Garry Tan: 'I am 90x of myself' — G Stack hits 60K GitHub stars, 30K daily users as Tan evangelizes AI-native software development

saying that hard hard companies are hard

full analysis company by company.

This is sort of a cool launch video. It feels like a blog post as

get a DM of the full

full video. Like it has a lot more there's there's like more of a thesis and meat on the bones.

He ratioed himself in the video. He got way more likes in the video than Well, we have Gary Tan here. Here we go.

Fantastic.

If all

Let's bring Gary Tan in to Alter. Gary, how are you doing?

Doing great.

Is it over or are we back? Is it over or are we back? Everyone's saying we're it's over. We're back. We're back and forth. How are you feeling today?

Oh, we're so back.

Okay, break it down. How are we back,

founder?

Oh my god, how could we not be back? I mean,

explain what's exciting. we are at the dawn of like being able like software is totally fungeible now and then basically the tricky thing now is like you just got to spend the money on the tokens and that's like one of the things that's I think people are very very afraid to spend money on tokens and

uh the things that you can do right now like you can literally make open claw

Yeah.

And uh the reason why Pete could do that was he's actually like really really successful previously. Sure.

And uh he could let it rip. Like I think one of the key things that people have to understand about like

the current moment is um

you cannot be precious about your like clawed code max account.

You have to like really just let it rip. And when you let it rip, you can create like these things that

like set the world on fire actually.

What about all the other things? I mean so YC is interesting because there was a you know for a long time it was about technical founders uh you know writing code oftent times still is right but it was like okay you're technical you are you are you are going to write code while you're here but also you're going to talk to users you're going to figure out a go to market motion you're going to find a hard problem you're going to figure out you know secrets of how this business works what makes your moat deep and whatnot and it feels like is AGI capable of that or are we in like the software

singular? We still need people. I mean like this stuff is it's like a you know it's like Jiminy Cricket man but you still got to be you know Jiminy Cricket's got to stand on someone's shoulder and it might as well be yours.

Okay. So don't

how so so obviously if somebody sent you an investor update a YC port co and said and the and the only number they shared was how much they spent last month based on what you're saying you might be like okay you're you're on the right track but that's clearly not a metric that's yeah it's not a metric to optimize for how you know what what should founders be optimizing for you know when they're locked in

something people want right it's crazy I don't know I mean I've been I feel like I'm at the center of a lot of like weird controversy like did the guy lose his mind? What's going on?

I've been thinking about it and it's like uh it's not actually about me. It's about like people's relationship with their craft

and that has to change. And so uh I think my response to the haters is like have fun coding at 1x speed, bros. Well, I don't know. I don't know that the I don't know that the haters are are not also using the tools and not and not generally excited about it. I I think I think it's just primarily

uh

like I'll give you a concrete example. Like I noticed yesterday I threw I had a throwaway comment cuz I I created GStack. It's got uh 60,000 stars now. Like I'm actually growing faster than OpenClaw by stars on GitHub right now. And uh you know I mean they had their moment and like

I'm you know on my way, right? Um about 30,000 people use GStack every single day. I've been getting emails that are pretty awesome actually. Like people trying to start consulting firms and they just want to like uh you know feed their kids and maybe they lost their job and then they literally are like starting this consulting firm. I've never done this before. I like sit down with a client uh you know I just open office hours with GStack and as they talk I type in you know what what we're talking about and then I'm like live talking to you know and they sign a customer on the spot within like 20 minutes of talking to someone because of the GStack like open uh office hour skill

to build custom software for that.

Yeah, exactly. Okay.

Yeah. And then maybe you know the thing is custom software you know I was uh at Palunteer when you know Stefan Cohen and Sham Sunker basically invented the idea of um

your forward deployed engineer right so everyone's now

that's what you're doing everyone can be an FTE okay

and the thing is like if you don't know how to do it like this thing is like the training wheels that will teach you

and then you know it was the training wheels for me as I was learning how to do it like the original prompt I had posted on Twitter uh right when I was working on Gary's List and like those things like both both my uh engineering prompt to uh like shake out all the bugs and get to 100% test coverage. That was my plange review that is in GStack today. And then the other one was a plan CEO review. I call it my Brian Chesky review. Okay.

Like it's like having Brian Chesy sit on your shoulder and be like I mean that would be a heavy person to sit on your shoulder

but like he's going to ask you are you in founder mode? Are you moving fast enough? Are you talking

It's like what what's the 10star experience, right? If you've ever seen him talk about the 10star experience, like that's what having CEO plan CEO review in the Gstack skill feels like to me.

Okay.

And um

I really think that that's awesome. Like I want everyone to have that. And uh

so I I was like having a throwaway thought like you know one of the core things I made was a browser plugin, okay, that uh wraps playright. And so, uh, and you know, the reality is like I'm just solving problems for myself. Like I was building, um, Gary's list and I found myself, you know, the plan CEO review worked well. The plan end review worked really, really well. And then I found myself like uh, you know, I'm using conductor. I have like six or seven windows open at that time. I'm up to 20 right now. And um,

wow.

I I found myself like running between windows uh, just doing manual testing. And I'm like, great. like I worked so hard to like build all these skills to like get way faster and then now I'm just a blackbox QA engineer for the robots. Like this is so boring. How terrible. Can I

uh automate it? So I opened another conductor window and I was like, man, why does claude in co, you know, claude in Chrome MCP suck so much? I don't know if it sucks now. Like it's been a month or something. Maybe they fixed all the bugs, but

it literally couldn't do it. It's like two or three seconds, 5 seconds, crazy context bloat. it was unusable. And then I said, okay, like, well, I know Playright exists. Can we make a CLI that's like a thin shim over what Playright does? And then basically it did it. And then that was what the first V1 of GStack was. It was those two scripts and uh like a Playright browser plugin that like allowed me to not QA anymore. Like I could just say, okay, now QA like you know what we did in this branch. We have all these plan files. like go line by line and act like my QA engineer. And it did it. And I like the first time I used it, it was like a 100 milliseconds. It was doing it. It could log in. It could like click on things. It like took screenshots of things that were broken and then it would fix them immediately. And I was like, "Oh my god."

Yeah.

This is how to do it. Like this is how you build a software factory. It's like I automate each of the steps. It's like I've built so many pieces of software over the years and I've helped so many people do that and it's like this is the process and so now there are 30 of these skills

and then you know now you still like I think we're still in manual mode like I built an auto plan so that now like all of these different pieces like there's a recommendation in each of them um and like if you really want to be in the weeds and understand what's happening you can just like read the recommendation you can talk with it you be like well I like option A but like option B sounds better for X Y and Z reason like let's talk about it, right? Um, so yeah, it's interesting because like I'm arriving at similar things to what I think like Devon or other automated software factories are doing, but I'm doing it in the open open source MIT license like anyone like if you don't like something just fork it, man. And like,

you know, yeah, just make it yours. And I have like 200 PRs I have to look at right now, you know,

PR review. MD, we'll take care of it. Uh what what was the is is Gary's list uh more of just an experiment to put this to work because I mean it it's

because in some ways like you didn't have to build Gary's List you could have used like a

Substack or or fork something that's already open source but building it from scratch like are there clear benefits where you're like I'm really happy or was it more like

experiments I'm having right now that you can't you know if you use package software or you can't do is um

all the emails come from me directly like the from line is from me and when you reply to it so I don't send you know how like you sign up for New York Times or something like that you get a inbox like a daily digest the same digest that you everyone gets you get a personal one

oh you know you do have to apply to get into Gary's List and you have to like sort of tell me what you're into and interesting

but you know all the emails come from me directly and I don't write them the AI writes them but I do read them

interesting So everything that and you know when you reply you're it's going to my own inbox.

Wow. and AI is helping me read it. Yeah. And it's like going into the personalization system I built so that we can send you things that are more relevant to you, but

like in some sort of weird parasocial way like Gary's List is my experiment and um

communication at scale personalized I can be your personal assistant on like hey I'm mad about politics or like my kids getting you know can't get a good education in school. What should I do? Yeah. And like obviously I have a chat and it's like you know state-of-the-art like Opus 4.6 six thinking like all that stuff like

you know it has data sources you know there have been users on there that have um gone online who are like just engineers like living in Pleasanton like there's this one guy who uh never got involved in politics before and he's setting sending letters to all of his uh representatives based on the policy things that are like screwed up in his life like his kid you know

his kid's school is messed up like there's all the you know and you know he's having you know a hundred conversations a month with our agent about like this, you know, and then he emails me about it. I'm like, "Oh, cool. Well, I like this and maybe you should send this email to, you know, such and such person."

Um, and so I don't know, this is I I think all of these are just fun experiments. Um, but I'm learning a lot. Uh, you know, it's it isn't, you know, I am trolling on people about the lines of code thing,

but I will say that there's a a grain of truth here where I am not about lines of code. I don't work for anyone like really, you know, I I work for the uh institution of YC as the steward of it.

But aside from that, like I'm not here to like max lines of code to like, you know, for to good heart law something. You know, what I've learned is that like

the machines don't aren't goodart lawing like lines of code either. M

uh so that means that if you are 100% of your code is written by uh you know the machines and you are earnestly trying to solve the problems of your users and yourself like I am with GStack in the open like you can look at it and I'm doing 36,000 lines of code a day and not a single one of them is like for me to be able to say that it's actually like I'm just having the time of my life building software like we're about to launch something really awesome at YC that

Oh I was about to ask can you tell us the story of Bookface and then where you think software development will go at YC?

Yeah, I mean Bookface we have I mean we have you know 20 engineers uh at YC and you know book one of the funnier things we uh we found out was like actually a lot of people seem to have access to bookface. Bookface is the internal social network used by um we have about 16 17,000 people who are YCMs now who have access about 40% of them use it every single day. Wow. And uh I've started to realize actually anyone who has a friend of someone who works who uh

went through YC

you know yeah like people basically share their login

share their login

so I mean which is cool actually like the reality is we should probably open it up like we probably should have a like if you're a builder I don't think I want absolutely anyone like I I want like techno optimist people who are psyched about making the future and like I think bookace should be open to people you know That might be one of the things we work on this year, you know.

Yeah, there's so much interesting uh like wisdom and stories in there. Some of the stuff is maybe private and people don't want it to be screenshotted and shared, but uh but there's definitely a process to open it up.

How uh how are YC founders in the in the latest batches adapting to the change of pace of software development? There was a guy yesterday who made the claim that he rebuilt the whole batch with with agents. Congratulations. God bless.

Respect. And of course,

incredible troll.

I'm not sure he made anything that people want.

A good demo for his product. It got some attention.

I love shout out to Tim Draper, you know, big respect, man. Like you, you know, you got to do what you got to do. I love it.

Yeah. But but in a world where uh a YC founder might have in a previous era been very proud of the system that they built, the code that they wrote, even if they're not tracking lines of code, they said, "Look, this piece of software did not exist. I spent three months grinding with my co-founders. We built a piece of software that solves that problem, and now that's, you know, five minutes of work." Uh where will the Moes come from in the future YC companies? I mean basically the future like what's funny is basically all the people who already did that like you got to speed up like it's time to let it rip guys like software is too

uh we treat software like it is so precious and in the new world that we're going to it is it is going to be much more funible but that's the good news

like that's where taste and agency and trust matter a lot. Uh this is something that only crystallized recently like we've been talking about agency and taste for the longest time. Agency is being able to prompt and taste is being able to do eval right. Uh and then the third part that um I only started thinking about I think there was a tweet I reposted today that like said it really really well that like trust is the third thing.

Mhm. And so, you know, that I mean that's probably the thing that's the most important. Like when you have an enterprise company that is selling to real businesses and they're relying on you, like that's your moat. Your moat is actually trust. Like they went to the effort to try you to get you onboarded and to incorporate you into how you know you they work. And then, you know, I think that either you can hold on to that and that's a real moat and you can build something over the long haul. Like someone gets promoted or someone avoids being fired because you exist.

Um, and they're never going to switch and they trust you. Like that's like even more important than those other things. And so, you know, yeah, like what is this story about? It's not about like is someone going to try like if someone came along to you like this you know this impressive like uh Draper company like troll on YC companies is a good case study in that right um like I I'm impressed by the volume of software and they clearly built a software factory

um but you know maybe you should open source it like that would be cool um aside from that it's like are people going to trust things that are like trolls like I don't know I mean

I hate to bring you know, in let me let me state plainly like I actually really respect Roy. He's always been really cool to me on the internet and I feel like I've been unfair to him.

Uh, you know, I think that Cleuli legit is like we talking about at YC partner lunch today. It's like, oh yeah, like the guy actually has real revenue

and the idea is good.

Mhm.

Like not you know the troll marketing part like

you have problems with. We talked about that a lot.

Yeah. The interesting thing with with the product from the beginning, I think some of the criticism was, okay, if this product works, like there's going to be some big companies that roll it out. And so that's like this other question, right? And

that's the trust thing again, right? It's like there are things that you can get you here, but they're not going to get you there,

right?

Is sleep overrated in the is sleep overrated in this moment? Is it is it a waste of time? Are you adapting Sleep is great. I'm starting to like last night around I don't know 2 1:30 a.m. I was like getting foggy and I probably could have kept going, but I'm like let's go to sleep, man. Like I'm going to wake up in the morning and I'm going to dream about uh a bunch of the things I want to do and I'm going to wake up and I get to wake up and go to my computer and go to conductor and claude code and

you know, let's see what the workers are up to.

I love it. I love it. I have two uh two questions. one. Uh, what was your stack like back when you designed the Palunteer logo? What was the Was that a tasteful exercise? Was that like what went into designing the Palunteer logo?

Yeah, I mean, basically what's funny is um yeah, that was such a weird formative year like uh I'll paint the picture like we're in Paige Mill Road in uh PaloAlto, like the iconic um downtown Palo Alto office. We it we're a year away from even moving into uh this was like um across from where I think Tesla headquarters in Palo Alto is now on Page Mill.

Um we only were 12 maybe 10 or 12 engineers at that point. And uh I was working on the hiring site for Palunteer, you know, Palunteer.net or something. I think we didn't have the com.

And uh the original logo was created by my high school buddy Stefan Cohen who's still a Palunteer and a billionaire many times over. And um you know basically I was like how do I get engineers to want to work here?

The logo kind of like it was like six hexagons I think and people came and they thought it was like a biotech company or something.

Oh interesting.

And then we just basically I took maybe I don't know a week. It wasn't more than a week of like working on it but um we made like a thousand different versions of it. Um which is really funny because like

and that was pre GStack.

Yeah.

Yeah. Now in GStack, you can go slashdesign-shotgun and you can say design me a hundred of these things and then it'll pop a window. It'll use codec.

That's amazing. Wait, so yeah, I mean how do you think do you think uh AI is changing design? Do you think there's still room for taste in this world? Do you think that all the same uh rules apply or is it somehow different?

Oh no. I mean the machines um don't have a point of view like they'll give you lots of options and you know I think if you don't have taste and you're using these tools you'll make something that's like marginally better but it'll still just be like you know marginally better than average whereas I think it's clear that like

you know I mean there are tools like variant uh UI that's like YC company that we funded is like awesome like you just go in then it's like that's the real version of design shotgun like you know you can you will have design tools that have taste and if you use those you will have you will have a leg up. Um but I don't know I guess it just goes back to the agency part again it's like you know one of the principles I built into GStack recently that has served me really well is um before you start coding or before you start making technical plans at the engineering level uh I tell it always search the web. M

it's like figure out what's you know you want to someone else has figured it out like there's something open source there's something that's out there go all the way to the edge and then you know basically don't don't try and reinvent the wheel right

interesting yeah I mean I think

a world where it reimplements Gary's list in Jackal or something

who knows really I mean we'll see

it might it might

I mean it's using a lot of open source software obviously he didn't write his programming language is probably didn't write its own database. It's using off-the-shelf tooling there. Uh but the core app, I mean, I guess with the with the right flexibility and the tools that you wanted and the features that you wanted like it makes sense to it it we're we're almost it's not build versus buy anymore. It's like it's like fork versus inference or something like that. There's like a new

weird I mean increasingly you don't even fork like I mean you know it's funny like all of these things are very strange. We're in Bizarro land right now. Like we're in mid-transition and like

all of the transitions are not in distribution yet, which means um like if you uh try to make a product right now and you ask it to estimate how long the agent is going to take to make it, it says human terms. So it'll be like it'll take a human a month to do this and and then you have to ask it explicitly, well, how long is are you going to take? And it's like, well, I'll probably take like 15 to 20 minutes.

Mhm. And it's like literally across the board like I am coding 90 like my output in terms of like real products not just lines of code is like 90x that of what I did when I was in 2013. I made Post Haven that year. I made Bookface that year.

And um it's you know I was at a CEO conference uh last week um

at JP Morgan 100. And I was talking with a bunch of like former tech

uh they you know very technical CEOs who like now are full manager mode

and like I was like good news man it's time to come back time to time to you know you're Peter Parker right now time to put on the Spider-Man suit because this is the time and I was blowing their mind and I'm like they're like oh no like I shouldn't do this it's not worth my time. I was like dude it is worth your time. Like you can still be the manager during the day, but by night

you can be Spider-Man or Batman. Like you can dawn the suit and pull out the Gstack.

That's a little sleep. Yeah. Pull out the G-Stack and you you know and my pitch to them was like look I am 90 times I am 90 of myself. Like

you're manipulating time.

Yeah that's remarkable. I love

and you you know you could have 90 of yourself working on things like tonight and be the CEO, right?

It's fantastic. Well, uh you want a ton of people over in the chat. They were skeptical and they're and they're very fired up. So, thank you so much for