Ship Flawless AI Code Daily

Jake Aaron Villarreal: I'm Jake Aaron Villarreal, born and raised in Silicon Valley, here to take you behind the scenes to share what it's like to be a startup founder in an effort to make our lives better. I'm excited to have with us today Amar Goel, co-founder and CEO of Bito, which does AI code review. Amar, welcome to the show.

Amar Goel: Hey, thanks, thanks for having me, Jake. Excited to be here.

Jake Aaron Villarreal: Well, thanks for jumping on. A little bit more about Amar: He is a serial entrepreneur who has previously founded PubMatic, which went public in 2020, and formerly worked at Microsoft, McKinsey, and Harvard University. He's excited about using GenAI to power the next generation of how software gets built. Um, before we jump in here, where are you uh joining us from today?

Amar Goel: I'm calling in from the Bay Area. So I'm in Menlo Park, California, right next to Stanford.

Jake Aaron Villarreal: When I started this podcast, I had this vision of going behind the scenes to share what it's like to be a startup founder in Silicon Valley, where I'm born and raised. And weekly, we dive deep in the trenches to find the wins, the losses, and the lessons learned with the hopes that it helps you run a better startup. But here's the thing. If you've gotten any value out of these conversations or been inspired at all, hit the subscribe button. And I get it, your podcast feed's probably overflowing. But to stay up to date what's going on in Silicon Valley, stick with us because at the end of the day, we're all just trying to build something that matters and we're better when we do it together. So, hit that subscribe button and follow us wherever you find your podcasts. Now, let's get back to the show.

I love that area. There's a lot going on today. Transformation happening across the world. I happen to have worked at Oracle for a number of years and I'm excited to see today. Today, Larry Ellison became the richest person in the world. Hey, Oracle stock went up 40% or something, which is crazy. Yeah, it's like the one mistake I made when I worked for them, which was I kind of forgot to sell my stock and you know, 20 years later, I'm kind of happy I did that.

Amar Goel: Oh my [laughter] god, good for you. Congratulations.

Jake Aaron Villarreal: Yeah, thank you. So, anyway, uh excited to be...

Amar Goel: Yeah, you know, I, I was actually born in the Bay Area. I was born in Mountain View in El Camino Hospital and uh, and so you know I think in some ways it's kind of been in my blood a little bit to, to think about technology and the startup ecosystem. I mean, I remember in college talking to one of my friends who grew up in Minnesota and his dad worked at General Mills and you know, he said, 'Well, I started a company in college that I was running after, after college, an e-commerce company.' You know, he was asking me about how I started. I was like, "Oh, yeah. I just, you know, kind of started or whatever." And then he was like, "Dude, you grew up in the valley. Like, it's in your blood. It's in the water. You drink it every day. You know, every day when you read the newspaper, uh, the Mercury News," which used to be kind of the dominant newspaper in the Bay Area, "was about this startup got funding and..." It looked, tech was a lot smaller in the, you know, in the 80s and 90s for sure, but at the same point, it was still what Silicon Valley, the Bay Area was about.

And so, it just was kind of imbued the, the, how you, how you thought. Both my parents worked in technology, my dad was a semiconductor engineer, my mom worked in software companies and hardware and SAP. So, it was just kind of... and I guess I was really interested and really excited and started doing programming in high school. Uh, yeah, it's just part of, part of the water.

Jake Aaron Villarreal: Yeah. Well, we work with a lot of companies that are in Silicon Valley that are not born in Silicon Valley, which is the title of this podcast. You happen to be one of the lucky ones that, you know, you might have a different perspective. You know, we saw the transformation of the internet in the 2000s and we saw, you know, AWS and cloud and the transformation of that, what that did for the industry, and then mobile, and now we've got AI, which is just incredible what it's doing and the pace it's moving. So, we've got experience. You took a company public, which had to be an amazing experience.

Amar Goel: Really was. Yeah. Really was amazing.

Jake Aaron Villarreal: When you do something like that, do you have the mindset of 'I kind of built it, I achieved it, I got to the mountaintop, like it's time to...'

Amar Goel: Yeah, I mean I think you know going public was super exciting uh you know for Pubmatic and the team and everyone, all the co-founders who had built it. And I would say it was a little bit like mentally hard after that for a little bit you know, trying to process that. And you know, it was kind of this brass ring you're going after and as you start a company it's definitely an aspiration that you have. And so that was exciting to get there. And then it was a little bit like, 'what now?'

And so I spent some time really thinking about that. And I, I guess what I kind of came to the conclusion was that I thought it was... I really like building things. You know, I always, I still, I have a very bad long-term memory. Maybe that's what lets me keep being an entrepreneur. But I still remember like the fir- looking at the first mockups, you know, of, for PubMatic, you know, like printing those out actually. This is like pre-Figma and all this stuff. So like, you know, looking at the mockups of PDFs and I wanted to print them out so I can make comments and all that. And it's like 2006. But yeah, so like I still remember that in my mind, you know.

So I really like building and I really like getting product out there and seeing kind of people's reaction. It's not like always like the funnest experience in the sense that, you know, it's not like when you launch a product, everyone's like, loves it. You know, a lot of times there's a lot of problems and a lot of things to work through, but I just like that process of kind of building and innovating and creating new things, you know. So, that was kind of one thing.

And then the second thing is like I thought I, I really do believe in hard work and I believe that that's an important part of, you know, achieving things and getting things done. And so I felt like it was really important for my boys to see that, you know, and, and just see them like, my kids see me working hard, you know, and, and building. And you know, and so I think I wanted to show them that. I kind of felt like if I just check out and play golf or, or whatever [laughter], that's really checking out. You can actually do that seriously as well. But, you know, I just I, I didn't want to just be kind of like, 'Oh yeah, I'm just like hanging out now,' you know. So, I mean, there will be a point for that for sure, but, you know, I wanted to keep building and I wanted...

Jake Aaron Villarreal: Yeah, I've got two boys, too, and I know exactly what you're talking about. You know, they're coming in my office all the time asking me what do you do and why are you doing it? And, you know, they make me think about like my purpose and how am I helping the family and helping them. And you know, the other day we were out building a fence and you know, I was trying to teach my kids how to build, and as I was going through it, I started to think, "You know what, what am I really doing here? I could just pay someone to come out and do this." [laughter]

Amar Goel: And totally, totally. And uh, and I'm not super handy either.

Jake Aaron Villarreal: But what I thought about what I was really doing...

Amar Goel: Yeah.

Jake Aaron Villarreal: ...because actually showing him can have, if you build this and you do it the right way. And you know, they were complaining, but at the end we were all pretty happy. And...

Amar Goel: Right. Yeah.

Jake Aaron Villarreal: ...kind of the art of going through the process, you know, and sometimes the outcome is not as important as like...

Amar Goel: Exactly.

Jake Aaron Villarreal: So, you went to Harvard, right? Congratulations. Great school. We've had a lot of people on our show from Harvard. And you know, it's interesting what we've heard is, and what we've learned is that, I don't know what specific degree you got and where your focus was, but there's some courses that really are very purpose-driven about helping the world, about inspiring, you know, careers where...

Amar Goel: But that's a good question. You know, critical thinking is probably a big, big part of help. How to take a problem, how to break it down, how to build an argument, right? How to really research, how to think through those problems, right? So, I think that was one thing. I mean, most of your classes, I think the actual material that you learn, you do forget, you know, 20 years later, right? I mean, I remember taking a course on Sub-Saharan Africa and I remember the professor saying the first day, and this is what I still remember. He said, "If you learn nothing else from this class, remember that Africa is a continent, not..." and you know it's so true actually.

Because I think uh growing, I grew up in America and I ended up living overseas for a long time that really changed my perspective about a lot of things. But growing up in the US, like I think it's very easy to just think of many parts of the world as very homogeneous. And you know, if you think about a lot of American companies, they have the US as a territory, geography, and then they say "rest of world." Like "rest of world" is one thing, you know. And you know, Europe is not one thing, right? There's a lot of different cultures and ways of doing business and that's totally different than a bunch of different countries in Asia and Africa.

So anyway, so that that you know, so you don't remember that much of the content, getting it back to the point of your question, but I think like for you, I know I studied, I studied economics and computer science, but for computer science I think a lot of those, because that ended up being a big part of my career, like that you do remember a lot of those concepts, right? And like you know, how to build products and how to, you know, how to think about scaling technology and, and whatnot. So, you know, I think that part was very applied for me, but I think the, you know, like when I think about English and all that, you know, it's more about, that was probably applied too, like I said, learning how to write well and communicate well, but a lot of the rest of it, you know, practicing those skills and at the time learning makes sense.

Jake Aaron Villarreal: Well, let's fast forward to where you're at now. So, AI is here. You've raised, you know, eight plus million in seed. Is that accurate?

Amar Goel: 35.

Jake Aaron Villarreal: Yeah.

Amar Goel: Well, I think you know it's been a journey for us too as AI has evolved and you know kind of what we are focused on. We actually started you know with some AI chat products you know for developers back in 2022, right around when ChatGPT started. But where we, you know, what we thought was happening was that as AI writes more and more code, a lot of the downstream processes like code review, like testing, you know, like deployment... there, a lot of pressure is going to get put on a, on, on these, on these processes because developers are going to interact with agents that are going to write a lot of that code, and then, or maybe all of that code, and then now they're still responsible for making sure that it goes live with quality, you know.

And so we started by building a code review agent. And really that code review agent is fundamentally built on a focus on really deeply un- people's code and engineering teams understanding their, kind of, their code and their systems so that then you can review their code well. And so what it does is it integrates into like GitHub, GitLab, Bitbucket, you know, all the tools where engineers tend to kind of do their pull requests and, and review their code. And it tries to, you know, analyze the changes they've made, analyze their code deeply, and then post feedback, you know, and it's, you know, try to be really spot-on about like high signal, kind of low noise feedback. And now it's also available right in the IDE. So if developers want to get feedback as they're coding, they can, they can do that. And, you know, in just a minute or two get a bunch of feedback before they even get, when their code gets reviewed by a senior engineer. So anyways, that's kind of what we're, we're focused on. And you know, we're really going after and working with a lot of enterprises and seeing a lot of traction.

Jake Aaron Villarreal: That's really cool. Yeah. You think about code and you think about code generators and you think about, you know, engineers that today, you know, they're using these tools to help them do their job better or faster. And, but the code review around the quality isn't something that you think too deeply about on the front end or on the outside looking in.

Amar Goel: Yeah. Yeah, I mean I think you know what we have believed is that the need for this solution is going up as developers are using AI to write more and more code, they're less and less familiar with the code that they're writing. You know, if I mean... though sometimes the way I think about it is, it used to be that if you asked a developer like, "Hey, I have a question about this feature that you shipped," and you could ask them any, and they, they know they wrote each line of code so they know they can answer the question. Now they're like, "Uh, let me check. Like I have to go check the code because I didn't write it. You know, I scanned it maybe."

And sometimes developers who I think are not doing a great job are like, "I didn't even really scan it. I just kind of saw that AI outputted it and you know, I ran a basic test and looked like it worked so I'm good." You know? And so now all of a sudden like we're hearing engineering teams say things like, "Well, when we have a bug in production, it takes us forever to figure it out now because nobody's actually that familiar with the code, right?" So it's a little bit of, it's a little scary if you think about it. You know, it's like you kind of have this code getting shipped into production and it's nobody really knows what the heck is, is going on.

So, so anyway. So I think you know, we see the need to like more and more deeply understand your codebase as really an antidote to that, to really help be able to provide that important detailed specific feedback. So, are shipping high quality code because at the end of the day, any system at scale does need to work. It needs to work. Reliable debt is building up where bugs are and you know nobody wants to be interacting with your favorite website or app or whatnot and have it crack.

Jake Aaron Villarreal: Yeah, it makes sense. I mean, it's similar to just not code, but just content. So, you know, you can be a writer and you can have a lot of content created and it looks great and you know, you're going to put this social media app in.

Amar Goel: You're totally, totally. I was actually just reading an article today about kind of like, Sam Altman, I think, has been commenting on like, "Oh, there's a lot of the internet is becoming like AI fake," you know, it feels like AI generated fakeness, you know. Reddit, a lot of fake threads and bots, and Twitter and or X or whatever. And then even just the internet. And some people were kind of saying that's the pot calling the kettle black a little bit. He's the one enabling all this. But, but maybe that's an, that was an aside, kind of funny comment.

But what people were kind of saying is like the, you know, there, there's a lot of like low trust places that doesn't matter the slop, AI slop that they kind of create. And so whatever, like you just said, you know, there's all these mistakes. But if you are a high trust brand, if you are the New York Times or if you are Conde Nast Traveler, you know, you may be using AI tools to help your writers write better, write faster, whatever. But people still expect and require that the end product you create is high quality, right? They're not willing to accept that, "Oh yeah, you just made a bunch of factual mistakes in this article." So, I think it's the same for software products, right? If you go to amazon.com or if you are, you know, salesforce.com business user, you expect that product to work and work accurately and not have bugs, you know. It's not like some one-man show who just kind of vibe coded an app, you know. So what we kind of like to say is like, vibe coding does not equal vibe engineering, you know. Like what vibe coded app is production-ready app? And so we're trying to help people...

Jake Aaron Villarreal: A lot about vibe coding and agents...

Amar Goel: Totally, and I think that's great. It's a great first start and it's certainly helping. I'm not trying to say vibe coding doesn't have its role, it absolutely does. That, but like you said, that's more at the prototyping stage. "Hey, I want to show you how this idea works" or "I want to get something out there in a basic form and see what people say about it." But then to take that...

Jake Aaron Villarreal: Yeah, you know, we're in the business of hiring and helping companies hire engineers and product folks and sales, go-to-market teams. There's some fears around AI and how it might replace the job. But we're seeing a lot of companies that are hiring, but they're only hiring if you actually have AI tool experience. How...

Amar Goel: Yeah, I mean what we see happening, and what I think is happening, and it's still really early days, right? And a lot of this depends on how fast the models are improved. You know, that's kind of the under intelligence, right? We had storage, we had compute, now you have sort of intelligence as a service, right? And so like how much, how much cheaper, faster does that, that get? But I think what we're seeing is like, you know, critical thinking is still super important, right? Like communication skills are becoming more important, but not necessarily verbal communication, but written communication, right? That's how you communicate and that's how you get something out of AI, right? Is like, you know, you got to break a problem down. You got to give it small and comp-, you know, tokens, not tokens, but small components of what you want it to build.

Like maybe you have a huge feature you're trying to build. It's like, "Okay, well, can I break this down into the five or six key things I need to do and then maybe take one of those and break it down into a couple pieces, and then feed that to the model? And now I get back quickly what I need," right? And so now it all kind of works because I broke down the problem. But, you know, some engineers are doing a really great job at that. Other engineers are not. They're not thinking in that way. They haven't adopted those work practices. They're not great writers. They're not... So I think that's a skill that's really you know kind of, kind of coming up.

And in some ways this is kind of the next level abstraction, right? If you think about 30, 40 years ago, you know as an engineer maybe 40 years, 50 years ago, you had to, you know, assembly language, right? Assembly language was what language like, you know, any programming language was uh compiled down to assembly and then that was down to machine language. And so now you have assembly. And so people had to know assembly. Now nobody knows assembly. And now you have these higher programming languages that don't require you know, you to know things like memory, garbage collection, and you know, thread management. There's maybe a next level of abstraction which is like, "Hey, engine, you know, maybe English is that programming language."

But just like with a higher level language, you had to be able to clearly think about how to structure that, what classes you need, what modules, and then if you did that wrongly, you wouldn't scale. That's the same maybe in like how you communicate in English to the LLM to get the right program. So I think it's really just that next level of abstraction. And so I think it's both the communication skills and maybe just one other thing. I think you still need to understand how the underlying system works, you know. So you can't be an engineer and be like, "Well, I don't really understand anything about scaling or you know, observability or how I should think about debug, you know, what a data structure does." I mean, because you do have to debug those things. You do have to think about trying to, trying to scale.

I mean, just give you an extreme example: Google. You know, sort of famous for like super fast search and doing it you know, really low-level optimizations, like how the disks had to be able to make search super fast and super cost-efficient. If you were just an engineer that sort of said, "Hey, I gave an LLM to build me a search," you wouldn't really understand super deep level optimizations to make. So I think that those are still skills that are...

Jake Aaron Villarreal: Yeah, there's a lot of complexity and a lot of innovation going on that everyone's trying to figure out, and learn new pricing models for compute and training.

Amar Goel: Yeah, I mean we're really finding a lot of success with kind of mid-market enterprise type companies, right? So, you know, I would say broadly 50 to five... mostly they tend to be software companies or, you know, services companies that, you know, are doing software. I mean those are kind of like our you know, customers. And so folks like that who you know, are kind of leading edge, cutting edge customers, companies like Gainsight, Fetch, Whatfix. That these are companies that are really building great software but need high quality and really want to make sure that product that's shipping well done. But not, we're not yet going really after super large companies and you know, we're not also really going after startups.

Jake Aaron Villarreal: Yeah. Well, you had a company that went public. What are some of the key...

Amar Goel: Yeah, I mean I think you know, in some ways you, you know, with AI people say everything's changed. And I think you know, the market is moving fast and you know, companies are going wide and trying to do lots of stuff. So I think that feels like it's changed a little bit. At the same time, I would say not a lot has changed, right? It still requires like going deep. I think doing things well, you know, focusing in on a problem and really how do you build a great, great quality, high quality product, you know? And so you need a great, a great team to do that. You need to have good processes about how you guys work and, and how the team works together, how you communicate. So it's a little bit of like, everything's changed and nothing's changed.

I mean probably, you know, the AI speed of the model improvements and the number of models, I think that has been probably the, the biggest change, you know what I mean? Like that you're like, every year you're seeing like meaningful changes in how, how sophisticated are the solutions and how sophisticated are the problems that you can solve, you know? And I mean, we won't get into all the technicalities of the models, but just you know, they're getting better and faster. And so now you can do more and more stuff and just better. You know, I was just, you know, I think about our code review agent, you know, the types of issues we're able to find and the way that we're able to explore and understand your code. It's just a step level improvement, you know, today from a year ago, you know. And some of that's the model and some of that's how we analyze your code, right?

And so I mean, I kind of like to say we probably added points of IQ, you know, to the agent, you know, so somebody might have gone from 100 IQ to 150. I mean, you're talking about sort of average to genius level, you know, in the last year in terms of what we can do in code review. Or, you know, I was, I was messing around with Gamma the other day. It's like this kind of presentation use. And I remember I was using it in April to try, you know, build some slides and I was kind of like, "Hm, you know, okay, interesting. Like I, I was able to use a couple slides, but I had to make a lot of changes and all that." And then I was trying it again a couple days ago and like, the, the quality of the output and what it was able to do was like so dramatically different, you know. Like literally I gave it I think a five-line thing about what I wanted and it outputted like 10 tons of detail and all that. And so it's just, maybe I gave it 10 lines of like detail and interesting and images and all that. So I was like, "Wow, that is like a step level change."

Jake Aaron Villarreal: Yeah. Yeah. That's crazy, you know, with AI agents becoming so prevalent now. You know, I was on this earnings call last night with Oracle just because I was wanting to see what was, what was going to happen with the stock knowing that it might go up like $20 today, and it went up so much more than that. I was like, you know, firing off text messages and emails to my friends saying, "You guys, you got to jump on this at like 6:30 in the morning, Pacific Standard Time, because Oracle is gonna like go up. It's probably going to go up maybe 20 bucks and, and so get in now." It went up significantly more than that.

But the reason why I bring Oracle up was because, you know, we were looking at, you know, what do these big enterprise companies do that have tons of engineers and tons of developers and they're, they're building enterprise applications for the world? Oracle, uh, but as a user, you get these applications and what do you do with them now? Like it's just software and apps and dashboards you log into and you put information in. But with AI, what they've done, what they're doing, probably going to get this wrong, but they've basically created this AI studio, what it allows you to choose, you know, within whatever workflows you have within your applications, let's call it CRM or marketing or finance. And then being able to choose the AI agents that they've already created these templates. And you just drag in this one. "I want to use you know for my back office this app, this AI agent, and this one for my front office, front end office and or front office." And ultimately it's, they're being integrated and orchestrated without having to actually do the building.

So they're like code gener- apps or AI agents and then you can make tweaks to it. But I was like, well, where, where is everything going to be heading if, if you've got big enterprise apps that have the agents that are now being offered by the companies that have built the apps? And then yeah, you can do modifications to those, but then if there's like potentially voice AI where you can just say, you know, "I want an agent that can do XYZ in my financial, you know, financial app," and the agent will create the code, build the agent, and now you have agents building agents and optimizing how you operate the company. And now you don't need 20 people in, you know, accounts payable or... so like that's their, their vision of how they're seeing it, but that's like just one ecosystem of software, you know. You've got multiple other applications and you have homegrown applications. And so there's opportunities for a lot of different areas to develop, but code review is I think a great...

Amar Goel: Yeah, I mean I think the way we kind of think about it is we want to help that code that you wrote with AI, how do we help that get into production, you know, with really high quality and really... and if you think, jump ahead a little bit, like software development teams are going to be writing like thousands and thousands of lines of code per day per develop, you know, a year or two from now, right? Because they're going to work with some agent. It's going to write a couple thousand lines in an hour and now you have a job of like making sure this works properly and checking it out, you know, seeing if it's all good. And so I think that is kind of how [we] think about things.

But I think as the, as software goes faster and faster, you know, the, the average engineer will understand less and less about their code. And so they really need, they're more aware of that codebase and more help that codebase to help. So, I think that's, you know, what we are, are focused on is how do we help you ship quickly? Yeah. I mean, it's maybe like a boring answer, but it's just kind of keep going at it, right? I mean, I think I'm really excited about some of the new capabilities we have coming out and it's exciting time with AI. Lots going on like you said in the valley, building a company, you know, with, with a great team. So, yeah, it's just kind of keep going at it and keep putting one step forward every day.

Jake Aaron Villarreal: Very cool. Well, I want to thank you so much for coming on and sharing your story, and for the listeners for spending your time with me. It means a lot. I'm your host Jake Aaron Villarreal signing off for now. I can't wait to catch up with you all on the next episode. Until then, Amar, the world. Take care. If you like what we're doing, don't forget to subscribe. Leave a review on Apple Podcast or wherever you listen and follow us on YouTube where we go behind the scenes to learn what it takes to be a startup founder.

‍