Episode 112
· 38:28
22 year old tech CEO, Daksh Gupta, says an 80 hour work week is a lifestyle choice.
Radio2:It earned him death threats and job seekers.
Jack Bridger:Daksh Gupta is the founder of Greptile which is an AI code understanding API. Dax is not afraid to do things a bit differently. They launched their own brand of greptile energy drinks, and they hand delivered boxes of cookies with Steve Ballmer's voice inside when you opened it. We talk about the value of doing marketing completely differently, why a lot of people misunderstand what sales actually is, and we talk about how big enterprises are buying AI tools and why that's a bit different to how they buy most other tools. It's a really interesting episode and Dax is a very interesting guy, so enjoy.
Jack Bridger:I think we were talking a lot about like the importance of being different, and just before coming on here James from Posthog was putting this announcement out that they're gonna let someone else design their billboard.
Daksh Gupta:Design their billboard.
Jack Bridger:Yeah. And so I just put 2 submissions. I just wasted like half an hour trying to like come up with ideas.
Daksh Gupta:Amazing.
Jack Bridger:And it really I I feel like that was a really good, reminder of how, like, no one cares about your billboard. But suddenly, if you let random people decide design your billboard, it's, like, super interesting.
Daksh Gupta:Yeah. I like that a lot. That's very, I I I think there's a there's, like, realization that there are no rules, because, like, you think what I think the you a normal person would be sitting in a boardroom and they'd be like, we should let anyone design the billboard. But you'd never say that out loud because it just sounds so crazy. But I think the really interesting ideas come when you like push against the skepticism.
Daksh Gupta:You're skeptical of the skepticism when you say, okay. Well, what like, what look what's actually wrong with it. Like, what is actually wrong with letting anyone just design our billboard? Like, we'll still vet it so they don't put, like, a slur on it, but, like, well but anyone we can let anyone make it. And, like, we can make it a competition, and they're and then that's allowed.
Daksh Gupta:Yeah. I think, there's some alpha in, like, not, shooting down your own wildest ideas.
Jack Bridger:Yeah. Tell tell me about how the, the Griptal energy drinks came about. This episode is brought to you by WorkOS. At some point, you're gonna land a big customer and they're gonna ask you for enterprise features. That's where Workhorse comes in because they give you these features out the box.
Jack Bridger:Features like skin provisioning, SAML authentication, and audit logs. They have an easy to use API and they're trusted by big dev tools like Vercel, as well as smaller fast growing dev tools like Nock. So if you're looking to cross the enterprise chasm and make yourself enterprise ready, check out WorkOS. We've also done an episode with Michael, the founder of WorkOS, where he shares a lot of tips around crossing the enterprise chasm, landing your first enterprise deals, and making sure that you're ready for them. Thanks WorkOS for sponsoring the podcast and back to the show.
Daksh Gupta:So the Grubhub Tile energy drinks, came out of YC office hours. We were thinking about how we basically had built a product which if you're the right person for whom this like, if you're if you have the problem that this product solves, which we'd proven that that person exists, that they would use it and it would it'll work for them. Like, we'd spent the time to do that. So, we wanted to start focusing on getting just more people to hear about the product. And, the energy drink idea just was given to us by our, group partner.
Daksh Gupta:He said, you know what would be crazy if you had, like, energy drinks, which is, like, energy drink for coders. And then we said, oh, what if it said, like, drink this to ship faster?
Jack Bridger:That's so good.
Daksh Gupta:And so in the first pass, we just printed out a bunch of stickers and then we bought cold brew cans from, from Costco and then we sandpapered the outside of it and then put the new label on it manually. And they didn't look very good, but they were a good start. And, like, we we dropped them off at a hackathon. So we, like, did do things that don't scale. I'd I literally sat and wrapped a bunch of cans.
Daksh Gupta:And, but to see the concept would work. And the concept worked, the people really liked it, and it was memorable. And long after that, people still come up to us and be like, oh, I got I saw your energy drinks at the, at the hackathon. One thing that we did wrong was people thought we were an energy drinks company. And so, I think we should have been clear on the packaging that we're primarily a developer tools company.
Daksh Gupta:We just also make energy drinks. So we fixed that a little bit in our next label. So this time we're like, okay, this works, so we should like, try to scale this up a little bit. And so we found this manufacturer in Illinois who would make us a custom energy drink. And so we just made like 200 milligrams of caffeine, lime flavor, and like get them wrapped in or like and like made a design in Canva and just like did that.
Daksh Gupta:That was the second iteration. Again successful. And then now we're doing the 3rd iteration. We had a designer actually make a proper label for it to do. It wasn't just hacked together in Canva.
Daksh Gupta:So like at every step, we just said, okay, we've proven that this is worth additional marginal effort, and so we'll take it to the next step. But the realization that came from the energy drinks thing, I think that was actually fairly pivotal in in how we decided to then go about go to market because the energy drinks idea is so outlandish. It kind of like opens up everything as a possibility of what you can do to get people's attention, get people to use your product. Clearly there are no rules if you're allowed to as a developer tool company just make energy drinks and ship them to to people. And so I think there are a couple so that was the first thing that came out of that.
Daksh Gupta:And the second thing was, I started to look at go to market, especially like marketing, the marketing level of go to market, which is just getting more people to hear about your product as a essentially a multiple of, how many people see it times how many people retain it. And I think people index a little too heavily on the first one, which is why Google Ads seems so appealing because for like very cheap, you can get millions of people to see the thing. But people probably won't retain it because there's nothing novel about a Google Ad usually. And you can get creative with a copy and increase the the mental retention a little bit, but it's still not very, very memorable. The energy drink footprint is much smaller.
Daksh Gupta:It's just everyone that physically saw the energy drink, which is maybe a few 1,000 people, and you spend 100 of dollars on it. And so the cost per impression, so to speak, was very low, but you never forget an energy drink from a developer tools company. It just like, it's so it's so weird. You will not forget it. And so we took that to the extreme this time around.
Daksh Gupta:I don't know if you saw this, but we we built a box, and we filled it with cookies. And the box has a grubtel logo on top, it's green. And the inside flap is a picture of Steve Ballmer's face, which is a frame from his famous developer, developer, developer speech. And then we bought a bunch of, light activated speakers, from AliExpress. And we and we recorded this the actual sound clip from the from the speech on it.
Daksh Gupta:So when you open the box, you'd see cookies, you'd see Steve Ballmer's face, and you start hearing the speech. And we made, like, 50 of these and started and we actually I think we set we sent out half of them yesterday, and we're doing the other half today. And then again, the footprint is pretty small. Like, that's the only like, it's it costs us like 30, $40 per box to make this including the cookies. Because we also wanted to get like really good cookies.
Daksh Gupta:Good. You want us you want people to associate you with like deliciousness. You don't
Jack Bridger:wanna be disappointed.
Daksh Gupta:Yeah. Like, it's gonna be like, well, if you just get Costco cookies and it'll be like 10 times cheaper, but then then people associate you with Costco cookies, which are good, but like they could associate you with like fancy downtown San Francisco French bakery cookies. And so we decided to go down that path instead. But like at the end of the day, it's again, like you you drop us off in in an office, people see it, they won't forget it. It's like low number of impressions, very high retention percentage.
Jack Bridger:Yeah. It's actually it's so funny that you talked about, you know, for people that haven't seen Greptar's logo, it's like bright green. And you you had, like, the lime energy drink flavor. You wanted to get that right. And now you've got you wanted to get the right cookies.
Jack Bridger:You're obviously thinking not just about the color of your brand, not just how it looks, but how it tastes as well.
Daksh Gupta:Yeah. I feel like these details matter a lot because they I think the details are sticky and they can improve your attention in terms of like how often people remember you because there's like little things about it. The cookies thing actually came from BrainTrust sending out pizza to a bunch of, companies across the cities. So BrainTrust, they do LM observability, they've launched recently. And part of their launch, they shipped out like whole pies of pizza to companies across the city.
Daksh Gupta:And so the interesting thing with those guys was they could have just gotten any pizza. There's tons of popular pizza places. They found a specific tiny shop in the Tenderloin that makes just the most incredible pizza. And that's what they picked up. And they did the work of, like, like, it's kind of out of the way for everybody, but, like, they went and got the pizza from there.
Daksh Gupta:And the number of times that, like, people commented, not only that Braintrust bought them pizza, but it was, like, some of the best pizza they'd ever had. And, like, now your Braintrust and, like, that's Braintrust Pizza. I'm not even sure what that place was called. I think it was called, like, out of sight or something. But, like, I think of it as Braintrust Pizza.
Jack Bridger:Yeah. I I wanted to go there, but it's, like, it's in, like, the dodgiest part of San Francisco. Yeah. It's, like, you just Right. They're actually solving a problem for people.
Daksh Gupta:Right. Yeah. And it's and and like so there's 2 things. 1, they went into, into like kind of a difficult part of the city and they they did the work of like finding the really, really good pizza and they were like, yeah, we care enough. So which like, there's a couple of things there.
Daksh Gupta:I think one is now I associate them with delicious pizza, and I'm talking about it way more than if this just brought me some pizza, because the details are made and sticky, and I'm still talking about it weeks later. I probably would not be still talking about pizza weeks later had it not been some incredible pizza. And second, I think, when people, especially when companies buy products, there's like some amount of, and this is true for people too. The things that you get right at the start are like indicative of who you are. There's like Patrick Collison quote where I think he says more of Stripe's success than people think is, like, downstream of it being a beautiful product.
Daksh Gupta:Because you use it, you see the details, how beautiful it is, and you're like, well, they must really care. I think the the word you used was like it shows people that, that you really care, because you cared about what was on the surface. It's a good indicator of what's underneath. Not in all cases, but it's it's a good indicator. And to someone, there's like this tiny subconscious thing where you associate Brain Trust now with like good pizza, but you also associate them with like the type of people that go out of their way to create a great experience.
Daksh Gupta:And like that might be the case with their software
Jack Bridger:too. Yeah. 100%. Did you get any initial reactions with the box?
Daksh Gupta:Yeah. Some people tweeted it out. There's only people talking about it across platforms. Not too many impressions yet. I think probably about like 35, 40,000.
Daksh Gupta:But I think we can get it farther up. We also were very targeted with who we sent it to. So we sent it to existing customers and we sent it to customers that are like very ideal fit for us. Surprisingly, so the number of companies that are at truly in person is surprisingly small, that are like over 50 people. So, we actually, I think are pitting pretty much all of them.
Daksh Gupta:I think every software company is like 50 to 500 people that are like in that have San Francisco in office presence. We are basically hitting all of them.
Jack Bridger:Yeah. That's how did you just, like, curiosity, like, do the research on like that? Because that seems, like, quite quite hard to figure out, actually.
Daksh Gupta:Yeah. It was difficult. We used, Apollo and Clay and a bunch of stuff. Actually, the first half of this is pretty similar to a normal sales problem. And then I don't know if you've used Paradigm.
Daksh Gupta:They're like an AI spreadsheet company. They build basically, it's like a spreadsheet and every cell has an AI agent in it. So we basically put all the companies down, on the 1st column of the spreadsheet. And the 2nd column, we were, like, find their address. And 3rd column was, like, are they in person most days of the week?
Daksh Gupta:I'm not sure how it finds this information, but, like, it does. And it's pretty accurate. Worked. So yeah. Wow.
Daksh Gupta:So it it's it's it's kinda crazy. It's like a spreadsheet with a person that can operate a spreadsheet and built into it. It's really crazy.
Jack Bridger:Okay. That's super cool. I didn't know about Paradigm. Yeah. Yeah.
Jack Bridger:That's amazing. And so you are doing a lot of, like, quite, you know, you're doing sales, right, like, how how has that been? Have you, have you got any, like, lessons that you've picked up while doing that?
Daksh Gupta:Yeah. I I saw I'm sort of a reluctant salesperson in some ways because when we started the company, it was me and my 2 co founders and all 3 of us are software people. I was the worst of the 3 of us at programming. So that means I had to be the one that had to do all the sales stuff, which is probably for the best. I think the product would have come out a lot worse if I was one of the ones that was engineering.
Daksh Gupta:But I started doing sales, I think at first I was sort of, especially if you're a technical person, you maybe don't look at sales in a very positive light, you're picturing with sort of like a crafty, like car salesman. But I think there's a couple of mindset shifts that happen. One is if you like winning, like as just like a thing, as an experience, you probably will enjoy sales because there's this like, you work really hard, you try to solve a problem. The problem you're solving is there's an organization that you think a product can benefit and that you have to will the organization into like starting to adopt it. It's actually a surprisingly tricky problem.
Daksh Gupta:It's an engineering problem, it's also a human problem and a people engineering problem. You see, you start to enjoy it for that reason. And second is, salespeople are actually, their job isn't to, I guess, convince people to buy their product, like push like down people's throats. The actual problem that salespeople have is finding the people that want your product and then conveying to them what your product does and how it can solve their problem. If your product actually helps people, then a salesperson job is to find the people that need the help the most and, and then like help them.
Daksh Gupta:Like it's actually just connecting supply and demand. Especially, and that is again, if your product is actually good and it helps people. So I think that the mindset shifts are good and over time, like you get good at it if you learn from the backdrop because the speed at which you learn that you're bad at sales is immediate. Like you go into a sales call and you're pitching someone your products, at the end of it they don't really get it or they're not into it and you're like, okay, that was a negative sample. My loss function value is high.
Daksh Gupta:I need to, like, fix something. So within a few sales calls, you should probably start to learn the patterns of, like, what people respond to and what aspects of your product they resonate with. And, like, what are the indicators you can find about people before you talk to them that will tell you whether or not they're the right person to be using your product? So I think the the iterative learning aspect of it and then the problem solving aspect of, like, getting an organization to to adopt your product is actually very interesting. I'm I'm trying to enjoy a lot.
Daksh Gupta:I'm I'm very much not looking forward to the day that, like, I have to eventually let go of it and have someone else do it instead.
Jack Bridger:Yeah. You'll probably still be doing that. Always do the big deals. Come in and
Daksh Gupta:Yeah.
Jack Bridger:Close. Close Amazon and stuff.
Daksh Gupta:Yeah. Yeah. I guess. Yeah.
Jack Bridger:Yeah. That's that's really cool. I I guess when you get to, like, 50 person plus companies perhaps there's a few people involved in how have you kind of managed to do that? Because I know that's something that can be quite tricky for founders.
Daksh Gupta:Yeah. I think there's, the good thing is that in every company, there's the same type of person that's involved. And, and so you start to learn patterns and what each person cares about and how you can best serve their needs. There's always going to be the engineering leader that really cares about your product and wants to adopt it. In some cases, that is the engineering leader is the VP or the CTO.
Daksh Gupta:In other cases, they report to the VP or the CTO. In the latter, then there's also a CTO or VP who's very pragmatic and needs to be convinced that this is something that's actually going to help them, and it is going to be worth the money and the time that they put into it. And then there's usually going to be someone on procurement and someone on infosec. Procurement just essentially needs a pricing that's predictable that they can trust and, like, they will want some sort of discount if it's a larger company. And infosec just has a set of questions they need to have good answers for.
Daksh Gupta:The good thing about infosec is everyone kind of has the same questions. Like, we don't really get their own net new question. So after you've cleared out the basic things that companies care about on a security level, especially because we're fairly invasive in terms of we like we're scanning your code base is the most your most valuable IP. And so it's very important that we have good answers to all the security questions. And so once you've answered them for like 1, 2, 3 companies, you've basically covered like the p 99 and every new company will just have the same question that you can answer them.
Daksh Gupta:So it becomes fairly formulaic after a point for this size of company. I think the complexity starts at like when it's a few 100 people and there might just be like a new sort of process you have to navigate. And there are enough people and then you now have to worry about how they interact with each other. And so the problems get like this, the number of nodes goes up, the number of edges goes up like it has to be quadratic. And so like it makes the problems like much more complicated when you get to larger sizes.
Daksh Gupta:So in those cases though, you can actually, like really good software companies are often really just multiple companies in a trench code in some ways. And so if the you just have to find the like smallest independent unit that can make their own decisions And, and just work with that first. And then, like, essentially go piecewise instead of trying to, like, get in the entire like brute force away the entire organization all at once.
Jack Bridger:Yeah. I love that. I love the multiple organizations in a trench coat. That's so good. Say your quote.
Jack Bridger:That's so that's such a good one.
Daksh Gupta:Yep. I I guess the trench coat framework is popular.
Jack Bridger:Okay. Yeah. That's good. And then one thing when we spoke last time, you were talking a bit about, like, how companies are buying AI stuff and, like, how they do it now might be different to how they do it a bit later.
Daksh Gupta:Yeah. I think there's it is a couple of things. We've already seen pretty meaningful differences in how engineering leaders are thinking about buying AI developer tools. About 6 months to a year ago, they were maybe like dipping their feet into trying Copilot or Cursor or something of that kind. Like the core generation is the first thing that they try.
Daksh Gupta:So when we were trying to do sales at that time, pretty much no one was interested because they were just starting out and something that we were a little like sort of in a second order effective AI, which is we're doing like code reviews and diagnostics. We're essentially giving people the framework to build their own custom internal AI developer tooling. And that isn't something that is immediately obvious to be useful to you until you're done adopting Copilot or Cursor. So once that happened and, like, Copilot and Cursor became like or, you know, Codee, other tools that are code generation tools in your editor, once they became super popular and virtually all of the people that are target customers for us were adopting them, the next thing they wanted to do was maybe augment code reviews, do test automations, and then it so the process becomes significantly easier for us over time. There's a couple of things that are apparent.
Daksh Gupta:I think 2 things need to be basically necessarily true for someone to buy developer tools, for at least to buy our product. One is they've already adopted 1 or 2, because I think you, if you never, it's hard for us to imagine as like AI people in San Francisco, but like there's people that have never applied AI to their day jobs and most people haven't, and they can't imagine just the absolutely ridiculous productivity enhancement that it is. And so you have to try it. And then second, you have to come out of it with, like, conviction that this is gonna change your, your productivity incompletely. In many cases that comes from like senior leadership.
Daksh Gupta:So we have customers where the CEO just has a mandate and says, look, this is going to transform everything about how we operate this company. We need everyone to be like looking for the best vendors that can help us do that. And if that's the mindset someone's coming into your call with, then you don't have to convince them of underlying value. You actually just have to tell them that, like, you are different from everything else and better, and you are the perfect person to be. It's it's, like, a different problem than if you were trying to, like, tell someone, hey.
Daksh Gupta:You have this problem. We can help you solve this problem in this, this, this way. You first have to convince them to have the problem, and then you have to convince them then. Because they may not know that they have the problem in many cases, which used to be the case pre, but now they're like, okay, we program software. It takes a long time and here's a like solution space of tools that are designed to make that faster.
Daksh Gupta:It's like intrinsically useful. And the question isn't like, is this going to help me or not? Is it, is this going to help me? Like, is this specific thing good enough to help me?
Jack Bridger:Yeah. Yeah. That makes sense. So, yeah, it's evolving. What people so, like, even yeah.
Jack Bridger:You said a year ago, it's like they're not they're starting to adopt copilot.
Daksh Gupta:Yeah. Every quarter it changes, I think.
Jack Bridger:Yeah. Where where do you see it growing? Do you see it changing?
Daksh Gupta:Yeah. It's hard to predict, but I have a couple of frameworks that make sense on my end. I think the a fully automated AI software engineer, which is like instructions in software out, I don't see that existing, not because I don't think it's technically possible. I just think that if you can build something that can create perfect software from a prompt, you probably don't wanna sell it as a software engineer. That's just, that's pretty much AGI.
Daksh Gupta:Like it's like, if I could build something that can create perfect accounting software, for sure, just from saying, here's my like type of tax paperwork that I want to process, make software to do it. That's just an accountant. Like why would I go through the work of having like why would QuickBooks need to like why would TurboTax need to exist and they buy this thing that generates offer? Like it sounds like I'm maybe a step away from this like fully automated everything instead of having the intermediate step of building software first and then like packaging and pricing and selling the software and then having people buy the software. It seems like a lot more work.
Daksh Gupta:What I see happening in the interim and maybe in the longer term too for increasingly complex pieces of software is like everything that humans Only humans can do will be the only things that humans do and everything else gets automated away. And that leaves all the creative work for humans to do for software engineering. They don't have to worry about, like, how much time is spent refactoring, debugging, like, on deployments, like, all this stuff that we should just be able to make go away, and only the creative the creative aspect should should be around. I'm trying to think, there are crafts which have achieved this in some ways, which is like, so I used to produce music in high school. And if you produce music, you realize that like you can do it now with just a computer and you can basically produce like studio quality music with nothing else other than a computer and like, and a piece is a couple hundred dollars worth of software.
Daksh Gupta:But you needed way more apparatus back in the day. Like, you needed to have, like, an 8 track and a studio and, like, an IVocalization booth. Like, you needed so much stuff to create basic music that you would can now do on a computer. And the thing that changed was that we were able to digitize music, and, like, we were able to have it in, like, these completely malleable file formats. That was, like, the main thing.
Daksh Gupta:And, of course, like, computers became very powerful. I think that's the parallel here, which is there AI takes away all of the apparatus that was required to to engineer within the next, like, I think couple of years that'll happen. At the same time, I think we structure software companies not as a collection of developers, but rather as like developers and the teams that are made up of them. And the reason I draw this distinction is we often think of ourselves as all AI developer tools are either above or below Git. If you're below Git, you are a developer's tool.
Daksh Gupta:You are cursor or you're copilot. You're something the developer uses to execute and do their job. It a good indicator of whether a piece of software is below or above git is if it's below git, then everyone should be able to use a different one. You don't need consistency across your organization. If it's above git, everyone needs to use the same one.
Daksh Gupta:And above git doesn't just mean post git because like linear is above git. You can think of the journey of a ticket starting at like a ticket is created in linear, it like is assigned to an engineer and then it dips below Git. And now the engineer does stuff in their editor, whatever their setup is, And then they floated above Git by making a poll request. And now it's in GitHub and then it's gonna go here and then it's gonna go there. And you'll have incident management in Sentry, you'll update your documentation or Notion, all this stuff is above Git.
Daksh Gupta:And there are really good vendors below Git. You know, we use Cursor, Corie is really good, Copilot is really good. But above Git, you're no longer selling to developers, like your customers and developers, just development teams, because teams by GitHub, not software engineers. And you, sorry, you might make the decisions, but the customer is the team. The reason GitHub exists is you wouldn't need GitHub if you just had 1 engineer.
Daksh Gupta:Their justness is reliant on there being a team. And so we build developer tools for teams, which is the key distinction here. And I think that that's the mindset shift where I think developers realize that AI was gonna change everything. And of course, organizations are slower to realizing than the way those are. And so now for the first time, I see organizations realize that they need developer tooling and they need developer tooling not for their individual engineers, but for their teams.
Daksh Gupta:They need to augment the stuff that's happened above Git because it's taking so long. I think that's that's been, like, a key shift that that has occurred in the last maybe, like, 6 months.
Jack Bridger:Yeah. That makes sense. I like the way you've you frame it with, like, above and below Git. May maybe this is, like, I don't know if this is really, like, the topic we usually talk about, but digging into, like, the creative work, I just kinda wanted to go back to that as well because, like, I it's something that I think about all the time is, like, what actually is the creative work of a software engineer? Like, do you think?
Daksh Gupta:Yeah. I think there's this is a sort of a question of abstraction because you could say, well, you know, the design of a of, like, a database schema is creative work. But you could say, well, with the correct abstractions, it isn't. Like, there does there exist a perfect answer or database schema? Arguably, yes.
Daksh Gupta:Like, for any given well defined well enough defined problem, there's a perfect answer for it. So I think that's actually it is, creative work is anything where there isn't an objectively correct answer, at least in the realm of software engineering. So if something has an objectively correct answer, there's an objectively correct way to deploy something that is just the most performative. Like, we've well defined enough problem, there is an objectively correct answer. I think all of those cases should probably be something that AI does or computers in general do.
Daksh Gupta:And everything where there's, you know, I think the word taste gets thrown around a lot in Engineering Now, but like, I think that's actually it. There's a case, there's no objectively correct answer. There's just like well thought out opinion. Obviously, user interfaces usually are opinion. It's hard to define like perfect UI or perfect UX.
Daksh Gupta:Perfect form factor is also hard to determine. Just as hard to define the problems in the 1st place. And even if you did, then it would be hard to define, like, what the suit the quality of a solution looks like. So I think that's the where I would draw the the distinction. Engineers should just be doing things that there aren't perfect answers for.
Daksh Gupta:Because if there's a perfect answer, then we should be able to, like, deterministically arrive at it with a computer.
Jack Bridger:Yeah. That make that makes sense. So if we were launching, you know, if if you and I were to do a start up tomorrow where we're gonna allow people to create their own boxes with, Steve Bulmer, you know, voices or whatever they want. They can they can pick their own voice. They can upload their own voice.
Jack Bridger:I guess, like, in the future, how are we just gonna describe exactly what we want? Like, how do you how do you see that playing out?
Daksh Gupta:Yeah. I'm not sure. I I guess it would depend on at what point the model stop improving at the rate that they're improving today. Like, it will eventually happen. I don't know if it's gonna happen this year or 10 years from now, but at some point, the model will stop improving at the pace that they are today, and will arrive at some, like, generally stable point, whereas yeah.
Daksh Gupta:This is about as powerful as this this thing is going to be, until there's the next major breakthrough. So I don't know what that stopping point will be. If it's very, very high, I can see it being, like, everyone has this perfect machine guard accessible to them in their homes. And you don't need a startup for this at all. Every individual person can be like, I conceived of an idea of a Steve Ballmer box of cookies.
Daksh Gupta:Like, if I can describe it well enough, then out comes all of the software and all of the maybe 3 d printing instructions I need for it. So maybe that's what it looks like. You you don't need, organizations for execution of the of building software. You need enterprises for everything else. And I do wonder if that to to an extent might happen to software before that too because, like, it is way easier to build software today than it was, 30 years ago.
Daksh Gupta:This the abstraction available to us, high level programming languages, the existence of like pay as you go cloud, the tooling that exists around building software is extraordinary compared to even like 10 or 15, 20 years ago. So this is just another step in that direction, which is why I also I'm not convinced that, like, this is fundamentally different either because, like, the the journey from like punch cards to Python was surely greater than the journey from like Python to natural language in terms of how much easier it makes to makes it a program. I mean, imagine trying to build like a, like, a snake game and punch cards versus in Python, you can do it so easily. And then with the cloud, you can do it in one shot. Like you just describe, hey, generate snake game in Python.
Daksh Gupta:That's the all it's the end of your job. So but it seems like punch cards to Python was the bigger jump there. So this is just like another higher level of abstraction. So the question is like, what abstraction does it stop? Like, can we really make it so we just give it our general goals?
Daksh Gupta:Like, what if you take it beyond the snake game because, like, hey. I want I want you to entertain me right now, and then it, like, decides to build you a snake game. Like, that's the next level of attraction here. So I wonder at what point it stops.
Jack Bridger:Yeah. And then also with, like yeah. I guess, like, it's like, with developer tools, it's like so let's say you you wanna build your own, you wanna build your own pull request, custom, like, pull request reviewer, then, you know, I guess it's like, do you just is there a can you just say the right prompt and then you're gonna get your own version of in a sense of, like, it's like, and they'll they'll go, like, deploy all the infrastructure, and they'll just do everything, and you know, I get because that's the that's the question that I think about, Because some people ask me, like, why interested in developer tools? Like, developers are gonna be obsolete, you know, and so are the tools.
Daksh Gupta:Yeah. And I I think that is a real possibility, and I am worried that, like, people are underestimating. I think people generally underestimate the odds of bad things happening and every person that has an occupation today underestimates the likelihood of it going away as an occupation. And when I hear developers say, yeah, no developers are not going anywhere. It's like, I think you might be overestimating like what you do, but more importantly, you're underestimating what software can do.
Daksh Gupta:I think you're probably right that you're not gonna get replaced because what developers do actually does require enormous amounts of creativity. It's a very difficult job.
Jack Bridger:By the way, I actually don't have a view on this.
Daksh Gupta:Yeah, but you'd be surprised. There are a lot of software developers that would say, you're not, this is, and it often means that they're rejecting AI as a concept entirely. They're like, well, AI generated code is just slop. And it's like, well, I trust you that you are writing better code than AI is writing, but like, look around you and tell me that the median developer is writing better code than you're, than like AI is writing. And that's today, like this is where, we're so early, like in the curve of, I mean, we, like, we haven't even maxed out our own benchmarks for software engineering quality from AI.
Daksh Gupta:Like, we can only picture what a really good version of this could look like and there's gonna be a day when we'll max that out and then there'll be another benchmark that we'll wanna hit and then it's gonna be, like, this is the very first, like, SuiteBench, we were at like, I think 4 like 43% is like the state of the art for 3 bench. We're like barely halfway to our own benchmark, on how good this thing can get.
Jack Bridger:Can you explain what 3 bench is actually?
Daksh Gupta:So this set of this is how it started out. I think they actually modified a little bit, but it started out as like a set of, real world GitHub issues. And then at the other end of it, there are like like good pull requests made with those issues. And it's a benchmark to see AI models, AI software engineers, how close they can get in one shot, or I guess in the case of agents, a few shots, how close they can get to like that perfect pull request. So, I'm not entirely sure how it's scored, but I would assume it's something to do with, like, how close the code was and whether it worked or not, whether it did the thing that the the issue intended.
Daksh Gupta:And 43% is currently the the the state of the art, which at least I don't know if if that's a verified number but that was like the the most recent high number that I saw.
Jack Bridger:Mhmm. And is it growing fast? Like, how how long ago since it was like 20%?
Daksh Gupta:Yeah. I I think Devon when it launched was at 13%. And that was maybe 3 or 4 months ago. So it's going pretty fast. There's also, there's a distinction between like verified results and non verified results because I think that there's potentially, not to say that the 43% people did this, but there's potentially some out of leakage from training data into going into, like, the, like, the test data going into training data and stuff.
Daksh Gupta:I don't know what the highest verified is, but I know it is going really fast. Like, I think we're well on our way, to be to, like, max out this this benchmark from what it looks like. There's also, like, there's it's an enormous economic opportunity. There's tons of hard people working on on, like, Max and SpeedBench. I would hate to do that.
Daksh Gupta:I would hate to be one of the companies that are trying to yeah. It's it's extremely competitive, to be on that path.
Jack Bridger:Yeah. That's, because I guess you're kind of competing with OpenAI.
Daksh Gupta:Yeah. No. No. I see. In in the sense of, like, I would hate to be a company that is trying that where, like, my success hinges on being the best at SpeedBench.
Daksh Gupta:Yeah.
Jack Bridger:Where,
Daksh Gupta:like, I'm building an AS offering, there's some sort of code gen tool. It's just that there's too many extraordinarily smart people that would, like, 1,000,000,000 of dollars behind them that are trying to solve that problem.
Jack Bridger:Yeah. Yeah. Yeah. No. Totally.
Jack Bridger:So I just wanted to ask you, like, 2 more questions. The first one is what advice you have for other founders?
Daksh Gupta:Oh, god. I guess the the only ones would be, 1, you should make something that you or someone you know really wants and will pay for launch fast and charge people early. So they you can charge them like last later, but you should charge them more now just to prove that they actually want the thing you that you want. Second would be, you should find great co founders. They'll make your life much easier.
Daksh Gupta:Even 1, like just pick your closest, smartest friend and that could be the person. And 3, don't raise money too early. Wait until you actually need it. And it's like the only thing that's standing between you. It might be it actually might be day 1 in case, like, if in in our case, we had to raise a little bit because we were, like, coming out of college with no savings and needed to stay alive.
Daksh Gupta:But, I think we still probably probably raised too much. Like, I would looking back, probably would've done less. And so that's probably what I'd recommend to people is don't do it too early. And I definitely see a lot of founders that where because capital is available, they take it. And it's a capital scarcity mindset, and, Silicon Valley has not had a capital scarcity problem ever in its existence.
Daksh Gupta:And so I don't think that that's something to worry about.
Jack Bridger:Yeah. If you're doing a good if you're building a big business Yeah. There's gonna be money.
Daksh Gupta:There's always, like, infinite money for good founders and zero money for bad ones. And so if you just become a good founder, you probably don't have to worry about it too much.
Jack Bridger:Yeah. 100%. And then the last one is, besides Gravtel, of course, what what other dev tools are you excited about?
Daksh Gupta:I've been really oh, there's so many, but, obviously, I I use a lot of them. I use middle of a lot. I love that so much. Browserbase is one that I've seen recently that I wanna start playing around with. If Raycast counts as a dev tool, I kinda use it like one.
Daksh Gupta:So I guess it it sort of does. Yeah. There's there's tons that I'm excited about. Cursor obviously I've been using cursor a lot and it's really good. I do wish it was lighter weight.
Daksh Gupta:It just feel a little it's starting to feel a little clunky so if if anyone who works at cursor or knows them
Jack Bridger:Yeah. You tell them to remove some stuff. Has a clunky clunkiness problem.
Daksh Gupta:It's hard. Yeah. Yeah.
Jack Bridger:Yeah. Yeah. That's a that's a really good one. Yeah. Amazing.
Jack Bridger:Well, thank you very much. And where can people learn more about Greptal and about you?
Daksh Gupta:At greptal.com. Every all all links lead to greptal.com.
Jack Bridger:Yeah. That's awesome. Yeah. Well, thanks very much for, your time, and thanks everyone for listening.
Daksh Gupta:Yeah. Thanks, Wash, everyone for listening. Thank you.
Jack Bridger:Yeah. Amazing.
Daksh Gupta:Thank you, Jack. It's great talking to you.
Listen to Scaling DevTools using one of many popular podcasting apps or directories.
