EVERY SPOKEN WORD
30 min read · 5,888 words- 0:00 – 0:02
Intro
- SPSpeaker
[upbeat music]
- 0:02 – 1:06
Panel setup and why investors are coding again
- SPSpeaker
Please welcome to the stage co-founder and CEO of Clay, Varun Anand; co-founder and CTO of Emergent, Madhav Jha; co-founder and CTO of CFO Sylvia ProCap Financial, Shane Noor; and partner at Sequoia Capital, Lauren Reader. [audience cheering]
- SPSpeaker
All right. Hello, everyone. Thank you so much for joining us. Um, I am delighted to have this wonderful panel to talk about what it's like to build with AI. Um, I am one of the partners at Sequoia. My name is Lauren. I was a former engineer. I wish that I had Claude code back when I was writing a lot more, uh, software, but now I can do it again, which is great. Um-
- SPSpeaker
How much, how much code are you writing right now at Sequoia?
- SPSpeaker
Every day, actually. [laughs]
- SPSpeaker
Okay. Well, what are we building? Let's get some, let's get some investing alpha out the door, you know?
- SPSpeaker
Okay. So a lot of our job is how do you find the right founders? How do you figure out what are the right ideas? And so we write little scripts, little products, you could say, to help us with that.
- SPSpeaker
That's pretty cool.
- 1:06 – 2:44
What each company is building: personal CFO, app-building platform, and GTM AI
- SPSpeaker
But we're gonna dive into what you guys are doing next. So maybe we could start with a round of intros from you, and then I'll kick off the panel. Shane, do you wanna start?
- SPSpeaker
Sure. Um, hi, everyone. I'm Shane Noor. I'm the, uh, CTO of ProCap Financial, founder of Sylvia. Uh, we're building, uh, an AI personal CFO. Um, you can kind of think of it as connecting all your bank accounts, your crypto, your real estate, all that stuff, all in one view. Uh, Americans on average spend, you know, less than two minutes, uh, a day on their finances, and we think that, uh, there needs to be something else out there to, uh, to solve that.
- SPSpeaker
Cool. Hey, uh, I'm Madhav. I'm the co-founder and CTO of Emergent. Uh, Emergent is an app-building platform which lets people build, uh, production-ready apps. Uh, we launched about nine months back, and, uh, we are at, like, a hundred million in run rate revenue. Uh, like eight, eight and a half million people have used our platform, uh, ten million apps have been built. And, uh, the main sort of distinct-distinction factor, uh, between other app-building platforms is that, uh, the app that comes out of our platform are more production-ready, ready to ship to your end users. So the kind of people who use our platforms are small business owners, and they, uh, they're, they're building these apps for more business-critical needs. Uh, and you can build a full stack mobile app, uh, and it's, uh, uh, you know, just like thanks to Claude, we've been scaling pretty, pretty, pretty well. Uh, and, uh, yeah, excited to be here.
- SPSpeaker
Thanks.
- SPSpeaker
Great. Um, hi, guys. My name is Varun. I'm the co-founder and CEO of, of Clay. We're a go-to-market AI product, and we, we help people, um, find their best customers, help them grow them, and help them find more like them. And, uh, we were talking backstage and, and the energy we wanna bring to this is, um, uh, our, our top objective is to have fun. And so that's our, uh, that's the energy we're bringing into this conversation.
- SPSpeaker
[laughs]
- 2:44 – 4:13
Capabilities unlocked by modern models: enrichment → action
- SPSpeaker
Thank you, Varun. Um, yeah, so we're gonna have fun talking about what it's like to build today. I think both what kind of products we can build has changed dramatically and is still changing dramatically, and then how we build products is going over the same, uh, transformation. Um, and so I would love to hear maybe from the three of you, tell us more about how you've built what you've built today and what you see in the models that have unlocked new capabilities you could never do before if we're thinking six months ago or two years ago.
- SPSpeaker
Um, yeah. So I, I think like, you know, uh, Clay originally started with the intent of how do you bring the power of programming to orders of magnitude of more people? Somehow that became a, a data enrichment product at, uh, like four years ago. And then, um, actually the product really was able to expand, um, due to, um, APIs from Anthropic. And so, um, you know, uh, the initial thing we saw that was pretty cool was like, okay, people were able to take these APIs, scrape the web, and get very custom, unique data points that are relevant to their business, right? So maybe I'm selling security compliance software, and I can go scrape someone's website and figure out if they're SOC 2 compliant or not, right? That's, that's kinda basic. But then what, uh, what LMS allowed was actually how do you take that data point and do something with it, right? So how do you take that and actually send someone, uh, an email or send someone a very customized, personalized video or something that actually helps you target someone at the right moment, at the right time when they need your product and you need them. Um, and I think what's happening now that's way cooler is people's-- you're a-actually able to do things that, that are, that are never before, like, seen.
- 4:13 – 5:10
Creative AI-native GTM in the wild: the Waste Management dumpster-color story
- SPSpeaker
You know? That's like, that helps, helps you do the impossible. And so I was, I was sharing like a fun example last night at dinner where, you know, uh, uh, this is kind of an out there example, but, you know, Waste Management is like a trash services company, and they're one of our customers. And, and they use Clay actually to look at Google Street View images of people's homes and their businesses, and they use Anthropic's, um, APIs to actually an- use image an-analysis and analyze the Google satellite views to look at the color of the dumpster outside people's businesses. And the dumpster color indicates whether... what, what competitive trash service they might be using. And if they're not using Waste Management, they use Clay to create like an automated direct mail campaign to convert the business into a Waste Management customer. And that's like very creative and helps people take their human creativity and bring it into practice, and it's just wouldn't have been possible before AI.
- SPSpeaker
That's an amazing story.
- SPSpeaker
Yeah.
- SPSpeaker
I had no idea Waste Management was using-
- SPSpeaker
That's the cutting edge-
- SPSpeaker
[laughs]
- SPSpeaker
... of go-to-market Waste Management, Lauren.
- 5:10 – 7:43
Empowering non-technical builders: production-ready apps and workflow discipline
- SPSpeaker
No, I mean, love this example, and I think like that's, that's exactly what we are seeing in... at Emergent. Like, these, uh, small business owners and who you would think that they're not as sophisticated, but given the right set of tools, they're able to like, you know, outshine your best developers as well at some, at some point. Like, I've seen examples of people, uh, you know, in-- Like, I have an example of a person who is based out of Norway, and he built a real estate sort of management platform similar to Airbnb. And, uh, it's, it's like much more sophisticated than Airbnb because he managed to put, uh, agent into it so you can interact with it while, while you are doing it. And they have learned the ropes of the game where they know, okay, you need to break things down into, you know, small workflows. Only then the hallucination can be controlled, and all of that without having any coding background. So like that, that kind of empowerment that you see is pretty incredible to see once you give them the right set of tools.
- SPSpeaker
Yeah, we're seeing a lot of products built for non-tech builders. And you do too, Sha-Shane. Like, you're working with people who maybe wouldn't know about Claude otherwise.
- SPSpeaker
Exactly, yeah, yeah. Most of our users are definitely high net worth individuals, you know, thirty-five plus, and so most of them obviously know about ChatGPT. They somewhat use Claude and all that stuff. But, uh, you kind of have to handhold them throughout this entire experience. And so the biggest insight that we had, you know, very early on was that don't assume that the independent investor is, is dumb. Uh, assume that they are smart and that they can make the decisions, uh, themselves. And so, you know, better tool calling reliability, code execution environment, uh, bigger context windows, uh, all that stuff played a, played a factor for us.
- SPSpeaker
Yes. And Varun, tell us a little about like some of you-- you guys have each built new products in the last six months with these new models. Tell us what are some of the new things you guys have built that have been enabled today?
- SPSpeaker
Yeah, yeah. I mean, I think what's, what's cool is like actually how can you build on top of these models to have agents that do work on your behalf, right? So it's like, okay, how do you use, um, agents to, to not only do web research and take action on that, but maybe actually account agents that go into your, um, uh, CRM. And, and by the way, if you have a CRM, my God, it's a mess. Every company, uh, has a CRM, and it's a total disaster. And like you have sales reps, they, they don't fill out anything. And so you go in, and then you're like, okay, an agent can be like, okay, why are none of the, the closed lost reasons filled? Okay. They never are, by the way. And, um, but now you can use, um, like AI and Clay and Anthropic's, uh, uh, APIs to actually go look at gong call recordings and fill all those in and then take that information and be like, "Okay, this is why it was closed lost seven months ago. These factors have changed, and let's take an action autonomously to go get that customer back." Right? And so these are, um, examples of things that, uh, agents are, are-- can do autonomously now that can kind of independently try to drive revenue for you.
- 7:43 – 9:07
Ambient agents and the token-burn lesson: designing proactivity in consumer finance
- SPSpeaker
Yeah. And then, Shane, you guys are doing some proactive work too on behalf of your customers with agents.
- SPSpeaker
Yeah, that was another huge unlock, uh, pretty early on. I think, uh, especially with consumer apps, uh, proactivity is probably key. Um, you realize that consumers probably don't want to continuously log on every day, especially seeing a dashboard with their net worth that's already, you know, been done. They want these sort of like ambient agents, is kind of what I call them, um, to run in the background for you and then reach out to the user whenever there's an insight, whenever there's an alert on your transactions, you know, w-whatever that the user wants to set up. Um, specifically, it's, uh, you know, semantic crons is what we were doing in the very beginning.
- SPSpeaker
Tell us more about how you build those ambient agents. So-- and then not blowing up your token spend while having an agent that's always running in the background.
- SPSpeaker
[chuckles]
- SPSpeaker
Like, that's not an easy thing to do.
- SPSpeaker
Hu-hundred percent. I think, uh, well, we, we made a big mistake early on, which was assuming what the user would want as their semantic crons. And so we had, uh, what you would call like a radar, but it was running twenty-four/seven on a user, and it blew through token usage for us. Um, you know, I think we've publicly said it was more than half a million in a month, uh, of Anthropic spend. So, you know, I'm sure Anthropic's happy. Um, [laughs]
- SPSpeaker
[laughs]
- SPSpeaker
But, uh, we slowly started to transition that and say, "Hey, how about we look at what users are doing and give them the capabilities to do everything manually first?" So give them the tools, and then you can sort of learn off of, you know, what they're doing and then build on what should be predefined and, and default, uh, to the user when they're onboarded.
- 9:07 – 11:00
Building agent infra: autonomy + rapid feedback loops over heavy scaffolding
- SPSpeaker
Yeah, interesting. Madhav, I'm curious how your architecture's changed over time too when you're building whole applications. Has that evolved, changed? Was it very expensive at the beginning? Has it gotten better now?
- SPSpeaker
Yeah. I mean, I think we have like, uh, you know, riding the trajectory for model improvements themselves. You know, models have become pretty good at like doing things autonomously, long-running tasks, right? And so I think one of the early decisions we took was that we did not try to like hard code or like overengineer things around the model. Like, let the model be in charge of like doing the work and let the-- uh, sort of give the autonomy to the agent itself, right? And you enable, uh, sort of like rapid feedback to the agent. Uh, so, and, and your o-agent is only as good as the feedback that it gets. So one of the things that we did early on was we built our own infra from the ground up. So o-our sandbox container technology is built from the ground up. The agent is built from the ground up. And, and all of that is done in the interest of like give-giving rapid feedback to the agent, right? So when, when agent is building some prototype or, or backend, it needs to get all the logs from the database. It needs to, you know, like, be able to look through it. And so we kind of like thought, "Hey, what would it be if you-- if-- how would a human developer do it," right? The human developer would run, uh, you know, a lot of developers in the cloud would run like a front end and a back end and on various ports, look at the database. So basically, we sort of like said, "Hey, what would an agent do if, if they were like given a laptop in the cloud," right? And so that kind of enabled us to let the agent drive most of it. And, and then it almost becomes easier when a new model comes in because you're not over tied to your, your own architecture. You're basically letting the agent do the job, and, uh, it would do basically just as well if, if a new model comes in.
- SPSpeaker
Yeah. We've seen a lot of our companies take off a lot of the scaffolding that they built-
- SPSpeaker
Yeah
- SPSpeaker
... to supplement the models. Now that the models get better, they can do so much more, but then you have to push the product farther.
- SPSpeaker
Yes.
- SPSpeaker
You have to go proactive. You have to go deeper into the customer context and do the integrations. Um, and like it's really just keep running.
- SPSpeaker
Absolutely.
- 11:00 – 14:31
Pricing and packaging in AI: separating cost coverage from value capture
- SPSpeaker
Um, one of the questions we get a lot from our portfolio companies is how to think about pricing, packaging, and how do you drive towards outcomes, not just price with tokens plus a small margin. And I know Clay has done a really nice job with this. I'm curious to tell the pricing story behind your company, Varun.
- SPSpeaker
Yeah, yeah, we can talk about it. I think, um, you know, a lot of pricing-- basically a lot of pricing until SaaS companies was like how you price this bottle of water, right? And so how do I-- if I were to price this bottle of water, what would I do? I'd figure out how much does glass cost. It feels actually quite expensive.
- SPSpeaker
[laughs]
- SPSpeaker
And, um, uh, and how much does the little cap cost? And then I would charge more, right? It's called cost plus pricing. Um, and when software came around, um, people were like, "Wait, we can actually charge more money." We could, um, we could have like value-based pricing where we're like, um, let's, let's charge by seats and, and seats will kind of simulate the value we are, we are extracting from this customer, and so let's charge on that. And then when AI came around, people like-- people actually went back to the bottle pricing, right? Because they got all, all the, the AI customer, they're like, "Okay, let's do cost plus again. Let's just cover our costs." And then over the last couple years, people are feeling this push to get to outcomes-based pricing. And I think there are a couple of industries, namely like support, for example, where that's maybe more possible, but it's extremely hard and very challenging to execute. And so, um, what we just did and, and what I actually think more companies will probably start to do is separate these things. And so, um, we kind of bifurcated the pricing. Now, bifurcating pricing going from one metric to two metric, no pricing consultant will tell you that's a good idea because it's more complicated. And, and the... Like, one of the cardinal rules of pricing is to keep it simple, because it is, uh, like the more simple it is, the better it is for customers. Um, but we chose to make it more complicated, uh, and we think that generally pricing is becoming more complicated overall for, uh, and, and so that's just becoming increasingly a problem. And so we chose to make it more complicated, but we're covering our costs on one side. So we have one type of pricing that's just kind of low margin. We're not trying to make that much money on it, and it's trying to cover our costs. And then there's another metric that's trying to represent the value that we're creating, and that's more, um, trying to tie to the value that we're creating for customers, and that's high margin. And so that's kind of actually what we've tried to do, and we're starting to have early success and what we think a lot of other companies might do as well, uh, because outcomes is too hard to get to in the short term.
- SPSpeaker
How do the two of you think about it?
- SPSpeaker
Yeah, I mean, for us, I would say there is a similar sort of a style because, uh, we also allow people to deploy apps. Once you have built the app, you also deploy it. And so we see that there is a build phase where, uh, there is a token... With, with... Your build kind of like mimics your token ex-expenditure, like how much tokens do you spend. Uh, but there we have taken the, uh, you know, effort to make sure that you don't waste tokens. You know, what we have seen generally in the customer support tickets that we get is that if the agent does something and takes a lot of tokens and does a good job, nobody complains. But if the agent does something and it screws things up, people are up in the arms, you know, like, "I want my refund," right? Like, so I think what we, what we have seen is there, there's also like a psychology at play, uh, here. And, uh, I don't think anybody has e- you know, sort of like completely figured this out yet. Uh, obviously, like the build phase would be token heavy, and it would, it would have to be, you know, spent in terms of token. And so like we price it in terms of tokens usage and all, all that. The deployment side is, is more around like SaaS revenue where you deploy and like you offer services around deploy- deployment side and, uh-
- SPSpeaker
So you also do the value-based pricing and some of the cost-based pricing-
- SPSpeaker
Yes
- SPSpeaker
... just to get over the hump with the large token budget.
- SPSpeaker
Yes. Yes.
- SPSpeaker
Yeah.
- SPSpeaker
Absolutely.
- SPSpeaker
Oh, very cool.
- 14:31 – 14:51
Service is software: positioning, free tiers, and services around AI products
- SPSpeaker
Um, for us, we are a free product.
- SPSpeaker
Okay.
- SPSpeaker
Okay. [laughs]
- SPSpeaker
Well then... [laughs]
- SPSpeaker
[laughs] But I think the, uh, the new norm is definitely gonna be, uh, offering services, uh, around, uh, your product. Um, I think Jensen recently said that, uh, for the first time in history, service is software and software is service. Um, and I think that's gonna be a, a big trend.
- 14:51 – 16:56
What’s surprised them: smarter models can be cheaper, but trust/brand standards persist
- SPSpeaker
Mm-hmm. I'm curious if we dive back to some of how people are building and using models. What are the things that have surprised you from model behavior or how you've built with it in the last six months?
- SPSpeaker
What has surprised us?
- SPSpeaker
Yeah.
- SPSpeaker
Uh, I, I, I think like the, the, the... What's a surprise to me is the... is even, uh, you know, somebody who is so embedded in AI is also pr- you know, pleasantly surprised when a new model comes in, you know, what kind of capabilities it, it unlocks. It's almost like you always underestimate what the new model is, is going to be capable of. You always think, "Okay, yes, it's gonna be follow this trend. It's gonna be doing this, that." And then you see this new model and then suddenly... Because even the price point, sometimes you think, "Oh, this is model is smarter. Is it gonna be more expensive?" But it turns out that a smarter model can be actually cheaper, uh, because it doesn't waste too many cycles, right? Like doing, uh, stupid things, right? And so some of that plays out, you know, where you think that, "Hey, smarter model is gonna be more expensive," but it turns out to be cheaper and all of that like... And then you can just pass on all of that, uh, to your customers. So that has been surprising.
- SPSpeaker
Okay. I think, I think, uh, something that surprises us is, um... I mean, uh, what, what you said is definitely true, and so I think we're definitely feeling that. I think, um, how, um, reliable the prompts have been is, have, have like surprised us. And also, you know, I think one thing that we're starting to watch is in our market, we're in the go-to-market space, right? And so we... A lot of... Uh, one common use case is using Clay to automate highly targeted outbound campaigns to, to customers. And so one surprise and challenge is like dealing with, um, enterprise customers who have like high standards for their brand and their voice and what they're saying to customers. And actually, that level of scrutiny hasn't changed at all, uh, despite the, the, the, the models', uh, substantial improvement over the last three years. Um, and so people are still doing things with the same as, as human review that they were doing two and a half years ago. And so, um, that's surprising, and I'm, I'm, I'm interested to see like how, um, you know, how that changes as the models keep improving and, and, and, and practices change on the go-to-market side.
- SPSpeaker
Mm-hmm.
- 16:56 – 17:51
Memory and file systems: personalization as a major product shift
- SPSpeaker
Yeah. The most surprising thing for us, at least, uh, recently, is definitely the, uh, concept of this whole file system, uh, memory. Um, like you guys probably saw yesterday, there was, uh, Claude's, uh, auto memory. I know GPT and, um, others are also adding this, uh, new concept of file system. It's kind of like Karpathy's, you know, wiki or creating your personal wiki. Um, we integrated that, uh, maybe like a month ago, and now, you know, Silvia can fully manage your own file system for you as you're chatting. And obviously, our users don't necessarily, you know, know what, what that exactly means, but that just means that there's hyper-personalization, right? Whether it be skills, memory, preferences, all that stuff being stored for you and managed by, by Silvia. Um, so much so that I think that it's definitely gonna be the trend for, for 2026, and we created a, a new product specifically for developers to go, go and use it called, uh, Trove, Trove Files. And, uh, you know, in three lines of code, you can get your own, uh, spun up, you know, fully managed file system for you.
- 17:51 – 20:34
Product affordances + adaptability: building evals and rebuilding constantly
- SPSpeaker
Very cool. So we talked about surprises. I'm curious about what you have on your wishlist.
- SPSpeaker
Well, I think just building off his point right there on, on memory, I'm very excited to see-
- SPSpeaker
Yeah
- SPSpeaker
... like that continue to improve. I think especially for us, right, like the reason that's relevant is because, um- You know, we want to like learn from what y- things that you do that work to get you new customers and things that don't work and create like recursive loops that help you get better and better at selling your product and help you get better and better and faster at, at getting new customers. And so we're, I think, really excited about how that like advances, um, to make it easier and easier for customers to experiment quickly and come up with new ideas.
- SPSpeaker
Yeah. I really like the concept when thinking about products and AI of affordance. So how do you build a product that's purpose-built for a given task?
- SPSpeaker
Yeah.
- SPSpeaker
And I think that's something you guys do really well at Clay is being purpose-built to go find the next customer to run whatever-
- SPSpeaker
Yeah
- SPSpeaker
... campaign, where instead of sitting on a blank page, you give everyone the tools-
- SPSpeaker
Yeah
- SPSpeaker
... to help them be creative.
- SPSpeaker
I think it's like a balance of like how do you be purpose-built but also be flexible enough-
- SPSpeaker
Yeah
- SPSpeaker
... where you have, um, where you are mostly giving people underlying primitives that they can do for whatever thing and, uh, but, but it's purpose-built, but it's flexible enough that as models improve whatever, um, that, that it can kind of mold to that particular situation.
- SPSpeaker
Uh, uh, adaptability is huge.
- SPSpeaker
Yeah.
- SPSpeaker
Uh, we've re-architected, uh, plenty of times.
- SPSpeaker
How many?
- SPSpeaker
I think, uh, uh, four or five times, and that was just for-
- SPSpeaker
In the last year.
- SPSpeaker
That's just for retrieval o- of user data. I mean, we went from hard-coded tools to, um, you know, obviously integrating with skills and now this whole file system with, with POSIX as well. Um, I think there's a... Y- you have to be adaptable. Um, every month you kind of have to scrap what you're doing, and I think the biggest thing that, uh, anyone in this audience or, or watching at home could probably do is build evals, right? And I think internal evals has been the biggest unlock for us that we should have started, you know, way earlier. Um, but we could literally model out several different versions of Silvia, uh, to see how she performs in different tools, different skills, different preferences. Um, and then obviously you look at the data and then, you know, if it says that you have to start over and, and scrap what you just built, then, uh, you, you kind of have to do that.
- SPSpeaker
Yeah, we've seen a lot of people starting over. [laughs] Um, how about Madhav, your wish list?
- SPSpeaker
Wish list?
- SPSpeaker
Yeah.
- SPSpeaker
Uh, smarter model, faster model, cheaper model.
- SPSpeaker
[laughs]
- SPSpeaker
[laughs]
- SPSpeaker
Wish.
- SPSpeaker
Yeah, I would, I would just like, you know, index on... And, and also like given a choice, I would index on, on intelligence. You know? I, I always feel that, uh, people can like sort of work around, uh, speed, the latency, and the cost aspect of it to some extent, but the intelligence is where like I feel like given a smarter model, I would always prefer a smarter model over a, a, a slightly, uh, weaker model. But yeah, I mean, I, I could-- I cannot get enough of intelligence, to be honest, you know?
- SPSpeaker
I think you're gonna be happy-
- 20:34 – 23:14
Choosing bets in an unpredictable market: stay close to customers and close the loop
- SPSpeaker
Um, I think one thing that we see is it's, as models get smarter, it is very hard from the outside as an investor to predict how things are going to go. How do you guys figure out what are the right bets to be making as you're predicting how your cust- what your customers' needs are changing and then where the market's going and where the models are gonna land?
- SPSpeaker
You gotta, you gotta make all bets in time.
- SPSpeaker
Yeah.
- SPSpeaker
Um, it's too, it's too tough to predict now. Um, I mean, like I said, we changed our architecture so many times. It's changing every month. You just have to be, uh, open to, to adapting and then building towards it.
- SPSpeaker
Yeah. So I think like, you know, just as an example, a discussion that we're having right now is, right, like do we want to, uh, open up an API? Do we wanna have a CLI? Do we wanna do these things? And I think it's like this natural tension between do we want people to be building in the product? Do we want people to be able to access it where they are? What-- How does that compromise the business in the long term? Um, just thinking through those natural trade-offs. Uh, and so, you know, for us, it's like we also know where the puck is going. We also know that our underlying kind of, um, you know, uh, principles of the business are we've built it in such a flexible way that it's, it's very-- we're very open and, and as things change, we can adapt to that. Um, but I think generally we're trying to help people do things where they can do them. And Clay, we are not like, we're not like too proud or whatever of like people should be in the product. It's more like we just wanna help you grow. Whatever way that happens, um, we just wanna serve it in, in, in the way that's best for them.
- SPSpeaker
Mm. So you're very flexible with the surfaces-
- SPSpeaker
Yeah
- SPSpeaker
... and how you interface. Yeah.
- SPSpeaker
Yeah, I mean, I think, you know, just staying closer to your customer actually like is, is, is where, where you get all the insights from. And, and basically, you know, you might say that, "Hey, uh, you know, Opus four point seven c- came, what should we do?" Et cetera, et cetera. But once you go and like look at what customer is complaining about, you'll see very quickly that all the hype around AI doesn't always reach them the way you would imagine it would reach them. Right? They're still struggling with some basic stuff. They're still struggling with like s- the basic things to work, and even to make things more reliable at their end, right? Like just to make things reliable because even out of the box using all these great models doesn't, uh, solve the problem until you finish the loop, until you close the loop, right? You cannot expect your customers to basically, you know, like build the app and then like find the platform to deploy themselves, right? You wanna sort of like help them close the loop, help them... They came on the platform with a very, very, very simple goal to solve their problem, and if they walk away solving the problem that they had, uh, is, is where you, where you kind of bridge the gap. And I think as long as we are like following that as our sort of, uh, you know, focus, uh, it keeps us moving in the right direction.
- 23:14 – 26:23
Advice for founders: follow curiosity, don’t fear crowded markets, embrace discomfort
- SPSpeaker
Yeah. That's great. Um, I know we have a lot of aspiring founders and builders in the audience, and so one of the questions that I wanted to ask was what advice you would have for someone starting from day one today. How do you think about where to start in terms of building a company, building a product, and using models in a way that you will grow your business rather than be trampled by them over time?
- SPSpeaker
Hmm. Um, I think this is like, um... It's, it's, I feel like I, I, I talk to a lot of early-stage founders right now, and like it's not like easy to pick something in this moment in time because it feels like, um, you know, a- anything is, uh, is, is like, like roadside meat for Anthropic, basically.
- SPSpeaker
[laughs]
- SPSpeaker
And so, um, it's like, you know, it's challenging, and then the competition, you know, Sequoia's funding every company, so it's like, it's, it's like, uh, on both sides, Anthropic and Sequoia-
- SPSpeaker
Sorry. [laughs]
- SPSpeaker
... it's like hard to, hard to find something. Um- I don't know. I, I, I think, like, this is gonna sound generic, so it's probably gonna be deeply unsatisfactory and help- and un- uh, unhelpful. But, you know, there's this computer scientist, um, um, who, who wrote this book that g- great- greatness cannot be planned, and it's like it, it, it, it, it, it... I think it applies to this topic but also applies to anything ambitious you wanna do, which is just that I think people, uh, oftentimes just don't, like, follow their own instincts and curiosity and, and try to, like, work backwards from a prescribed goal, which is just not very, um, common and, like, as a way to, like, doing, like, unique and great things. And curiosity al- oftentimes just helps you take, like, take it from one stepping stone of imagination to the next one. And so my, like, honest advice is just to, like, follow that curiosity and, like, spirit, and then, and then you will kind of find something interesting and then, like, not think too far ahead.
- SPSpeaker
Yeah. Fr- fr- from my side, I would just say, like, you know, get your hands dirty and, uh, do not assume that if there's a big player in the, in the market, you know, that that market is shut off. You know? Like, when we entered this app building space, it was... it seemed pretty crowded. Uh, but when you look and, uh, you know, try to work with your customers, you see the pain points that they have and you see the problem is still pretty unsolved, right? And so behind all the, you know, media glitz and marketing glitz, you'll see some, you know, like you'll see that... You'll, you'll almost, like, write off a particular area that, hey, this area is already written off. Nobody else can, can, can sort of enter this space.
- SPSpeaker
But most of those products don't work.
- SPSpeaker
[laughs]
- SPSpeaker
Like, they really just don't work. And if you just, like, use them or talk to people, like, you'll find that they just don't work. And so just destroy them, you know? Like, go for it.
- SPSpeaker
[laughs]
- SPSpeaker
[laughs]
- SPSpeaker
Pretty much, yes.
- SPSpeaker
Yeah.
- SPSpeaker
It's still very early. Anyway, Shane.
- SPSpeaker
Yeah, yeah. I mean, uh, curiosity, adaptability, seek, uh, discomfort. It's kinda like the main thing. If you're uncomfortable wiping out, you know, your entire architecture, it's probably the right decision, uh, right now, but yeah.
- SPSpeaker
Yeah. Yeah, I think my main advice is just keep going. The world's changing, and if you're, if you're moving with it, like, you will find pockets of opportunity. Absolutely.
- SPSpeaker
Yeah.
- SPSpeaker
All right, well, thank you all for joining us.
- SPSpeaker
Yeah, thank you, guys.
- SPSpeaker
Thanks.
- SPSpeaker
Thank you. Thanks, Lauren.
- SPSpeaker
Thank you.
- SPSpeaker
Thank you. [audience cheering] [upbeat music]
Episode duration: 26:32
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Transcript of episode HrEy6MmqI-g
