How to Build AI Agents, From the Guy Who Wrote the Framework

You've already built two open- source JavaScript framework companies. How many people on Earth have done that? Probably less than five. Why?

Open source is incredibly valuable thing for the world to do, but it's there are also so few people that know how to do it really well. For anyone who doesn't know what Ma is and who it's for, fill everyone in. So, Ma is an open-source TypeScript agent framework. People build customerf facing, userfacing agents within their SAS, right?

People build internal agents. people build like developer platform agents like AI SRRES with MRAA in TypeScript. How old were you when you started Gatsby? Few years out of school.

So I think I was 28 or 29 when we founded the company. You know the things that you sort of learned the second time. You're just a little bit more confident in your own judgment and your own capabilities. There are a lot of things that I wish we had known the first time but we couldn't have known the first time cuz we were doing it the first time.

Is there anything you shipped recently that you're especially excited about? So recently shipped this agent controller which lets you build your own kind of like clawed code. You're the author of principles of building AI agents. It's become one of the go-to texts for people building this space.

This is crazy. Like it's insane. Were you expecting this? Sam, welcome to the Agents at Work podcast.

Thanks for having me here. Super exciting. So, what where are we right now? We're at Y Cominator.

And what does this mean space mean to you? So we were we went through IC in winter of 2025 and so much happened that was like really important to MRA and our journey. So like this is one of the most meaningful parts of San Francisco for me. It really is and this is my first time here as well.

So this is very exciting. Okay. And so for anyone who doesn't know what Ma is and who it's for, fill everyone in. So Ma is an open-source TypeScript agent framework.

People build customerf facing, userf facing agents within their SAS, right? People build um internal agents. People build like developer platform agents like AISR with MRA in Typescript. And so you've already built two open- source JavaScript framework companies.

How many people on Earth have done that? Probably less than five. Why? It's this open source is incredibly like valuable thing for the world to do, but it's there are also so few people that know how to do it really well and yeah and you you're one of them apparently.

So very exciting. So are you ready? We're going to go inside and do the full interview. Okay.

So actually like when we were in the batch, we were standing here handing out like 20 boxes basically of Pete Cass because Angie's founding sales books. It was one of the things that inspired us to write our own book. Oh my gosh. Okay.

I'm really excited to get into your own book. Sam, you have a really big audience on LinkedIn. You have 50,000 like somewhere around 15K on X. Tell me more about that, your strategy and content.

I mean, I don't know if it's a strategy or but like we we basically like like I mean I have this belief that like content wants to be frequent in the same way that like information wants to be free. And so if you can post like helpful things that are like good quality at like a good cadence. um you know the like no you know there's a limited audience for like your personal jokes and whatever but if you're really thoughtfully posting things that and like making things something that's like hard easy and you just do that over over time um you know there that's like a um there's there's other there's other nuances and bits but like that that are sort of like the specific tricks of like how to post things that do really well on on on Twitter or how to post things that do really well on LinkedIn that are very platform specific but I think the general The general thing is just like be be generous with your expertise and knowledge and be helpful to other people and like that you know the good things will follow. And so you um you mentioned about like shipping like the tiers like shipping every day.

Hey, tell me more about So, right now that the the team has gotten to the place where they're shipping a tier one feature every day or, you know, maybe it's like every other day sometimes, but um and so we're doing these releases on on X every day and and so um it's it's it's actually interesting because the you know, it was actually you know I'm actually like the bottleneck of the or has moved to like how do we release all the cool stuff that we're building which is like a really great place honestly to be in like hey we've just done all this really cool work and now we need to shout about Um, you know, I think that like, you know, we're we're now sort of like moving to a world where like we want, you know, now we've figured out how to like release um release in public and share in public what we're building like every day and and you know, we're we're now trying to figure out, hey, how do we share like a a user story? And this is a specific kind of thing that people are building with MRA um every day. Okay. So, you have user stories, you have like case studies, you have like smaller like tweets like bigger ones.

What kind of other like formats are you experimenting with? I mean I I it was actually interesting like sort of like around the time of the Netifi acquisition I wrote sort of like a long article about this and I think there's like there's different groups of you know you sort of have like community developer community type stuff whether there's like a GitHub issue or Discord you have like documentation and and and sort of like tutorials and how-tos. You on the other hand have like this you know marketing kind of case studies which is like I got a 3x improvement from this or it saved me like 70% or whatever. In the middle you have this like solutions engineer type of content which is like here's a architecture deep dive of like how we built this or and how you might build a thing or you know a very like uh very sort of like technical deep you know it's like a reference architecture.

It's a technical deep dive. It's something that's like really specific to a particular use case or a particular industry. It's like very grounded. It almost, you know, I think it like hearkens back to my journalistic background where it just feels very, you know, the reader reads it and they feel like, well, this was written for someone like me because it was written by someone that really understands like their particular situation.

Wow, that's incredible. And how much did you raise? Um, so we raised um in our seed round, we raised a $13 million seed round um coming out of YC and then we raised a then recently earlier this year, we raised a $22 million series A. Congrats.

That's incredible. Thank you. Okay, Sam. So, tell me some of your other exciting moments here in YC.

Um, so the room that we're in now is the the room that um that all the YC founders come to after the batch is is done. And you know, we have about an hour where everyone is saying like thanking other, you know, other companies, other founders for like, you know, things that they did during the batch. And during that like three founders like called out like something like I I had done or like we had done as Monster to like help them. And it was just like the most you know flattering and like it was like we just felt like we had done something like not just for ourselves and their you know users but like for the the rest of the the batch.

Um were you uh also looking at other ailies? What made you choose YC in particular? I mean there's only one YC, right? So, um, so I was, you know, I've worked for three different YC founders.

Um, and so my first company, Gatsby, we'd sort of applied to to YC back in in in 2017. Um, and and you know, honestly, we were just not very good at explaining what we were doing back then. Um, but, uh, you know, we ended up raising money elsewhere. Didn't get into YC, raised money elsewhere.

But, you know, as this was taking off, we wanted to be in YC because YC was ground zero of of building agents. And we just felt like if we came to to YC, you know, we would learn the most and we would build the best product, we felt like the community and the network were just second to none. And and you know, sentimentally, like having worked for three YC founders, having loved all that experience, I just I kind of really wanted to be a YC founder. Right.

That's awesome. What other really cool people you met while you were here? Um I mean there's there's so many right so there's like a few different um a few different types of things like one you just kind of have you know you have evening kind of events every week and then you know the Brian Chesky's of the world or the Tony of the world come and kind of share you know their stories and their experience but you know on a on a onetoone level you know you get to know like a lot of your other batchmates people who are building incredible things sometimes in like very different domains than you are um and you know sometimes in very similar domains And so some of the companies that we you know work very closely with like um assistant UI for example like we we met in in the YC batch uh um and then you you get to know like the partners and you get to know you know the alumni and sort of like the community right so um Gustaf um was our our our YC partner and you know Gustaf has this incredible growth mindset and he's able to break down any problem into sort of the analytical capabilities and walk you through like a decision framework of how you would sort of like choose is the right answer. Um, each of the partner has like something they bring to the table.

Brad Flora like is like the most encouraging person and always has like the pathy like moral but practical lesson. And so like one of our values um is generosity breeds luck which was you know originally like a line that he said to us as we were kind of like you know was kind of in in a in a like in one of our group office hours. So there were so many like people and that like shaped our journey. That's so amazing.

And so when we were talking for this the prep call you mentioned you said you've gone from turnbased strategy to real time strategy. So what does that mean? So I think that like you know have this is my reflection having second time founder right but it used to be that product development was like let us think in a very analytical frame about what is the next 3 months or 6 months or 12 months of product development and you know you'd have these like you know here's the next four quarters and like let us analyze all the user requests we have and you know now it's just more like you know life comes at you really fast and so you're lucky if you can plan you know two three months in advance um but your your goal really is just pump as much information into your network into your system as possible, right? And then like act as quickly as possible.

So it's really like, you know, it's if you're a Starcraft player, um, you know, it's really about like actions per minute that that that you're doing and it's less about like, hey, I have this like perfect strategy for like winning, you know, and and like after we do these like seven steps in a row. Wow. What does it change to how you run a company dayto day? Um, you know, again, it's it's sort of like about acting quickly.

um in as opposed to you know doing the perfect thing. So I think like George Patton had this like quote right like a violently a plan a good plan violently executed like tomorrow is better than the perfect plan executed next week. I think that's always been true, but I think it's even more true in a world where like development sort of speed is up by like 3 or 4x, right? Your take is that the people with an edge right now are the ones who've done the specific thing before.

So the UNs, what does experience um beat fresh eyes in a moment everyone calls unprecedented? So I think like you know again I think it's about actions per minute. So there's like two ways of having sort of like actions per minute. One is like and it really just is like being in contact with the ground truth.

If you've been doing something for 10 years like we've been doing open source, JavaScript, TypeScript frameworks, um we know our domain really really well, you know, better than maybe almost anybody in the world, hopefully better than everyone in the world. And so we just have a really clear mental model of like how to succeed. And that means that like we can react to events really well. Um but also you know if you're sort of um you know you've just dived you know deeper into this frontier of like AI development or ML research more than anybody else um because you you know you you just spent the last four years obsessing over it while you were in college.

Like that's also like a domain type of expertise and and you may have an advantage that like you know people that are in their mid-30s like me have never seen that like particular you know niche specialty. You may you may be the world expert on that thing. Um but ultimately like it's about moving quickly and maximizing your actions for a minute. Yeah.

Are you really happy that you kind of um started at a time that you started? Um how old were you when you started Gatsby? So So you went let's see when I was when we started Gatsby I was a few years out of school. So I think I was uh 28 or 29 when we when we founded the company a bit before when I started working on the open source project.

um you know the things that you sort of learned the second time you're just a little bit more confident in your own judgment and your own capabilities and is the judgment that kind of comes from from experience. There are a lot of things that I wish we had known the first time but we couldn't have known the first time because we were doing it the first time. Yeah. What are one of the things that you wish you have known?

Um I think there's a lot of um so so for example like sort of like how to optimize you know spending time optimizing onboarding flows you know that what are the nuances of making a really great um API name for example um what are like how do you have sort of progressive disclosure of complexity of your framework where like it's easy to get started but also that doesn't block you from like the more complex path. Um, de like developer experience is this like weird niche that like unless you've like worked for a long like long time in dev tools, it takes a while to kind of gain an intuition for it. How do you like talk about what you're building in a way that is like sort of accessible for beginners but not kind of a turnoff for for experienced folks and like in in as much technical detail as possible where people really understand here's what I do. How do you um how do you build a community?

How do you um how do you sort of uh get get people answers to their questions and deal with some of the weird nuances of of a community that lives on GitHub or or Discord and the sort of this like you know there there's just like a lot of nuances that are like very specific to the domain that we've spent the last 10 years in. Yeah. And I remember listening to one of your podcasts. " Like you got to like dig deeper and like get people to really tell you kind of.

So So I I'll tell you the story. So um we started working on on Monster on October 1st of 2024. Okay. Um three weeks later we showed up at uh at Nexcom.

Um and and like Shane Abby and I we we built a little bit and we went around to get feedback. We talked to about like 11 or 12 people and it we got a lot of oh that's that's interesting you know and and it turned out that like what we had done is we had built our workflow engine in a config first way rather than a code first way. And so when you showed the API to the developers, which is our, you know, target market, right, or target user, um, they were like, well, that could be cool for someone that isn't me and their sort of mediocre level of of like enthusiasm. And so, you know, about like we were going to try to talk to 20 or 25 people, but by like 1:30 or 2:00, we'd had enough.

We'd gotten the feedback we needed. " Right? like and then then like we kind of figured it out within like an hour or two like we we like debated various things and we're like okay we think this is what we up um and then we spent the next three weeks um like fixing the APIs um taking the good things that we' done and and sort of like then we relaunched um three weeks later and and then like that that had like really positive feedback right but we knew when someone said oh that's interesting that like they were actually like not actually that interested Oh, okay. Wow.

And so you also tell founders it's a quit their idea and go join something that's scaling. Yeah. So I think like you know there's a there's always a balance like there are people that work on like the idea for like six or seven or eight years. Um and they can work on it you know you can work on you can give up on an idea too quickly but you can also work on an idea past the time when you could reasonably be successful.

And I think it's okay if you think that that's the case for you. Um to say like look, I've given it everything I could. You know, I'm I'm I'm it's not time for me to be a founder right now. I might be a founder again in the future, right?

But it's but um I think it's okay to say I think it's okay to say like, hey, I'm going to do something else for a little bit and we'll we'll come back to this later. Yeah, I agree. And so walk me through the agent stack as you see it. Where does a framework like MRA sit and what plugs into it?

So the way that we see this is that there's sort of a you know a domain complete framework like Monster that lets you do all the things that you need to do um to build uh an AI agent or application or harness. js, you still might want to use something like resend, right? Um you might still want to use an O provider like like Better O, right? Um and so, you know, in the in the agent world, um you know, there are a few, you know, for example, like browserbased or exa um you know, there are these different categories of of of tool providers.

Um so browser use and web use is one. Um you have like integration. So Composio's a part of that. Um, and then then there are these other things.

So agents, you know, need email. You know, agents sometimes need a voice. And there there other bits and pieces along the way that are sort of these utility kind of categories that like plug into the central uh domain complete framework. And is there anything you shipped recently that you're especially excited about?

I mean, there's a lot of things. So I'll I'll sort of like um I I'll I'll kind of like call out a few um I'll call out a few of them. So we recently shipped this agent controller uh which which lets you build your own kind of like clawed code. Um so that's like really valuable if you're like hey I want to put openclaw or I want to put cloud code in a box but gosh it feels like it's giving it too much power and out of control.

Now you can kind of build your own with with MRA which is like really exciting. um our our agent builder which is um our our agent builder which is a uh kind of a a basically a way that you can um developers can configure a set of tools and a set of integrations and a set of permissions where nontechnical users can actually build agents. We shipped that recently and then we added aosio integration for the the tools as well. Um and so here we have your book.

You're the author of principles of building AI agents. It's become one of the go-to texts for people building this space, which is crazy. Like, it's insane. Were you expecting this?

So, no, it was really an accident. So, and it's funny that we're here at YC, right? Basically, like, you know, we started um during the batch, we were actually um we were we're handing out these founding sales books to people. So, Pete Kazandi, the author, sent me a bunch of boxes of books and we were just handing them out to our batch because we're like, hey, you know, like everyone kind of needs to learn how to sell and so like we'll just help our batchmates.

It's a good way of getting to know our batchmates, right? It was kind of a nice thing that we could do for them. But everyone loved the books. And then um then after that we were like, well, like because I'd written a book at at Gatsby.

Um and so so was like we we're having a bunch of like, you know, whiteboarding type uh sessions and and after one of them, my co-founder Shane is like, you know, what if you wrote a book? It's like maybe I should. And I was like, what if I wrote a book? And then we brought it to demo day.

And then and so that was this like really um and so it was almost in some ways like kind of a practical joke, right? We're like, well, nobody's ever done this before, so wouldn't it be kind of funny if we did this thing? So then I just kind of like locked myself in a room for a couple of weekends and then like I write really fast. I used to do journalism in school, like you know, I can pound out 600 words pretty quickly.

Um and so so um so then came out of those weekends with a you know, a 90page book. We figured out how to get it printed pretty quickly. um you know brought it to demo day and then people loved it at demo day like investors would like read copies of it and then they'd tell our batchmates you know based on this book you know I have some product feedback for you and we were like okay like this is then we started bringing it up to meetups and again it was just like this huge hit and then you know this was in March of last year and then soon you know we're just we we brought it to startup school we gave out 2,000 books in the course of I don't know like a couple hours basically and and it was just it was almost like I was like a human vending machine. It's like reach down, hand out book, reach down, hand out book.

Right. Um and and it just like one thing led to another and it just kind of snowballed. That's amazing. So, I got this book when I was at a hackathon and and I was like, yeah, like I've seen it online.

I started reading it. I read halfway and then I I didn't finish reading it but then as soon as I started prepping for years I and this was like after a few months later like I read it again and I like because I was like I've been in this space for for a few months now um working here full-time uh I like started picking up things and things were starting to make more sense compared to the first time I read it. is like as a non-technical, I definitely recommend you guys read this book as well because it really just it's just such an easy read. And this is Oh, this is the second edition, but you have three editions, right?

Yeah, this is Yeah, this is so we we recently came out with a third edition as well. And it's because things are always changing and moving fast. You always have to keep Yeah. Is that why?

Right. It's It's in the age It's funny because it's kind of ironic that in an age where development is actually moving so fast, we made an old school print book. And so, you know, to do that, we've had to like update it like pretty frequently and keep printing out new versions and sharing it with folks. So, when's the fourth version coming out?

We'll see about the fourth one. A third one is coming out probably in November or December. So, I think I'm we're going to we're going to work on that first. And so, has this is this almost like kind of like an growth engine for you, too?

Like, yeah, absolutely. I mean, I think there's like sort of like two two pieces, right? Um there was a there was something we learned at at YC um which is um something our one of our partners said to us which is generosity breeds luck. Yeah.

Um and that became like a core value of ours. And so part of this is that like look like everybody is trying to figure out how to build agents right now. And so we have built an open source framework to help people build agents. And and sometimes the catalyst for people like you know building is that we like you know print and distribute tens of thousands of this copies of this book every every month and like a significant fraction of our of our users hear about us through the book or like oh you're the guys even if they like they found us another way then they realize that like oh we're the people that wrote the book and they're like you know wow and so you'd mentioned how you were a journ you were in journalism like you were studying journalism before tell me more about that like what did you learn there so I maybe there's like a literary streak in my family.

My grandfather was um in in Mumbai was a relatively prominent publisher there. My father is a scientist so he publishes a lot of papers. Um you know I I was just I loved um working for my student paper when I was in school. Uh you know I really just loved immersing myself in a problem and getting to like go around and understand something and then be the person that kind of like narrated that to other people.

Now, I was originally like class of09, which mean and and like, you know, was kind of during the great financial crisis and and journalism was sort of like in very rapid decline. It was pretty obvious to me that it wasn't like a great, you know, even though I love doing it, it wasn't like a great like path, but it's something that I've always had kind of a a fondness in my heart for. I love writing. Um, and I love I love the thinking that you get to do while while you're writing, right?

And so, you know, I just ended up being in this really unique place where I got to write a book kind of describing like the this huge transformation that was happening. That is so crazy. And I feel like now we're almost entering this new era of like journalism. I mean, new media has been going around.

What are your thoughts on that? I mean, I think people really like authenticity, right? there's this sense of like, you know, we're talking about this and, you know, you you you were going around San Francisco putting up those MCP versus CLI signs which just like so much captured the zeitgeist of of you know, that that moment in time and and I think like, you know, there's this like, you know, rather than sort of like traditional media where everything's kind of repackaged and reprocessed and like nobody's really like, you know, you're able to like go really deep on something, you know, really well and then produce it in a way that like everybody kind of like consumes it. And I think people can sense the authenticity of of of both the storyteller and the subject in in that kind of format.

Yeah. So yeah, be authentic. And so be authentic like this is this is also like a big principle of YC, right? Like is like if you are authentically yourself, if your company is sort of an authentic expression of yourself.

I think people very much sense that like you know it's not just a product, it's like part of who you are, right? Mhm. Yesterday I had this conversation with someone at this event and he was telling me that there's actually a lot of founders they don't actually do it for the passion. They do it for like more of the money and then to get acquired and then yeah like I was surprised by that actually.

I wouldn't say there's as many founders like I the way that I would think about that is that like being a founder is a much safer career track than it used to be. So like you know like I don't think a lot of people are doctors for the money. they're they're doctors because that's what their parents want them to become because their parents perceive it as the safe career track. And now that being a founder is more of a safe career track, there's an increasing tendency of like, well, I'm just going to do the, you know, the thing that's kind of like the default.

You know, it's a good thing and a bad thing, right? It it increas means it's more acceptable to people, but it also means that, you know, when back when tech was the weirdos and the outsiders, there was like a sort of purism that you know is is maybe more rare these days. Yeah, that is true. And so you write that our job may shift from choreographer to architect as models do more planning.

What does that mean for people building agents today? Yeah. So I think you should like be thinking about like I think let's let's talk about like the difference between context engineering and harness engineering. So you know when Dex wrote his piece on on context engineering last summer, which is seems forever ago, right?

It was about like how do you structure your prompt? How do you inject the right things into the model's context window? Whereas harness engineering is much more about how do you empower your harness to create the right sub aents to get the right information that can then be injected into the context. And I think there's like a fundamental distinction between like these two types of things.

And one is like you know carefully crafting it and the other is like how do we design the system that that does the the sort of like does the right thing. Yeah. You say hand coding evals is dying in the era of clawed code. what replaces it and why do you think eval are more important than people realize?

I mean, I think this is like, you know, hand coding evals is is dying in the same way that hand coding anything is dying, right? Um, so I think if you're looking for sort of like a driver that to me that's the the biggest driver. Um, I would say that uh, you know, increasingly we're turning to these sort of like continual learning or sort of like synthetic evals. there's a few different names for it where you're actually able to like look at the traces um look at the trajectories look at the conversations um and then you know a human is kind of you know maybe selecting like okay let's create an eval from this or maybe you know an assistant is suggesting you know hey here are some here are five conversations you may want to create evals for or like maybe a human is like labeling uh hey like this is inaccurate and then you create an eval based on that but you know it used to be much more of a sort of like hey like, you know, you'd have a domain expert entering in question answer pairs into a a notion spreadsheet, let's say.

Um, and and now it's much more and now it's much more sort of a volume produced based on like real trace data. Yeah. And so now, can you walk me more about your YC winter 25 takeoff story? The dog patch apartment, you and Shane and Abby.

So, so, so we're so let me let me sort of set the stage. So, there are three co-founders, myself, Shane, and Abby. Um, you know, Abby and I were living in this, you know, two-bedroom apartment in in Dog Patch, right? Shane was living nearby with his his partner and and their their two kids.

One of the kids with had just been born. He was about six weeks old. And so, Shane, she's partner, she had some maternity leave from her job, so they're they're in they're, you know, they're there. were here um every you know we we get to YC um you know beginning of January 2025 and and we'd launched you know and sort of had a little bit of lift um which had kind of like gone down again as people found all the bugs in our thing and sort of started turnurning but we we're like well we think we've built the right thing and we get to YC and the first three days of YC are this like you know it's almost like um you you just spend them like getting to know all the other founders in the batch right building community um and we spend a lot of time getting to know our the fellow founders in the batch and we also spend a lot of time hey so you're you're building an agent for you know Swedish accountants can you can you show us your code can you show us how you're doing this and like this kind of like you know multiply that by like 20 um and then and you know we would just like get people to like walk us through their codebase and how they were doing things and at the end of the three days like you know we kind of came together to like regroup and we're like guys we we missed like seven features that we need to have and so we spent the next like three weeks like furiously building that stuff.

Um, and then after that we kind of like then went kind of started going around the batch and we're like, hey, you know, we're we're an open source TypeScript agent framework, you know, are you building agents? You know, do you want to use Monster? We'll we'll help you build your agents. We were doing these like whiteboarding sessions in our in our apartment where we would ask people to like, yeah, let's like let's let's design it together.

And, you know, we were still learning about this stuff, but like it was where we we got to learn from the people we also that that were coming in and and and sharing their thoughts. And we were also helping them like think through the problem and and we got a lot of knowledge. Um and then you know some of them we got to we got to use MRA. Um and you know you can sort of see if you look at like our npm download chart in early February you know both from us being you know live on the internet on on GitHub right on npm as well as like within the YC batch like our you know our downloads starting to uh creep up.

Um but then there was this one moment we launched a couple times internally within YC. " Um, and so, um, we we spent about, um, we spent about, uh, you know, I spent about a day writing this like show HN post, and then like sent it over to to Dan Gley, who's the, um, the HN mod. " Like I spent the next day like furiously like revising it. And then the day after that um we this is Wednesday I think February 15th if I remember the the date correctly we post um we we of 2025 right we post our show hn um on on on hacker news and and all of the sudden like you know it just keeps getting like up votes and and likes and and and we ended up on the front page all day and then like maybe like through the night probably like 20 23 hours we we pointed it to our our GitHub, right?

And so people then went from from the from the from Hacker News homepage to the GitHub repo and started starring our our repo and we got probably like 500 stars overnight. Um and we were like really excited and then one of us noticed that we had gotten on GitHub trending because we'd gotten so many stars overnight. And so then we stayed on GitHub trending for the next week. And and in that week we went from like 2500 stars to about like 7,500 stars.

and you can like sort of like watch our our npm download growth accelerate. And it was one of those things where it was like before that we were like we had like a small group of like you know user and after that there was just like we had so many people like flooding into our discord like using monster like giving us filing a bunch of issues and and it like there was like there was a sense that like we were this like experimental thing beforehand and then like we were this thing that was like still new but like growing very quickly. Wow, that's incredible. And with all this growth, did you have to um did you notice more like things you had to like maybe fix or improve?

Did that really help you figure out? Absolutely. So, like one of our biggest pieces of advice like is that like you need to be able to like you need to be able to build the machine that pumps user feedback back into your organization. Like you can't you everybody has to be a project engineer.

Um, so whether it was GitHub issues, whether it was Discord, whether it was like Slack shared, Slack channels that we started creating with our users, um, you know, we we like we just tried to have as many touch points as possible. Um, we'd actually we'd actually brought on board some of the engineers that worked with us at Gatsby. And so it was like really easy whether it was like one of us, you know, writing a a fix or building a feature or was one of the team, you know, if someone mentioned something in a shared Slack channel like, oh, that's good to know. and then like suddenly you like you build a small feature, you make a small bug fix like whatever it is.

Um I think that actually a lot of what you what traction is valuable not for the up and to the right chart but because users that use your product tell you how to make your product better. Um and and yes, taste helps. Taste gets you started. Taste puts you in the right spot.

But to make your product from good to great, you need that feedback. You cannot make it, you cannot polish it. It cannot shine without like a constant stream of of of these like, you know, of of like 10,000 little like, hey, I was confused by this or that, you know, that didn't feel right. Right.

Yeah, that makes sense. And so before Gatsby, you were an econ major in Michigan who came west in 2005 and ended up engineer. So, how did you get from there to founding an open-source framework? So, when I graduated from school, I worked at a couple of we'll just call them like more corporate jobs that I didn't really enjoy that much.

Um, and you know, I' I'd done a lot of programming, but I I wouldn't have like, you know, just considered it more of like a side hobby or or whatever. Um, and you know, then I ended up getting a job at a startup as an engineer. Um, and I like that much better. Um so so I worked for a few different um YC startups uh and just you know as a full stack product engineer um you know built in JavaScript some of the time built in types built in Python some of the time um you know around like 2016 um my best friend had was working on this open source project and it Gatsby and it started to take off and he this is my friend Kyle Matthews he is you know as it was kind of taking off he was trying to figure out what to do with it and he was like I say do you want to work on this together?

I'm like this looks pretty interesting, right? Um and you know, we started working on it together, started building the community together, ended up going and raising um some money together and then starting uh starting the company. Um I I think my journey is interesting because I didn't sort of start out in mind with like this is exactly what I'm going to do. " Right?

I think that like it's okay. It's okay that if you haven't known since you were like 10 or 12 what you wanted to do with your life, it's okay to get pulled, you know, kind of like a little bit of luck, a little bit of circumstance into like what you kind of discover is your life's calling. Yeah. And so Gatsby was able to grow to real revenue and then and then what happened?

So I mean it was a it was a journey, right? So, so we started with a um we started with a very popular framework. Um we weren't sure like how to monetize it. This was 2017.

This like open source monetization was very new and we ended up you know sort of like the framework kept growing. We ended up building a hosting platform. Um we ended up um you know getting few uh a few hundred maybe a thousand customers. Um that journey took a few years.

Um and and by the time that journey was you know few years in um you know the market had shifted and it was like pretty clear as we went out to to fund raise that like we weren't going to be able to we could have probably raised some more money but like not on the terms we would have wanted to raise them on not to the place where we felt like we could keep innovating and pushing the frontier forward and so uh we decided that like that was the right time to to sell the company. Um so we you know the natural sort of like acquire was was Netleifi because um a large proportion of Netleifi sites were actually Gatsby sites there. Gatsby was the underlying framework. Um and so you know we as a team you know we joined Netlefi.

Um and something that I like to tell founders is that you know when you sell your company it's it's not your company anymore. And so you know there there there good parts there there are bad parts. Um but like you know like what ended up happening was basically like we had this Gatsby cloud product um hosting product. We'd hoped to sort of port those parts to to Netlefi um and sort of like improve Netlefi and then sort of like shift our customer base over um you know and then sort of like what ended up happening was that like that that sort of feature migration didn't happen.

So when we shut down Gatsby Cloud, the the the customers didn't transfer. Um and then there were some rounds of layoffs at Netfi and the you know a lot of the engineers that were were in sort of the core team ended up getting laid off, right? Um and um I mean I think like you know I it was and this is just something that happens during you know we we're still on very we love Matt the the Netlifi CEO you know um we're still on very good terms. uh you know the acquisition didn't go the way I think like you know on either side it would have hoped to be um but you know this is kind of like what happens as part of an acquisition you know you are now a part in in a different in a different machine you are no longer the person that that architects the machine um and so uh you know like that's that's just part of the trade-offs.

Yeah. Do you know if uh a lot of founders they when they get acquired they just continue working and then they end up starting something of um of their own again is that like is there like a common pattern with that like getting choir and start Yeah. Yeah. I mean, I think I think sort of like, you know, a lot of people, you know, that there's kind of like a decision point that happens.

Like you you kind of learn after you've been a founder for a while. Like there's things that are great about, you know, about sort of like being a IC or being an a manager in a in a well-run startup, right? And there are things that are great about being a founder. And if you're a founder that sort of sold your company, you know, you kind of have this moment where you're sort of deciding, do I want to go off and start a new thing?

Do I have conviction on a new thing? or do I want to like find something that I think is great that I think is going to impact the world in a positive way and do I want to help them uh you know with their their scaling journey? Um you know for us uh for us I think you know because we kind of didn't have this great acquisition experience um we didn't really feel we kind of all left Netifi at different places and then it was almost kind of like a natural thing for us to to start off and do something new. M um advice for people like other companies getting acquired like how do you make sure you don't feel this like pain or this maybe you know you don't get yeah I mean it's tough in every situation is individual I think there's like a few things so the the advice that I'll give that is um you know maybe um maybe a little bit silly but like is is also true is that the more like you know our our acquisition basically returned the investors capital, but we we ourselves didn't make any money.

We basically got like a signing bonus. Um, if if an if an acquisition is sort of a significant positive acquisition, that means the company has invested a lot of sort of their internal and real physical capital into this acquisition. And so the more they invest into it, the more likely they are to like want to, you know, keep it working, right? Yeah.

Um, you know, if if you sort of, you know, also there's like an aquaire where like you kind of know that your thing is going to get shut down. And so if you come into that with with that being your expectation, I mean, you're not going to be disappointed when your thing gets shut down. But I would say like it's a lot about like alignment of what you're trying to build and and and what you've been building and what the acquiring company has has built, right? So if if they really understand it and it's like very close to their domain, you have much better chances than if you're, you know, if it's just something completely different.

Yeah. And so then your team went off to build sales agents and you said you've you've forgotten you were great at open-source dev tools. Yeah. So um this is something I tell to founders a lot.

Um you know the Rumsfeld's quote about like there's like known unknowns, right? So he thinks that there's like three quadrants. He thinks there's like known knowns um you know so things that you know you know. He thinks that there's like unknown knowns uh so like or known unknowns.

So like things that you know you don't know. and he thinks there's unknown unknowns. So things you don't know that you don't know. Well, actually like those that's a 2 x two quadrant, right?

But he's actually missing one which is the um unknown known. Um the unknown known are are things that you know but you have forgotten you know. Um and so a a lot of my founder friends that have you know you know founding is an up and down journey and and you know I I've more than one founder friend I've like hey look you are really good at this thing you were earlier in it than like anybody else and yes the market caught up to you and passed you by but like you were earlier and you have this gift of like understanding and like building and and you you know I know you're really down that you sold your company and like you know you got kind of like but I want you to remember that like you know and so so I think like and the reason that like I I'll I'll sort of like tell my friends that you know in their moments where I see them you know see them kind of like down is because I remember this moment when when we were down right like so so let let me kind of like set the stage right so Gatsby's kind of like fading as a framework you know next is kind of like clearly won you know we got acquired by Netlefi and like you know but now even like Verscell is like beating Netlefi right and so like we start wondering ing, you know, were we actually good at this thing? Maybe we were just like lucky.

And so we're like, you know, it really I think especially for for myself, for like Shane, for Abby, like, you know, losing sucks and we feel that like really viscerally, right? And and so for us, um uh when we kind of our emotional reaction, our first emotional reaction to that was to kind of like, well, let's do something else like let's work on like sales technology. Let's work on these like this AI sales stuff. And we tried a couple different ideas and and whenever we were working on that, you know, it just kind of like we would we were like, well, we want to do this thing like let's spend a week talking to people about how this should work.

Then we'd each spend a week and like okay, like let's kind of like come back and like we'd each have like different versions like we could try it this way or we could try that way and you know we'd figure out what we thought was the right solution. It might be like a week or two later. Um and and then you know so we we a lot of time went by and it just felt like we hadn't built a lot you know and we hadn't gotten a lot of like you know users um and then you know we decided to we just kind of felt like this isn't the right thing and so we tried a couple other things and all of a sudden in October of 2024 when we started working on MRA all of a sudden like you know and we had nothing keep we we we'd spent like a you know a few months or working on this thing and it was like it was a relatively complete product or whatever and through threw it hallway and we're starting with nothing, but all of a sudden it felt like we were moving really fast because whenever we um we should add this feature. Well, I think the API should look like this.

Yeah, I agree. And I think the functionality should be this. Yep, agreed. And then like here's how we'd like describe it.

And you know, it's just like and like here are the nuances. Here are the options that like this this here are the different options that we'd have for this key value pair and the the object, right? And so so we we just like we be so in brain sync on all the like little details of nuance that make that make any that make a framework great but that really make any product great, right? Um and and so it just felt like we were running really fast.

And so, you know, even though we were running out of money and and um you know, we we had nothing and and we had thrown out this thing that we had spent like months on um and then you we went to this conference and like people didn't really like the we still had this like confidence that like we could figure this out because it it just felt intuitive to us again and and we we figured that like look if it doesn't work we'll go out doing something that we feel like we're good at doing and that we like love doing and and you know then we we came to YC and then it all started like it was like almost the perfect story. Oh, that's awesome. Thank you for sharing. I can see the passion in your voice and it's just it's incredible.

And so, what would you tell a founder who's in that kind of dispirited place right now? Yeah. I mean, I've had founder friends that like, you know, they've I've seen them, you know, circling the strain on this old idea and, you know, they then like they'll come back, you know, a year later working on something completely different, having having gone through this journey where they were trying a bunch of different things, maybe consulting for friends or like, you know, just like learning and exploring in a way that like paid their rent, right? Um, and and like, you know, you can be so maniacally focused on one thing that you get blinders on and you maybe tune out the world.

And that's good because you need to do it. But like sometimes if you're maniacally focused on the wrong thing, you know, it's just you're going to go nowhere, right? And so like usually it's usually the advice whether it's kind of like is just kind of like zoom out. Try to zoom out a little bit.

You don't need to like, you know, and then there's founders that will zoom out and try to evaluate the entire idea space of everything they could build. And I think that can be counterproductive too because it's like look, you know, there's, you know, but like try to zoom out to like a relatively, you know, decent thing of like, hey, we think we'd be good at this, but but I think it's usually like, you know, if you kind of get dispirited, it's usually because you're narrowed so far in on a particular thing. Okay. So, insider versus outsider.

You say founders choose should choose deliberately. Where did you choose to be and why? Yeah. I think there's a, you know, there's kind of like this.

So, so you know, open source is an area that is typically filled more with outsiders than insiders. And that's just because the decentralized nature of it. You know, if you're somebody that like made a popular project, you could live anywhere, right? Um and you know the open source community hasn't typically if you look at open source companies uh from you know Hashi Cororb to elastic to GitLab to you know a lot of times they they they draw their contributor base from all over the world um a lot of times they you know their employees might live all over the world their customers are kind of all over the world you can't sort of like you almost you can't kind of like target and so inherently like inherently when we when we started building MRA, you know, we were a bit of outsiders be like our team is distributed and remote um because it's the same team we had to Gatsby which we had hired from everywhere around the world.

Um and so so I think that like one thing is that like you start in a you start in kind of like a specific place um and and we started much more with an outsider person and I don't think it's bad to be an outsider. I don't think it's bad to be an insider. I don't have like value judgments around one or the other. Um, but you know, it kind of shapes the type of company that you you build.

You know, if you're if you sort of um are more of an outsider, you may build products for for more of outsider. If you're more an insider, you might build products for for more of insiders. Um, and and I think like you should just but but I think the thing that sort of can can be really hard is if you you don't know who you are and you want to be something that you're not and then you're just building something without your heart in the right place. And speaking of remote work, you said it's very underrated in the age of AI.

I mean, a lot of people, a lot of startups want in person, people to be in person. Yeah. And I think there's a few reasons for that. So, um here's the here's the um steel manning of the inperson uh of the inerson argument.

So, in person in San Francisco, the steel manning of that is change is happening so fast in AI. Um change is is kind of is sort of like distributed differently. So it's San Francisco is ahead of New York which is ahead of London and like certain pockets of San Francisco are ahead. And so the the biggest argument for being in person in San Francisco and then the other argument is that um coordination is easier in person.

Um and uh especially like if you are a first-time founder, it is a lot easier to like keep sort of like keep everybody on the same page if you were all sitting in a room than if you're like coordinating people in in different places, right? Um and so so that's the the best argument for building in person in San Francisco. Um the uh the best argument for building for sort of building remotely um or or is that uh everybody or a lot of people um you know there's been a huge shift towards in person. Um and that means that it it is hard to build a team in San Francisco with great people.

It is hard to keep a team in San Francisco with great people. Um, and there are a lot of people that are uh you can hire uh people of a higher caliber if you're willing to expand your scope accepting that there's some tradeoffs in that. Um, and it's always funny, you know, when we do a an offsite or when we get together, whether that's a whole company off-site, we all wish that we could spend more time together in person, right? And so we love working with each other and we like wish that we could you know this is like even even as a remote company we you know we we we want to spend more time in person than than we're like physically able to do by the constraints of travel.

Yeah. No, I agree about the like talent caliber. Like it's just like there's just you find a lot more people like especially like you know people who don't necessarily have to be like in person in person like maybe designers or editors. It's just yeah it's easier like that.

And also there's like companies like Shopify, they're like remote first which is really cool and then they do like these team bursts every few months with the whole team. They come together so it's like very dedicated. Yeah. And I think you have to be like we we happen to the the thing though is that like you you have to understand that it is a degree of difficulty higher.

Um the the way that um someone put it to us once in 2017 and they were like an engineering leader. They're like, you know, someone that might be a a director in an in-person organization might only be a manager in a in a um remote organization. It is a higher degree of difficulty. Um so so you have to be, you know, it rewards sort of like people that can kind of like document their thoughts and and writing, people that can sort of also like keep in mind that remote does not mean async.

You can be remote, but you can still make all your decisions on Zoom calls. That's totally okay. So don't assume that remote means async. Don't assume you have to hire people from everywhere in the world.

It's it's much easier if you're like remote but like in North America or or or North American time zones or remote but like North America and Europe, right? Now remote and and all you know or or remote and and North America and APAC, right? It's really tough to have people in all three time zones because you can no longer have a standup where everybody is on the call at the same time which means that you know the inerson experience the sort of the bonding is lower. The other thing I'll say and and and like I have as you can tell I have a lot of strong opinions on this is that we have this really weird advantage which is that we went through this like crucible moment together you know 60 70% of the team at at Gatsby together.

So there was a real sense in which we were already in a a bunker together like you know with with our like firearms right like uh you know shooting out at at the at the enemy. Um and so we we didn't as much need that experience um with with Monsterra because we'd already had that together, right? Um so so the team bonding is still important, but like it's you know because that's the thing that like the team loyalty and team bonding if you don't have that like it's it's really challenging. Yeah.

No, it really is. And so what I think is really sweet is that you dedicated this book to your kids. They're growing up talking to models and agents as much as to people. And so as a parent building this, what do you feel about the world they're walking into?

Yeah, I mean, um, it's interesting, right? Um, you know, my my kids are nine and seven, so um, and you know, like as the models get better, right? They have more conversations with with, you know, hey Ekko, like tell me about the 10 different types of harbor seals or whatever, right? Like they have more of these kinds of conversations.

um you you know I I think there's I don't think that um that LLMs are a fundamentally in terms of like the human computer interaction paradigm. I don't think that they're a fundamentally different type of thing than the things before. I think they're a natural extension of search of like mobile types of things. So I we got used to like asking Google questions instead of like yeah I bet that like that answer that like you know that the biggest thing that ever happened was like a thousand people in the same place, you know, whatever, right?

Like we we we got used to like asking Google questions instead of like wildly speculating about stuff. We got used to and we got used to like using our phones in the same way. I think it's an extension and the you know the the trade-off um the trade-off is always that like the more time you spend interacting with computers especially as like a a kid the less time you you maybe spend interacting with other people. No, I think there's like paradigms like a lot of people that do gaming where you're both interacting with people and with computers at the same time.

I think that's the best option. Um but I think it's you know I think there's great thing about this is like they're learning to they're they're they're learning to get all the information that they want whenever they want it. Uh do you do your kids have phones or tablets or No, I mean they're they're too young for that. Yeah.

Too young for that. Yeah. Yeah. That's that's really amazing.

Yeah. Because I think a lot of parents they they give their kids like technology very early on as well. I I tend to like so so they do have they do have um small computers. The reason I give them computers and not not phones is that computers have a you can see this in like think about the level of gaming that you can do on a mobile device versus a computer, right?

like the the interaction patterns are like much more limited on a mobile device versus like I can do this with keyboard and I can move a mouse around like you can play like a real game on a computer versus like you know what are you like versus what you're playing on a phone the the the complexity is much more um and so so I do like they will play like games on on on a computer um but but I think that like you know now when you sort of like now this is the interesting thing right like if if you're talking to a model again it's a much more conversational paradigm um it's much more it's much more social in a way. You're talking to the models the same way you talk to a person. Um so so I think that like you get a lot of the the pro-social benefits without the sort of like you know anti antisocial costs. And so we talk a lot about like the positives with like AI but what about the negatives other than the common you know people are going to be like AI is going to replace like jobs.

What are I mean I think this I think this kind of comes back to like the question is like whether human computer interactions is a substitute for human human interactions um and and if we get better at human computer interactions if that if we'll reduce our skill on human human interactions I think that's just the fundamental trade-off of of computing um and and so like uh I think that's to me that's the biggest the the biggest question is like how do I you know for as as a parent is how do I encourage my kids to you know continue to develop their social skills. Yeah. Exactly. And also, I mean, the fact that you can like ask, you know, Claude for anything, for any help, but just like for example, if I like need some help with something, I still come to my dad with it just because like yes, I could easily maybe search it up.

Yeah. But it's still like it's really it's still really nice for maybe for him to maybe explain it or for him to like help me out with it even though I could easily just talk with Claude. Right. Exactly.

And and I think like I mean this is true both I think personally and and like you know one of the things that we've noticed is that like you know like this time around um people are a little bit more lonely working than they were last time because instead of like we don't pair program as much anymore, right? We don't we used to pair program a lot when we were at Gatsby, but we don't like because like what you're going to both watch them the model, you know, like we're each driving our own agents, but like we you know that that that makes less social touch points for us as a team. Um I don't know what to do about this yet. I honestly don't, you know, like hopefully we'll figure it out in over the next few months, but um you know, it's it's more of a like I think this is a I think this is a real challenge in in both Yeah.

I think this is a real challenge we have to think about. Yeah. And so you close the book with, "We're all perpetual beginners. To build something enduring, you have to stay humble.

" Um, I tend to think that, you know, people are either humble or or they aren't humble, right? like there's like you know some people are just sort of like you know and then but like once I I think the challenge is kind of like the thing that I think the thing that probably keeps us humble most is is our is is our user feedback channels right um because like it just constantly reminds us that like you know hey when are you going to release this thing or or like hey I ran into this bug or um I really love this like little improvement it would be you know, much nicer. And I think like that's the kind of constant stream of feedback that that you need where it's like, you know, when other people start liking your product and and you know, um, and then like you there's like a you some people can let that can kind of let let that go to their head. Um, I I think there's also like a sense that like for for myself like I'm you know like mid-30s.

Um, you know, I've had a had a couple kids. Had a couple kids. I've been divorced. My co-founder Shane has a couple kids.

He's been divorced. You know, like no matter how much uh how well like Monster goes, like it doesn't make it any easier to convince my daughter to like brush her teeth in the morning, right? Like it's just like, you know, so so like um there there's like a lack of of you know, portability to some I've also like gone through ups and ups and downs in my life. And you know, life isn't um you know, I end up just having a little bit of, you know, equinanity or or or whatever or perspective or something.

Um and then also just I think the experience of having having sort of like built something that was like pretty cool, very widely used, and then still sort of had a a pretty unsatisfying ending like makes you a little bit almost paranoid about like, hey, what's the next, you know, cloud on the horizon? um that that that you know I would say keeps you humble, keeps you keeps you like at the top of your game and what's the next cloud of horizon for you like what are you looking forward to with I mean it's both sort of like a cloud of excitement and a cloud of worry which is that so so yesterday I was speaking at AI engineer about my talk was called every harness will become a claw and it's about almost this you know agentic spectrum from LM to like now it can do actions it's an agent now it's like durable and dogged so It's a harness. It's like not just a local harness, but it's a cloud harness. It's always on.

And then, you know, now it's not just a harness. It it kind of becomes a a claw. It's like responsive. It takes initiative.

Um, and and like you know, every point along that journey, right? Like as a domain complete framework, we keep building more and more primitives at different layers of abstraction. Our goal is to kind of keep helping the users to do that. You know, if we' stopped building in, you know, a year ago, like we would be dead now, right?

Like so so the thing that's exciting is that we can build better and more advanced and more longunning agents and applications that can do more work in our lives. And the the challenge is that like we ourselves as a you know we have to keep up with that otherwise you know we're going to be toast. Yeah. And final question.

So what's the thing you believe about agents that you haven't found the right conversation to fully work out yet? thing I believe about agents that I haven't had the right conversation to work out yet. Um, I think that uh, you know, it's going like, you know, if we look at like the 2010s, what happened was that there's a lot of different categories of mobile apps that kind of took off, you know, hey, you can like use maps, I can hail a ride, I can order food or whatever. I don't even, you know, I don't think it's like I don't think we really know what the categories of agents are yet.

Like I have some ideas. I don't think we have a sense of like what types of agentic frameworks are better for things. We don't necessarily know this is these these parts of these kind of use cases are going to be you know some like hosted model hosted agents like managed agent services are going to win this open source are going to win that like I don't think we really know how the there's the agent world is so big right and it's expanding so fast I don't think we anyone really has an idea of how how it's going to shake out. Yeah.

Well thank you so much Sam. This is incredible. Appreciate I'm glad we could do this. Yeah, this is super fun.

And yeah, grab a copy of the book and stay tuned for the Yeah. third edition soon. Yeah.