Want to join the conversation? Connect with us on Tribe.
Find “Founders in Arms” on Apple Podcasts, Spotify, and YouTube.
Transcript of our conversation with Sahil Lavingia:
00:00:00:00 - 00:00:18:23
I learned over time that basically like the sort of the Doge agenda has, three stools, to the three legs, to the stool. One is cutting contracts, two is helping with the reduction in force. And then three is shipping software. So actually, like, and they're, they're all kind of related because I think without shipping software, it's hard to cut a lot of contracts.
00:00:18:23 - 00:00:38:07
It's hard to like reduce the force. But unfortunately a lot of the Doge sort of like the bandwidth. And I think the stuff the Trump was okay with them focusing their time on was a more on that first two. So I didn't end up doing as much software. I kind of just went kind of like on my own and was like, hey, VA people like, can I work on stuff?
00:00:38:07 - 00:01:01:03
And I kind of got like shown a bunch of different software projects they were working on and I saw this one VA GPT thing they were building. I saw this VA chat bot stuff that was like kind of aligned with what I was already doing, and what I had a lot of experience with. And then I just started working on those projects, as just like a basically a pro bono software engineer.
00:01:01:05 - 00:01:23:19
Hey, everyone, welcome to the Founders in Arms podcast with me, Immad Akhund co-founder and CEO of Mercury. And I'm Raj Suri, co-founder of Lima and Tribe. And today we have with us Sahil Lavingia, founder and CEO of Gumroad Sahil Welcome to the podcast. Thanks for having me. You and I met at a Pinterest event, probably around 20. It was a 2011, 2012, maybe that type of zip code.
00:01:23:19 - 00:01:43:03
You were one of the first employees of Pinterest, I think, at the time, and you were like a drop, like a high school dropout or like a college dropout. You were really young, I remember. Yeah, I was a college dropout. I dropped out of USC, and moved up to the Bay area. I was, I think probably 19 at the time, and joined Pinterest as the employee number two working on the iPhone app.
00:01:43:05 - 00:01:55:17
And you've had a really interesting journey since then, so we'd love to dive into that. But I wanted to first, like, ask you that question because, I, I remember it was remarkable to me meeting you of like, wow, you know, this guy dropped out of college to join a startup, you know, at a very young age.
00:01:55:17 - 00:02:09:13
And it was still rare. Then it happened, and then it became even more rare after that. And now it's happening again. So it's kind of gone through the cycle. Tell me a little bit about that decision. How did you, like, get into startups in the first place? Yeah, I mean, I always felt like I wanted to build stuff.
00:02:09:13 - 00:02:26:21
I learned to code and do some, like web design, graphic design. When I was in like high school. And it was it was still like, I think early, early enough where I didn't I didn't know if this was like a real career path. And so my, my inclination was to go get UX degree, and then work at Google.
00:02:26:21 - 00:02:56:13
That was like the dream job at the time. You know, everyone wanted to, like, work at Google as if you were a software engineer. That was like the beanbags and ping pong tables and all that kind of stuff. And, and so, yeah, my first plan, I guess, was to go to college, graduate, you know, go to Google, learn what goes into building, you know, a high scale website, you know, and then eventually, like, work my way down to, like, working at smaller and smaller companies and eventually, like, starting my own thing.
00:02:56:15 - 00:03:18:21
And when I got to USC, I was in California. I was, like, so much closer to Silicon Valley. I went to startup school, Y Combinator, startup school at Stanford, and, posted things I had built on Hacker News. And, you know, it's kind of funny how some things, like, haven't changed at all. Like, I do recommend this path for, for even today, you know, 15 years later.
00:03:18:23 - 00:03:37:03
And so, you know, I would post my stuff on Hacker News and it turns out there was like this massive talent sort of war going on where, like, you know, if you're a Pinterest, for example, it's hard to compete with Google and Facebook and these big companies, you know, paying decent comp, hiring like the top tier talent, all the new grads.
00:03:37:03 - 00:03:55:07
And so as a, as a sort of freshman in college, I started getting emails from all sorts of different founders being like, hey, do you want to work for, you know, Mixpanel or Pinterest or, Reddit, maybe a monk or something like that, you know, like a bunch of these kinds of companies and like a lot of the wiki sort of adjacent companies back and back in the day.
00:03:55:09 - 00:04:11:13
And, and I basically was like, well, this is kind of like a, you know, it's a lot, a lot of steps to like, work at a startup. If I can just skip all of those steps, you know, and, you know, just go straight to like, sort of like the penultimate step of being a founder. Why not?
00:04:11:13 - 00:04:28:22
And also, college is expensive. Pinterest is paying me money. So like, the just the math was like, this is kind of like a really great opportunity. I think in hindsight, I think now that I'm like a little bit older, I have some gray hair, like I think I am a little bit more moderated about like that decision than I used to be, like when I was a kid.
00:04:29:01 - 00:04:45:04
And this is obviously what I should be doing. You know, I did, I think, lose out on some sort of like social interactions and like building a network at college and like there would have been probably some utility to that. Like the old, the older you got like sort of like all these alumni networks matter a lot of for certain people.
00:04:45:06 - 00:05:06:15
But it was just like, yeah, this like accelerated pace of learning. And it felt like a moment in time that like, if I didn't join, you know, like, this was like this unique moment with, like, the iPhone, the App Store, all these mobile apps going from 0 to 1, that if I had like waited two more years or 3 or 4, etc. like a to your point, like it kind of like phased out a little bit.
00:05:06:15 - 00:05:26:13
And now we're getting back into this resurgence of like AI and and craziness in San Francisco again. So I think in that sense it was probably the right call. How quickly did you go from kind of Pinterest to starting your first startup and you went your first trip was Gumroad, right? Correct. Yeah. Yeah. I mean, pretty fast.
00:05:26:15 - 00:05:45:10
You know, I joined Pinterest in like January and I started Gumroad in September, so I. Oh, really? Okay. You were really accelerated, I know. Yeah, it was a bit too fast to be on. And then we raised our series a like February of the next year. So I went from kind of like being a high school student to like having a series from Kleiner and yeah, about like 18 months or something.
00:05:45:12 - 00:06:08:03
Something like that. Two years. Why did you say it was TV? But, it was there, was there things that you kind of regret in terms of, like mistakes you made or something like that, or just like, I just think that I never really was. So I was like, so focused on the future and just focused on, like, trying to, like, you know, kind of like hit this crazy personal month over month growth for myself that I never really I think of.
00:06:08:03 - 00:06:25:05
I spent more time at Pinterest, when I would invest in some stock that would have been nice and but I probably would have just like, had more experience and, and gone into gumroad with like a little bit more of a network and of like a point of view on how to build things properly. You know, what?
00:06:25:06 - 00:06:42:10
What are the right or wrong things? I mean, I ended up working out, so I have no, no regrets, but I was kind of on this accelerated path, that I was just really trying to, like, get to the next milestone. I'll get to the next milestone. And, and Silicon Valley, you know, there are all these there's a lot of this now vibe coding, bootstrapping type stuff on Twitter.
00:06:42:12 - 00:06:58:12
I think the nice thing about VC and sort of this quote unquote traditional approach is that there's like such a clear path to define of like what you're supposed to kind of be doing and like what success sort of is supposed to look like, which kind of just make it kind of like college in that sense when you're like, oh, I'm just going, I'm supposed to be doing this, you know?
00:06:58:12 - 00:07:15:13
And then I'm supposed to be raising our, you know, I'm UIC, and then I do demo day, and then I do sit around and then I do or Pre-seed seed a, you know, that, that sort of thing. You know, and so, yeah, I was just kind of like trying to kind of like get through the hurdles as as quickly as, like as I, as I could.
00:07:15:15 - 00:07:34:15
I remember in my early 20s, like six months felt like a long time, like I remember vaguely like, wow. Like I just got to do this so quickly. And now that I'm like 40, I'm like, wow, six months is like a breeze. I don't even know what happened to 2025. It's just gone. You might do you hire people like dropout, dropout out of college?
00:07:34:15 - 00:07:50:14
I'm curious of what's your philosophy on hiring folks and and funding founders. You know, who are that young? So in terms of funding founders, I mean, I don't I don't really ask, like, what's your age? I guess sometimes we will say we dropped out. It's it is, I would say, still a rare thing that it's a college dropout.
00:07:50:16 - 00:08:12:18
An early college dropout, like doing it in your last year, which is like, in my opinion, basically the same as, like, graduating. So I think it's actually rare. But, you know, I wouldn't say I this discern, I think one thing which, you know, it's probably part of a longer conversation is that junior engineers, like, it's actually kind of hard to hire them now with AI and things like that.
00:08:12:18 - 00:08:41:13
Like, I think having I guess maybe it's bifurcated, but like, generally speaking, I feel like senior engineers, like more valuable now that, like, you can quite quickly do, like, basic code and like, this more advanced stuff, is more valuable and more senior engineers can be more tanks, engineers, if that makes sense. There is probably a bifurcation where there's some just like really good junior engineers that like very deep in AI and can like show that they're like extremely productive.
00:08:41:13 - 00:09:03:23
But, I do feel like the current job market is like much, much harder for junior engineers than it used to be. Like five years ago, even, I don't know what you think. So, yeah, I completely agree. I mean, I think it's a super interesting phenomenon that, like, effectively coding agents, you know, the sort of the technical engineer is often like a result of having nothing else to do.
00:09:04:04 - 00:09:39:03
Right. And just being so focused on learning a specific kind of skill set and spending your nights and weekends like getting really into it. Could you been open source, like staying up to date on whatever is happening and being the first to learn this thing and apply it? And nowadays I find that, yeah, if you like, what I really care about these days is hiring people who's seen a lot of stuff and have a lot of experience, not necessarily people who are willing to work nights and weekends, because you can now kind of queue up a bunch of background agents, you know, via cursor or Devin Cloud Code Codex, like there's like several options
00:09:39:03 - 00:09:58:22
now, that's something that happened in 2025, you know, like, like, and, and the bottleneck is not, is basically not code monkeying, like, not actually typing the code, testing the code, making sure it works. It's, Yeah, it's like actually understanding. Like what tools even exist. Like what what what, like software architecture should we be using?
00:09:58:22 - 00:10:17:08
Where should the data live? How should we, you know, should we go single tenant, multi-tenant. And that, I think, comes with just a lot of trial and error, a lot of trying things and seeing how things have evolved. So and to be honest, like a lot of sort of these college dropout type personalities just like me at Pinterest, like they don't make great employees, like they make great employees for like a short amount of time.
00:10:17:08 - 00:10:36:15
They make great interns by the end of the day. Like, they clearly want to go do their own thing. Right. And so you're kind of this stepping stone, to them going out and doing their own thing. And their priorities are not necessarily going to be aligned with, like, writing the best possible code that's going to scale, you know, and be super performant.
00:10:36:15 - 00:10:56:17
It's going to be like just shipping feature feature, feature learning as much as possible and then like building relationships and then starting a company. Right. And so I I'm open to it. But it's rare you mentioned a bunch of tools that it seems at least from your, from your ex post, that you're like very into them and you, you get a lot of value out of them.
00:10:56:19 - 00:11:32:05
As in you get, a lot of productivity from your engineering. Is there one tool that you're like, hey, this is it as a cursor or or is it like, try everything and use everything? I mean, it's it's crazy how fast things change. You know, it's kind of like models where, like, my default coding model was like, first, you know, I started writing code with AI only as recently as, like, not even a year ago, with cloud 3.5 and cursor, you know, that came out about June of 2024, I guess July 2024 almost been a year then I was that was kind of my daily driver for a long time.
00:11:32:07 - 00:11:49:07
And then I started using Devin. And Devin was awesome because I could basically be on slack on my phone, and be like, hey, you know, fix this bug on the homepage or like, update this, you know, help center document to include this new feature. And it was kind of this, like asynchronous model of communication, which I really liked.
00:11:49:09 - 00:12:07:14
And so that's kind of and then, and then my coding model changed, you know, we switched to Gemini 2.5 Max, Pro Max, which is awesome. Now for opus and Sonnet are there. So I don't really O3 apparently is really good too. I haven't tried Codex yet. I've heard really good things about Codex. I've heard really good things about Cloud Code.
00:12:07:14 - 00:12:32:19
I just started trying it out today, actually. But generally I find that I'm either using cursor, like, if I'm in front of my laptop and I'm the one shipping the code, I'm using cursor, agent mode and if I'm not on that laptop, I'm using Devin via slack. And that's kind of like my paradigm. I think eventually Devin will kind of personally, I mean, I am really interested in this idea of like as coding agents get better, like, do you even need an IDE?
00:12:32:21 - 00:12:49:02
Like, why are you staring at the code? Like, what's the benefit of staring at a bunch of code that you don't you didn't write, don't want to read? I mean, tell me like the success rate of like Devin. Like you tell it on slack, like, hey, go fix this thing. Is it like 95% of fixes that, like 80% of fixes?
00:12:49:04 - 00:13:09:00
Like how much how much back and forth is it afterwards? Like, did your code review and then tell it like, this bit's wrong or like, go fix it yourself? I think it it depends on what you're building. But I think there's like this bell curve of, of stuff that it's like of complexity and or stuff that it's sort of like you can definitely do the stuff on the left, like the stuff that's like really easy.
00:13:09:02 - 00:13:28:06
And you know, one line fixes, you know, typos, things that aren't really like, sort of conditional business logic changes. Yep. It can do stuff that like, like a lot of, like API type changes too, because you can like run the tests and see those like the Json object is changed in this way. It can do a lot of front end stuff too.
00:13:28:06 - 00:13:50:21
That's like, you know, add a checkbox. So when the person hits the checkbox, pass this thing to the back end. But I think the architecture piece where I a lot of features end up being like this, where like a single feature you're building requires like a migration, a schema change. You know, this like multi-step planning that is, you know, front end, back end API docs help center doc changes like a big feature update like that.
00:13:50:21 - 00:14:07:16
It's not going to be able to do. But if you're willing. And this is kind of how I think about what what the job is now is that if you're willing to break it up into a bunch of atomic steps like that ends up becoming like what I hire people to do, and then at that point, you can kind of farm out the tasks and say, hey, Devin, update the API in this way.
00:14:07:16 - 00:14:30:06
Hey, David, you know, update the the front, end it this way. And cursor now also has like background agents or cursors is going to start you know go and I apparently they have a slack integration now too. So you know I do think this will become like the premier way of shipping code that like you will have like a chat interface with an agent and that agent will write the code for you, test the code for you, run the code in.
00:14:30:10 - 00:14:46:22
You know, it's kind of interesting. You're saying like for now, the the most important skill is being able to break down your problem into like, tiny pieces that like, it has a high accuracy and like being able to do and I guess like the size of the pieces will get bigger as it gets smaller. I would say there's, there's, there's fundamentally like two skills.
00:14:46:22 - 00:15:06:10
I think one is that one is breaking things down into there are smaller components, like for example, it's like, hey, I want like a green screen algorithm, right? Like okay, like what does that actually like how do you do that? You have to like iterate through all the pixels on the screen, you know, you like there's some like data structures and algorithms type thinking there that's required.
00:15:06:12 - 00:15:27:09
But then the other side, I think, is knowing what tools even exist. I think this is like a lot of where I find that I still have to kind of help people figure out what to do because I'll, like, do the right thing, but they'll, like, not realize that like, a sword exists, you know, like like they'll they'll do it by hand and they won't realize that there there's actually like a really great open source project that solved this problem really well.
00:15:27:15 - 00:15:54:10
And so I think that is also it that it's really hard to know what you don't know. And so, you know, if you're asked to solve like a problem and you don't know where ffmpeg exists, you know you're going to spend a lot of time. And so I think research basically, I think research, and product management, I think those are like the two fundamental skills like software architecture, product management and research, which ironically, are kind of like the the sort of like there's like jobs that like people like to make fun of PMS for not being really that valuable.
00:15:54:14 - 00:16:19:07
But actually, I think in this new world, effectively that becomes like the most valuable stuff, which ironically requires like a lot of meetings, a lot of, like, sitting by yourself, like doing research, the coding aspect and even the design aspect, if you think even about what design often is. Right. It just like looking at how other people have solved this problem in a bunch of different ways and then like, sort of like transmuting into, like what makes sense for your context.
00:16:19:09 - 00:16:39:11
It's not like going to a museum and like, you know, like painting or like, you know, grab, you know, crazy graphics type type stuff anymore on the design piece, I think there is an element of, of the both. Right? You have to not understand what context you're operating in. I think you also need to think about like if you want to reinvent something, then you have to understand things outside the current context.
00:16:39:11 - 00:17:04:15
And then, you know, you want to maybe bring some of those ideas into what you're currently building. And so there's there could be a, there's an idea of like, okay, yeah, what have other people done? But there's also like, what should this be and what should this be. Could actually come from some other I, think so. I do think designers are still very valuable in this age because the, you know, AI is not going to necessarily come up with the brand new ideas that can sometimes change the game.
00:17:04:17 - 00:17:20:05
But I have one question for you on the code stuff, which is like, how do you interview engineers and in this, in this new age group, we've done a bunch of interviews at tribe and, you know, everyone's using AI now, obviously, which they should, but then how do you evaluate, like, you know, the best engineers?
00:17:20:07 - 00:17:44:12
Yeah. It's hard. I mean, honestly, I think the the conversation is like a lot more casual. You know, it's more about like talking to them about just trying to understand like how they spend their time and what they think about, and how they would solve a problem, not necessarily solving the problem, but like, you know, if you're asked to, to do something, you know, like I done, right, for example, we would often ask, like what?
00:17:44:13 - 00:18:01:21
Like, you know, in the in the beginning, the way I built Gumroad was it was every, every, every user had a balance. Like on their database. Like in their, in their, in their row. Right. So like if you made money the balance went up. If you, if you got paid out, the balance went down, which is like, you know, any, any account it would like have an aneurysm.
00:18:01:21 - 00:18:18:04
Like if that's like what what about logging. Like how do you know, double entry it like all this kind of stuff. So how would you design a system that would like, basically allow for, like, for example, like you, you, you get paid, but then you issue a refund. And how do you map this over, you know, multiple months?
00:18:18:06 - 00:18:37:18
And how do you you know what what what you know, what would be the right approach for that? And not necessarily like the answer has to be correct, because I do think at this point, like if you kind of feed that into like an AI, like it will do a decent job, frankly, of doing that. You know, if you give that to a one pro or like oh three and it does research and web search turned on like it'll it'll do a really good job.
00:18:37:19 - 00:18:55:10
But I think listening to like having them do it extemporaneously and having them like what kind of like the chain of thought part of the AI, right? Like having them walk through like where, where, where did where does their instinct take that. Like, oh yeah. Like, oh, I have this experience. Why we did this. That's not exactly the same.
00:18:55:12 - 00:19:17:06
But it does have this like similar element to the problem where you need logging, you know, like security, logs or, you know, something, something like this, like a HIPAA compliant Soc2 compliance or something. And maybe there's like and and just trying to get a sense of like, how interesting is this? Like, has this person's life been, you know, like what life experience do they have and just and just how generally intelligent they are?
00:19:17:06 - 00:19:39:05
Because I think AI is really good at like helping you on a specific domain, but not necessarily giving you the, the high level, like interpersonal skills and like the high level problem solving and business skills. So yeah, a lot of it is like, okay, I know you're a good engineer, or you can be if if you're given these tools, but like, what have you, what have you built, what have you struggled with?
00:19:39:05 - 00:19:58:22
I know Elon, for example, he asks like, like what's the hardest problem you forked on? Like walk me through that. You know, the other one is, like, just explain your career. Like, explain your decisions. Like what? Like what leads you to this specific opportunity? Because I think that's a huge one, frankly, it's it's like, doesn't matter how good you are, but, like, what is your alignment with the problem that we're trying to solve?
00:19:58:22 - 00:20:23:01
If you spend like 15 years helping creators make money on the internet, then you probably like a pretty good fit for what we're trying to do. If you're just optimizing for like, total compensation, like we're probably not the right fit for you, maybe it's, good fortune to talk about Department of governance, of government efficiency. So, you were you were a software engineer that, for a bit, like, tell us about, like, how you got into it.
00:20:23:03 - 00:20:39:02
I only found out, like, six months ago that, like, random Silicon Valley people were kind of going to Doge, which I thought was kind of cool. How did you kind of join it? And, and then I guess, tell us, like, what you worked on and what the experience was like. Yeah, totally. Yeah. So I joined in March.
00:20:39:02 - 00:21:05:10
I started talking to them in December, mostly because I and I was yeah, I was a software engineer. I was working for a, VA the Department of Veterans Affairs, which I learned is like the largest civilian agency in America. It's like the budget is $350 billion a year. So it's a huge agency. It's number two, the Department of Defense basically of in terms of scale and, and and yeah, so I got connected through a friend of a friend.
00:21:05:10 - 00:21:21:21
This is kind of how life often works. I did apply it through the like, the traditional recruiting means, but I didn't really hear back. So I got I had dinner with somebody who mentioned who knew I knew someone working at Doge. And I was like, oh, that sounds awesome. I would love to do that and see if I can, like, have an impact.
00:21:21:23 - 00:21:38:10
But you're operating a company at the same time, right? So, like your founder, CEO of a company, how does that you know, it's like, you know, am I are you going to go pay for doge? You like, is it, you know, so how did you decide? Maybe he's announcing that soon. Yeah. No, I mean, I think I think I think it's a it's a good fit for certain founders.
00:21:38:10 - 00:21:50:08
I mean, you know, I would not recommend it, like, you know, if my portfolio company founders were like, hey, by the way, I'm going to go join does like, what are you doing, man? Like, you got the company to run. Yeah. But I do think, like, you know, Gumroad is just like this weird, unique company where I started at one.
00:21:50:08 - 00:22:10:15
It's super old now, you know, it's. I've been doing it for, like, 14 years. It's profitable. It's been profitable for several years now. You know, it's it's sort of like not on the venture capital train anymore. Just profitable issues, dividends. I still own, like, the majority of the company. And so, you know, I sort of, like, solved all the firefighting.
00:22:10:15 - 00:22:30:10
There's no firefighting going on. Like, we're shipping features. There's still stuff that we're doing or, you know, building features for creators, but it's not the most sort of like on the ground creative startup. Basically. It's not really staffed by guys anymore, you know, just like a boring technology company. You know, we, we help creators make $150 million a year.
00:22:30:10 - 00:22:50:03
It's been roughly flat for, like, 2 or 3 years now. So it just felt like an end. It was a temporary thing. Well, you know, I wasn't going to go join and, like, leave that behind completely. But if I could do this for, like, my plan was for 60 days to go and join 60 days and sort of have this, like asymptotic learning curve where I could, like, learn a lot very quickly and hopefully have an impact very quickly.
00:22:50:05 - 00:23:19:16
Just like your Pinterest is. Right. It's a yeah, I guess. Yeah. And I had a newborn the new day. Why why would be 60 days. That seems like like, can you really have an impact to learn a system and like, implement something in such a short time? I mean, I think if you I think so. I mean, I think that at least if the promise of Doge is met, that it is like there's a lack of a certain kind of skill set and knowledge that if you introduce it into the system, you would be able to make at least, at least sort of an inflection point where like the work would need to
00:23:19:16 - 00:23:38:22
continue, but you could have some sort of like non-zero amount of impact in, like a in a relatively short amount of time. And the 60 days thing was me basically being like it was already starting to get pretty late, and I had a newborn coming my first that with the due date of May 22nd. So I just told them, like, look, I can only really do this, like, once I have a kid.
00:23:38:22 - 00:23:55:11
This is like, life is different. I'm not going to be able to, like, leave everything behind and, like, go join a startup for, you know, 24, seven and that sort of thing. At least not for a while. So I can really commit. Now, this is kind of my, like, last chance at doing something like this, for 60 days.
00:23:55:11 - 00:24:17:08
And I'll move to DC and I'll just work, you know, all the time. There's nothing else to do, really. And so, Yeah, that's kind of like the the concern. And I honestly, I think I did get a lot done, like, I joined VA, there was like a sort of a series of quick signal chats where I talked to, like, you know, 2 or 3 different people, up to Steven Davis, who who kind of, like, ran Doge for a while.
00:24:17:11 - 00:24:32:12
I think he's now out as part of, like, the whole 130 day limit thing. I don't know, I don't really keep in touch too much with them. They're pretty locked down. And then they said, hey, we have a spot for you at VA. You'll be the first software engineer. VA. You'll help us do like the Doge agenda at VA.
00:24:32:14 - 00:24:48:01
I was like, cool. That sounds awesome. And then I just showed up, you know, may, may, March. 17th and got my laptop. And at that point, I had no real clue, even, really, what Doge was going to do or what the mission was. Besides, like the stuff you read on Twitter where you don't really know, like, what's real?
00:24:48:01 - 00:25:07:05
What's not real, like what's propaganda, what's media bias, etc. so I was like, I'm just going to go and learn, resolve my FOMO and hopefully have an impact, like hopefully like use cursor, you know, and like ship about how many software engineers showed up at VA. I was the only one. You're the only one. Yeah. For the whole time it was the only one.
00:25:07:05 - 00:25:32:00
I was the only one. I mean, there's some, you know, VA employees, right, that are software engineers. Maybe, maybe a hundred or so out of the VA, by the way, has 473,000 employees. So it's a huge organization. It's massively. Yeah. It's shoot. So you show up and then one. Yeah. And then the one Doge person who's already there, who's on, was kind of more of like a business mind, was like, you know, we need your help.
00:25:32:00 - 00:25:48:17
I was like, awesome. What you want me to do? And they were like, oh, you know, this is what we're working on. We're like, reviewing the VA's contracts. We have 90,000 contracts we're reviewing. I was like, that's a lot. He's you know, he's like, yeah, we're basically just like reviewing the manually, right now. I'm like, okay, well, we should probably use like AI to do this, you know, like at least a first pass.
00:25:48:17 - 00:26:16:11
You could use AI. And so that was like one of the first things I did was like basically write a script that I published actually on GitHub that anyone can read and feedback. And some people, every once in a while tell me, like, I'm evil for publishing it. It's great. But, you know, I basically just wrote the script that's like a pretty simple, like, you know, this is kind of a good example of like, I think even if you're a really good engineer, like, and you're very smart, if you don't understand what tools exist, you may not come up with this kind of this, this piece of software where it's not very
00:26:16:11 - 00:26:46:17
complicated on its face. Basically, it just plugs into OpenAI's API. And, you know, VA has like an internal to VA, version of, GPT four. And so we use that, and, you know, just basically loops through every single PDF in the system, which I had to get access to. And, basically just extracts the text and then runs it through an Lem that says, hey, you know, this is the kind of contracts that we're looking to cut in this sort of way.
00:26:46:19 - 00:27:04:07
Should we cut it or not? And it would return, you know, basically this big CSV of like, you know, about 10% of the contracts we should sort of potentially terminate. And so it's kind of just like helping, you know, I learned over time that basically like that's sort of the doja gender has like three stools, to the three legs to the stool.
00:27:04:09 - 00:27:22:18
One is cutting contracts, two is helping with the reduction in force. And then three is shipping software. So actually like, and they're, they're all kind of related because I think without shipping software, it's hard to cut a lot of contracts. It's hard to like reduce the force. But unfortunately a lot of the DOJ's sort of like the bandwidth.
00:27:22:18 - 00:27:39:02
And I think the stuff the Trump was okay with them focusing their time on was a more on that first two. So I didn't end up doing as much software. I kind of just went kind of like on my own and was like, hey, VA people like, can I work on stuff? And I kind of got like shown a bunch of different software projects they're working on.
00:27:39:02 - 00:27:57:16
And I saw this one vague, petty thing. They were building. I saw this VR chat bot stuff that was kind of aligned with what I was already doing, and what I had a lot of experience with. And then I just started working on those projects. As just like a basically a pro bono software when you would, when you were looking at a contract, like, how do you decide?
00:27:57:16 - 00:28:14:06
It's like, cut a bull. Is that like, is it just it's very expensive. Is it like hardly anyone's getting utility out of it. Like, what are the what are the metrics that you're looking for to decide? Yeah, I mean unfortunately like you don't have a lot outside of the you know, you were kind of kind of coming into the system with like, not a lot of context.
00:28:14:06 - 00:28:31:22
Right? So to, you know, to to branch point, like, you know, it's helpful to know, like how other people have done or like what's happening outside of the exact context that you're in, right? And in this case, we have like these PDFs which have a lot of information about the contract itself, how much money it is, like what it's for, the different, line items.
00:28:32:00 - 00:28:50:12
But we don't the vendor etc., but we don't have like, like what is the utility of this? So you then have to like go find some other data set that says like, okay, this is like the result of this contract and like what is the value? And it never really, you know, you could ask the contracting officer for justification so often that's like that would be the next step.
00:28:50:14 - 00:29:14:11
But for, for in the beginning we were just kind of doing it topical. Right. We're basically saying what are the kinds of contracts we don't want to renew that we want to terminate, basically things that don't affect veterans. So if there's a bunch of contracts that are kind of like internal support, kind of like social media posting, recruiting services, video production, you know, we would kind of sort of like come up with this script.
00:29:14:11 - 00:29:32:07
And the way I would do it is I would basically like, do my own best attempt with, like, the doge person who was already in the room, we'd run it on like 10 or 20 contracts, and then we would kind of like disagree or agree with the results, and we would edit the prompt and we'd kind of just keep doing that at larger and larger amounts, different contracts.
00:29:32:09 - 00:29:52:04
But we brought in some VA career people to kind of like, you know, say like, do you guys agree with this kind of stuff? Because effectively, at the end of the day, like, I, I think it's like it's just like replicating what human judgment would have done. Right? Like if you have a really good prompt, it's just kind of doing what the 300 contracting officers would have taken like six months to do because there's just so many contracts.
00:29:52:06 - 00:30:08:21
And just like saving them 80% of the time, like it's not full self-driving, right? It's not going to do all the work for them. But it can like at least take one of their tasks and automate it. And so, yeah, we kind of just iterated on it and we had like sort of we ended up with like 300 lines of English that I later asked Elon.
00:30:08:21 - 00:30:31:19
I was like, hey, can I open source this, such that, you know, in my opinion, that, you know, people may disagree and agree with you with this sort of the Doge approach. Maybe it's too aggressive on certain things, etc. but if we published the these like let's say every agency has like a similar 300 line prompt, that's a kind of a first pass of this like chaotic contract process.
00:30:31:21 - 00:30:50:01
Wouldn't that be good for like the average person to just see and know about? Right. Like just publish on GitHub and say, this is the you know, this is the prompt like help us improve it, show it to career employees. Unfortunately, I think Doge kind of took a little bit more and he was cool with that. So I did, but I think Doge ended up taking a it was a it wasn't enough.
00:30:50:02 - 00:31:11:11
I think in, in my opinion at least, like it wasn't like working with the organization as much as it was like kind of coming in. It was got more like a little bit more like a hostile takeover, versus like, you know, like, let us we have some skills that you guys don't really have, and we're willing to work a lot longer than you may as like you've been doing this job for 20 years working, you know, roughly 9 to 5.
00:31:11:13 - 00:31:38:23
Let like, let us help you do what you want to do. And it was a little bit more like, you guys are not aggressive enough. You guys are not cutting contracts that you should be cutting. So we're going to tell you which contracts to cut. And if you don't cut them, there's nothing we can really do except kind of like, you know, try to like, tag someone in at like, a higher level of leadership, using our doge aura to be like, hey, look, like Trump said this, but, like, you know, Doug Collins said this, but, you know, the contracting officers don't really seem to want to do this.
00:31:39:01 - 00:31:57:23
And that was kind of a lot of the, a lot of what the effort was spent on a ton of it was and this is more valuable was me sitting in on a lot of it, spend meetings there, called for Tara reviews. But this was like a law that was passed relatively recently. That basically meant that the CIO of each agency gets to approve every IT contract.
00:31:57:23 - 00:32:31:06
All IT spend has to go through the CIO. And so the CIO was like, hey, Sol, you can like sit in on any media you want related to it, spend and you're in. And most of the time I was like, I was maybe all the time. I was like basically the only software engineer in the world. So like, you can imagine the VA has, let's say, like, I think it's about $8 billion of IT contracts, you know, with IBM, Booz, you know, Deloitte, all these Accenture, it's that are like all these people, basically almost all the software that that the federal government ships, like any software that you use to pay taxes to do whatever,
00:32:31:08 - 00:32:48:00
is an outsourced service generally, like the government, some of the best software engineering companies in the world, right at Deloitte. IBM. Yeah, exactly. Yeah, exactly. Deloitte. Right. Deloitte. I used to think Deloitte was like a toothpaste brand. You know, I was like I was like Deloitte. What? Like what are they doing? It's like they're a software company.
00:32:48:00 - 00:33:05:11
Like that's what they do. That's what IBM does. Like that's what the center does. Like I, you know, people always joke about like a shaving company presentation. So yeah. Exactly. Yeah. Yeah, exactly. It sounds like a exactly. I guess how much money, you know, how many contracts did you end up canceling? I don't know if it's public, but how much money did you save?
00:33:05:11 - 00:33:29:13
And then also, like, is it continuing? Is there like a new cycle there that's like a new software engineer or is it just like now it's like stalled that you've left? Yeah. I mean, unfortunately I think it is kind of like unlike, you know, once you leave, there's no one really. They're picking up the work like that. Just the, the amount of sort of constraints on those are just like, there's a lot to be, you know, you got to move to DC, you got to, you know, it's it's a very uncertain role.
00:33:29:15 - 00:33:42:15
A lot of people don't really like you if you do it. You know, like there's like a lot of stigma associated with it. So I think that that's the team was a lot smaller than I was expecting. I think there are certain projects that will do really well. Like I think Joe Gabe is working on, like the OPM digitized retirement stuff.
00:33:42:15 - 00:34:00:00
I think that will continue. That was a project that was happening pre Doge. And so they just kind of like massively accelerated it. GSA is working on like centralized procurement. I think that will do really well over time. But I think that yeah, I think at VA it's just too it's just huge, a huge it's like a cruise ship, you know, it's just like this massive thing they should.
00:34:00:00 - 00:34:16:01
And what I recommended was to hire software engineers. Like, I like kind of like the number one most important thing I could do would be like recruit and filter your software engineers and like, actually, if you have ten people who can code that, I don't need to be in these meetings. And then you would save a lot of money.
00:34:16:03 - 00:34:36:14
I think I ended up I mean, the, the Doge number on VA is about $1.6 billion in contracts that were cut, but most of them were cut, actually, even before I got there, all that low hanging fruit, the in terms of the contracts, I was able to cut, maybe like 10 or $20 million a year worth of contracts, which is still impressive because it's like I had to burden that work and do it myself or whatever.
00:34:36:16 - 00:34:56:12
But, you know, it's yeah, you know, it's like more than the revenues of gumroad, but, you know, but but it's not they like, doubled my impact on the world or whatever. But, but it is. Yeah. This like one and done thing. And now you're seeing what the big beautiful bill that like to be honest, like, you know, doge is like a really nice marketing thing, but but, you know, it's kind of a scapegoat for Trump, I think, to a degree.
00:34:56:12 - 00:35:13:06
And I think Congress doesn't really care. Like and I think, frankly, the average American like, you know, think they like the idea of, like, you know, saving money and making the government more efficient and better. And hiring is often years of modernizing. But if it comes at the behest of like, other things they really care about, you know, it like, you know, take the debt, for example.
00:35:13:06 - 00:35:36:04
Like, it doesn't seem to me that like, people care that deeply about it. And, you know, unlike Elon, who seems to really care or care about that. So, yeah, I think I think, you know, I hope and I've heard that Doge will spend a lot more time like on this sort of the shipping software piece after the easy wins are done, you know, kind of like a, you know, private equity buyout, you like, cut costs first and then you start to rebuild and build new stuff.
00:35:36:06 - 00:35:51:00
I just worry that, you know, at the end of the day, Doge is not really in charge. Who's in charge is Trump. And I don't know his appetite for actually being like, hey, you know, we should have an app in which you could pay your taxes. You know, I don't I don't know if that's kind of the kind of thing that either party is that excited about.
00:35:51:02 - 00:36:06:18
It's sort of ends up becoming more of an ideological battle. And the carriers are the ones who actually care, right? They're the ones who actually work in the VA who actually care about veterans. And so that's kind of why I when I joined, I was like, I just wanted to hang out with the career employees and, like, help them become more competent.
00:36:06:20 - 00:36:21:10
You know, I got curser, I got V0, I got Devin, I just gave them a bunch of AI tools. I like, would do demos for people in the engineering org and just be like, hey, by the way, this is how fast you can move if you're willing to use V0. And some of these AI tools, you can prototype stuff.
00:36:21:10 - 00:36:46:17
You can, you know, and I think that's where most of my impact was, which is unquantifiable. All right. It's hard to say. Like what's the value of sort of upskilling a bunch of VA people on AI? But I'm hopeful that that that impact is like is quite massive over time. You know, there's something to be said about this idea of like, private, you know, company leaders and like, you know, talent going into government for a while and having an impact.
00:36:46:17 - 00:37:07:22
I think that's something that could be really valuable if it were structured correctly. And if we had a system for doing this. So, like, I think this is not a bad idea for like, you know, you know, anyone in private industry could be, you know, Ahmad or me or even going to the government for 60, 90 days and doing something and then having some sort of, some sort of tail to that, like maybe once, you know, you come in for one day a month after that.
00:37:07:22 - 00:37:23:13
Right, just to see how your initiative is going. But I do agree, it's a hell they can't be with this like aggressive like, you know, you guys screwed up. Like we're coming in to like, fix your problems. Like it has to be much more of a corporative start. It has to be corporate for sure. For sure. And there needs to be buy in from the president, you know.
00:37:23:13 - 00:37:48:12
So yeah. Yeah. You need you. And you know, I think I think the credit that that Trump and Elon, everybody else deserve a lot is that they got us in the room I think like, you know like I heard from people that like Biden obviously like, you know, generally the average tech person is probably leans left. But just like, you know, when you actually go to DC, there's like just zero interest in, like having software engineers, like build software like it was it was very policy heavy.
00:37:48:12 - 00:38:09:09
Right. Or, administration and generally is. So I think that was really impressive that led that Elon and convince Trump that like, no, we need actual people who build software in these in these meetings, you know, in these agencies. And I think that I but yeah, I think it's sort of like it becomes because it became so ideological, it just alienates most people.
00:38:09:09 - 00:38:33:06
Like the smartest people aren't willing to, like, pick a side. Right. Like you just want to make things better. And I try to be as apolitical about it as possible. And effectively it meant that I wasn't like a good fit culturally for Doge. Eventually, like it just didn't work. Because I think that's what they at the end of the day, that's like the job they were really signed up to do was to kind of just like hatchet the, the federal government to, to, to a degree that most people would not be willing to do.
00:38:33:08 - 00:38:49:00
And, you know, you could even agree with a lot of it, but just like the political flak is just too great. But yeah, if there was a system, if there was like a really great way for like all of these amazing tech founders to spend a day, a week or, you know, two hours via zoom, like, it doesn't even have to be moving to DC, right?
00:38:49:00 - 00:39:05:19
It could just be like like every agency should have access to private industry experience in a way that, you know, I think the what they would probably tell you is like, oh, it's, you know, corruption and like, you're going to like self deal and like all the, you know, like the flak would be it would be, would be a lot.
00:39:05:21 - 00:39:19:04
But I think the, my solution to that is like and what I, what I told viewers, what I told Doge was like, why don't we just live stream the meetings? Like, why, why do these meetings even have to be private, right? Like, yeah, if, if, if like, Patrick Johnson is like helping the Treasury think through a problem.
00:39:19:06 - 00:39:39:02
You know, it's not like, like what's national security risk about, you know, like, why not just put that on on the internet? Yeah. And that's a great idea. And I yeah, I think over time that's what will happen. You know, the government is is actually required. OMB has this rule where the 20% of the code the government, pays for has to be open source.
00:39:39:04 - 00:39:55:07
Oh. Really? There's a lot of open. Yeah. So this is it's slowly happening, you know, and I think that, like, there'll be a doge, you know, usds like 3.0, 4.0 like, this is like, you know, it's kind of a startup within the federal government, and it's it's it's kind of this was like the series B or something like that.
00:39:55:07 - 00:40:15:21
Right? And like hopefully it, you know, hopefully, you know, had like 40, 50, 60 people and hopefully, you know, it doesn't get next by the next administration. It it continues to like thrive and over time like they just keep changing its name. Exactly. Yeah. Like that's like the it'll always be usds but they'll always come up with like some new acronym for it.
00:40:15:23 - 00:40:31:10
The government loves acronyms like I think that's like the best thing. It's kind of like getting a startup to, like, every, every become naming every business. Yeah, exactly. Naming is so fun, right? Like, it feels like you have a real stamp on the world when you get to name something. I worked at ExxonMobil for an internship, and they love acronyms.
00:40:31:12 - 00:40:50:15
Well, I can also suggest names I want to touch on one thing, as I like. So, you've been through this journey where you raised money from from Kleiner, right. Series a and that, you know, and then you got off the venture track. So this is a very interesting thing where you've gone from like, venture to non venture, which is normally very difficult.
00:40:50:15 - 00:41:06:03
Right. Like because the venture capital is expecting something. No big companies just die when they just go from like venture to non venture and the founders start a new company. How did you manage that transition I guess Gumroad probably wasn't didn't have this escape last year is another round or something. But maybe tell me tell us how that.
00:41:06:04 - 00:41:21:13
Because I remember the round was quite hyped at the time. You know, I think Mike Abbott was the, partner, and he was also, you know, a hotshot new VC, right? Yeah, totally. Yeah, it was a it was a hard I mean, I remember back in the day, $7 million series A was like massive, you know. Yeah, yeah.
00:41:21:15 - 00:41:38:09
Now, now, now it's a little bit different. Yeah. But yeah, I mean, basically we fail to raise our series B, so we really wanted to, you know, it was not my plan to get off the venture track. You know, I was like, no, stay on this thing and like, reach this like, why can you raise it? You were you weren't growing fast enough or basically.
00:41:38:09 - 00:42:00:11
Yeah, we were growing. I think that year we grew like 60 or 80% year over year and thrive, I think, had just put money into Patreon at a really high valuation. And Patreon was growing, like much faster than we were, you know, had recurring revenue, you know, by, by definition, and I think we were just never we were always doing good, but never great.
00:42:00:11 - 00:42:17:19
We had, you know, we had like a great product, great brand, great team, great investors. Like, we never like I think the good but not great is like the real VC trap. Yeah. Like if it's great or bad, that's easy. But good is really difficult. Yeah, exactly. I think we were, we were we were, we were doing good.
00:42:17:19 - 00:42:35:19
And that's just, you know. Yeah. Not good enough. Right. I guess normally in that situation VCs would push you to go sell it. Right. So like yeah. How did it come to like you actually like got them to like basically like sell the stake. Right. Like what was the what was the I mean, basically I was like, I don't want to sell the business.
00:42:35:19 - 00:42:56:18
So I really I, you know, I had a lot of faith in it. Maybe I was just stubborn. Maybe it was just too hard to do it, but I just really didn't want to didn't want to go down that path, especially in, like, a forced sale, you know, like at some point, like, strategically, it maybe it makes sense, but, like to just do it, you know, to generate like $0.30 on the dollar or whatever.
00:42:56:18 - 00:43:18:05
Like it just didn't feel like, a great outcome, like, you know, especially if it led to the product eventually getting killed by the acquirer. And we did talk. We we talked to Patreon, Kickstarter, stripe, square like those kinds of companies, media. I think, you know, they offer like some, some amount and, and but it wasn't it was like an acquire more than it was like a product acquisition.
00:43:18:07 - 00:43:35:18
So I was like, this doesn't make sense. Like, I'm just I'd rather just like, you know, let almost everybody go and just, like, run it profitably and see what happens. And the product is still growing 60, 80%. So like, maybe eventually it'll be something interesting. You know, Covid happened. That was like a big moment for us. So like the some of that panned out.
00:43:35:18 - 00:43:53:23
But but basically I was like I'm just. And Mike to his credit was like, dude. Like just do what do you want to do? Like, I'm not going to pressure you to like, force yourself to sell or force yourself to shut down or, or, you know, or whatnot, like there's some opportunity cost your, you know, like you, you only have a certain amount of time to, like, do this kind of thing.
Share this post