah. And soon people gonna be like follow you and as well it's like Sher's doing Shar's doing B toC. Why could I do BDC? You know, I I try my best. Um I I post stuff out there. I I try to like share as much of my learnings as possible. I'm definitely not one of those founders who are like, "Oh, like you shouldn't put your idea out there. Someone's going to steal it." I'm just like, look, if someone can take my ideas that I post and do a better job, like I don't deserve to work on this. Like I should probably find something I'm better at if they can, you know, off of a tweet build something better than me. I know, right? I know, right? Cool. Cool. All right. So, I have some more questions here. Let me jump. Yeah, of course. All right. You posted this awesome tweet on X. That is so wrong to say. We can keep that in there. I'm so wrong to say. I said we posted awesome tweet on. You awesome post on Twitter. I'm still calling it Twitter. I call it Twitter, too. Long live Twitter. Long live Twitter. And you said that you're betting say convets you betting on convets for peak. Yeah. You said you know the old Postgress world feels like passing notes back and forth. What we want is the responsiveness of a game engine. All right. What does that mean? What does that mean? Right. Yeah. Yeah. So maybe I'll like start out with like the user journey, right? And this is where consumer is different than B2B where um most of the time when you're building a consumer product right now um it's not like you are going to lose your job tomorrow if you don't make a sales quota right it's yes you're solving a problem but you also have to like manufacture joy and enjoyment when people are using your app like think about the most popular consumer apps like Tik Tok or Dualingo whatever like um there's this responsiveness of like I do something I get a dopamine hit when there's confetti reward points all of that, right? Or I see something novel and interesting now. We want our app to feel that way as well. Um, in the context of you chatting with our agent, right? Um, now just to explain some of the mechanisms, right? If I'm sitting normally like if I'm doing my I guess customer interviews, I'm sitting down asking people about um how they feel about like a I don't know shopping spend that they made and then they share like oh this was a deal. This was a sweater for my mom uh and I feel really happy with this purchase and I enjoy shopping. These are what we call like artifacts that we remember about this person, right? Um, now we also want to show the user almost like traces like what's happening, what's peak remembering about you. So it feels like it's this like um two-way street and that peak is building this almost like a scaffolded identity of you. Right. Right. Oh, I'm understanding myself more as I'm sharing about that requires uh real-time responsiveness. Right. you completely ruin the experience if I'm sharing something and let's say I'm just like in a real human interaction you're like sharing something vulnerable and I'm like okay cool I realized that was a gift for your mom right and so that's why we want something that feels like a like you've got that quick responsiveness but also b something that when our you know team is building it's a much more delightful experience than having to deal with like three four layers in between like a database and a front end client, right? And so it just made sense to us. Um we used to um have like a more traditional Postgress backend uh with AWS and everything. Um and then just recently we did a migration um to Convex. It took us a week and a half. Um, which honestly it's it's great because nowadays with the tools available, you can kind of orchestrate and do all of it much more easily. Yeah. So, what was the migration like? Right. A lot I think a lot of people when they think about switching databases. They're like, oh, it's going to take it could take us down. We don't want to impact our customers. Like, what was the whole migration like? Was it a stressful process? Because it was an email chain we had going on. like like we're about to migrate. Any tips you have, let us know. Yeah. So, how was the migration journey? Like, so actually the very first thing is we got rid of a bunch of stuff we no longer needed to use. Um, so again, this this came from like us sharpening our own product vision as well cuz we used to like um store things like uh net worth, balances, like all of this stuff on the investment side when now really most of the product is about spending, behavior, cash flow, things like that. So we just got rid of a lot of stuff. We're like why would we migrate stuff we don't use, right? It's like when you're moving a house like you don't just move all your junk from one house to the other. You throw out like probably 50 60% of the stuff and you are like this is a clean slate and that actually this is not even like a convex specific thing. This is just generally good practice is just pretend you're starting over again. What do you what services do you actually need? Um and then we essentially just kind of built those roundup, right? Um and honestly the even though it's um you know not written in SQL like your you know traditional database it's pretty logical I would say. So when we kind of broke it down to the key things we needed okay we need to be able to bring in plaid data for spend. Okay. Um we need to be able to uh categorize right we brought the categorization engine over. Uh, but to be honest, a lot of the stuff that I talked about, which is the behavioral artifacts, the the memory system that we have to make sure we remember what Wayne tells us about his, I don't know, weekend spending habits. Yeah. Motorcycle parts. Motorcycle. Yeah. Uh, that those were all built new actually. Like we didn't have them the same way. So, um, it was actually a lot easier than expected. The thing that we wanted to make sure was that our data integrity on the spend side was inaccurate because that is like the number one way you lose trust as a fintech product is if you tell someone you spend $4,000 when it was$400 or $2, right? So that's all like we just really focused on what we needed and then um everything got rid of a whole bunch of stuff and then um um a lot of the newer the the real time stuff that was actually just built from scratch. Nice. Nice. And and with combats so far, so with combat so far, how has the experience been? You notice the speeded difference if anybody on um front end app is like, "Wow, this is great." Because they see the difference before after you migrated. Yeah, I mean it's a it's a new product that people have. So it's there's not exactly like a complete before and after because we didn't get the artifacts the same kind of way. Um but I mean myself, I can notice a huge difference, right? Like I mean it's it's almost uh when when you were in a world before convex it's like you expect latency as just business as usual. Like that's fine, right? Um and you don't know the difference until you're presented that. Um, and I I think that's just so cool when um you're able to like uh you know message click on something and immediately when I open up the convex dashboard, I just see it there. It's just I think there's some magic that comes with that. Um I think it also makes uh at least for us from a building process like it makes it so much better as well when we're like testing and seeing where where things break and everything. Yeah. No, I appreciate you saying that. That's not a paid advertisement. No, it's not at all. Not at all. Yeah. Um, and I often tell developers like like why why should I try conference? Why should I migrate? I'm like, give us just go do the demo 10 minutes and if you're not you don't like it, let us know. We work to improve it. But yeah, no, that's good. That's good. Um, earlier you said you did workshops teaching people about combat. So here you go from hackathon side projects migrating to comx to now teaching people how to use convex and what happened what was that journey like like how did it go I I get a picture on the weekend it's like we're doing a convict workshop like what happened yeah uh so um you know something I do on the side um a lot of side projects um is uh AI education um basic mostly focused on helping people who have no technical background know how learn how to build an app in like 48 hours right using tools like convex bzero cursor etc um and uh we do this on the weekend uh it's like a boot camp and try to make it like accessible right this is not really people who are necessarily like I'm trying to become a developer tomorrow but they might be doing it as a hobby project or they generally want to like upskill uh so how this came up um I mean basically Ally uh in the beginning of this year Algrim and I who I teach with uh we were like let's kind of revisit what we want to teach cuz the space moves really fast and we felt like convex uh was a better solution than what we were using previously and not to name names but um people were really struggling on some things that feel really basic to uh feel very basic that don't really that can be very frustrating if it isn't done right but then doesn't create much joy if it is done right which is for example like just getting off set up right like imagine you're trying to trigger a triggered word right um uh or even like you know trying to explain to somebody why like you have to like run like we were teaching this in Vero but it's like um sometimes when the integration doesn't work well why you have to like click on run SQL table and have to open up another page and valid validate that's there, right? Like you're learning all this from scratch. Like why are these things in all different places? And um when you're teaching, right? Um you want to try to limit cognitive overload. And one of the ways that we found it to be much easier to teach databases is just to have it in the code itself, right? Um and again when you are you know a newcomer to this um you can ask these kind of questions because you you haven't been accepting what the norm or the world had been for a long time. So um it makes it easier to debug right and whatever agent you're working with be it the cursor agent or you know manis or vzero um you can basically have it direct to those files right and it just makes it so much easier. Yep. Yeah. And so what was the feedback from learning convlets from the people? Yeah. I mean simple. Yeah. Right. Nice. Um it makes people not have to think about the things that like they have to think about middleware issues. Is that the thing? The off loop problems. Um you basically like work on plan mode. Explain like what you're trying to do. It will if you um mention convex as a solution, it will create the schema based on how you're describing your app. Um your files will be created. You do have to run like a terminal command. The agent can kind of get you there and that's it. You don't have to set up an MCP if you want to like mutate anything like all those functions are there. It's like it takes away the things that you have to think about that stop you from, you know, focusing on, you know, creating the delight. your app experience. Oh, that's a great story. That's a great Yeah. You can focus on building. You can focus on building, talking to customers. Don't have to worry about all this other infrastructure because we got you. Yeah. And like imagine like you're new to this and you're like you're seeing all these bugs and then you're like, oh man, like what's the difference between a storage bucket and like the tables and anyways and what's what's SQL? Why is this in a different language than ne like you know what my next.js app looks like? These are just complexities that take away from the core joy, right? Yeah. Especially in today's time where the the mindset in Silicon Valley barrier to now more everyday people, not like Silicon Valley Bay Area, not everyday people, but the mindset is changing in the world where it's like I want an app, I can build it fast. Yeah, I should be able to build it fast cuz it's 2026 and I should have modern platforms to help me build it fast and not and help me ship fast without have to think about all these other layers. Yeah. Yeah. I mean that's a good way to describe it. I I felt I still feel like Convex is a very modern solution, right? Um there are just so many more like I think when solutions are not modern it's a reflection of how humans used to organize themselves around software engineering where you had like your backend person then you had your front-end person and that's how a lot of things were but now like we all know that the lines kind of blurring like you know you are a lot of people are very full stack and work across both and all that stuff and um I think the solutions kind of need to reflect something that feels more modern right and the other part of it is making it very very LLM readable and friendly. Yeah. Right. Because now you have another colleague, your AI agent that's building and if it's not able to understand how to navigate um what you've built, that makes it very hard, right? So I think it's very cool. Um we've had people like as young as 13 years old in our class like being able to build. So I think that's just like a testament to like how seamless it is to set up something that used to feel so arduous. Yeah. No, that's great. That's great. So, you are clearly our builder. You're passionate about building. What's your stack? What What do you is I heard cursor. I heard Vzero. What? Okay. Before we jump to stack, what's your current model that you use? Yeah. Um I mean, my preferred one in the last uh couple days has been uh 53 codeex. Yeah. On the medium or the high mode? By the way, people are like very like extra high and I'm like I mean I'm like, "Yeah, sure. if you if you like hate money and want to spend it all. But um yeah, it's been um that's been my prefer preferred choice. Uh but I also feel like I'm very much um somebody who isn't like a bandwagoner or like I'm a like oh you know I'm when when Claude Code was the hype like I was not like feeding into that. I'm like a bit more like let's let's kind of evaluate how much incrementally better this is. how much is it real? How much of it is it hype? And I just kind of keep it open, right? Um at the same time, if I find something I like at the moment, I'm also happy to keep it for the time being, right? So you you're a little that means you're loyal. You have some loyalty to the model. I I would say um I'm loyal to what's good. Well, that's great. Some people are just loyal to a brand even when the brand does not deserve it sometimes, right? Um I'm like look if you keep delivering and you um you know raise the bar um and there's clearly a lot of craft and thinking behind what you do I have no reasons to switch right like I think something else needs to be actually quite significantly better for me to like want to switch over right um so that's that's the way I see it uh in terms of what I use uh I am I would say a a cursor preferring person Um, I learned cursor uh in 2024, which was sort of the earlier days. Um, it's it's so weird to say I adopted cursor a little earlier because I feel like a bit of an impostor saying it cuz I wasn't like even a proper developer at the time, but I was like, you know, I just learned it because uh it was the main thing that was available for me to try out. There was no cloud code or any of that. Oh, that's what I Why does 2024 seem like such a long time ago? I think it was honestly, you know, there was there's no agent. So, when I learned cursor, this is May April, May 2024. Yeah. There's no agent mode. It was all in line. So, it was like traditional IDE. It was a straight straight IDE. Tab. Tab. Command K in line. Tab was the thing. Tab was a thing. I learned on tab. We thought tab was going to take over the world. Oh, tab. It was tab. And guess what? cursor didn't even have the ability to search the internet. So, guess who had to go on Stack Overflow and chat GPT to ask questions. Um, here we are. Here we are. Here we are. So, it was like the basic VS for it. It was um it it helped people 10x productivity within the paradigm of what coding was at the time. Yeah. Right. It it it really I think the first um step change was I think when sonnet uh 3.5 came out and then cursor launched composer which now it's just called agent that multi like that uh orchestration across multiple files like that was like groundbreaking um towards the end of 2024 like that was groundbreaking right um so it's it's been really really fun to evolve with the space um I think one of the reasons why I've continued to like cursor is that they really put so much emphasis on the user experience. Uh again, this is not to say like don't use other tools. I use other tools as well, but there's just something around the way they are able to like make it feel more intuitive to how a non-technical person would like, you know, use something that feels familiar, not like a CLI, right? Um, that's the part that I think um keeps me on cursor. Cool. Awesome. Awesome. So, Bay Area, Mountain View, San Francisco, now you're in Singapore. How we compare the two tech ecosystem? Yeah, they're very different. Um, I'd say Silicon Valley SF is uh obviously a lot more mature, right? Um, you've got, uh, you know, decades of history and now like you've got this like renaissance, right? um coming back where and not that the taxi never disappeared but I think there was a time like right around co it really was quite sad but yes it was very sad so um you know I I am from here so this feels like home so I come back here very often I saw you like two months ago I think um so it's it's great being here and like every day if I want to meet somebody who's building something cool I can right and um that's where I get a lot of inspiration when I'm here um I'd say single Singapore is the place where I feel the greatest motivation to build the ecosystem because I think it's a very very talented market that is underexposed. Um you have a lot of people who are like great builders. Um but they don't have necessarily a direct line to learn about what the companies are building here. um they're only just now starting to showcase and that's part of what I was trying to do as well with um you know the cursor hackathon and we did the Gemini hackathon to just like show how cracked the builders are here right or in Singapore not here but um yeah and and it's it's fun and um Singaporean builders are so funny as well I think there's this like false stereotype I think it's false that like you know Singaporeans are more like rule- abiding and like you know they're they're you know they they follow the rules they they kind of check the boxes but I've seen some of the most unhinged funny projects at the hackathons. I mean there are like there was these 17year-olds. They're like kids and they they built this app called F A K A uh U sorry F A K Y U AI. [ __ ] you AI which is an AI that roasts you. Oh, when you were slacking off. So, it's use this video and the two were just like in front of like judges from like Mattis and OpenAI were just like giggling like but it was really it was like a very like Gen Z take on like productivity which is like we don't want these boring productivity tools. You need an AI that just bullies and roasts you. And I was just cracking up, right? I mean there's a lot of these examples of like oh these these are some really creative kids, right? given like the right environment to really tinker and play around. Um I think there was a really cool convex project that came out of um our cursor hackathon as well. Uh so yeah it's I would say like that's why I my I feel like my heart belongs to two ecosystems and um and I think that's actually okay and that's that's a big part of who I am. That's great. That's great. So, um, when Mattis was acquired by Facebook, how did that like impact the Singapore tech ecosystem? And everybody was like, "Yay!" or everybody was like, "Oh, we got one." How was it? What was the vibe? Yeah. Well, I think it was definitely uh validating in some way cuz um you know there's there are a lot of companies that are built um outside of the Bay Area that have been doing cool stuff, right? but they don't always like I would say make it to maybe the center of conversation as much. Um which is why um you know when the whole acquisition was announced I think a lot of people on Twitter were like what's Manis like what the hell is this right? And I was like yo Mannis has been making my slides for any presentation I make for the last six months. I haven't touched like Google slides in so long or even gamma for that matter. Um, so, uh, it it was this like, uh, I would say it's one of a few different events that have happened recently that have made people really excited about Singapore as an ecosystem cuz it's it's this really interesting place where you've got this like balance between like east and the west in in a very neutral environment, right? Um, and you can actually be able to go to an event and see like amazing Chinese labs and uh companies that are born out of like Singapore, etc. Um, and also you've now got like companies from the valley like all in the room, right? And I think it it's very um it's very interesting that Singapore is starting to become sort of that place. And then um as a result, I think there's just more excitement in general for uh Singapore as this AI hub. Yeah. So that's great. Yeah, that's great. We're going to try to do a comet meet up. I mean, you're already teaching the course already, so try to do a com meet up in Singapore, so that'd be cool. Yeah. I mean, just you tell me like we do these things like what, one, two days in advance. Like we we'll we'll get you a crowd. Let's make it happen. We need to get you out of Singapore sometime as well. All right, Jamie. James going to Singapore. All right, cool. All right, cool. So, I have the saying. Well, it's not my saying. It's someone else's saying. It's about touching the grass, right? You got to you got to unplug. You got to got to like get out of the matrix, you know? Opus Gemini. They'll still be there. They still be there. So, what is your what is your unplug moment? How do you touch the grass? Yeah. Yeah. Um, well, I try to have unstructured time at night. M um this is by the way completely different than how I used to operate which was uh just work all the time and fill in the gaps. Um but now um I try I have an earlier schedule. Um each day I try to work I start work at like 7 a.m. Um and then I work till like 6 p.m. I work out. I try to do that every day if if not like four times a week. Um and uh yeah I just my evenings are whatever I want it to be. Um, so that's one way I unplug where I basically like I'm like, let me just do all my high leverage thinking in the day. Um, my brain's kind of tired by the end of the day because I think our cognitive work also um is faster. So we consume our energy faster, right? So like you know, you can be like uh you can train like LeBron James with your your mind. Okay, he doesn't you know train for 12 hours a day, right? So um that allows for me to have some time to unplug. Uh, in terms of hobbies, I mean, I try to keep interests, uh, outside of just like Twitter and tech and everything. So, um, I I love music, so uh, I DJ. I love house music. Well, wait a minute. Wait a minute. You DJ? Yeah. So, this could have been the DJ set. We could have party. Oh, yeah. DJ was like, I'm coming to visit. I'm still interviewing. You You said I bring the DJ. You We could have jammed. Yeah. I mean, if you have a I mean, I I don't have my deck with me. I know a friend. I have a friend. Yeah. Okay. All right. All right. Next time, next visit. Next time, next time party. Well, what do you what's your what's your jam? What's your music? Um, I play recently a lot of like uh indie house, sorry, indie dance. Uh, so some of the I guess um they call it Macabi House, but it's kind of a lot of the guys come from Tel Aviv. Um, I also like Tech House, so it's a bit more of the high energy, very bass heavy stuff. Um. Okay. Yeah. So, a lot of like I I would like um to call it like, you know, daytime, nighttime crossover. So, stuff that like feels like a 5:00 p.m. to 7:00 p.m. kind of set time. Okay. Okay. Do you remember day parties in tech where it was like the party start like 9 7 a.m. There was like a party. Oh, like the like the coffee uh matcha parties or the Yeah, it was something like that. It was like 7:00 a.m. like office there's a DJ set people party. All right. So when I come to Singapore Oh yeah, let's have a we can do a daytime party. Yeah, we'll we'll wake you up with the caffeine and like some some some beats. All right, cool. Well, Sherry, yeah, it's been great chatting with you. Thank you for building with ComX and um Yeah, we appreciate you for sure. Thanks for having me. All right.
When Sherry Jiang rebuilt Peek, her consumer AI fintech, she migrated from one century of backend thinking to the next. The old stack was Postgres on AWS with the usual middleware around it. The new stack is Convex, and the rebuild took a week and a half once the team cut what they didn't need.
This article is about why that migration made sense, what a real-time backend for consumer AI apps is, and why an AI-native product has to feel less like a database and more like a game engine. The conversation is drawn from Wayne Sutton's interview with Sherry on designing an AI tech stack for consumer apps in 2026.
Why modern backend means something different in 2026
A real-time backend for consumer AI apps is one where state changes in the database propagate to the client automatically, without a separate websocket layer, queue, or sync service stitched together by hand. The reactive model treats the UI as a live view over server state, so when an agent writes something, the user sees it immediately. That's the shape modern consumer AI products need, because the agent and the user are co-editing the same surface in real time.
Most teams default to Postgres with a queue, a websocket layer, a cache, and a vector store bolted on. That stacked-services pattern is a legacy of how engineering orgs used to be split, with backend specialists owning one half and frontend specialists owning the other. The lines between those roles have blurred over the last few years, so the tooling that assumes a sharp split between them is starting to feel out of step. Sherry calls Convex "a very modern solution" because it collapses those layers into something a single developer, or a single agent, can hold in their head.
Sherry Jiang describing Convex as a very modern solution
The shift matters because the kind of app being built has changed. A CRUD dashboard from 2018 could survive on a relay of services because users tolerated a refresh button and the occasional half-second pause. A consumer AI app can't because:
The agent is writing to state continuously
The user is reading from state continuously
The user feels any latency between those two things as awkwardness rather than as engineering.
The architecture has to reflect that, which means the database, the sync layer, and the client transport need to share a model rather than translate between three of them.
There's a second reason the stack matters more now than it used to. The agent writing the code is also a stakeholder in the codebase. If the backend is spread across four services with four contracts, the agent has four times the surface area to misunderstand. If the backend is a single coherent system with one language and one schema definition, the agent reads it the same way a new engineer would and writes correct code more of the time.
The old Postgres world feels like passing notes back and forth
Sherry borrowed a metaphor she'd seen on Twitter to describe the legacy pattern. Three or four layers sitting between the database and the client means every state change is a relay race of notes being passed back and forth. Every handoff adds latency and demands its own contract, which is just another place for a bug to live. For a CRUD app that's annoying. For an AI product where the agent is updating state continuously, it's a structural problem.
The note-passing metaphor lands because it's literally what's happening at the wire:
The database notifies a sync layer
The sync layer publishes to a queue
The queue feeds a websocket gateway
The gateway pushes to the client
The client's cache is reconciled against the new payload.
Every one of those hops is a place where a developer has written code to translate between formats, retry on failure, and decide what counts as "fresh." The reactive model collapses the whole relay into a single subscription contract, so the developer writes the query once and the framework handles the rest.
What game-engine responsiveness means for an AI app
Consumer apps that work, the ones people open out of habit, manufacture joy. TikTok's video swap is instant. Duolingo throws confetti. Reward points tick up the moment you earn them. Sherry frames this as game-engine responsiveness, where the system has to give the user a visible reaction the instant they act. An AI product that pauses for half a second after a vulnerable disclosure breaks the spell the same way a real friend would if they paused awkwardly. The interaction collapses, and the user stops sharing.
Game engines solve this problem with tight render loops and shared memory between the simulation and the display. Consumer AI apps need the same property at a higher level of abstraction, which is what a reactive backend provides. The model is that the UI is always rendering the latest version of server state, and the server is always pushing new state as soon as it changes. There's no polling, refresh, or waiting for the next chat response to surface something written three seconds ago. The result feels less like a web app and more like a piece of software running locally on the user's machine.
What Peek is and why behavioral AI needs real-time UX
Peek, a consumer AI fintech, is building behavioral intelligence around the way people spend money. Sherry spent a decade in consumer fintech before founding it, including work on Google Payments in India, where her team ran behavioral-science experiments around variable rewards. Scratch cards with 35 to 65 percent win rates were one of the levers they used to make saving and spending feel like a game rather than a chore.
That background is the reason Peek behaves more like an attentive friend than a dashboard. It talks to you, learns what you care about, and reflects it back so you can see yourself more clearly. The product class demands real-time state because the emotional contract with the user depends on it.
Sherry Jiang on earning your first $100,000 through behavioral consumer finance
The variable-rewards work matters as context because it shaped how Sherry thinks about engagement. A money app that hands you a number and expects you to feel bad about it will always lose to one that turns the same data into a moment of recognition or surprise. Peek inherits that thesis, which means the surface has to react to the user the way a thoughtful friend would, and the underlying state has to support that reaction without lag.
Building a scaffolded identity through conversation
Peek's internal term for the things it remembers about a user is artifacts. You tell Peek something about how you think about money, Peek extracts an artifact, and the artifact appears in your interface immediately. Seeing the trace appear is what makes you willing to share the next thing. The loop is conversational on the surface and reactive underneath, so the backend has to push the new artifact to the UI the moment it's written.
The reason that loop works is the same reason note-taking apps with live previews feel different from note-taking apps with save buttons. The user is forming a model of what the system knows about them in real time, which is only possible if the system shows its work as it happens. If the artifact takes two seconds to appear, the user has already moved on, and the chance to anchor the next disclosure is gone. The reactive backend is what makes the difference. Without it, the identity never feels like a real conversation, just a form being filled out.
Sherry's analogy lands hard. If you share something vulnerable with a friend and they pause for a beat too long before responding, you feel the pause and don't share the next thing. An AI agent operates under the same social physics. A two-second wait between a user message and a visible response damages the relationship, not just the performance metrics. Real-time is what makes this product class possible in the first place.
This is where the comparison to B2B AI sharpens. A sales-ops copilot that takes three seconds to summarize a deal is fine, because the user is in a working posture and treats the delay as compute time. A consumer agent that takes three seconds to acknowledge a disclosure about debt is broken, because the user is in a vulnerable posture and treats the delay as social withdrawal. The same engineering latency carries different meaning depending on the emotional register of the surface, which is why consumer AI can't borrow B2B's tolerance for round-trip time.
How Peek migrated from Postgres and AWS to Convex in a week and a half
Peek rebuilt its backend in roughly a week and a half. The precondition was a hard look at the existing surface area and an honest cut of what the new product no longer needed. Behavioral artifacts and the memory system were built fresh on Convex, whereas categorization and Plaid ingestion were ported across directly. The whole sequence was designed to protect the integrity of users' financial data while the underlying system changed beneath it.
A week and a half is fast enough that it's worth being honest about why. The team wasn't translating ten years of accumulated business logic. They were cutting what wasn't core, rebuilding what was reactive, and porting the parts that were closer to standard pipeline work. The number isn't a benchmark anyone else should expect to hit without doing the same pruning first.
Step one cut what you don't need
The previous version of Peek had net-worth tracking and balance features that weren't core to the spending-and-behavior product Sherry was building next. Migrating dead surface area is the moving-house mistake, where you pack the boxes you should have thrown out and pay to move them. Cutting first is good practice independent of which backend you're moving to. It just happens to be especially valuable when the target is a reactive system where every table you keep is a table you'll model intentionally.
The cut went deeper than features. It also touched the implicit assumptions baked into the old schema, since each one of those would have had to be re-justified in the new system. Net-worth tracking, for example, implied a particular shape of account snapshots and a particular cadence of refresh that didn't match how the new product would think about user state. Carrying that shape forward would have meant the new schema inherited the previous product's mental model, which is the opposite of what a rebuild is for.
Step two rebuild what's actually real-time
Once the surface was pruned, the truly reactive parts of Peek, the artifact extraction and the memory system, were rebuilt rather than translated. Reactive state is something you design for from the schema up, so a lift-and-shift wouldn't have produced the right shape. Categorization logic and the Plaid integration moved over more directly because they're closer to traditional ingestion work. The split was deliberate, with the live-feeling parts of the product getting native treatment and the pipeline parts getting a clean port.
The distinction between rebuild and port is worth slowing down on, because it generalizes. Anything that the user is supposed to feel in real time should be designed for the reactive model from scratch. Anything that runs on a schedule, processes a feed, or pulls from a third-party API can usually be moved with minimal change, since the latency profile of that work isn't user-facing in the same way. Drawing that line through your codebase before the migration starts is what keeps the timeline honest.
Protecting data integrity in fintech
Peek deals in real money, so a categorization error that shows a user $4,000 in spending instead of $400 is a trust event, not a bug. Sherry sequenced the migration so the spend-accuracy guarantees stayed intact end to end, with the new system reconciling against the old before customer-facing surfaces switched over. The same logic applies in any domain where the data itself carries trust. The migration plan has to treat correctness as a precondition rather than something to verify after launch.
The old system kept running while the new system was filled with the same inputs and asked to produce the same outputs, with any divergence flagged for inspection. The new system only got to drive the UI once both sides showed the same number for the user's last six months of spending. That sequencing turns a risky cutover into a series of small checks, each one able to fail loudly without the user ever seeing a wrong number.
Why Convex works for vibe coders and 13-year-olds alike
An LLM-friendly backend matters because the AI agent writing your code is now a coworker, and coworkers need a codebase they can navigate. Sherry has taught over a thousand people to build apps zero-to-one, and her bootcamp uses Convex because newcomers ask the right questions out loud. They want to know why the database lives in a different place than the code that queries it, and they haven't yet absorbed the historical reasons that became normal. The answer, when you're on Convex, is that it doesn't.
The bootcamp signal is more interesting than it first sounds. Beginners don't have priors about how backends are supposed to be structured, so their confusion tracks genuine cognitive load rather than habit. When a thousand of them keep asking the same questions about a stack, you've learned something about the stack. When a different stack stops generating those questions, you've learned something about that one too.
Sherry Jiang on the best growth levers in the product
LLM-readable backends are a design choice
When the schema is defined in TypeScript next to the functions that read and write it, an agent reading the repo has the whole picture in one place, with no separate console, migration tool on a different surface, or out-of-band SQL to reconcile against the application code. You can see exactly this design principle in how Convex's query engine works, where the absence of certain SQL affordances is a deliberate choice in service of a single coherent model.
The single-language property compounds. An agent that already understands the TypeScript in your queries also understands it in your actions, your schema, and your client. There's no moment where the agent has to switch dialects, learn a new ORM, or hold two type systems in its head at once. That property is what lets the agent generate code that compiles and runs the first time more often than not, since the surface it's writing against is the same shape everywhere it looks.
Removing the cognitive overload of middleware
A beginner, or an agent acting like one, gets overwhelmed by middleware before they get overwhelmed by logic. Convex skips the separate MCP wiring, the SQL console in a different tab, and the queue debugger that lives outside the editor. Sherry's youngest student is 13 years old and shipping apps in 48 hours. That's only possible when the surface area you have to hold in working memory matches the surface area of the problem you're solving.
The middleware tax is invisible to engineers who've already paid it. If you've spent five years learning to keep a queue, a cache, a websocket layer, and a SQL console in your head simultaneously, you don't notice that you're doing it anymore. A 13-year-old does, because they haven't built the muscle yet. They ship working apps in 48 hours because so much of the standard backend stack turns out to be incidental complexity rather than essential, not because they're unusually precocious. Removing that incidental layer makes the same problem solvable by people who would have bounced off the older stack entirely.
Sherry's AI tech stack with Cursor, Codex, and a loyalty to what's good
Sherry has used Cursor since April or May of 2024, back before agent mode and before in-editor internet search. She stayed with it because every iteration kept being incrementally better than the alternative she'd test against. Her current model preference is GPT-5 Codex on medium for most of her work. The tools matter less than the evaluation discipline. For every new model or AI tool, she asks how much of the launch is real and how much is hype, and only switches when there's enough real there to outweigh the hype. The same loyalty applies to Convex. She's stayed because it keeps delivering.
The evaluation discipline is more transferable than the specific tool choices, because the choices themselves will be different in six months. The question Sherry asks of any new release is whether it's better than what she's already using on the work she's doing, not whether the demo is impressive or the changelog is long. That filter is what keeps her from cycling through tools every time a new one trends, and it's what makes the loyalty meaningful when it does land. That kind of loyalty only counts when the tool keeps clearing the bar, otherwise it's just inertia dressed up as conviction.
The same logic applies to backends. The traditional relational stack earned its place over twenty years of CRUD apps, and for a lot of products it still clears the bar. The question for any team building consumer AI in 2026 is whether their AI tech stack is serving the work they're doing or working against the product they're trying to ship. Sherry's read is that for an agent-driven, behaviorally-rich, real-time consumer surface, the older stack stops clearing the bar, and the reactive model starts to.
Where to start if you're building consumer AI
If you're building a consumer AI product where the UI needs to reflect agent state changes in real time, model your AI tech stack around a reactive backend first and add inference around it, rather than the other way around. Peek's migration worked because the team cut surface area before they moved, treated the live parts of the product as a fresh design problem, and protected data integrity through the cutover. The same playbook works for any consumer AI app where joy and immediacy are part of the contract. You can spin up a working backend in about ten minutes and feel the difference yourself, then explore more Stack articles on how teams are shipping with reactive infrastructure. If you want to compare notes with other builders, the Convex Discord is where most of the consumer AI conversations are happening.
The broader pattern worth carrying away from Peek's story is that the constraints of the product shape the constraints of the stack, not the other way around. A behavioral AI fintech where the user's willingness to disclose depends on real-time reflection can't be built on a backend that treats latency as an engineering metric, and a consumer surface that needs to feel like a game can't be built on a stack that thinks of itself as a database with adapters. Choosing the right foundation is the choice that makes the rest of the product possible, which is why Sherry treated the migration as foundational rather than as plumbing.
The other lesson is about who the codebase is written for now. Five years ago, the answer was the team of engineers maintaining it. Today, the answer includes the agents writing alongside them, which means the codebase has to be legible to a reader who doesn't have the history. A backend where the schema, the queries, and the client are written in the same language and live next to each other is a backend an agent can navigate. A backend stitched together from four services with four contracts is a backend that even a careful human gets lost in. The shape of the team has changed, so the shape of the code should change with it.
Frequently asked questions about real-time backends for consumer AI
Q: What is a real-time backend for consumer AI apps? A: It's a backend where state changes in the database propagate to the client automatically, without a separate websocket, queue, or sync service stitched on. Convex is built around this reactive model, so when an agent or a user writes data, the UI updates immediately.
Q: How long does it take to migrate from Postgres to Convex? A: Peek completed the migration in about a week and a half after pruning surface area that wasn't core to the new product. The migration window depends heavily on how much you cut first and how much of the existing system was reactive versus traditional CRUD.
Q: Why does consumer AI need real-time more than B2B AI? A: Consumer products run on dopamine, and the emotional contract with the user collapses if the interface pauses after the user shares something. The same way a friend pausing awkwardly after a vulnerable disclosure ends the conversation: a half-second delay in an AI agent breaks the loop that makes users want to keep engaging.
Q: What makes a backend LLM-friendly? A: Schema-in-code, a single language across the stack, no out-of-band consoles to context-switch into, and a file structure the agent can navigate end to end. When the database definition lives next to the functions that read it, the agent reads the whole picture in one pass and writes correct code on the first try more often.
Q: Is Convex good for beginners or non-developers? A: Yes. Sherry Jiang's bootcamp has taught over a thousand students to ship apps zero-to-one on Convex, including a 13-year-old who built a working app in 48 hours. The lower middleware overhead means new builders spend their attention on product logic instead of on plumbing.
Q: What's the right stack for a consumer AI app in 2026? A: A reactive backend like Convex, an AI-native IDE like Cursor, and a current frontier model such as GPT-5 Codex. Evaluate new tools by whether they're incrementally better than what you're already using, not by whether they're trending, and stay loyal to the pieces that keep delivering.
Q: How should I sequence a backend migration for a fintech product? A: Start by cutting features that aren't core to the next version of the product, since migrating dead surface area is the most common mistake. Rebuild the reactive parts of the product from the schema up rather than translating them, and port the pipeline-style work directly. Run the old and new systems side by side and reconcile their outputs before switching customer-facing surfaces, so any divergence in financial data is caught before a user sees it.
Q: What does game-engine responsiveness mean for a consumer AI product? A: It means the user gets a visible reaction the instant they act, the same way a game responds to a button press. For an AI agent, that translates to the UI reflecting new state the moment the agent writes it, with no polling, no refresh, and no perceptible delay. The standard is whether the interaction feels emotionally continuous to the user, not whether engineering latency hits a target.
Build in minutes, scale forever.
Convex is the backend platform with everything you need to build your full-stack AI project. Cloud functions, a database, file storage, scheduling, workflow, vector search, and realtime updates fit together seamlessly.