Codeium’s Vision for AI-Driven Development with Varun Mohan | E2064

But that is what a VC is thinking. You're worthless and weak to their goal of having a 3,45x fund. They need a lunatic who can get two other lunatics or at least two other samurai to go out on a crazy mission. This is elite Jedi stuff. Is everybody a Jedi? It's an elite pursuit. It is not for everybody. Everybody can try to be a Jedi. Most people cannot go and fight a Sith Lord because you will get your arm cut off. And even the best Jedi, the most powerful ones have have lost limbs. Luke Skywalker, Anakin. There's a long list of lost limbs in Jedi land. It's a brutal pursuit. Even if you are elite.

Anakin's not Luke. I thought God you you Star Trek.

People make me crazy.

I I Darth Vader Spoiler alert son one of Sorry. I haven't seen the Jar Jar Binks movie since high school.

This week in Startups is brought to you by Squarespace. Turn your idea into a new website. Go to squarespace.com twist for a free trial. When you're ready to launch, use offer Code Twist to save 10% off your first purchase of a website or domain. LinkedIn Ads to redeem a $100 LinkedIn ad credit and launch your first campaign, go to LinkedIn.com thisweekinstartups and prize picks Run your game the best place to get real money action while watching your favorite sports. Download the app today and use Code Twist to get $50 instantly after you play your first $5 lineup.

Hello and welcome back to this Week in Startups. This is Alex and we have an awesome show for you today. So we are going to talk to a startup CEO that I think is super interesting building something awesome. And I put it to the test and we're going to show a demo in just a couple of minutes. And then after that we had some live Q and A with Jason, which is always good fun. But before we get into all of that, I want to talk about Perplexity for just a minute. Now, I'm sure you've heard of the AI search startup. It's been in the news constantly this year, not only because it's growing very quickly and has had dust ups with major publishers, but also because of its insane fundraising cadence. And I'm not exaggerating here. Let's just talk through what the company did according to public data in 2024. So Perplexity announced a $74 million Series B in January of this year and that pushed its valuation up to the $500 million mark, which is a lot of money, but it was just getting started. Then the company added to its series being April, closing around $63 million at about a $1 billion valuation. And again, it was only kicking the tires on its year. This summer, Perplexity raised $250 million at a $3 billion valuation. And then this week, Bloomberg reported that Perplexity has raised $500 million at a $9 billion valuation. From 500 million to 9 billion in less than a year is insane. So what's going on? Well, after handling around 500 million queries in 2023, an impressive number, perplexity served 250 million in July alone. And more recently, in late October, the startup said that it had scaled its query load to 100 million per week, or about 400 million per month. Now, SEMrush reports that Google handles about 5.9 million searches a minute. So Perplexity is still orders of magnitude smaller than the search incumbent. But here's the thing. I don't think that really matters. What matters is that search is incredibly valuable. And, and that means that every point of search market share that a company can control will generate an enormous amount of wealth. To put this in very, very loose mathematical terms, 1% of Alphabet's market cap today is $23.2 billion. Now, of course, Google, Alphabet, they all do more than just search these days, but that just goes to show what the value of search is. And with rising search query load, an AI first approach that is, I think, cleaner than what Google offers today, a subscription business and a nascent ads business, well, you can kind of understand why VCs are excited about the company. I view it as kind of a binary bet. If you value Perplexity on its fundamentals today, it's probably a little bit expensive at the 9 billion mark. But if you think it's going to get 1, 2, 3, 5, 10% of the search market down the road, then it is absolutely on sale today. All right, enough of all that. Let's talk to Varun Mohan, the founder and CEO of Codeium, and see what I cooked up using their new ide. And then, as promised, we'll throw some questions at Jason. All right, let's do it.

All right, next up on the program, we have a great guest. One of my venture capital friends told me, hey, you know, they whisper to me in the whisper network, I've got a really great company. It's kind of. And I tell my VC friends, hey, if you ever have or angel investor friends, you get a really great company you're excited about. Number one, email me, maybe I'll invest. Number two, you know, always great to have people who've been validated by the venture community on the program. In this case, it's the latter. I didn't get to make a bet here, Alex, but we have a great guest and so let's introduce him. It's very topical.

Yes. So we are going to talk to founder and CEO of Codium. Codeium is a company that you may have heard of at very, very basic form, Jason. It's an AI coding tool. There are a number of these in the market, but I do think that Codium is actually unique and pretty awesome. Um, we always like to talk about who's backing the company. So Codium has raised about a quarter billion total. Kleiner Perkins, General Catalyst, Green Oaks, Founders Fund, you know, lots of names that we all know and love. And the reason why I wanted to have Codium on the show, not just because of the Whisper Network, but also I've been keeping tabs of what people are building inside of this niche. And one thing that I'm excited about with Codium is that they started off building extensions so that way they could bring their AI intelligence into wherever it was you were writing code, whatever your IDE of choice was. Now what they've done is build their own ide and I got to play with it. It's super cool. But please welcome to the show Varun Mohan, the founder and CEO of Codeium. Varun. Hey.

Hey, folks, thanks a lot for having me.

Thanks for coming. Varun, is this your first startup?

Yeah, this is my first startup, but we had a little pivot in the middle. So I'd like to consider myself a second time founder, but first startup indeed.

Well, congratulations on that. Tell us a little bit about what the product is today. And since you brought up the pivot, man, we always love a good pivot here. So where did you start the entrepreneurial journey? What did you learn and then how did you pivot into the new company? Let's start with that. What was the first idea? And then pivot. Pivot us right into the second idea.

So I worked at this company before this. After graduating from mit, I worked at this company called Neuro. It's this autonomous vehicle company based in Mountain View. From there actually felt that deep learning was going to be very pervasive. Right. The idea of machine learning was going to not only pervade these high tech spaces like autonomous vehicles, but would also be there in financial systems, in healthcare and defense, and the company actually started out as a GPU infrastructure company. So a company that made it easier to run these deep learning applications. We ended up managing upwards of 10,000 GPUs. We got to a couple million in revenue, but our target customer base at the time was only these autonomous vehicles and robotics companies. And I guess the middle of 2022 rolled around these transformers which were the underlying architecture of these GPT models started to pop up. Right? They started to become really, really popular. And what we felt was it was going to usher in brand new, almost the same way the Internet ushered in a brand new set of applications. So in a world in which we were selling to a niche, we felt that, hey, what if we could actually build the next almost Google or Amazon, which sounds fairly silly, but we thought a company of that size could be started on top of these generative AI applications. So we actually took our infrastructure expertise and actually went out and vertically integrated. And that's where codeium came about. Right. So we pivoted the company wholesale and completely started working on a product to actually provide AI code assistance, right? Entirely.

Varun. Can I just ask a question for the other kind of founders out there? So the first company was called X a function. When you decided to pivot the company, did you have investors at that point? Was there pushback? I'm just kind of curious at the operational element of deciding to change course.

Yeah, we did. So we had actually at that point even raised a series. Our key investor at the time was Green Oaks Capital. So we had raised around $28 million. I'll just be very honest, we pivoted over a weekend. We actually pivoted over a weekend even though we had a couple million in revenue. Because, hey, we're a venture scale company. It doesn't matter if we get to 20, 30, 40 million in revenue. If we didn't see line of sight to make this a billion dollar revenue business, then we have to go work on something else. Right. And I think one of the things that is a big learning lesson for me for companies is just, you know, when you're in school, all else equal, you work harder, you're going to get a higher grade. All else equal, you study harder or you're smarter, you'll get a higher grade. But in startups, you actually need to ask yourself the question, am I taking the right test every time you take the test? Right. And if you feel like you're taking the wrong test, there's nothing wrong in moving on to something else. There is some shame that you have but to be honest, no one in the world cares about your ship. No one really gives a crap about your ship.

My goodness, what a great insight. You do have this attachment, sunken cost fallacy. You know, you've spent money, you, you pitched everybody with all your heart. This is the future. You raised the 28 million and now you kind of go on this exploration and then you do have this doubt or shame. I told you this was great. Green oaks gives you 28 million. You hire a bunch of people and then you realize, you know what? Hey, I went. And it's a dead end. And you're responsible as the founder for taking people in the wrong direction in a dead end. But what great leaders do is they define reality and framing is everything. What I love, Brun, is how you frame this, which is you came up with your own framing, which is, simply put, I want to ace the test. But is this the test that I need to ace? If you were, you know, going to have a future in robotics and you ace a test on, I don't know, Roman history, does that actually the right test to take, the way I frame it is I tell people, hey, you went to a beach to serve. You found a great beach, Alex, but there were no waves and the tide was out and the waves don't break properly. There's some rocks under there, there's some great whites. You probably don't want to surf that beach. There's no other surfers there. Okay, you found another beach. Hey, there were two or three surfers there. Competition, but great waves and everybody's surfing them. And that's where, you know, you, you can have an easier time pivoting. And for investors, we, we bet on teams and we bet on founding teams. You love the idea that a founding team comes to you and says, you know what, I took everybody to the beach. The waves suck. Or we took a test. This test is not going to result in a billion dollars in revenue. But by the way, I found one over here that is going to lead to a billion that actually builds your credibility. This week in startups is brought to you by Squarespace, the all in one platform to build a beautiful website and grow your business. Whether you're just starting out or scaling up, Squarespace makes it easy to create an incredible site, engage your audience and sell anything, products, content, or even your time all in one place. Here's what makes Squarespace a game changer. First, their design intelligence feature combines decades of design expertise with cutting edge AI. It's like having a high end designer right at Your fingertips helping you craft a stunning personalized website in just minutes. Second, Squarespace Payments makes it so easy to manage your payments seamlessly with fast onboarding and support for popular methods like Apple Pay, which I love, and Klarna. And you'll be up and running in no time, right? Ready to take payments however your customers prefer. And third, let's talk about Blueprint AI. This is wild. Just answer a few quick questions and Blueprint AI automatically builds a polished on brand site with curated content tailored to your style. So here's your call to action. Check out squarespace.com twist for a free trial. And when you're ready to Launch, go to squarespace.com twist to get 10% off your first website or domain purchase. That's squarespace.com twist. The shame, Alex, that people feel needs to be reversed. And what it is is it's not shame, it's maturity. And it's being a great leader is saying we tried something, it got modest success. You know what the biggest impediment to great success is?

Little success.

Modest success.

Modest success.

It's not failure. People fail. That's awesome because it makes it easy to move on to your next thing. It's moderate, moderate success. And I assume you had moderate success in the previous company, Maroon.

That's right, yeah. So that would say, I would say like, I would say actually that if you were to look at our revenue at the time we were only eight people. So I think there's. And we free cash flow positive. So I think a lot of, a lot of ways you could look at it would have been like, hey, actually if we keep trooping on, this could work. But, but I actually totally agree with you. I think the hardest parts are when you really want to believe you can keep going. When it's so clearly a failure, you have no other option. Right. But I think level setting expectations, I think one of the things, Jason, that I feel is quite important is always trying to keep the team very, very lean. Right. Like at the time, even though we were doing that much revenue, we only had eight people. And that was fundamentally because the market, I didn't know how to grow it. Right. Why did we need way more people? Like what was that actually going to accomplish for the company? And now it's completely different. Right. I think we have a massive opportunity in front of us and the company has grown materially in that period of time, but we still make strategic changes all the time. Right. I think that's the biggest advantage a startup has is just constantly changing.

Right.

Change is the best, is the best thing for us, because that's how we could topple an incumbent.

Well, and Alex, that. That threads with what we were just talking about on a recent episode where I said, you know, sometimes the best management philosophy, we're talking about this in relation to the United States being we have federal laws and then we have state independence, the laws. And there's this tension between the two Varun, obviously where, like should education be handled locally or should we have a department of education federally? You know, gun laws, abortion. You. You pick the topic. You know, people will have this debate, local or federal government. Sometimes the best thing to do is to just move it from one to the other and get, get another shot at it and change the management philosophy. The management philosophy that I see that works is constantly changing the management philosophy while keeping the goals and the mission of the company the same and evolving those a little bit more slowly. But you can change how management works. You can say, you know what, this person's in charge for the next 60 days. It's your right to do that. And you could swap people out. So I think those are all great insights, Alex.

Yeah. So I want to talk about where you went next. So clearly, exo function set aside and then the company, as far as I understand it, built a number of extensions that worked with other IDs to help people code, kind of in situ varun. So can you take me through from the pivot until you were in the market with what code was up until Windsurf came out?

Yeah. So basically what happened was we ended up vertically integrating and building out this AI code assistant platform. We actually trained our own models and random ourselves. Because of that, we were able to offer the Codeium extensions entirely for free, right across all the IDEs. And the IDEs are the platforms where developers actually write software that offer autocomplete capabilities, chat capabilities for the largest enterprises. We were even enabled enabling them to run the product entirely within their firewall. Because code is their most important sort of crown jewels internally. And we made it so that the suggestions that we gave to them were personalized to the private data internally. Like for the extensions. We have over 800,000 people currently using the extensions. We have over a thousand enterprises using it. Companies like JPMorgan Chase have thousands of developers internally who were inducted into their hall of innovation 22 months ago. Which is actually an interesting strategic decision we made by going enterprise grade very, very quickly in the company because we felt that we could actually target with our infrastructure background these large, large enterprises. But Alex, what interestingly happened was two months ago we saw that, hey, could we deliver a best in class experience? And because of that we ended up deciding we were getting limited by the IDE applications that already existed. So we ended up creating our own. And that IDE is called Windsurf.

I want to get to Windsurf in a second. Cause I'm gonna give. I played with it today, but talk me through building your own model and how that allowed you to offer your extension for free. Because my read of that comment, varun, is that because you rolled your own model effectively, you had a lower cost per query and therefore you could essentially not lose all of your stored capital by offering what could be an expensive service for free.

That's exactly right. So I think one of the big things for us, having been a GPU infrastructure company, was we could build our own compilers and all of these pieces to run transformers really, really efficiently on our own servers.

Right.

And this made it so that some of these expensive workloads, like the auto completion workload, that is literally every keystroke you are typing, we are doing trillions of computations in the cloud for you for every keystroke you're typing. Right. And that was actually what enabled us to build a massive, massive free individual product, which enabled us to also ab test a lot of features, which is critical if you're going to go and actually work with the enterprise. Right. Enterprises don't want to be guinea pigs for whatever products you build.

Okay, now there is a pivot or an extension of what you were building in the new IDE that you guys built. And I'm curious, did you actually decide to build that two months ago? Because it sounds like you were trying to tell me that you started this project very, very recently.

Yeah. So the IDE itself is actually built on a lot of things that we, that we did for all the extensions. Right. This is not a zero to one sort of projects. The company has been working on Codium now for well over two years now. Right. So it's a, it's an amalgamation of everything. It provides sort of a really, really high quality sort of UI workflow. But on top of the ui, the thing, you know, you're probably going to showcase is this idea of Cascade, which is this agentic experience, right, where the AI is actually able to write more and more of the software. Maybe. One thing I will share is with code, right now, in the extensions, we are seeing around 50% of all committed software generated by code. Right? That's what we're seeing, I think with Windser, we are probably going to see well over 80 or 90%. Right. And we thought we needed to give a different experience to provide that workflow.

Was that a demand from the market itself or was that you guys saying, look, what we have in the market works, but we want to take it up a notch and we're going to give people something they don't even know that they want yet.

It's definitely the latter. I think in one of these cases, and this goes back to startups, if all you're doing is incremental things, incumbents are going to be fine. Right, because they have more distribution, they have more capital, they have more resources than us. The only way for us to succeed is we are offering 10x better experiences to our users every year. I tell this to our company. We need to cannibalize the existing state of our product every year. And that should be the expectation internally.

Alex, you've been using this product. There's a free version, I guess, and maybe you could show us what you did. And I'm very fascinated with Maroon agents and how these agents, we all understand how a copilot works. It's, you know, not super dissimilar to what happens when you see email or Grammarly suggesting things to you as a writer. English and code are just two different languages. French and Latin are also languages and they all kind of work the same. We know. And we can predict what you're going to do with a lot of certainty in the next word or five or, you know, when it comes to code snippets. My Lord, it could be a couple of paragraphs or it could be something you would have previously found on GitHub, you know, a repo of something that you're going to just drop into your code. So, Alex, show us what you did. And then I want to get into the agents and how the agents are different than the copilots, because we've talked about copilots for years. Have you built something awesome? Are you ready to bring it to market? Well, that's great, but you're probably asking yourself, how do I get to my target customers? Well, here's an easy solution for you. It's called LinkedIn ads. And with LinkedIn, you can precisely target the professionals most likely to care about your product. You can use LinkedIn to reach people by job title, industry, company location and more. You know, all the stuff that you have on your profile page. Imagine you're looking to build your B2B audience. Maybe you want mid sized companies, maybe you want large ones, maybe you want emerging small businesses, right? Small and medium sized businesses, SMBs. Well, you got to build those key relationships and you want to drive results. Well, you just do that with LinkedIn ads and you're going to gain access to over a billion members of LinkedIn of which 13%, 130 million of them are decision makers. That means they can pull the trigger on buying your product or service. And 10 million of them, 1% of the LinkedIn audience, are those elite C level executives. You know what I'm talking about? Chief Technology Officer, Chief Strategy Officer, Chief Revenue Officer, Chief Executive Officer, Chief Financial Officer. They're all there on LinkedIn. And you're going to get two to five times the return on your ad spent. 79% of B2B marketers say LinkedIn delivers the best paid media results for a reason. So here's your call to action. Start converting your B2B audience into high quality leads today and we're going to give you $100 credit to start your next campaign. LinkedIn.com/this week in startups to claim that credit. That's LinkedIn.com/this week in startups. Terms and conditions do apply.

So everybody, I'm going to show my screen which is essentially Windserve this ide we're talking about. I'm going to sportscast as we go through this. But here is Jason and Varun what I did. So as you can tell up here at the top, I told Cascade that I wanted to build an app. I was literally like, what if I was a complete idiot? No knowledge, no. I didn't read any directions. I just started and to my credit of Codium, it said, I'll help you. That sounds great. What do you want to build? So I said, I want to create a very simple messaging service for myself and a few friends. Web based is fine. Then it began to message very quickly and came up with lots of ideas, lots of things it wanted to do and often for an I had to click Execute. Essentially the Cascade system said, hey, this is what we want to do. Do you approve? So there was definitely a human in the loop component of this that was pretty easy and honestly pretty cool. So things went along. It made many things for me. I had to install Python. I did not have Python installed. It told me what to do, I installed it. I then ran it. It said, fine, very cool, very slick, very easy. And then it says, look, here is the app that's now running. Then I ran into some issues that access to localhost was denied. Here's the cool thing, Jason. It said, oh, okay, let's figure out why that's happening. Ah, I see the issue. Here's the problem. Let's fix it. And then it did. It was incredibly, incredibly cool. So essentially, as I went through, I talked back and forth, I registered, I logged into my new app, ran into some errors. It said, cool, we'll restart the server. And then on and on and on. And I tinkered back and forth. For example, when I send a message, it doesn't appear in the gui. How can I fix that? And again, Cascade walked me through it. And then I got to the very end for it. I ran to a very interesting message entitled error rate limit exceeded for model upgrade to Pro account or try again in about an hour. So before I show my little app that I built, I'm curious, how much credit did I burn through in this?

How much did that cost you?

This is going to come off as strange. I'm happy to talk about agents, but our product uses a combination of our own models as well as anthropic models. Anthropic models. Our own models because all the local edits that are made, if you notice, actually Jason and Alex, we make a lot of changes. This is not just a copilot that is making five changes. It actually goes and edits many, many files in a code base and deeply, deeply understands the code base. So all those edits and retrievals, we are actually doing ourselves. Right. But the problem is for anthropic, we ended up taking them down multiple times. That's how kind of popular this product has gotten. So effectively, we would have loved to give, be a lot more sort of, I guess, giving. And obviously we've raised a lot of money. But the problem is we are literally running out of sort of capacity issues for all of Anthropocene.

No, no, you should. You should totally throw up a roadblock at some point if people are getting value. Just curious, like, what do you think that interaction cost you as a company? Because you would take a hundred of these users, and if one converts, you would say, well, that's our acquisition cost of that customer. So walk me through this model. Yeah.

This interaction that you just saw here. Yeah, the edits, a lot of it is a function of how large the edit is that was ultimately made. It's kind of interesting. It's almost. Our expense is similar to the amount of value it's giving you. If you largely align value with the amount of code that we wrote for you, which I It's not necessarily the case. Maybe some code. Even if it's one line, it's very, very valuable. And there's no way for us to attribute that. I would say this is probably low tens of cents, right?

That's how much value this was.25 cents.50 cents.

Yeah, that seems reasonable. Yeah.

Okay, so we put it at 50 cents. If 100 people do it, that's 50 bucks. What do you charge? What? When he hits upgrade premium, what do you charge?

We charge $15 a month for the pro and then we have an ultimate tier that is $60 a month.

Great. So if, yeah, you, you blend that to 25 bucks a month, on average, you're at $300 a year. You need only convert. I mean, gosh, out of 600, you convert one, you're, you're already at break even. So you could, you could have a hundreds of people doing this and then convert. So it's important for people to understand that when you give something away for free, you're doing some sort of calculus on, on the back end of the. How many times we give people a sample of the milkshake and then how many people actually bought milkshakes. The profitability of the milkshake. You, you do keep track of that?

Yeah, yeah, of course. Yeah.

Okay, I'm going to show you my milkshake. Which brings all the developers to the yard.

Absolutely. You got the joke before I did.

Wow. Ladies and gentlemen, behold my login screen. And I've already made an account, so I'm going to log in as myself here.

Great.

And by the way, it turns out passwords are cap sensitive in this. Didn't know that. Got locked out of my own application and had to figure out how to get back in. Watch this.

Ta da.

It's my. Oh, never mind.

Look at that. You're. You're now 1% of the way to building your own threads. Twitter.

I know.

So this is, this is stuff that I sent in that I then had to debug. So I didn't clear this because the first couple of tests I sent didn't appear. Went back to Cascade, said, hey, they're not showing up. It fixed it. And this is the same thing that I built over.

Did you ever get trained in being a developer? Have you ever taken a development course?

I have C. High school, self taught, lots of those huge books you used to buy at Borders. And I've done little bits of things here and there like this, playing with stuff, trying to test and learn. But I wrote no fricking Code I literally straight up, amazing. Spoke to Cascade and it works. So, like, you know, test Jason, give Alex a raise and see it there and it's safe and you can log out. I was going to have it create multiple accounts, multiple kind of multiplayer chat, but then I ran out of credits and I had to prep for the show.

I mean, this is the future. It's. I mean, what you're seeing here, folks, is the promise of no code. We had a moment in time and we had a lot of people saying, hey, listen, WYSIWYG will allow individuals like Alex who maybe have dipped their toe into coding, or myself who did it in high school when it was Pascal and BASIC and didn't even get to C. That's how old I am. And so they're going to start to be able to use natural language English as opposed to WYSIWYG and dragging stuff around and writing, scripting. That is the future.

Arun, in your mind, I totally agree with that. If I can share one anecdote about the product before we launched it on what was very, very exciting, our company is over 100 people, right? We have a lot of people in our go to market team, largely because we actually sell the enterprise. Everyone at the company built an application, literally everyone. And I'll tell you a couple of things happen because of that. Some people on the go to market team are actually building apps all the time. Now. Some of them are personal apps, but some of them are actually apps that are useful for the company. Like you can imagine, there are lots of these go to market tools, right? An example is a quoting tool. Why do you need to build a quoting tool if you could very, very quickly iterate entirely with a tool like Cascade, with a human in the loop to kind of get the final output and it's tuned perfectly for your business use case, right? And that is like the crazy thing, right? Because some of these quoting tools are literally six figures. And on the other hand, we are potentially building a better version for ourselves for dollars, right? That is a crazy, crazy sort of outcome that we are seeing internally.

Tell me about agents. How are they going to. And I don't know if we have a demo here of an agent, but maybe define for people who, you know, are curious about how these are going to work and might want to start using English to work with an IDE and start making their own code. They're inspired. Or somebody's like a hardware developer. What is the agent going to do for each of those people?

That's exactly right. So an Agent is effectively when an LLM is able to do almost work on itself, right? You can imagine there are two ways to almost use an LLM. One is immediate prompting. You construct a clever prompt and you just look at the outcome of the prompt. But what happens if you could make the LLM do something and iteratively use the output of what it just did to go out and do the next thing? Right? That's almost doing what humans do when they do reasoning, right? They say if this, then that, and if that, then then this other thing. And they keep going down and down until they finally get the best quality answer. So I think here's the way that agentic workflows kind of work in a product like Windsurf in Cascade, right? We were able to do agentic retrieval, right? Almost going through your entire code base to find, hey, given this question, where should we actually be looking to make the change? And the AI itself can then go make the high level change in those locations as well. So this is where it's like very, very helpful for real developers that had been developers for many, many years, right? Because they're operating on code bases that are millions of lines of code. It's very different than building a zero to one app that Alex just did. But for them it is not sufficient to just rewrite the entire app, right? You actually need to modify it. So with tools like Cascade, it is actually able to do deep retrieval. Because we've actually invested a lot internally in actually deeply understanding the code base, right? That is actually systems that we've built with our own models that are really, really good at that. And that is actually where the agent is very useful. On the other hand, for the 0 to 1 use case, super quickly the agent can go out and very, very quickly execute code, which is what you saw Alex just did, right? It installed all of the requirements for him. That's just busy work, right? Looking up manuals and installing things is not something that actually requires a ton of brain cells. It just requires some amount of knowledge, right? And all of those kinds of operations, we would just like to move entirely to the agent to actually end up doing that. So ultimately what I think is this is going to be a bar raiser for everyone. People that are non developers can now build apps and people that are that work at large companies can actually modify them very quickly.

So I have a request, a feature request. So sorry to put my own needs here at the forefront for room, but what I would have loved is if there was like a 10 or $20 a month option to have you guys spin up a little server for me and allow me to use one of my URLs that I already own and actually take my little thing off of my local machine and put it out into the wild. I presume that for, you know, Alex's personal web app, the cost to host that would be de minimis and I would pay you 99.9% gross margins if you kind of made that super simple. I know you're thinking about the enterprise and not dweebs like myself, but I would have loved that.

No, actually, so interestingly, Alex, we're not only thinking about the enterprise. And the reason is actually because we are a developer product. We're a little bit different than products like a database where only a handful of people inside the company interact with it. When we're used inside a company, everyone in the company needs to use it. So I would actually say Windsor is built so that every developer in the world and people that want to build use the product. So that's a great piece of feedback. Like, we'll take it. We'll actually take it and see if we can run with it.

All right, sports fans, let's talk about prize picks. It's the best way to get real money action while watching your favorite game. I love it. Here's how it works. You just pick more or less on at least two players. So, like, I love the Knicks. We've got these great players. We got Karl Anthony Towns we traded for. Do I want to say he's going to have more or less than 20 rebounds, 10 rebounds, or sometimes they'll put together two or three stats like rebounds and assists. And you can do this on all kinds of sports. You make your selections, you lock them in, and then you can win up to a thousand times your cash. You heard that right? 1,000x. I love these wagers. They're so much fun and they keep you really engaged in the game. I was able to put my picks in in under a minute. And then you have flex pay where you can still cash out even if just one of your picks doesn't hit. So you can pick as an example, if you just go for your top three, you're going to get 12 to one or something. But if you say I want to get eight to one, if I hit all three, but I want to get half my money back, I could still cash out and get like a smaller percentage on the second. So it's a really to kind of get a safety. And Prizepix now offers MasterCard for quick and easy deposits into your account. I use PayPal. It was so easy. Join now because this holiday season prizepix is giving away two free picks in December and they're giving away 30 plus million dollars in rewards during Pixmas. Download the app today and use the code Twist to get 50 instantly after you play your first five dollar lineup. That's pretty fun. That's right, 50 bucks guaranteed. You don't even have to win. Download the app now and use the code twist to get your $50 instantly after your first $5 line. Run your game. Satya Nadella was on the BG2 pod last week and he had this clip where he talked about the future of SaaS, essentially saying there is no future in SaaS. Let's play the clip here. We will talk to you in 1 minute and 50 seconds.

I think the notion that business applications exist, that's probably where they'll all collapse, right? In the agent era. Because if you think about it right, they are essentially CRUD databases with a bunch of business logic. The business logic is all going to these agents and these agents are going to be multi repo crud, right? So they're not going to discriminate between what the back end is. They're going to update multiple databases and all the logic will be in the AI tier, so to speak. And once the AI tier becomes the place where all the logic is, then people will start replacing the backends, right? That's what in fact, it's interesting as we speak, I think we are seeing pretty high rates of wins on dynamics backends and the agent use and we are going to go pretty aggressively and try and collapse it all.

Do you believe with the central premise that the SaaS layer is going away and it's going to be replaced by agents just hitting a database in the back end?

I think a lot of these simplistic applications that are basically, as Sathya said, a database access and some logic on top of it, it's going to be extremely easy for companies to build themselves and an agent to actually work on it. And maybe the right way to sort of look at it is in the past. The reason why they didn't is just because the company on the other hand is able to maintain it, maintain it very easily and then beyond that they have enough R and D to go out and actually continue to improve it with time. But a lot of that R and D is probably going to improve a bunch of use cases that are not your own, right? Like you have a Particular use case, like, let's say a company has a kitchen sink of features, you might only need five of those features, right? So actually your maintenance cost is not nearly as high, but you just don't want to take on the engineering effort of doing that. But now with some of these agents and some of these tools, I just think these simplistic apps are going to get replaced. I agree with that. I think it's a completely correct statement. I think some of these more complex apps, I think they will take a little bit more time, like apps like Salesforce. I genuinely believe that it is more than just a tool for an agent to go out and do things. It's actually complex workflow software that fundamentally affects the way the business runs and that will take much more time. But for some of these more simplistic apps, I think Satya is completely right.

So one of our live audience members sort of points out Campbell is their name. Kids of the future are screwed. Hey, they won't have to learn to do anything. Agents will just do all the manual labor. It builds character to install packages and set up your own environment. What is your thought on the future of being a developer? Because now, if developing isn't limited to the top 2% of humans who have a propensity towards it, a desire, a motivation, and it opens up to say, let's not even say a hundred, let's say it just opens. It goes 10x to 20% of people who maybe understand basic logic, math or whatever, and they can get in there with these tools. It'll. It'll get them good enough to make these. Do you see this being incredibly deflationary and the. Let's just keep it a buck here. The salary of a developer has been artificially high because of the demand and the supply you 10x supply with these tools, demand is going to get. In some ways, the demand should stay the same. Even if it increases, it's not going to increase 10, 20 fold. It's going to increase to a 3x maybe, in which case we will have a lot of supply and parents right now. Or, you know, people who are thinking about going into development are saying, oh, is there 150, $250,000 salary for me at a big, you know, uber, Airbnb, Google, etc. Or am I going to be, you know, making just but a hundred thousand dollars competing against people in India, Mumbai, Ukraine, you know, and Sao Paulo for this work, which they're already doing. They are competing with them already, but now it's going to be even More pressure. How do you look at that?

Yeah, I think that maybe empirically speaking, every time software development has been easier, we've just gotten more people doing it and salaries have gone up. Right. Like you look at, you look at going from assembly to C. Right. That is, maybe that's what the package installation question almost sounds like, which is that, hey, when you went from assembly to C, programming became much, much easier. By the way, C is considered hard now because you went from C to Java and then from Java to Python. And Python is considered much, much easier than, than Java, C and Assembly. Right. And we've just had more developers now. Maybe this is like a little bit of the technological optimist in me. I just think if technology becomes easier to create, what companies are going to do is they will get more return on investment from investing a larger percent of their revenue into technology creation. Unless a company believes that they cannot become better, there's like a limit to how much technology makes a company better. That may be the case for some companies, but I would say for a large chunk of companies, that is just, I think, a unwillingness to dream in a lot of cases. So here's what I do think happens. I do agree with you. The number of developers does increase in the future. On the other hand, I think knowing more and having more taste does make you more valuable. Does this mean that there are a class of people that have less taste? And by taste I just mean, hey, like, you know, when you do any sort of creative pursuit or you do anything that has design decisions, there are people that just have better trade offs, understanding the business and understanding the technology that will make better, higher level trade offs. Like when you talk to an AI and AI is not going to know about this conversation that you and I are having here that is affecting the way in which. Not yet, not yet, maybe in the future.

It does know that Zoom is recording this and that if this was a staff meeting, it could very easily be fit into your.

It could very easily. You're totally right. But you kind of see what I'm saying here, right? I do agree with you though, that there will be a long tail. I don't think salaries are going to be equally distributed, but I believe in this deeply that there are 10x people compared to other people. And for people that get deep in the weeds, there will always be great opportunity for them in this space. That is what I was, and I.

Was very politically correct. You were very. You have a future in politics. Being a politician, perhaps there. But you know, Alex, I'm going to call BS on it. I think that we're going to see deflation in developer salaries. I just don't see how it's. That doesn't happen because you we saw it happen with remote work. Now. It used to be very rare that a team like Matt Mullenweg shout out to Matt, you know, was one of the first people to have like a really distributed team. And he in way before code a decade, he had work from home. Everybody in the industry thought it was crazy. They thought it would be have been $100 billion company instead of a 10 or $20 billion company had he been in an office and you know, pursued it harder in their minds. But he benefited from having a global workforce and little tiny remote offices where people could hang and they still have that philosophy where they just, they're all nomadic. He paid less. He was able to pay less salaries by having people in Uruguay and Paraguay and everywhere in between or everywhere outside of there. They're actually next to each other so nothing in between them. So I just think we're going to see the developer. Elite developers will always be elite developers. I don't disagree with that. But I think for the average developer, we're going to see salary stay the same, if not go down from the peak. We may have seen peak developer salaries over the past couple of years or the next couple of years and then.

It starts going down after that peak median developer salaries. I think that this stuff just dissolves crummy developers. Like if you've been always a C plus or B minus developer, I think you're going to get just steamrolled by the market.

Okay.

A plus crew are probably going to be able to extract even more value, Jason, because if you can be someone who is absolutely at the top of the pyramid and you have more tools, you're amazing and you're more impactful. Shouldn't that give you that?

Yeah, I'll go with that for sure. Yes. I think that is like the elite becoming a leader. We saw that in Hollywood. The people who actually could draw people into a movie theater with Avatar way of the water. Like he just, he's able to get budget. Robert Downey Jr. Is still going to get a huge salary, but then the average person, maybe they're just a commodity now and they're doing TV anyway. This is all incredible stuff. We'll have you back on the program soon. And I know you're hiring a ton because you raised a ton of money. But you know, I'm curious, your internal team, do you need to hire a ton of people or do you just think is it better to go slower on hiring and, you know, at this point, as opposed to doing what Google, Facebook and everybody did for the past couple of decades, which is just hire, hire, hire, figure it out later. Feels to me like something's changed where people are very selective in hiring. Maybe don't expand their team size and just work on professional development and make people who do work for you 10% better, 5% better every month. That's what I'm pursuing. I'm curious what you think of that.

No, I think it's exactly right. I think the expectations on what can be done in a given day have gone up right within the company as well. We're building these tools ourselves. I will say the ambition of the company is to fundamentally change the speed at which we can build technology. So I think we could use more people. Now, granted, we have a large rule at the company. The only reason why we hire another person at the company is everyone at the company's underwater. Right. That's the only reason why we hire another person. If we think. If we think people are not underwater, we're not going to go and hire people. This has always been the month throughout the company.

So you believe hard work is what startups are about. You believe in intensity. You believe in long hours and sacrifice to build something great.

We're all in person. We're all in person at the company.

Oh, really? How is that? Has it always been that way? I mean, obviously you took a look.

It's always been that way.

Yeah. Do you. Do you have always been people who just won't apply? I guess, because you demand people.

We may, we may. But I just think if you're building something that is. That is going to be material to society, I think it's hard to do it if everyone is remote and distributed and, you know, people are doing their laundry while other people are doing work. Yeah.

Tell me about young people coming into this environment. Are they stoked that you're in person? Are they reticent? Are they excited to be amongst other people? Because a lot of developers are like, I would rather be at my ski house in Tahoe. I understand, you know, like me saying all this sounds quite hypocritical as I explain. I'm going to go ski for an hour in the middle of the day. But I'm curious how young people look at this effort that you're pursuing.

I think great young people want to grow really quickly and they want to work with high caliber People and discuss how we could be better. Right. And I think that's antithetical to it not being in person in the very beginning of your career. Right. Like, that's the hardest time to onboard to a complex technology. So I think they're hungry for it. Actually. The new young generation had to go through college with COVID Right. So where they couldn't actually even meet their friend. That was not fun for them. That was not fun for them.

That's a very interesting observation, Varun. I think you've got a lot of good observations outside of the business you're running. I think this is a super interesting insight, Alex. The COVID generation, people who lost their high school graduation or, you know, their freshman, sophomore, college experience, whatever we took from them, this disease and also the lockdowns, which were, in hindsight, you know, too long in most places for young people who, you know, we obviously found out that they weren't as impacted. The death rate for young people was so low, thankfully, man, they gave up so much that now maybe they. They actually, Alex, crave it. And so this whole work from home thing might be predicated on people like you and I who have kids and, you know, who value it differently, but maybe young people were underestimating their desire. Alex, what do you think?

I. I think that's pretty much dead on. I think people like getting together. And my favorite example this is no matter what happens in the world, people love going to see concerts live because there's something special and magical about humans together making music. And that's always going to be the case. No VR headset has taken even a 1% share from the concert industry. I think for people that were stuck online, saw what it is like to work or learn entirely remotely, I bet they have like the best resume to explain why some in person work is the way forward. Though I will say as now, you know, working on my geezer Dom, Jason, as you know, I'm a little bit different, but If I was 22 right now, I would not be in my backyard in Providence. I would be where I was, which was in person in San Francisco.

Absolutely. That's how you built your career. That's how you got the slot here on the number one podcast in the world for startups this week in startups. And you earned that. Varun, you're amazing. Congratulations. We've added you to the Twist 500, which is a totally meaningless at this time designation. I believe it will be super important next year. Basically, we're collecting the 500 best private companies in Our estimation, in order to put one on, we got to take one off starting next year. So right now it's a little bit easier to get on. But you made it because we think what you're building is super, super important. And we're going to do an event for the Twist 500 founders in Austin. Do you like barbecue, my friend?

Yeah, I could. I could do. I could do veggie barbecue.

Oh, you're vegetarian. Well, do you like vegetarian food, is my. My question. Actually, I'm.

Yeah, I love vegetarian food.

You do love vegetarian food. Okay. Because in Austin, we've got this great vegetarian place, Korean vegetarian place, and we're going to have a big party and one of the parties is going to be at this vegetarian place. We look forward to inviting you to Austin to come to the twist 500 party or conference or something. We have a get together for the top 500 private CEOs in Austin. I hope you can make it. And welcome to the Twist 500. Go to twist500.com and we'll see Varun soon. Hey, will you come on next year and give us an update on the business?

Yeah, yeah, we'd love to.

Perfect. We're going to put you on the schedule one year from today. My team is going to send you a calendar invite because I thought you were a great guest. You should write some blog posts. You've got some good insights. Or just come on this podcast when you have a good insight. All right, thanks for coming on the pod Varun. Everybody go check out his awesome company. It's in the show notes. All right, everybody, another great segment. This is why you subscribe to this week in startups. And if you're a founder, this is for founders by founders. Alex is a founder now. I'm a founder now. And we have great founders on the program who are educating you on what happens inside startups. We don't dumb this down. This isn't like, here's how to build your side hustle, Alex. This is how to build a large sustainable startup that changes the world. Do we have any questions from the audience? Because it would be great to throw it and ask Jason here and bank it.

Dubiela wants to talk about what VCs really look for when they're vetting a company for investment. And they asked, is it quality of product, user numbers, or revenue? Clearly you want to have high quality, high user numbers and high revenue, Jason. But if you're thinking about a seed stage company in the consumer space, why don't you share a couple of Things that you think are indicative of the early seeds of success.

Sure. So a startup at its core are three things. There's a team that's assembled to build a product that delights or services, entertains or makes a market more efficient. Team, product and a customer. Right. This is just the basics of business. And so as a business becomes more mature, you will have things like the number of customers, how much each customer spends, and the profitability of the overall business. So we can look at Netflix from when they were sending red envelopes to now, and you can see how many customers, how much they're paying. And there's wonks on Wall street who are looking at it, you know, and they're even doing, you know, tiny little things like currency translations. How does the money they make in India versus South America, Europe and the us and tax treatment and money transfer. And how do all these things affect earning, impact, earnings? At the end of the day, that happens when you're at scale. But you asked about seed companies at a C company. What you're doing is you're looking for a certain type of team and you're looking for a team that is three members, not one, not five. Three is the magic number. It could be two co founders, it could be three, it could be four, one and five. Too little, Too many. You were looking for two, three or four. Why? Because one always quits, one always gets fired, and then one winds up running the company. I'm telling it to you straight. That's why you're here on this week in startups is to get the straight dope. The truth is, that's why an investor at the early stage is looking for three. If you cannot get a co founder, here's what VCs think you're too hard to work with, in which case it could be a sign that you're going to be very successful because you're just so crazy that you just want to be maniacal leader. But we would have seen that with your track record in the past, you would have done something before that, showed that massive leadership. And it's one out of a hundred founders. More likely. What it is is you're too weak to convince people to come join you on an unpaid sabbatical to go build the future. So in other words, you are not confidence inspiring enough. I'm in candid mode today. You're too weak, you're too worthless, you're too soft. You're not inspiring enough to get great people to join you on the adventure. That's what a VC thinks when they see you as a solo entrepreneur, I hate to tell it to you that way, but that's the truth. You can't get one person on planet Earth with skills to join you, to go change the world. You. You're Larry. You can't find Sergey. Sergey can't find Larry. You know, Zuckerberg had two or three partners. We only remember that Eduardo is not around. And who was the other guy, Chris, who bought a bunch of. There was a bunch of co founders there. There's a movie about co founders. Evan Spiegel had two co founders. One of them dropped out. There's another one. YouTube had three. Javid left early and then became an investor and went and finished college, I think. But you had two co founders there. It was started as three, went down to two, proving my point. So that's what you're looking for. You're looking for that founding team in the early stage. After the founding team, you might look at the product and the product roadmap, and you might talk to the pilot customers and see how acute it is. But it's team, team and also team at the early stages.

What about team, though?

Yeah. So you'd want to look at the team and maybe give some thought, you know, after you look at the team. At the team. Okay. And, you know, if you're not. If you're not inspiring enough to get a co founder right now, maybe you're inspiring enough to be the co founder. Maybe you're inspiring enough to join the team. So I said it to you in what a VC thinks. I'm not saying you're worthless and weak, you know, like some military drill sergeant, but that is what a VC is thinking. You're worthless and weak to their goal of having a 3, 4, 5X fund. They need a lunatic who can get two other lunatics or at least two other samurai to go out on a crazy mission. This is elite Jedi stuff. Is everybody a Jedi? Or do they search the galaxy looking for high Mandalorian counts to find the best Jedi? It's an elite pursuit. It is not for everybody. Everybody can try to be a Jedi. Most people cannot go and fight a Sith Lord because you will get your arm cut off. And even the best Jedi, the most powerful ones, have lost limbs. Luke Skywalker, Anakin. There's a long list of lost limbs in Jedi land. It's a brutal pursuit. Even if you are elite.

Anakin's not Luke. I thought you.

You Star Trek people make me crazy.

I know.

He's Darth Vader. Spoiler. Alert.

Sorry, I haven't seen the Jar Jar Binks movie since high school.

God, I had to catch you up.

We have another question. All right, Oleg Z. It was. It was a great answer. And I was only half kidding with my Star wars question.

All right.

I like a little humor.

Yeah, keep it up.

I'm an Asimov guy. Originally, I was never Star Trek or Star Wars.

You know, if I have to come up with my. I'm going to come up with an Alex impersonation. It sounds just like my Friedberg.

Yeah, I was going to say, like, I think I've heard that one before.

Directly a Friedberg. It's just my nerd impersonation.

I'm not even offended by that. All right. I mean, honestly.

Yes.

I don't. I don't regret anything nerdy that I've done in my life. I mean, all the books that I read, all the things that when I was being lame, it worked out great. Another question, Olegz says, how about hyper competition between the 4,712 companies that will deliver AI agent services late in 2025? I think the question here, Jason, is how do you stand out when people think they know where the future is and they're all kind of converging on the same point, in this case, AI agents.

Yeah. So if you're in the agent business, congratulations. And it's kind of like a crowded beach. Everybody's going to surf on those waves and it's going to be attrition. Sometimes if you survive, you thrive. So that is one strategy. If you can stay alive and other people give up because the water's too cold or somebody got taken out by a tiger shark, or the waves are too rough, whatever, there's rocks, people get dragged out. Those 4,000 people that you reference, surfing that beach will quickly become a thousand as people run out of money and give up. So what you're trying to do is find a customer base that loves you, that you can delight and you can just focus on. And that's where this ideal customer profile and your beachhead really matter. Finding those customers. Those first couple customers for SpaceX, they got NASA as a customer, right? Pretty great. For Tesla, they had rich geeks who wanted $150,000 toy slash sexy car. And they were going to buy a sexy car anyway, so why not buy the technology driven electric one, which was super cool to have at the moment, and be like the bell of the ball every time you pull up to a restaurant versus buying yet another Ferrari, yet another, you know, McLaren, yet another Lamborghini So you find that core audience, you delight them, in this case with agents. You know, we have tax GPT. They're going after a certain CPA and a certain size firm. And you know, you can hear those founders talk about it of who they're going after. Same thing with, you know, Grammarly. Grammarly went after students who, who, who has an acute need to have their grammar reviewed. Authors, journalists. Well, they have infrastructure for that already. Who else? And if you're very self selected, you're a graduate with an English degree or a journalism degree in all likelihood. So you probably have been through this torture. Oh yeah, how did you get tortured at school? So who did they go after? They went after schools. That's why Grammarly has 40 million customers. Because there's all these people at schools who are writing papers. 100% of people have to write a paper every semester. They have to write multiple ones. And what do they get graded on? Grammar is part of it. They're circling red lines. And you will lose points for bad grammar. In other words, you will gain points by having good grammar. Thus, they found an ideal customer profile. Students bad customer profile in terms of being able to pay. But students do not stay students forever. They eventually go into the workforce and they bring the tools with them to the workforce. And so that's how to think about it. Really focus on the ideal customer profile. Keep your expenses low, keep your burn low, and really have a great elite team that delights and studies a customer base.

Can you, can you actually add a little bit more on how to go about selecting verticals if you're going to take the niche approach that you just outlined, the go after the, that particular customer subset. Because Grammarly could have said, do you know who needs to write people who do business memos? And they could have picked that as another pain point or icp.

How do you find people who have to write business memos? How do you find them? Like that's the question, right?

Going to New York, throw a rock.

I guess. I mean, business memos. I mean, I would say the way you find people who do a lot of writing might be management consultants.

Sure.

So in business management consultants, what is their output? So I guess I would start to look at output. Well, everybody's got email as an output and everybody wants to have good emails. Okay, maybe they're good candidates for it. But I would say, I would go, McKinsey, they're writing a report, their output is a lot of words and people are paying for the word. So I would look for People where the output is words in a document that are paid for. Magazine, book, newspaper, a report, a memo. Seems like pretty throwaway. If people see an error in a business memo, do they rip it up and get freaked out? No. They see an error in a book, newspaper, or a management report they paid a million dollars for. Yeah, they get freaked out. Like, why is there. I'm paying you a million dollars to Write this report, McKinsey, and you can't use commas and you don't know how to use commas properly. And this, like, it just kind of the, the. What is it? The bloom falls off the rose or something. You're just like, God, this is.

Yeah, yeah, something like that.

There's some analogy here. So I, I do think for anybody operating in a crowded space, understanding the customer and surviving are two really good strategies.

All right, Oleg Z. There you go. That is Jason on how to stand out in the 2025 Agentix startup.

Boo.

And that's all the questions we have.

Awesome.

How fun was that? If you want to put your own questions to Jason, make sure to swing by our news live streams. We do them usually on Mondays, Wednesdays, and Fridays, and we make time for audience Q and A whenever absolutely possible. My request, bring your hard questions. We've had a couple of more generic questions lately. I really want to get to the nitty gritty to bring your business problems to the show. All right, I'm Alex. This is Twist. We're back on Monday. Lots more to come this year and of course, next year is going to be big. See you soon.