AI Coding Revolution: Exploring Bito CEO Amar Goel's Insights

0:28Welcome to the AI Chat Podcast. Microsoft, McKinsey, Netscape, and what I believe really makes him a leading voice in the intersection of AI and software development. So welcome to the show, Amr. Thanks so much for having me. Excited to be here. Super excited to have you on the show today. I wanted to pick it off and kind of ask you a little bit about yourself and your background. In technology, you've worked in this space for a while now. Is this something you always knew you were interested in, specifically like AI and stuff, or is this something that as your career developed, you kind of got pushed in this direction you know what what's uh what's your background yeah i mean i think i've been interested in software for a long long time i mean that's kind of i really love building products and building companies for a long time i started a company when i was in college and i kind of used to do randomly kid um but i i would say like ai is newer and the for me in the sense of you know i I'm a former company, Podmatic.

1:25I started, I used to run a machine learning team there. So something that, you know, I would say classically I'm more of like a product type persona. But building technology companies is something that I love to do for sure. So would you be able to tell me a little bit about, you know, what inspired you to start Bitto and kind of this new phase of what you're working on? Yeah. I mean, I think we started actually, we ended up doing a little bit of a pivot. We started with a focus on developers when we started the company, me and my two co-founders. And really our focus was how do we help developers collaborate better?

2:02And so we really wanted to be in the tools that developers use on a regular daily basis. But really, how do we help them, like I said, collaborate, kind of bring a Google Docs and Slack type experience to their IDE? And when everyone was like, oh, this is super cool, I'd love to use it, and then nobody choose it. So then we started, but we started listening to that feedback and we started trying to think about what are the different ways that we could help them. And they said, Hey, could you help us understand code, onboard new engineers? And then that's really how we started spending more time with large language models and AI and really, you know, ended up launching kind of the current, you know, Biddo in December of 2022, which is today is really an AI assistant that helps developers.

2:43So it's really a tool that developers can use. It kind of works in their IDE or their CLI, and they can use it to write test cases, write code, explain code. Actually, the number one thing people use to do is explain code, debug. You can just throw debugging errors right into it. It started out as kind of like a chat GPT in your IDE. Really, where we're going is that we're building all these capabilities that really help them do things in a more automated fashion. So we have agents that let you like automatically document your entire set of files or repository of code. You can write test cases.

3:21So all kinds of things like that. And we have about 100 ,000 developers who are using it today. Very cool. Wow, that's incredible. And I think, you know, big kudos to you for a lot of people I feel like starting a company do not have the, perhaps the vision when something isn't going exactly how the original hypothesis is to be able to make that pivot and take on that feedback. So big kudos to you for taking that and obviously, you know, seeing the reward of that 100 ,000 developers is incredible. What is, I'm wondering if you have kind of like a case example or, you know, of someone that has used this and been really successful or kind of the ideal person and maybe a success case using this.

3:59Yeah. So I can give you kind of two. So we have a very large transportation company that said, hey, we have all these apps that they're running, but they're not well documented. No one's really clear exactly how they work. Could we use Biddo to kind of automatically document them? Because we want to migrate them. They're older apps, and we want to migrate them to more of a modern kind of architecture. Well, first, we need to really understand what they're doing. And so they're using Biddo to just like automatically, you know, document those apps and they have like close to 50 of them. And so I think that's, you know, pretty, pretty cool that they're able to like do, it would have taken like thousands, you know, hundreds of hours and they're, you know, able to do it, you know, in 10 minutes.

4:49And then any changes you make, they automatically can update that, right? So I think that's kind of like one, you know, interesting kind of use case. I'll give you two more quick ones. One is some engineers at Pomatic were actually using it to like, they were saying, hey, how can I use it for testing? And so one of the things that they did was that they actually asked Bido to take some code and then said, give me ways to break this code. And then they turned those ways that it suggested to break the code and they turned that into test cases. So now they added that to the test suite. So, for example, they were like, oh, like it was accepting some input.

5:29The code is accepting input. So they're like the thing said, oh, well, it doesn't really do much handling to take malformed inputs. You know, like, you know, maybe somebody could upload a script that runs, you know, incorrectly. So then they use that to then turn those into test cases. And anytime somebody continues to update that code, they run these test cases to make sure that the code is more hard, you know. Very cool. So, yeah. So I thought that was kind of interesting. And then the third one, I'll give you really quick, is my nine-year-old. He's learning to program, but to be honest, it's funny to watch him.

6:01He's like an AI programmer. So he doesn't really understand how the code works, but AI helps him kind of advance and be a little supercharged. So he basically said, hey, write me a Chrome extension. Him and his brother were trying to trade Pokemon cards. So he said, I want to give you a name of a product. I want you to write a Chrome extension that like automatically searches eBay for those products and brings me back an ordered list by price. And I wrote the whole Chrome extension in like one minute for him. And then he just followed the instructions and set it up. So, you know, those are kind of, you know, three interesting use cases.

6:38Super cool. That's super cool. one thing I would love to ask you and get your feedback on is, you know, from your perspective and some of the things you've seen, right, you launched this thing back in December. What are the ways you see AI really shifting, like even from the beginning of the year till now, right? You've made pivots in your company, but what are the ways you're seeing the landscape of AI shift? What are some things you're seeing on the future? Yeah, I think that's a really good question because I think sort of the laws of gravity are occurring, you know, when it comes to, like, all the AI innovation that's happening.

7:13And what I mean by that, I don't know if you remember, but so far back as November, December last year, like, when ChatGD launched, right? I mean, you're in the AI space too, obviously. Like, it was crazy about, like, every day some announcement would come out, like, the whole world was changing. There'd be somebody would say, I just rigged up this amazing thing over the weekend and they would launch it on twitter x or whatever it was kind of like just seemed like everything was changing like overnight and people were talking about like gonna lose their job in 15 minutes and like oh the software is dead and like we don't need software developers anymore right there's no loser job but you know all these things were out there right and i was like it was really cool to like you could do things really quickly and you could rig up this little like 30 second demo of like hey i took one hour and i kind of did the thing and i think now we're you know here we are fast forward let's say like nine months if you look at twitter you know you look at like just there's a lot of the i announced what's still happening so i'm not trying to say they're not maybe instead of having three unbelievable things happening every day it's kind of like yeah one unbelievable one amazing thing is happening a week, you know, and maybe a year it'll be like, well, you know, one is happening a month, you know, because I think the bar kind of going up and, you know, and we're also in this phase now where to actually turn like that little quick demo that you wrote into like a real product that people can use that actually works repeatedly well, gives you kind of the similar output, you know, that is turning into a lot more work.

8:52And so I think what's happening is that people are realizing that like it's like everything's gonna change but maybe not in like 15 minutes it's gonna probably yeah some you know years of change and then maybe one last thing i'll mention about this i think users consumers business people engineers whatever like changing behavior doesn't happen overnight you know so people are um they are kind of used to doing things a certain way and adopting new ways of doing things isn't always like easiest thing. And so we even noticed that in our product. You know, people will be like, oh, this is really cool.

9:32It can do all these things. And then, you know, maybe we'll look at some users who stopped, fell off from using it and they'll be like, oh, I just kind of forgot about it. Or I kind of just ended up going back to the way I was doing it. Or if you look at like chat GPT, you know, there's been all these articles kind of saying that, you know, usage obviously went up like crazy and then it's kind of flattened and even coming down a little bit, you know, the last few. Yeah. And there's some different arguments about kids are in school, just not needing to cheat as much. But, you know, the point is that like you talk to a lot of people today who are like, oh, yeah, I have a chat.

10:04You need to scripture. I don't really use it. You know, so human behavior and even just understanding how to use these new tools takes time as well. Yeah, I 100 percent agree. I take the projects you're working on. Definitely. Yeah, I think. Like, I mean, of course, there's like the novelty of everyone wanting to try ChatGPT to just see what it can do. But then it kind of comes down to like who's, you know, actively using it in their workflow. Like I know for one, I would be I'd be doomed if I did not have ChatGPT to help me prep for podcasts and like so much stuff I do. It's kind of interesting because essentially what I've done is I've pretty much scaled up my outputs, my business outputs to like maybe like 5x what I was doing before.

10:50and it's like if I did not have chat gbt this would be unsustainable I would be I'd be completely doomed so it's kind of funny that I feel like that's like a big pivot but like again you know some people if they don't have it figured out into the workflow they'll just continue using it but if you find it for something and you it's like you can use it for something that's repeatable in your in your workflow like you're never going to give it up because it's too too valuable one thing I would I'd love to kind of ask you about um if I know from you is you know as far as like Bitbow's kind of AI code assistant goes.

11:19I'm wondering, like, how does this differ from, you know, competitors? Of course, there's like GitHub Copilot and there's these other, there's these other ones. How does this kind of differ in terms of like understanding the context of, you know, a developer's project and that kind of stuff? Yeah, no, it's a great, great question. You know, I think one of the things that we really are, you know, focused on is like, how do we really understand, you know, the developer and their, their context, like you were kind of mentioning. And so for example, like we index your entire code base and allow you to answer questions or work on your existing code.

11:52Copilot today is really just like as you type a line of code, you can complete it, but you can't really help get interactive chat. I mean, they're working on some of that stuff for sure, and I know it's around, but we're really trying to think about how do we help that developer really understand their code base and use AI on that, and we want to help you really understand kind of all that stuff going on in your organization and help you use AI on that. And I think that's one of the reasons we're seeing so much adoption is that it's a versatile tool. You can use it to do code completions. You can use it to do chat.

12:25And you can use it in a lot of different ways in a very flexible set of ways. Okay. Very cool. Something I'd be curious to hear your opinion on as well is obviously this is a big field. what are some of the main challenges that you guys are facing right now in enrolling these tools out and rolling Biddo out to people? It's a very ambitious vision, I feel like, for a company. So what are some of the big challenges that you're up against and some of the things you're doing to overcome them? Yeah. Well, maybe I'll mention two things. One is around ease of use, And then the second is maybe around cost of AI models.

13:13So the first one, ease of use. So we've taken a PLG, product-led growth approach, right? Which is that we want it to be super easy to get going and started with our product. You can literally sign up and start using it within like two minutes. And we're off and running. We wanted you to get to that aha moment very, very quickly. Now, cool. So, you know, one of the things is, for example, like when we try to index your whole code base, you know, there are some other tools out there that let you do that, but you have to set up a separate app and it, you know, can literally take like an org and like, okay, now publish an endpoint and set up this server.

13:47I mean, it turns into like a whole project, you know? We actually write on your machine, like deploy a vector database and build kind of all these, you know, index of vector index of all of your code and locally on your machine for security reasons so that it's not like in the cloud. But, you know, we try to make it so that the user doesn't have to change anything. Literally, they just had to say, accept, like, I want to do this. Cool. And so that's taken a lot of energy and work to kind of make that happen in a very seamless, easy to use way. I think that if you had like three hours of work that you had to do to enable that, I think that that wouldn't be, you know, a lot of users wouldn't be interested in trying it out, you know?

14:29So, yeah, today we only really offer that for our paid users, but we're trying to think about how to offer that for all of our kind of free users as well. And so that's something that goes back to kind of your question about like, hey, these are some of the types of challenges like in a PLG model, trying to make it super easy to use the product and not all this configuration, installation, kind of all that stuff. Second thing maybe I'll mention, you were asking, second thing I was going to mention was around kind of AI model cost. So, you know, these AI models are not cheap. You know, it's at a whole different level of cost than normal compute.

15:08You know, there's so many services where people are like, oh, yeah, it's free to use because they're like, oh, you're just using a little bit of like, you know, AWS compute time or, you know, AWS compute time. It costs us a couple cents a user a month or even that. Well, that's not really the case in these AI tools if you're really aggressively using that. You can easily spend several dollars a month for a user if they're really querying in a lot and running a lot of GPG-4 queries. Anthropic has a great model, Cloud 2, that we use. Those things can really add up in cost. And so people will be like, well, I want access to this thing, but I don't want to pay.

15:45And so we, of course, want to give them a really robust product. But we're all in a way that is cost effective. A is a price that they're willing to pay, but B also then, you know, hopefully over time, you want to have some margin, you know, to build some revenue for our company. And you do have, there's a long-term thing, which is like, you know, that there's a bit of a Moore's law happening on these AI models. And over time, the cost is going down as, you know, compute gets cheaper. But in Iran, you know, that's not always exactly the case. Right. Okay. Very cool. One thing that I would, I'd love to ask you about is what are some of the things that you get, you guys are currently working on that you're really excited for in the future?

16:32Like some maybe like future features or the kind of future direction you guys plan on going. This is a really interesting space. And like you mentioned, like there is a lot of competition. There is a lot of people kind of looking at it. I'm big players and whatnot. I'm like really impressed that you guys are like doing this as a startup and you're getting in there and like you've just done so much. I'm blown away. And I think maybe one of your competitive advantages is having that kind of more agile, scrappier team versus something like a GitHub co-pilot. You can get a lot of these innovative things done very fast.

17:02But just curious, what are some of the things you guys are looking forward to in the future and excited about there? Yeah, no, it's a great question. I mean, we're definitely trying to move really fast. And I do think that's kind of one of our superpowers, if you will, is to be agile and try to move you know, quickly. And, you know, there's just a few of us that kind of work to make some, some of these decisions. We don't have like a whole thing by committee or, you know, whatever, like a bigger company. So, yeah, I mean, I think a couple of really interesting things. So we actually have like a code completions product launching in October, which we're pretty excited about.

17:37So this will be a little bit of like, you know, as you type, we'll complete a line of code, or you can put a comment in your code and we'll write the whole block of code. And I think one of the things that we're really bringing to that, which is different than everybody else out there, is like we really do understand your code base. And so you can use that to actually really write much better code, right? We understand the functions, the methods, you know, the variable names that you're using. And so we can really use that to really write much more code that I think really is much more relevant.

18:08And so we think that, you know, we can help developers maybe, you know, write as much as like half of their code. So I think that's kind of something that we're, you know pretty um pretty excited about and then i think you know the other thing i'll mention is that we're really doing a lot of work on our agents so we have some initial agents you know that we put out there like one for i think i've just mentioned this briefly before but like for you know test automations or the right test cases now yeah one for documentation but you know those are still you know they're early they're kind of you know alpha kind of versions but we're doing a lot to like really improve those and make those more and more robust.

18:45And so every week we're trying to really improve those and add more, you know, things around that. We have some other things we're doing around code review agents that we're hoping to launch, you know, later this year. So, I mean, really excited about all these things. I mean, today our average user tells us that Biddle makes them 31 % more productive. Which if you kind of think about that, like I didn't think it was going to be that high actually um and we're just like surveying our users right we're like hey you spit out like how much more productive does it make you and we're a bunch of i mean 31's the average right so of course there are a bunch of people who said oh no it makes me like five percent more productive we give up archie users who said oh it makes i think that there was like more than 25 percent of people so it makes them over like 100 more productive oh wow yeah and i kind of feel like we're just getting started you know so what i think uh-huh maybe from where we are by the end of the year or maybe from sitting here in, you know, September 2023, where we are a year from now, I'm actually hoping that we can say, yeah, we make people like 100%, 150 % more productive, which if you think about that, like we're focused on professional developers and helping them in their jobs.

19:53Like think about that in the work context, like, you know, a developer, you know, company spend, let's say$150 ,000 a year on developer, like a, if you could be 100 % more productive it's like wow it's a huge return on roi but i think yeah maybe more interesting is that like for the developer themselves we think that we can remove a lot of the kind of drudgery of work that's been out there i mean there's been this whole kind of push the last like seven eight years called shift left in the development world oh we should include security earlier in the process then you write more secure code. We should make tests earlier in the process.

20:32So then you have more, you know, better tested code, more code coverage. And all that's kind of ended up on the developer's plate. And so the developer is like, I'm supposed to become an expert in everything. Like CICD, I'm supposed to become an expert in, you know, security, static code analysis, open source vulnerabilities. I'm supposed to become an expert in test coverage, test framework, unit test. So all of these things are like, they are making better code. We are having more secure, better performing code. But at the same point, the developer is like, I'm not getting that much time to actually just work on the stuff I want to work on, which is like thinking about how to design a good solution and then executing on it.

21:15And so we think that a lot of these AI tools can eliminate a lot of that work that frankly, a lot of developers don't consider like that core to what they're doing. And think clear of a lot of that documentation, you know, like all that stuff just be handled for them. And so developers can kind of get back to almost like having fun and doing the work they want to do. Very cool. Yeah, I think that's, I think that's incredible. You guys are doing some really cool stuff. So definitely, this is a company that I'm going to be following along on the journey and watching what you guys are doing. For people that want to, you know, find out more about you guys and, you know, maybe give Biddo a try.

21:52What's the best way for people to find Biddo? Yeah. I mean, just check on our website, Biddo, B-I-T-O dot A-I, Biddo dot A-I. And yeah, right from there, you can, like I said, sign up in one minute and get going. We have plugins for the Visual Studio Code IDE. We have plugins for all the JetBrains IDEs like PyCharm, Golang, Ruby, all the different languages that JetBrains supports, Android Studio, et cetera. We also have a CLI, so you can run it right from your command line interface. And we have a Chrome extension and a web app as well. So we kind of cover kind of all of your bases. Very cool.

22:32Okay, awesome. And I'll make sure for the listeners, I'll have a link in the show notes to Bitto. You guys can go check it out. Amr, thank you so much for coming on the show today. Thank you so much for sharing your story, the insights and some of the cool stuff you guys are working on. I'll definitely have to have you on another time. To the listeners, Thank you so much for tuning in to the AI Chat Podcast. Make sure to rate us wherever you get your podcasts and have an amazing rest of your day.