First Time Founders with Ed Elson – The AI Company That Codes For You
51:13 Share

Support for this show comes from Constant Contact. Constant Contact's award-winning marketing platform has powerful tools that make it easy to grow your audience, engage your customers, and sell more to boost your business. In just a few clicks, you can launch a marketing campaign that's tailored to your business and goals. That includes email, social, SMS, and more.

So you can sell more, raise more, and fast track your business growth. So get going and start growing your business today with a free trial at ConstantContact.com. Constant Contact, helping the small stand tall. Silicon Valley Bank is still the SVB you know and trust. The SVB that delivers human-focused, specialized lending and financial solutions to their clients. The SVB that can help take you from startup to scale up.

The SVB that can help your runways lead to liftoff. The only difference? Silicon Valley Bank is now backed by the strength and stability of First Citizens Bank. Yes, SVB. Learn more at svb.com slash vox. Scott, have you ever made a significant pivot at one of your businesses? And if so, how did you know is the right time to make that change? Yeah, I'm not sure there is a significant business that's ever been built without something resembling a pivot or

you know, iterating the business strategy. So my first firm, Profit Brand Strategy, was initially profit market research. And we used to go out and do surveys with the internet and computers and try and find a different way to collect data. And what we found is that what brands and clients appreciated was the interpretation. So we turned to Profit Brand Strategy and we became a consulting firm at L2.

we were totally focused on the luxury space. Then P&G called us and said, "Would you ever do this for P&G?" The name of the company was Luxury Lab. I hung up the phone and said, "Our new name is L2," and pivoted into consumer products. At Section, my online education started up. We initially thought we were going to be the Netflix of business, that it would be short form videos for B2B. That was going to be so expensive to produce that kind of content, so much of that content was available elsewhere.

that we pivoted straight into online education, focusing on upskilling the enterprise for AI skills. So I don't think I've had a business where we didn't pivot. I think you just look at the data and you are thoughtful about what are the opportunities. And there's nothing like facing the enemy

There's just no market research like launching a business and seeing what people are actually willing to pay for to inform your decision-making. And a lot of times clients will come to you and say, we'd love it if you could do this. And so that it's, you'll get signals from the market. And what I would suggest is have a solid board and have your kitchen cabinet of people that you can propose stuff to and bounce stuff off of them. And also talk to your colleagues and your employees. I'm trying to think if we've done a,

pivot here, I guess we're sort of, you know, we had our adventures in television. Now we're kind of, I wouldn't say all in on podcasts, but I think we're devoting the majority of the human capital of PropG Media now to podcasts. So the market is an unbelievable muse and advisor. And you just want to surround yourself with smart people who can help you interpret the data and make sure that you're not, you know, not speaking to yourself. It's very hard to read the label from inside of the bottle sometimes.

Welcome to First Time Founders. One of the most promising use cases for AI is code generation, that is doing the work of a software engineer. Already in the US, an estimated nine in 10 software developers are using AI coding tools. My next guests created one of those tools. And less than two years after launch, it's already one of the most popular AI coding assistants in the world.

Last year, they had less than 1,000 users. Today, they have more than 600,000. Now, after raising a $65 million funding round at a $500 million valuation, they're looking to take over the industry. Next up, compete with the likes of Microsoft and OpenAI. This is my conversation with Varun Mohan, CEO and co-founder of Codium, and Jeff Wang, Codium's head of business.

Varun, Jeff, welcome. Thanks for having us. How was the flight? You just got in today, right? Yeah, we took a red-eye and we, I think, slept one hour each maybe. Oh my god. And then when we got to the hotel earlier today, maybe we slept another hour. So we're in perfect condition to do this podcast. New York hotels are super fun. You get like this matchbox room, you know, you're right next to the wall every location you are. So, perfect. Well, I appreciate you being here and I appreciate you joining despite...

getting one hour of sleep last night. It's like two, I suppose. Yeah, yeah. Good enough. So you guys are the second AI company that I've had on this podcast. The previous one I had was an AI for finance company, more or less replacing bankers, though his argument is that he's not replacing bankers.

I'm just going to start this with the same question that I asked him, which is, are you guys at Codium replacing programmers? Our vision is actually to give developers the ability to dream bigger. I know that that sounds very vague, but I think there's one way of looking at it that we could just go out and replace low-skill labor or low-skill developers. But I think that that's not a very rich idea. We want to take the best developers and make them 10 times as leveraged.

And the reason why we think that this is only going to lead to more developers existing is unlike other professions, there's no limit to the amount of technology the world can actually consume.

Yeah. Right. I don't think you would ever just be like, hey, guys, stop making technology. If we can actually make it so that the next great invention happens 10 times faster, we will just only get 10 times the amount of invention in the world. So are programmers generally fans of yours, would you say? I hope so. Yeah. I mean, our message is not to replace developers, I would say. The way Varun actually got me on board was he said, like, everyone that touches a product, almost half the people...

continue using it. So when you have that, you know there's something there, you know there's some value, right? And some excitement. Is that higher within the industry? I mean, 50% retention basically of the customers. Do we know what it's like for other AI tools or is it too early? Yeah, I think for a lot of consumer products, it's usually in the, if you can get it above 10% at the long tail, that is considered very good. That's largely because consumers...

you know, unlike companies, they, you know, if they get bored of something, they throw it away very quickly. And these products are not collaborative also. This is like a single player product. So it's actually very easy to churn off of the product because you aren't chatting with other people. So it does mean it is providing a lot of leverage and developers stay more in flow state using products like Codium. Mm-hmm.

But yeah, hopefully we can make it even better. I mean, you're up against a lot of different tools. I mean, I feel like code generation was one of the first things that people said, oh, AI is going to take over this thing. But I feel like the biggest competitor in your space is GitHub Copilot. What does the competitive landscape look like in the AI code generation space?

There are reasons why we are able to compete directly with Copilot. One of them is that we have full repo context awareness. So as the user is typing, the results are highly personalized. And we're seeing a 30% to 40% boost in accuracy just having that code base be available and giving tools for the user to point to what they're working on and trying to figure out their intent. So just the quality of our code suggestions are very competitive.

And then the thing that we're really kind of trying to lean in on, though, is our ability to deploy onto like a private server. And people can host Codium inside their company or their work environment. And for example, if you're like in the defense space or like the finance or healthcare space, they can't use Copilot at all. And that's kind of where we are focusing right now, our efforts. But then we have some things down the pipe where we'll just be competitive even on the cloud front, too.

What does it look like for a programmer? I mean, it sounds like you're sort of typing in and then it gives you a list of things

auto suggestions for what comes next? What does it actually look like if you're a programmer? And please explain it to me as if I'm five, because I don't program, I'm not a coder. Yeah, so a little bit about the way software sort of gets built. So developers write code in what's called an IDE. It's this application that enables you to debug code. So if there are bugs, you could run it. You can actually see what the errors are, iterate on it, right? And then right before...

the code gets pushed into production, it goes through a review process. And other people in the company take a look at the code and actually review it. And then after that, it goes and it actually gets deployed into production. And it's on a website or whatever, where end users can actually touch the product in the end. And right now, where Codium is mostly focused is in the ID

So that's where developers actually write the code. It provides value in multiple ways. So as developers are writing code, it fills in, passively starts filling in more and more code. And because of the fact that we actually do train our own models for that passive AI, we actually found that around 50% of all software that is getting committed

by a developer is actually accepted and generated by Codium. So that's the amount of leverage that just autocomplete is providing to end users. But to add to that, we also provide a couple of other pieces of functionality that is super valuable despite what level you are as a programmer. So you can even chat with your code base.

And this probably seems like a very basic piece of functionality. But when you're a new developer and you're coming into a company and you have millions of lines of code, it takes a while to actually onboard onto that new code base. And what we're finding is even at the largest enterprises, the time it takes to onboard onto a code base goes down from four to six months to four to six weeks with a product like Codium.

I want to focus on how you started this company. Varun, you were a software engineer at a self-driving vehicle company. So you were kind of working on AI there, I feel like. And you originally had an idea to start a company not for writing code with AI, but for something called GPU virtualization. Completely different company from Codium. It also had a completely different name. The name was Exafunction.

Can you explain the first iteration of this company before it became what is now known as Kodium? So to add a little color there, so I graduated from MIT, worked at this company called Nero. It's an autonomous goods delivery company. And actually a lot of the learnings that I had from autonomous vehicles are actually making their way into the space we're in right now. And maybe to paint some clarity on that, in 2015, TechCrunch basically wrote, this is the year of AVs. And in 2024 now-

The quote is, is this the year of babies? And you can see how there are probably going to be a lot of parallels to generative AI where we are going to severely overestimate what is going to happen in a year. And one of the cool parts about generative AI is how easy it is to make a demo, but it is tremendously hard to make something production ready. And if you make a claim that you are going to get rid of a developer, that is a massive, massive claim. And in fact, I would actually argue that is a harder problem than autonomous vehicles.

Because ultimately, if you look at autonomous vehicles, all you need to do is press the accelerator or decelerator or turn a steering wheel. Think about the number of different things a developer actually needs to do. So I actually led a team to build large-scale deep learning infrastructure. So how do you run these models at scale?

And sort of left the company with this vision of deep learning and the idea of running these large models was going to affect many, many industries. And we had a small team of people. We had eight people managing upwards of 10,000 GPUs. We managed close to 20 to 30% of an entire data center. And we worked with a lot of these large autonomous vehicle companies when we started this company, Exafunction, because our mission was how do we make it easier to run deep learning models? But

But what ended up happening, and this is where startups can always get disrupted. And it sounds silly to get disrupted. We were making seven figures in ARR. But we realized actually most of the models would probably become these transformer-based models. And these are the models that underpin the GPTs, the models that OpenAI has. What is a transformer-based model? So the basic idea is, so we now know of prompting, right? You know, use ChatGPT. You pass in a prompt and notice how it streams tokens one at a time.

That's actually a property of these models that are called transformer models, that they are what are called autoregressive. They actually generate one token at a time.

And this is very different than a lot of other classification tasks in the past, where you pass in an input and it just gives you the entire answer all in one shot. But this actually like slowly generates the entire thing, sort of one word, one token at a time. And we started noticing actually that that was how a lot of models were starting to look like once OpenAI came out with GPT-3. And the beautiful thing about the model

that is truly crazy is because of the way it is trained, it can do it in an unsupervised way. So one of the things that was very different about models in the past is you needed a lot of label data. But these models are trained on the entire public internet. The label data is the internet. So because of that, you suddenly got these models that could take in

basically trillions of tokens of code or text. And this was not possible in the past. And this created these new sort of generative models. And in the middle of 2022, we had this business that was a GPU virtualization business. The idea was we made it simpler to run applications on GPUs.

And we found out that most applications would probably be these transformer models. And if all of what we were doing was running transformer models, we would largely become a commodity because they would become a race to the bottom, right? It would be the equivalent of asking, like they would ask Varun, how cheaply can you run this model? I'd say, I can do it for a dollar. Then they would ask Jeff how cheaply you can do it. He'd say 50 cents. And we'd go back and forth until no one made any money.

And this is the commodification of this entire space. But what we did see was we felt that this technology would be like the early coming of the internet.

there would be a brand new set of applications that would be created. And we were early adopters of GitHub's product, GitHub Copilot. And we thought that that was just scratching the tip of the iceberg of what the future would look like. And that's where Codium sort of came about. But it was, you know, as you can imagine, a very rough experience because we basically said, buy to all the revenue that we had. And we had to start all the way back down to zero. So, I mean, that to me is like,

the mother of all pivots, where, you know, not only are you changing the entire business as you know it, you're also doing it at a time when things are going really well. I mean, you said seven figures ARR. My understanding is that you'd also raise $22 million for this company. Like,

Things are going right. And then you turn around and you tell all your employees, actually, we're going to do a whole different thing. If I were a software engineer who worked at big tech and had quit to go work at Exafunction and the CEO told me that, I'd be a mixture of pissed off, freaked out, concerned, and

How did you rally the team and how were you able to make that pivot so successfully? So I'll say a couple of things about the composition of the team. Largely, they were people that we knew. Okay. And that's actually very important because they would be people that would go into the trenches with us. Yeah. They were people that knew the caliber of people that both me and my co-founder were. And also on top of that, we picked a problem space where we were all passionate about it. Mm-hmm.

And I 100% knew at the time there were products like Midjourney that were taking off. You know, a small team of people, eight people that were making tens of millions of dollars in revenue. And frankly speaking, when we decided to pivot the company, we knew for a fact none of us were that passionate about image generation, despite the fact that it is a very cool area. And if we had picked it, our team wouldn't have had, I guess, the mental fortitude to dig deep enough to actually build the problem space. And then I guess the sort of third part is actually that

We actually were able to take a lot of the infrastructure expertise that we had as a company to actually go out and build the application significantly faster. We were very quickly able to train our own models and run them at massive scale. And right now, Codium is one of the top five largest generative AI apps in terms of text in the world. And that largely is because the original composition of the team was these people who are effectively GPU infrastructure experts. But all said and done, everything that I said, it comes down to you need a very truth-seeking company.

And at the time, even if we were at seven figures in revenue, I did not know how we would 10x the amount of revenue. And we could continue to lie to ourselves and have a slow but certain death as a company, as the technology commoditizes, right? And we all run the same kind of models.

Or we could just say, hey, there is a high probability that we will die, but there's a space that we could be very passionate about and it could be very big. I think we just decided the latter was the more rational choice. It is very hard to make, but in retrospect, it was the obvious choice, right? Did you have any data showing you that you were going to crash and burn at that point or was it just a hunch? And part of the reason I asked that is because if I were your investor, I would want to be like,

Oh, yeah, yeah. It's very clear to me that you guys have to pivot. We were cashflow positive then. We were cashflow positive then. So you just had a feeling. We had a feeling. It's just because, and this is a little bit of a curse of being a venture capital, venture-based business. Right. If you're making millions of dollars in revenue, that is not a venture-backable business. Yeah. And if we can't figure out a path to get that to 100%

then that is not a business that we could build. We could continue to keep it as an 8% to 10% team. We have another mentality in the company beyond being truth-seeking that we are a very lean company. By the time we raised our Series B, we had barely spent our seed round. And I think that's just because we don't think capital is a limiting factor in building a good business. You have to build a great product that customers love. And that is usually not just you add more money. We'll be right back.

Support for this show comes from Constant Contact. The internet is a funny tool. If you run a small business, it brings countless new ways for you to get your name out there. So many, in fact, that actually leveraging those channels of communication can get overwhelming fast. It might even feel like you need a marketing degree and an extra day of the week to get any movement at all. That's why Constant Contact does the heavy lifting for you.

Constant Contact's award-winning marketing platform has powerful tools that make it easy to grow your audience, engage your customers, and sell more to boost your business. In just a few clicks, you can launch a marketing campaign that's tailored to your business and goals. That includes email, social, SMS, and more. So you can sell more, raise more, and fast-track your business growth. And you can count on Constant Contact's award-winning customer support for guidance along the way.

So get going and start growing your business today with a free trial at ConstantContact.com. Constant Contact, helping the small stand tall. There's a lot of things you might say when your small business has a problem. You've got to be kidding me. Come on. Well, I didn't see that one coming. But that won't get you the help your business needs. What you should really say is something that can help. Like a good neighbor, State Farm is there.

State Farm agents are ready to help you with your claim to help you get back in business. On the phone or in person, your State Farm agent is there to help. Like a good neighbor, State Farm is there. Support for this podcast comes from Huntress. If you're a small business owner, then the threat of hackers isn't just a threat. It can affect your livelihood. Small businesses are easy targets for hackers, and Huntress wants to give businesses the tools to help.

Huntress is where fully managed cybersecurity meets human expertise. They offer a revolutionary approach to managed security that isn't all about tech. It's about real people providing real defense.

When threats arise or issues occur, their team of seasoned cyber experts is ready 24 hours a day, 365 days a year for support. They provide real-time protection for endpoints, identities, and employees, all from a single dashboard. It's because their cutting-edge solutions are backed by experts who monitor, investigate, and respond to threats with unmatched precision.

We're back with First Time Founders. One of the things you mentioned is that you're training your own models for

I mean, this is an AI application, but most AI applications that I am aware of, they're purely building the application layer. And that is, they're basically using someone else's model, usually open AI, and they're tweaking and they're building off of that model and creating their own application. You guys are different. It sounds like you guys are, I don't know what you'd call it, full stack AI from the model to the application. Is that right? That is correct. Actually, even at the kernel layer,

We've done some, like we've rewritten some code, even at the infrastructure layer, so that we could set the models on top in an efficient manner. And the reason we have to do that is because of the latency issues we talked about. For example, if you are a passive AI and you take, let's say, one second to show up the suggested code, people are just going to stop using that. They don't want to get out of flow state and pause and wait for results. How many other AI startups are doing, what should we call it, the full stack, the...

infrastructure to application. Are there many other... I just, I mean...

Off the top of my head, I'm like anthropic, like open AI, but I guess they're mainly kind of infrastructure layer, right? I mean, isn't this super rare? I think in my mind, there's maybe a couple of unique things about code that make it so that you can actually do this here. And you're totally right. Most of the large companies that are even successful are largely built on top of an API. But we genuinely felt to build a best-in-class app here, we needed to become vertically integrated. And

And for us, it was also not a complex thing for us to do in that we have the technical talent inside the company to actually go out and do that. Maybe one of the unique aspects of code on why we can do this is code can actually be run. Like, let's say I am a legal AI tool.

and I'm redlining a bunch of documents. The only way to know if that is good is for a human to go in afterwards and actually take a look at it. For code, you can actually, if you make an edit to a code base, you can actually run the code and validate it is doing the right thing without a human in the loop at all. And what that means is there are ways in which you can close the loop in intelligent ways that you can actually, if you specialize on that application and you are vertically integrated, you can build an even better app for code.

And we've taken advantage of this in many, many ways. And we realized if we didn't do this, we'd be shooting ourselves in the foot. And I'll give you even a simple example of how this manifests itself. Right now, I mentioned this. Codium processes over 100 billion tokens of code every day, which is over 10 billion lines of code every day. If we pass that through OpenAI, we would have gone bankrupt.

And there was a recent article from the Wall Street Journal. And why is that? Sorry, because you'd have to pay for it. Because it would be too expensive. Exactly. Be too expensive and not the best for our particular application on top of that.

And if you look, there was a recent article on the Wall Street Journal about how GitHub Copilot was spending tens of dollars per user per month. That's actually because even GitHub, Microsoft's product, is not vertically integrated. They are relying on external models to build their application. And we view that as, hey, the model and the product and the infrastructure are so critical to delivering a great experience. Why would we not have control over every piece of that? My understanding is you guys don't use...

the actual data of your individual users. So how is the model getting trained? Yeah. So we actually do sort of two different things. And I'll let Jeff add on how this affects our enterprises and the customers that use the product. So first of all, we use permissively licensed code that is available on the public internet. And we also attribute it on generation time. So we actually take

sort of copyright and licensing very seriously as a company. But on top of that, when we release product from our user data, you're right, we don't take the data from our users, like let's say when they're auto-completing stuff and copy that code and put it into our training set. But we can see, hey, users are accepting these types of suggestions more and these types of suggestions less. So we have preferences on what users and humans like

And that actually informs us to actually build products that are better and more willing to use. And then that actually has a virtuous cycle in that now people are willing to try more complex things on our product because the easier things, they have high confidence that they work. And suddenly the frontier of what we are actually able to experience as we are able to give the user increase more and more, largely because we have a product that is so well beloved. We now have over 600,000 users that use our product. Yeah, it's unbelievable. I think one thing we might've glossed over in the beginning was...

Kodium made a very conscious decision to make the product free for individuals. And if you think about, you know, Varun just said Copilot loses $10 to $20 a month per user. That's after they've been paying a subscription too, right? So having that infrastructure background, being able to make it efficient to deploy these models, and then giving it out for free for individuals is

allowed us to build this very large user base, probably the largest user base for a coding assistant that's free. And just the data here, you started out last year with less than a thousand users. By the end of the year, you had half a million. Yeah, and we have over a million, millions of downloads across all the plugins. And the reason why that's very important is because of what we just talked about.

If we roll out multiple models, if we are changing the temperature or the thresholds here and there, we are getting so many signals as to what is the appropriate settings to tweak. And I think every hour we're getting like a million signals. So we can run all these experiments on these free users of all these models we train to really, really get the best results that you could possibly get. And then only when we've validated, like, okay, we've trained this model. This is better than all the other ones we've trained. These are the settings that make the best results.

then we could deploy that to our on-prem or enterprise users, right? Because after we've deployed it, we can't really get that much more information from it. It can be hosted even in an error-gapped environment. So I think that's a big element of why we are successful, of being able to deploy these models. Because if somebody's trying to start from scratch right now, how are they going to know that their model is good? How are they going to tweak the model, right, without a very large user base? So that's part of the secret sauce, is having that free user base.

Yeah, and it's still free, which is what I find pretty fascinating. And you said, I was just reading your mission statement, you are, quote, committed to having a free tier forever. The natural next question is, you know, how is it going to get properly monetized? And how are you going to maintain that?

that free tier into perpetuity? I think people underestimate kind of the demand for both just the on-prem instance, but also some of the Teams features we add on the SaaS product. So we have a free individual tier, but if you...

create a team and add onboard users to it, that is a paid product. And there are things like analytics and actually bigger models and seat management that are going to be better than the free product. But we are committed to making the free product the best coding assistant out there, no matter what. So whatever other coding assistants come to market, we will make sure our free product is still the best one. We want to make sure, you know, disincentivize others from entering, but we want people that always have the option. We don't want them to force, get forced to buy a copilot, for example.

And then maybe one thing to add to Jeff, like we monetize enterprises, right? So we have some of the largest Fortune 100s, F-1000s, even over 10,000 developers on our product. And those companies, obviously they want security guarantees. They want personalization for many, many repositories that exist. And they also want support across all source code management tools.

Less than 10% of Fortune 500 companies are on GitHub Cloud. And that is the competitor that we have. That is GitHub Copilot. And they have committed to making their product differentially better if you are on GitHub Cloud. So we want to take the approach of almost being Switzerland. We don't care what programming language you write. We don't care what IDEs you use. We don't care where you store your source code.

And ultimately, we also just don't really care what seniority the developer is. We will provide the maximum amount of leverage there. Whereas a lot of the larger players in the space are focused on being tied to another brand. And the reason why we don't think that that makes sense is we think AI is such an up-leveler. We think it deserves to be in a category of its own. You recently raised $65 million in a round led by Kleiner Perkins.

It valued you at half a billion dollars. Congratulations.

What is that money going to be used for? I think the way we would like to think about it is we have ways of spending cash, not only to train models, but to also make it so that we can build a better user experience for the end user. But one of the cool things for us is our product has such high ROI that we think that there will be a real payback period on that. Enterprises and companies will see enough value there that they will be able to eat the cost that we will need to spend upfront. But also on top of that, we want to spend a lot on making sure that we can become better partners for our customers.

We're onboarding some of the world's largest companies and we're onboarding tens of thousands of developers. That's going to take a little bit of effort to make sure that we do that properly. One company that we're working with right now that we have a multi-year engagement with, they account for 0.15% of all developers in the world. Just that one company, right? So, you know, I'm a little bit of a different type of founder in that I do not like the idea of spending money unnecessarily. But if it comes down to we are doing it because we make our customers more successful and our users more happy, we'll do it any day.

How have you guys restrained yourselves in terms of spending? Because, I mean, yeah, the story, the narrative in AI has been that it is an arms race. And the thing that I hear about in the venture industry, or at least what AI founders are being told, is just go out and raise a shit ton of money, as much as is even possible. One, because you just want to develop a war chest. And two...

You want to get the headlines. You want to be the AI company that's working on code generation, the AI company for finance, whatever it is.

So I guess sort of two questions for me here. One, is that accurate to your experience? And two, how have you been so, you know, responsible in terms of spending while you see all of these headlines of other companies spending so much money on talent and training their models? I think Varun mentioned earlier that we run very lean. And the reason we're able to do that is we hire people that are

generalists or ex-founders, and they're capable of doing many job roles at once. So our company size is probably a little misleading. It's probably way more effective, not just infrastructure, but the headcount also. What is the headcount? So right now, we'll be almost about 55 at the end of the month, I think. And the thing is, the people we hire...

are able to just slot into different roles almost out of moment's notice. It's like, oh, we don't even have a marketing team, for example. The product's growth has been organic, but we do want to do some marketing experiments to make sure that we are getting ourselves out there, just as an example. And then people, you know, randomly, someone will be like, I have an idea. Okay, go do it. And then we all of a sudden have a Google ad strategy. All of a sudden we have a bunch of blog posts. We're on podcasts like with you and Elsa. But I think that my point is, yeah,

part of the function of not spending the money is just being very, I guess, practical of where the money goes and running lean. And I think when there is a moment that says like, hey, we need to train a much bigger model and we need to spend this much money, we're all for it, actually. We actually are very conscious about what the ROI is of everything we do. Yeah. How do you maintain that culture of leanness and what would be your recommendation to other companies that are maybe not as big as you, but trying to be?

This is probably the hardest problem that we're trying to solve right now. Because we want to hire people that are very good, very technical, maybe they're generalists like I said earlier, but it's very hard to hire those people. So actually, this is probably one of the things we focus on in the next month, is what is our recruiting strategy? How do we hire the best people? Maybe part of that is getting Codium's brand name much more aware.

Maybe it's like a big push of user adoption. Maybe we're just going to have to be scrappy and be very creative of how we hire people. For example, I don't know if other companies are just like pinging every ex-founder on LinkedIn, but we are, right? So we are trying to scale with creative means.

One other thing is like, as a, as a company, I think culture is, we, we have some cultural principles and we run lean as one of them, but you know, who, who would want to say you don't run lean? So I think, I think, how do you actually live that out? We are a five days a week in person company. So we don't do remote work. We don't really do hybrid work either.

So people see what it's like to work at the company. And, you know, until very recently, our CTO was ordering snacks. And that's not to say that's a great use of his time. It's more just a, no one is high enough to

to not do some work to test a hypothesis. And we don't hire specialists at the company until generalists outgrow that role. And I'll give you another example of this. You know, when we ran that GPU virtualization company, even though we were making money, we never hired a sales rep. And that's not because I don't believe in enterprise selling. No, we have a great VP of sales now at the company. And I think we might have in terms of talent density, one of the strongest enterprise sales teams in the world, actually. Yeah.

But why didn't we hire someone? I just didn't believe that if we added one new person, I would be setting them up for success. Because the reality is, if I could not get $1 of additional sales, I cannot expect someone else to get $10. So this is one of those things where we have a mentality of, we try to do it ourselves, and then we try to eliminate ourselves from the role. We give away our Legos. We let someone else take that over that understands the role much more, but we don't do things prematurely.

And I think there's a tendency across people that the idea of building a scalable organization is really valuable. And I do see that, that you do want to build a scalable organization. But sometimes people get too excited about this notion of org building or fundraising rather than the idea of having customers, having users, because ultimately people that join our company don't care about how much money we've raised as a company.

as long as we will survive. And our customers genuinely don't care. Let's look at it this way, right? If you look at a company as big as JPMC that makes hundreds of billions of dollars, to them, does it make sense if we raised 100 or 200 million? It all looks like peanuts to them. It's all like 10 basis points of the amount of revenue that they make a year. So to them, what they really care about are companies of this size is, are we the best partner for them? Are we the best product to them? And as long as we're laser focused on that, we should do whatever it takes to build that up.

We'll be right back. This episode is brought to you by Shopify.

Forget the frustration of picking commerce platforms when you switch your business to Shopify, the global commerce platform that supercharges your selling wherever you sell. With Shopify, you'll harness the same intuitive features, trusted apps, and powerful analytics used by the world's leading brands. Sign up today for your $1 per month trial period at shopify.com slash tech, all lowercase. That's shopify.com slash tech.

We're back with First Time Founders. Sort of a more personal question, I mean, you started this company

in 2021, sort of just as the AI hype was bubbling up, and you now find yourself at the epicenter of the hottest industry, and you are one of the hottest companies in the hottest industry. Just a personal question for both of you. How does that feel? What has it been like getting used to being the guy in AI? I told this to the company yesterday.

I tell everyone, just get ready to get destroyed. Assume that something very bad is going to happen always. And this is where us having gone through that pivot is very critical. Things are going very well for us as a company. A lot of the reason why we haven't spent a lot of money is now we make money, which is a unique property about a lot of companies, apparently in this space where most companies talk about vision rather than actually building a product that people use. But I just, I tell everyone, get ready for something really bad to happen.

And this is why it's like very important that we hire people that, that are truly in it for the long run. And I tell people this when they joined the company, I think we could be, we could be a company that's worth over a hundred billion dollars. I think we can. And that's largely because of the total adjustable market of what we're building and the amount of impact that this can have given how important technology is could be massive, but also a series of bad decisions that we make could completely kill the company.

And that will happen very fast. I think this is where us building a very truth-seeking company. And that actually is very hard because people want to believe that what they're doing is correct and they want to embrace psychological safety. And what I tell people is, hey, if you feel something is wrong,

Lean into what you think is wrong and tell everyone. Tell everyone because we should not have pockets of people that want to report up the chain and tell their manager or tell me we have a very flat company or tell me things are going fine. I would much rather hear everything is on fire and have paranoid people at the company than people who are just happily going to work. And this is why I think startups are so much harder than big companies.

It's actually not that, you know, you're taking a massive risk on the monetary side. You still can make a six-figure salary, right? These are not people that are living hand-to-mouth. But the really hard part is the lack of psychological safety. We make, a bunch of people at the company make a series of bad decisions and the entire thing can go to zero. Whereas if you're at Google, you have very little accountability. If your team doesn't perform well, Google makes so much money that you are a rounding error. You will be shuffled to some other part of the company.

You never really need to deal with the impact of your decisions and consequences of your decisions. And that's why it's always a little bit of a funny statement when someone at a big company is like, I work at a startup at a big company. No, you don't. Imagine the idea of you potentially losing your job every quarter or every month.

Exactly. One thing that the way I think about it is radical transparency. And we have a lot of conversations within our company of, you know, let's be super transparent about everything. And even in my personal life, I'm like, I'm just going to be like, totally upfront. That's the cleanest, most kind of hygienic way to operate.

Do you ever feel that there could be too much truth? Do you feel that there's a possibility that, you know, if you're encouraging everyone to tell the truth, tell the truth, be transparent, tell me everything, that they'll kind of go overboard? I think this is the hardest part about a startup. You know, the two things that I say is a startup is really hard because you need to be both irrationally optimistic, because if you're not optimistic, the answer is always Microsoft is going to beat you, right? Right. It's the biggest company of all time.

They have the most capital. They have the most people. They have the most distribution. Why does any company win? But clearly that's wrong, right? There are companies that have beat Microsoft at different areas. We always say this. Oh, they have the most capital. They're going to win. It's very easy from my perspective to say that. Exactly, right? And then HBS, you know, I try to think about what Harvard Business School would say 10 years from now for us as a company. And I tell people this. It's like, you know, we fail.

And what they write is, in a world in which technology was changing, Microsoft had all the distribution in the world. They had this property called GitHub. It was inevitable that they would win. And because of that, they won. They won the entire space. And all of these startups, it was a fool's errand. Why did they even try? And then the world in which we win, they were going to write, in a world in which the technology was getting disrupted,

So materially, there were a set of companies, and one of which was Codium, that had such a technological advantage that there was no way a slow-moving gorilla like Microsoft could even compete. So what I think is very hard. Yeah, exactly. They will write whatever the future looks like, and the history will be written by the victor, and no one will know exactly what the pages of the book look like. And I think the hardest part about a startup is you do need to be irrationally optimistic to believe that you can win.

Because by default, if you don't, you will definitely lose. But then also uncompromisingly realistic, which is that sometimes, actually, there's no point continuing in a particular direction. This is where I think it comes down to what you just said, which is that if you are too truth-seeking and everyone is paranoid all the time, it could lead to paralysis.

And I think there's a fine line there. There's a fine line where one time we lost a deal and we actually every day for dinner talked about it for two weeks in a row for multiple hours. We talked about the implications as a company and that was very useful. But if we extended that out to an entire year, we would not go anywhere. Yeah. So there's, you're totally right. There is a, there's a level here. But what I do feel most companies do is probably err on the side of not being truth-seeking enough. Yeah. They become too complacent with what they're doing.

And, you know, I think the thing to really think about, and this is not true at a big company, which is why you really need to think about it at a startup is you do not win an award for doing the wrong thing for longer. So the sooner you can rip the bandaid off, the better your company will be. You will be way happier that you did it. It is going to be incredibly painful for like a week, but just do it. I love that. You guys have been generous with your time. So we'll begin to wrap up here.

Jeff, I'll ask you this question. What do you think has been the greatest challenge that this company has faced in the past couple of years?

I think the pace at which things switch is really challenging. We relied on some technology features to be the selling point, and then our competitors would come out, and it's like the same thing all of a sudden. And every time that there's a news release for one of our competitors, we immediately go into like a code red and go into a conference room, reverse engineer what they're doing. And this is constant. This is like every month this is happening. Like,

The question is, will this last forever? Will we just always be panicking? And the answer is probably yes, right? I think for people listening to this podcast, you two are the closest thing to AI experts as we've had. You're kind of on the front lines of this.

What would be your advice to anyone who is working in the AI industry or who wants to work in the AI industry? And I think that doesn't just have to be founders, but engineers, product managers, business operators, etc. What's the most important thing that they should understand about AI right now? There's an interesting property about AI right now in that it is actually imperfect.

When you use the products, it sometimes says the wrong thing. And despite that, it is actually very useful in some domains. That's not like anything in the past. When you used the internet and you ordered something off Amazon in 2002, it's not like they would ship you something incorrectly. Or maybe they would do that, but there was no expectation going in. You would buy one book and you would get a different one. And somehow that is still fine right now, which is a cool part of the technology, which is that it actually gets perfect.

It's going to usher in a brand new set of applications as well. But I think the key thing to really think about is not to think about go back from a demo and try to build that today. So think about what products can you build today that are actually imperfect, but still can generate a lot of value. And that is a lot harder than you would think because in a lot of domains, if you are imperfect, like let's say you're reviewing a legal document and it actually completely reviews the document, but 10% of the time it's wrong.

You can't give that to an end customer. So actually thinking about the trade-offs of how good does the quality need to be to ship your product, right? How fast does the experience need to be? In a world in which the quality is not perfect, it better be fast. There's no world in which I'm going to wait 24 hours for something and the quality is imperfect, right? And then if the quality is imperfect and it is fast,

How quickly can I correct it? And these are all important factors of a product. If you don't hit the sweet spot here, you will have a product that's a cool demo, but no one will ever use it. I think this is the most important thing to really think about. Adding AI to just any field that exists doesn't just suddenly make the product usable. It needs to be a product that is useful in its own right. I think if you give away a product for free and everybody keeps using it and a lot of people keep trying to get it,

Then you know there's something of value and there's product market fit. And I think if you look at a lot of these AI tools and AI demos, maybe they're not realizing that. Maybe they're just building something that is a cool demo or it's like a really... It pushes the limits of the technology, but it's not actually building something that people...

want, right? Or will find value from. I think that's maybe our biggest message to other founders or other people that are building products in the area is just think like, if I gave this away for free, is everybody going to want to use it and keep using it? I think that's maybe something people miss. Do you think maybe that's the model that

I mean, I was going to say software founders, but maybe all founders that you should just start with giving the product out for free and see where things take you from there. It worked really well for ChatGPT, right? Yeah. Yeah, exactly. And it worked well for you. Yeah. That's actually an interesting principle that Jeff just said. Obviously, in some scenarios, the reason why ChatGPT could do that and OpenAI could do that is they own the infrastructure. And if another company did that, they would have gone bankrupt. So...

You do need advantages in particular places to be able to do that. But at the very least, if you did burn money and you gave it away for free, if no one runs to use the product, you're probably in a world of trouble. That's a great place to end. Varun Mohan is the founder and CEO of Codium. Jeff Wang is the company's head of business. Guys, thank you for joining us. That was awesome. Yeah, thanks for having us. Thanks for having us.

Our producer is Claire Miller. Our associate producer is Alison Weiss. And our engineer is Benjamin Spencer. Jason Stavis and Catherine Dillon are our executive producers. Thank you for listening to First Time Founders from the Vox Media Podcast Network. Tune in tomorrow for Prof G Markets.