How Mintlify Is Rebuilding Documentation for Coding Agents

0:00Hey guys, here's your docs that are way better now on Mimovac. Take a look. and that's when I realized the beauty of having failed so many times, which was in failing so many times and building so many ideas that didn't work, when something did, there was no mistaking it. It's very important to solve one problem correctly, and even right now as we're experimenting and launching new products, we're hyper-focused on getting one person who loves what we're building as opposed to a thousand people who are going to just feel meh about it. It's still the small things that don't scale that really spark the customer love.

0:43And the thing that goes the extra mile, if you will, is how we can do that things, right? If you go the extra mile in a way that people don't expect, you know, imagine.

0:54For most of the internet's history, documentation has been an afterthought. Something you write after the product ships, something meant to explain what already exists. But what happens when the reader isn't human anymore? For decades, docs were written for people. They explained APIs, answered questions, and slowly drifted out of date as products evolved. That decay was accepted as normal because documentation was treated as reference material, not infrastructure. In the last few years, that assumption has broken. Coding agents support bots and internal AI tools now read documentation directly.

1:25Docs are no longer just explanatory, they're also operational input. When they're wrong or outdated, systems break. This creates a tension. Documentation matters more than ever, but keeping it accurate has always been one of the hardest problems in software. Static docs don't survive fast-moving products, especially in an agent-driven world. In this episode, we explain how documentation is changing, why static docs are failing, and what it would take to build documentation that stays reliable as software evolves. Our guests are Han Wang and Hanbi Li, co-founders of Mitlefi, joined by A16Z general partner Jennifer Li and A16Z partner Yoko Li.

2:04I'll start the podcast with this question. Given you're at the front center of seeing how agent has impacted coding because you're building a documentation product that's serving all the developer tools, the most popular ones, how have you seen the transition from the starting point of building Malify to what a role of a coding agent plays now? I mean, it's so different now, right? I think, and it speaks a lot to even how much has changed in such a short amount of time, right? I remember when Hanbi and I, you know, decided we kind of wanted to embark the journey of wanting to go build something that helps developers.

2:36Right. The first thing that came to our mind, this was like to paint the picture, like early 24, late 23, early 24. We were just like, let's go build something that could impact developers, help their journey as kind of like the impetus for why we want to start the company. And then landed on building like just better developer docs because we were like, let's go solve problems, you know, like we can relate to. And as people who've been building our entire lives and careers, we're like, we know that there's so many bad dogs out there. There's an audience we can relate to. It's a problem that we deeply understand and care about.

3:08Let's just go build that. And then, obviously, what changed from then till now was this tsunami of changing developer expectations, changing the way people build entirely. So what initially started was just like a very simple platform for people to build, write, maintain better docs, read for other humans, really start to transition to being something for humans and AI. And what eventually now it went from being an application to, it really feels like an infrastructure product, right? It's content that powers AI. It's content that trains support agents, coding agents, you name it. And I think kind of witnessing that change firsthand is not only a testament to how much the world has changed, but being front and center and having a role in something that we find particularly meaningful.

4:03Obviously, nowadays, you know, like even before we invested, we see Millefi all the time. Like every company we meet in the very early stage to enterprises, everyone is centralized on Millefi. Was it always obvious to you? What's the, if you look back, what's the story that led up to this point? Yeah, a lot of, a lot of pivots and a lot of failure along the way. And by the way, you know, my opinion, too many people use us. You know, we need more variants in the world. That's what I think. More themes. Exactly, more customization. The honest answer, and, you know, and I kind of want to, you know, speak on that is the fact that, you know, it wasn't the first thing that Hanbi and I came up with, right?

4:48In spite of how much we see and how obvious the problem is in hindsight. Hanbi - You pivoted eight times, right? And this is the eighth time. Yeah. Oh my goodness. So essentially the story of all the pivots and it goes, you know, for me and Hanbi, we started the company because we really know we want to build for builders, right? I was very fortunate to like started coding when I was very young, initially 11 years old. I was given a laptop, like a MacBook Pro back in the 2010, 2011 edition. And I found the joys of learning how to build products for people around me, community, my school, my friends, you name it.

5:27And that entire journey was self-taught like many other people who are very prominent builders in the ecosystem. I had to learn so much by myself. I had to go figure out through YouTube videos, a lot of bad docs, right? No cloud code, unfortunately. I wish that would have been great. It's going to help me figure out all the space. And then therefore, when we started the company, we're like, that's just a thing that we can relate to so deeply. And because we know that building a company takes such a long time, right? It's like the span of decades. If you want to build anything meaningful, you have to go pick a space that you can care so deeply about.

6:03And for me and Hambi, it's other people, other builders like us. And so that was the start of the journey, right? It wasn't all about docs at first. It was just like, how do we go build something that can enable people to build a little bit better, a little bit faster, because then they can go and impact people. And so then, you know, we started the journey and spent about a solid year and a half of wandering in the desert while chewing glass is probably the best way I can describe that. What did you run into in the desert? A lot of bad ideas. Too many to list and too bad to pitch at this point.

6:40But I think it was just a lot of things where, you know, we say like it was eight pivots, but, you know, truly, it really didn't feel like that because all of it was really surrounded with kind of the same kind of like idea space, if you will, right? It was always something for developers. It was always something about building for helping enable someone to go learn and build something faster. And so one idea after next, it went from something like, oh, using AI to auto, just describe what your code does to something like, hey, could you just connect static content to the code? And then that was a bad idea, by the way.

7:17I don't know if anyone is doing that. Don't. But then it eventually led us to where we are today, which again, in hindsight, is the most obvious thing, right? Oh, duh, I should have built the docs platform. But at the time, it was just about developers. And to this day and into the future, it still is. It's about enabling other people, and now we're growing beyond just developers, into learning and building a little bit faster. And I'm sure when you first launched Melify, this current version of the product, there was a lot more mature products out there and more fully fledged, like feature-rich products versus how Melify came out back then.

7:56Like how did you approach customers in the way that converted the first customer or the second or the third or some memorable stories around that time? Oh, man. Well, for starters, you know, the funny thing is the reason it took so long for us to even get to the idea that we're doing now is because we actually internalized that same opinion, which was, oh, there were so many of these incumbents, these existing companies that did it. There was a solution here, a solution that, oh, you could probably go build these things in-house. We just bought into that, right? And at one point and throughout the journey of these many pivots, we kept talking to people, right?

8:37Talking to users. And they kept saying, oh, well, while these other options existed on the market, I didn't find the right one for me. And here's why this is not great. And in fact, we kept hearing that again and again and just kept pushing that down because we're like, oh, no, it's actually great. You know, the market's already there and yada. And as we kind of start dugging into the user feedback and we start thinking about our own experiences, we're like, my goodness, like I've tried building this so many times. Here's why all of these things were bad. So if we're going to do it right, let's at least do the one that we wish we would have wanted, right?

9:15As developers, because we believe that's, you know, who we want to build for. Let's just go figure out what we would have liked. And so the initial version of what Mintlify is today actually came out just from one weekend. It was two days. We were like, because at the time we were like, all right, well, we just got to try everything and we'll just give it the amount of time we kind of have confidence in this idea for two days. That's all we got. And so we forced ourselves to build like this very initial version of it. And then the next thing I knew, like we built out this like prototype and we just want to get validation.

9:48So at the time, my roommate who was going through the YC batch, you know, we're building and pivoting into the API space, this company called HyperBeam. They were going from a consumer product to building an API product. So we went to them and we're like, hey guys, here's your docs that are way better now on MidRefy. Take a look. And that's when I realized the beauty of having failed so many times, which was in failing so many times and building so many ideas that didn't work, when something did, there was no mistaking it. Right. And so there's a lot of grace and failure, in my opinion, which is that like you learn a lot of perspective on things that don't work and the reaction customers get when they don't work.

10:32So in our case, it was many, you know, a year and a half of people who are like, oh, let's talk again next week. Let's, you know, you send them an invoice, they're not paying it. You send them a credit, they're like, oh, we don't have budget. It's 20 bucks, you know, to then the reaction. Yeah, exactly. To then the reaction of like, how do we get the setup right now? Like, I was like, okay, do you want to find time tomorrow to chat? No, like how do you do it right now? And then you send them like an invoice and they're like, they're done, like in a minute, right? And then for HyperBeam in particular, they were like, let's get going.

11:07I was like, bet. I'll set up for you right now. Give me access to your cloud for DNS configurations. I literally went and changed it over and that was our first customer. And that first customer led us to the second, then the fifth, the 10th, 100th, and now into the thousands. That's a great story. And that's still one of your current sales motion is to pre-build the doc and share it with the customer to see what the docs will look like in Millify, which is very powerful. Yeah. It's one of those things where I remember like maybe we were nearing like the first hundred customers and Hanbi and I were still doing the painstaking job of like manually migrating everybody.

11:46And by the way, back then, like before any of the bells and whistles for why someone should use us. You know, one of the reasons, ways we got people to give us a try was like, hey, by the way, if you switch to Millify, we'll actually go in and like actually help review your docs. We'll fix your grammar and we'll actually structure in a better way. There's a lot of the do things that don't scale. And we were doing that for a while. I remember at one point we had like an office hour, if you will, with Paul Graham. Founder of founders. Yes, founder of founders, the YC guy. and we were so many vibrators like how do you like do this i'm sure at one point when you scale it out you get to go do something a little bit different a little bit better he's like no this thing that you're doing now is going to be the thing that you're going to do forever just live with it i was like okay and then we hired a team and then built tooling and then you know obviously this it's you know ai has helped a lot in that journey too but yeah it's still the small things that don't scale that really spark kind of like the customer love and the thing that goes the extra mile, if you will, is how we think about things, right?

12:52If you go the extra mile that in a way that people don't expect, you know, magically. I'd love to maybe just point out a few of those going extra mile pieces because I would love to dive into more of the product intuition and also the products you have built so far because that's been an incredible journey, even in the short sort of lifespan of Menlify that you're just like nonstop shipping and with sort of AI's change of how developers coding and how they're leveraging documentation. Like you have came up with a lot of really great ideas from self-healing docs to sort of how to bring docs into context for agents to leverage.

13:28So I would love to sort of unpack that in the next few questions. And maybe we can start from where you start seeing these sort of areas that you can go further from the existing providers and how has AI impacted your thinking of building those functionalities? Yeah. Well, if I were to touch on some of the previous topics as well, I feel like a lot of what kind of goes into our product intuition and how we build is living and breathing the mantra of talk to your users. And also learning from our failures and being very data-oriented and being very pragmatic and MVP-ing things very quickly. That was kind of a lot more than I was expecting to say, But I think that that kind of really encompasses a lot of just how we approach product and strategy in general.

14:18And I think a lot of people who start from an idea that they have aren't starting pragmatic enough. And it's very important to solve one problem correctly. And even right now, as we're experimenting and launching new products, we're hyper-focused on getting one person who loves what we're building, as opposed to a thousand people who are going to just feel meh about it. And so we're, yeah, we're super focused on that. And then as for, you know, the future of the product and to touch on like what Han was saying about how AI has been impacting our product, like the future that we envisioned a year ago is here.

14:59You know, AI agents are the ones ingesting the content. And like, I'm constantly talking to founders who are talking to me about how like the documentation is the source of truth for making sure that their agents are working properly. And so this up-to-date documentation has always been important for developers, marketers, customer support, everybody within the org like views it as like a source of truth. And now there's an outsized amount of AI agents who are also relying on it. And so it's even more important now. And now AI is also here to help us solve that problem where humans like natural inclination is not to update the docs.

15:38You know, you're shipping code and you're, you want to say it's done when you're done, when you've shipped the last PR. But in reality, you need to, now it's even more important to update the docs. And so with this increased like importance, we were building tools to make the context stay up to date as well. Yeah. And in a different way of, you know, VC terminology, like your time expanded so dramatically now just because, you know, on the supply side, you have a lot more to serve of, you know, people who are building tools and people who are, you know, building dev tools, agent tools. But on your demand side, you have this new persona that are not just developers, but also coding agents.

16:18So the demand of docs and great documentation, accurate ones, up-to-date ones is getting much, much higher. Oh, I was going to say, and the vibe coders too. The whole new genre of developers are coming up, people who are so enthusiastic to build software. It's awesome. Yeah. And you guys are at such a vantage point of Silicon Valley because just because every single, you know, new AI company or established AI company, you know, use Mitlefy today. So I guess like from all the things you are seeing on the market with the customers, what surprised you? What are the things that people are now using with Mitlefy and the AI models I couldn't have done before?

16:59with mintlify there was a lot but that pales in comparison to the number of people who are building things that are cooler than beyond you know our imagination i was literally you know this is a little embarrassing to admit got my driver's license like not that many years ago that was literally two years ago even as an example they started on driving literally and then I started driving non-stop maybe you don't need to drive just use Waymo exactly or the robo taxi I was like literally preparing for the driver's test and I came across like this document that was like here's how you prepare for the written exams and it was on Millify and I was like I would never in a million years have imagined that this is like a place I would just come across and to your point like in a lot more places than I can even imagine And this is kind of like a common misconception I think people even have about Millify today, which is that, you know, they're like, oh, it's Millify.

17:56They're the docs company. And while it's not incorrect, of course, right, it's, of course, what we do, 25 has been a big year for us because it also dramatically expanded the surface area of what we actually cover. So there's a lot more help centers on Millify today. There's a lot of internal docs. In fact, actually, we recently replaced our entire internal knowledge stack with just Netlify as well, largely because of the automated AI tooling, the Q &A bots. So everything just asks a question on Slack. You tell it to go update things on Slack, right? You give a natural command and it just evolves over time is something that I've been personally getting so much use out of that I, and for many other companies that we work with, you'd be surprised by.

18:39They do use us for a public docs instance, and that's kind of obviously the thing we can showcase. But deep down, they have two sets of internal docs that we now work with for maybe their engineering team, for maybe even their HR team, who's talking about like their healthcare policies. And I think as we grow, it's going to evolve from not just quote unquote docs, but like more knowledge management more broadly. And I think kind of seeing that shift in 25 and hopefully more into 26 as well is something that we're very excited about. And why do you think the customers are using Minify that way?

19:10because the original and the most prominent branding is you have this very well-designed external facing dots, but the internal workflow could be pretty different. There may be more options of using Notion or Confluent. Why did people, even outside of the developers, are choosing MoneyFi for these use cases? Yeah, that's a really good question. So there's a lot of different things about this. The first and foremost is that you have to actually go build for obviously a great editing experience beyond just the developers, right? So when Hamby and I started, we took the opinion that it should just be developers.

19:43And so we really optimized for that flow. But of course, as we kind of grow in more use cases, larger companies, when you're talking like Fortune 10s, they're not just all developers. There's a lot of stakeholders. And so you try to build a better editing experience. So there's a point about product and features in their side. But the larger one that actually I think has been shifting in this is that a lot more of the world is also on the other side getting more technical, right? To Hanby's point earlier, a lot more people who are Vibe coding now. A lot more people who can understand Markdown now because the language of LLMs, right?

20:16So the syntax that might be somewhat foreign for most people is now commonplace. A lot more people outside of just developers are using Cloud Code and these tooling, right? Like the cursors to go ahead and do creative work, writing blogs. I remember, Yoko, you mentioned you wrote a blog, right? Using it as well. and all of that's done in like this developer-y way. And so especially now where AI agents are also powering so many different workflows, the engineers themselves too get also very involved in the process. So back then, if you want to go sell a support product, right, you're selling to the support teams.

20:55If you want to sell a help center, it's selling to the support teams. But now if you're basically like the support strategy is like 80 % of the tickets are built and managed by agents, the engineering team gets involved, right? And so then they have a say in terms of how you want to go build a platform, how do you want to maintain it? And so this kind of the early bet that we took, which was being for developers, also kind of became this tailwind for us as we saw. I wish it was more intentional than it was. It's my honest answer too. I guess, what was it like building it? We had this async discussion on Slack just very recently.

21:29there's just so many net new things that you can ship into the product, like when the coding models are getting better. So for example, workflow automation, self-updating docs, like the extra things that non-developers or like newly technical users can now do, how did you all wrap your head around all these new trends in AI and how did you build it? Still trying to, honestly. I wish I had a better answer for that. I think this is going to, you know, what Hanby mentioned, the pragmatic side about all this, right? Which is, on one hand, we can sit here and talk about the theory of it all, right?

22:05The general macro trends that enable products like this. Whereas, you know, from the, you know, the battle scars that me and Hanby had, you know, in this going back into the pivot days, is we also learned that it's so important to figure out how to bring that to the actual problems that people care about, right? Yes. Self-updating docs, very great thing to have, right? It's been asked for about 25 years now. And now there are a bunch of reasons why it can and should exist. But at the same time, there's about a thousand product decisions that need to be made from now until then. What is the actual way people do that?

22:40What is the actual experience of how those updates should happen? What sources should they connect? How are those connected? What is the actual flow of it all? Those are the things that you still have to get right. And so tying it back to it all, it's, you know, there's, of course, this big element that we want to pursue, but then it's going to come down to the day in and day out, the users and the problems they face and how they would like to see it. And that's kind of like this kind of the messy middle, if you will. Yeah, I feel like there's a lot of buzz going around in AI nowadays. And, you know, like even like people are scared because it's moving so quickly and they don't know what's going to come next.

23:20and like AGI, like doomsday, like sci-fi-esque scares going on. But the way that Han and I view it is that AI is really just another tool in the toolbox. And it's just so powerful and useful and valuable. And we just now have this extra tool in the toolbox and we should feel very open and embrace it in terms of utilizing it to help others and provide value for others. And I feel like we could abstract away, don't think about all the scary media around it and just view it as a helpful tool to help us get to point A to B, to help people understand products better, to help learn how to code faster.

24:03All of these, that's what AI is here for. Yeah. I kind of agree more with that. I think there's certainly a lot of things that coding agents and AI has surprised us of what its capability, but at the same time, there's still the limitations that we need great tooling to work around it with. And that's why context and documentation is so important to continue to feed the right pieces into context and for the agents to perform. And that also goes to sort of the maintenance piece that Han you mentioned earlier. It's always been the case where authoring is like a big chunk of documentation. It's an important part, but it's sort of one time.

24:40But the maintenance side of maintaining docs and keeping it fresh has been the biggest burden for organizations to make decisions on if this is the right version, this is the right version, what's the source of truth? What's the UI that's actually showing in the product today? And is it the right one showing in the docs? Like there's just a lot of thousand paper cuts that's making it so time consuming and cumbersome where now we can probably have a really capable assistant to help us do that a lot easier. And I think that's sort of with your self-healing docs you're trying to solve now and maybe not just for docs, but for a lot of knowledge work overall.

25:14So I'm curious and want to hear about what are some of the, I guess, bigger chunks of content you want to go into. Also, how you're solving that problem to reconciliate and come back to the source of truth, that kind of decisions you're trying to make now. Yeah, absolutely. Well, I first want to unpack that by breaking down, I think, a few different elements. The first and foremost is like why, quote unquote, docs, you know, have been historically out of date and difficult managed. And make no mistake, this is a problem that I'm sure everyone who is listening or understands is, you know, this is something everyone feels.