Great documentation with Han Wang from Mintlify

When something works and you've had the experience of knowing the many things I didn't Yeah. It's night and day. Yeah. There's, like, no mistaking it. Mhmm. Right?

Yeah.

Basically, so we we met Paul Graham, and then, you know, we start when we walked in the room, we introduced ourselves. We're like, hey. You know, I'm Han. This is Hanby. We're from Middlify. Like, immediately, the first thing he said was like, woah. Middlify? Like, what what a what a terrible, terrible name. You gotta change that.

Hi, everyone. You're listening to Scanning Dev Tools. I'm joined today in person by Han from Mindlify. Han, thanks for joining.

Thanks so much for inviting me. I am a huge fan of the the podcast. It's truly an honor to be here.

Thank you. And so Mindlify is docs as a service, would you say? Docs?

Service. Yeah. The modern we like to think it's, like, the modern, standard for documentation.

Yeah. Okay. Amazing. So we're gonna talk about your story, and we're gonna talk about how to actually make good docs.

Yeah. Sounds sounds like a plan. Should I start off a little bit more about the story of Millify and how we got here?

Yeah. That'll be amazing.

Absolutely. So Minlify as a company was originally founded, not even called Minlify. Fun fact. Nice. This was oh, man. It was a it was a hot second ago, very late of 2021. Me and my cofounder, Hanby, we've, you know, built products for most of our lives just independent of each other. So I started coding when I was 11. Hanby also was, like, a really early like, you know, like, just like everyone else in the Bay Area, basically. So Okay.

Wow.

The the

Is it people you use them?

Oh, yeah. Yeah. Yeah. Oh, no. That that product had PMF for sure. No doubt about it. Yeah. The the problem was there was, like, a market size of, like, 200 people. That was the one issue. And middle school students, I don't know if you tried selling to them, they their, their their budget constraints not not not fantastic. So much. Yeah. Not not wish it could have been better. It's a market sizing. Temp temp problem for sure.

Mhmm.

So the idea was very simple. Our college had it, like, a very decentralized like, no one really knew what they were doing.

Like kind of complex ceramic Yeah. To take.

Exactly. And then at the end of the, like, the 3 years, you're like, oh, like, am I actually on track to get my diploma?

Yeah.

And the school didn't really have any great services and it wasn't super robust and so we're like, I bet we can I bet we can solve that? And I met Hanby just being, like, hey. Like, this is a problem I think we wanna tackle and built it, launched it, and, you know, I think even to this day, I I think about, like, half of the, like, the the the entire college uses it. Like, I remember every semester. Yeah.

Which college is it?

Just in

case people are not using that.

Yeah. It's Cornell if, if you went there. I'd love to again, if and if you use it, definitely let me know. Yeah. And so it's always been around building for people that, you know, we could have understood.

Cool.

And then afterwards, we just, like, tried those and we're like, I don't know, man. Like, I don't think Big Tech is is is is is for is for us. I love building products that Yeah. People loved and, like, that we could've, like, you know, just genuinely helped especially people who are like us. Yeah.

That just

there was something just so truly special about that. So I literally called up Hanby right afterwards. I was like, Hanby, do you wanna, like, run it back? Do you wanna go do this again? And this time, let's do it for, like, developers because that's an audience that, like you know, we've been I mean, yeah, we were just engineers our entire lives. We've just been builders. So let's let's let's go. Let's go take a crack at that. And and for some reason, she was like, yeah. Let's do it.

That was the that was not the code one.

That was the code one. Yeah. Yeah. Yeah. So they just launched codecs and we literally got, like, access to the API because at the time it was all wait listed. And I don't know how we I think I just, like, begged everyone I knew who worked at OpenAI for access Well got access to it. And I was like, man, like, this is this is so cool. Yeah. Let's go build something for developers specifically for this.

Yeah.

And so we literally took the model and then we, like, built around all of these cool features that were like, oh, you can translate your your code from, like, Python to TypeScript. You could explain this code in English. You could do all these cool things with it. That it was felt fundamentally about like helping developers.

Yeah. We

didn't really honestly know what to do with the model. That was the really reason why we made it so big. Yeah. Yeah. Like, zero clue what problems we

solved. You're you're a tinkerer.

Yeah. Yeah. Exactly. Yeah. Yeah. It applies to the strategy of a lot of companies nowadays. But, but we we just were like, let's try this. We we we put it out there and it, like, blew up. Really? Yeah. It literally we launched on product hunt. It did super well there. It got, like, number 1 product of the day. Wow. And, again, like, we we we literally just, like, we had the model and then we just sprinted it up to to market.

1st day.

It was it was really, really bonkers because

It's really wild.

Yeah. We we never we never expected it, to do that well. We just thought it was like it's such a cool thing that, like, you know, we just wanted to see if it solved problems.

Yeah. And this was translating code from, like

Yeah. Yeah. It we it was built to be in a way where it was, like, kinda multifunctional Yeah. Where you could, like, like, you know, like, translate the code, like, describe what it does in English. Yeah. It could translate code. It could calculate, like, the time complexity. Yeah. A lot of those, like, kinda like more like, it's, like, individual use cases. It was, like, a lot of prompt engineering, basically.

Yeah.

By the time they interviewed us, our numbers were looking like just down into the right. Yeah. And we were just like, yeah, guys. This didn't work. Yeah. And but here's something else that could. Yeah. You know? That's insane. And and for some reason, they decided to take a shot at us and, you know, we we did a YC for the winter 22 batch.

Yeah.

It was always like, you know, how do we help them, like, understand a little bit more? How do we help them enable them to go build better stuff? But, no, didn't really get the right footing until until many tries in.

Yeah. And what what was it like the first, like, iteration of, like, Minlify in its current form? Was it Yeah. Would or was it like Yeah.

More or less what you do now? The the initial, I mean, the initial vision of Minlify was actually it was actually super simple. It was literally just like, Hanby and I, after basically pivoting 8 times, we were kind of like we're like really almost out of ideas at that point. We just were, like, you know, like, we've tried a bunch of different things that we thought worked. You know, some that some that we spent weeks on, some that we spent, like, you know, a month or 2 on.

Yeah.

Communication.

Yeah.

Right? They were like, hey. Like, I just wanna flag this problem by you. Like, just our docs are really bad.

Yeah.

And, you know, like, I I maybe this is something that's worth experimenting. And Hanmi and I kinda listened to that a lot of time. And we've had interview after interview of people, like, just telling us that. And then even we thought of our own experiences building, like, the docks. And we're like, yeah, it's not great.

Yeah.

And by the way, it, like, it it it had a lot of, like, the core features that Mintlify has today, but it didn't really work. It was literally just static pages. And we I just I just showed it to him. He thought it was cool, and then I was like, I bet. I'm just gonna I'm just gonna make it make it so it's on our Yeah. On our site now. He did it, and then the team loved it. Their customers really loved it. And they What what do

you think they loved about it compared to us?

Yeah. It was just about the fact that we made, like, the the the docs experience a lot more interactive. Mhmm. We really focused on, like, the quality of the docs.

Like running code or, like, snippets or

Yeah.

Kind of, you know, selecting lang is it that sort of thing?

Literally exactly that. Yeah. And, also on, like, the engineering experience, be like, hey. Like, now you can actually edit everything in markdown and keep it within, like, your your core code base. And a lot of, like, these like, the different things that, like, kinda make MiddleFi stand out a little bit today.

Yeah.

It's night and day. Yeah. And when we kind of hit that, it just really felt like this kind of really magical moment. I was like, man, people are jumping on board to to take more demo calls with me so they can get access to it before, like, more than I could realistically take. And that was so, so, so different than what it was before.

That's a really interesting journey. And so then what was like once you had this kind of, like, people are paying, did you was it just like a blur? Like, what what happened after that?

Yeah. I think it's always funny because, you know, you you talk to a lot of founders about, like, their stories

Yeah.

And they always tell you about, like, the early days. Yeah. Because I think the reason for that is because those were, like, the most memorable because they were the most painful. As the company grew, like, from, you know, the first 10 docs to now, like, you know, the 1,000, that was just more so, like, just repeating and optimizing this process. Interesting.

Yeah. We had

to figure out how to build onboarding. We had to actually fill that how to, like, build these settings. The the fun fact is even for our first first few customers, they actually couldn't update the docs. Like, because it was so hard coded, basically Yeah. That it was just these, like, static pages. Yeah. Yeah. And and I was so grateful that the first few customers actually did it on YouTube for some reason.

Yeah.

Because it took us, like, another few weeks to actually go build that. And I literally remember when we eventually actually go build those functions, they were like, oh, by the way, Uh-huh. I just wanna mention, like, we gotta go update the docs now. I was like, oh, great. Great. Like, you know, like, That's awesome. But if they'd asked it any time earlier, we would have we would have not been in the

That was like a lesson as well of, like, if you have something like, if you can't give them, like, a really clunky

Yeah.

Experience, then maybe it's not pressing enough of a problem.

Yeah. Exactly. Yeah. Yeah. Or you've launched too late or something. Yeah. But if it's not, like,

that's interesting. Yeah. And before before we, like, start talking about docs, do you wanna tell us why Paul Graham told you to

Yeah. Graham interaction? Oh, man. Yes. Yes.

That's so funny. Yeah. But it's become a it's become a big name in dev tools. There's so many great companies using it now. I know that people like, I know my friend, Flo, is a big fan of minnify. I've tried it. It's great. Yeah. And so you, I guess, become well, there's no guessing. You have become an expert in docs now, which is something really hard. So I wonder if you wanted to share a little bit about, like Yeah. I don't know. I mean, I'm sure founders ask you all the time, like

Yeah.

How do I make good docs?

Yeah. Yeah. Well, let me actually turn this around for a sec. Second, actually, I'm curious to hear your thoughts on this. When you think of, like, great docs, it's like a wonderful doc experience from any company who you might just try to build something. What do you think about?

You know what comes to my head more is, like, the bad docs. Yep. I feel like it comes Yeah. You know, like, you don't often think, like, wow. This is great docs. I mean, people talk about, like, Stripe when there's something, like

Yeah.

Outrageous. But, like, I feel like it's more like, can I find the thing that I want? Is there lots of, like, contradictory stuff? Right. Is it, like, unclear which version of the API it is? Yeah. This sort of stuff. Yeah. I would say that if I can find what I want Yeah. Then it's usually, like, pretty good. And then Right. Everything else is, like Yeah. I I feel like it's, like, tied up with, like, does it work? Like, does it Yeah. Do I get it? Am I able to do the thing fast?

Yeah. Right. And so it's, like, a big part of it is, like, just being able to get the get the job done. Yeah. Right?

Yeah. Yeah.

Yeah. And then I would say in terms of, like, that kind of, like, you know, any kind of advice or strategy, I think people do ask me all the time. It's like, how do I really maintain great docs? And I always say there's, like, 2 parts to this. The first the first part is that, like, there's not it's not like a playbook as much as it is just really having a good process and culture around it Mhmm.

Yep.

Right? Like a quick start guide, you know, these right links, you know, and all that stuff. But I actually don't think that's really the the meat of it. I think building great docs is very similar to, you know, just like building a good product too. You know?

Yeah. Yeah.

Right? And then how do you just iterate on over time with feedback? Right? Yeah. And just getting a team to even think of it that way and to make sure it happens Yeah. You're there, like like, you're, like, 90% of way there or even even if a 100 in my opinion because most people don't do that. Don't do it like that.

Why do why do most people not do that? Because this feels like it's like something that I guess we all know we should do

it Yeah.

That way.

Yeah. Yeah. Developers don't like to write. You know? Like, I I think that's the that's the that's the that's the really the meat of it. Right? Like, you know, it's just not in the training, I I would suppose. Like, the the, you know, like, developers got into this industry perhaps choosing to have, like, you know, it's when you're when you're, like, you know, in in school, you could have chosen the humanities route or you chose the STEM route. Yeah. Yeah.

Yeah.

Right? And so I think to that, like, you know, I always think about how do we make it, you know, really easy for the people who perhaps might not enjoy writing to actually just contribute very easily. Yeah. So this is one of the reasons why we at Middleify have always took, like, this docs as code approach

Mhmm.

Of keeping it within the workflow that they're familiar with. Yeah. Keeping it at markdown.

Yes. Yeah. It's a pretty interesting flow that you've got there. Yeah.

It's it's the one that we just kind of like we're like, this is how we would have wanted to contribute. Yeah. You know? We wanted to, you know, push out PRs with markdown file changes the same you know, alongside the features. And keep it within, you know, the the whole technical workflow.

Yeah. And does it seem it seems like there's kind of sometimes, like, is it important within the company? Yeah. You know, like Yeah. Having Stripe, they're they're kind of obsessed with documentation and books in general, I think.

Yeah. Stripe is very well known for their writing culture, is which in one part why they've done such a great job at it. Yeah. I think they've just, like, they have this crazy obsession about, you know, like, just being able to really put this proper tutorials and explanations out there

Yeah.

And, and that's that. But I I think in the question of in terms of, like, how important it is, I always like to think of it from this other perspective which is, like, imagine if you are a Stripe for a second. What would happen if your docs disappeared for a day?

Yeah. People are not gonna make many changes.

I don't know. Yeah.

Not too many people are gonna get started.

Yeah.

Only the really dedicated people that might

Right.

Dive in and actually go look at the code.

Yeah. Right. Or yeah. It's it's yeah. If if if you even could. Yeah.

If you even could. Yeah.

Right. And so, like, if because I would dare press. It's like if, you know, if the docs just disappeared for some of these, you know, big dev tool companies, like, it's almost to me as if the product has just disappeared, kind of. Yeah. People would stop being able to properly use it. People wouldn't figure out how to actually go work with some of the things. And under that light, it's so, so important to think of it like it's

the product.

Right? And that's always the like, I'm the biggest advocate of, like, you should just think of docs as a product. It's not like this little side thing that Yeah. You know, could, like, you know, that you'd want to, you'd want to just like have on the side to kind of like help some people.

Yeah.

But because if that's the case, then sure. Like if docs are spirited, you're fine. Yeah. But if you're like for a lot of these dental companies, it's not. Yeah. And if that's the case, then you definitely shouldn't treat it like that. Yeah. Right? And so, so I think that's just like one big, big kind of thing to I think that I've always been advocating for founders to think about. Yeah.

And so online tutorials, do tutorials sit within your docs ideally? Or, like I feel like sometimes there's kind of a bit of a blur between, like Yeah. Blog Yeah. Articles on their own or, like, full guides plus, like

Yeah.

You know, quick starts and you know, how do you do do you see people doing

Yeah. Like, the difference between the different content

and things?

Do you

think it's something yeah. I don't know. It just feels like sometimes it could be Yeah.

It's like it could be all over the place. Yeah. Yeah. I think my my opinion is, that it again, I think that, like, you can worry all about how to, you know, like, make it intuitive into, you know, structuring things together and making sure that a user goes to the right places at times for the things that, like, you know, making sure they're discovering the right content.

Yeah.

But I really again think that it it's the the companies that have issues with this stuff and don't do it at all, it's often that it's rarely the case that they have information, like, at too many places Yeah. And to spur places, but that they don't have it at all. Right? Like, that's, like, to your to your point, you know, when you when you ask about what makes good docs. Right? Like, it's bad when it's not there.

Yeah. Yeah. Yeah. Yeah. True. That makes sense.

It is I've rarely seen a case where it's like too many Yeah. Too many docs.

I guess because we use Google anyway. Right? Like, it's probably, like, if you're trying to solve a specific thing, you can

say,

how do I do this in JavaScript?

Exactly. Yeah. Yeah. That's why, like, search is such a I mean, like, functionality is a huge part of docs, good SEO. Right? Yeah. It's just crucial.

Yeah. With, like, you know I mean, obviously, you got your start with the AI stuff, with OpenAI. Do you think there's any are there any things which are making, like, seriously significant differences to, like, docs.

100%. Yeah. A 100%. I think there's going to be different waves of changes into the way people think and interact with docs.

Mhmm.

And I think, I'll I'll I'll narrate, like, kind of the ones that I I think first of all has already happened and I think I'll talk about the ones that I think are still yet to come Yeah. Based on the quality of the models, which is the first one that's already happened is that people are getting to an answer more directly. Mhmm. Therefore, actually structuring the docs in the way that they they they the car currently is starting to change a little bit.

Yeah. And

I'll describe what that means. Basically, the idea is whenever you're thinking about building a docs or the goal of it in in my view at least is that you should think of, like, you know, like, your docs structured in a way that's kind of like a tree. Right? You start from a home page Yeah. And then you branch out into your whether it's, like, you know, like, if you're looking for SDKs, your SDK references, your API references, or just some some sort of quick start tutorial.

Yeah.

Or they're trying to get to a set of paths within that tree. So for instance, you wanted to learn how to figure this out and, like, you wanna learn how to, like, the thing works. Yeah. Right? And this is, like, the the critical job of any docs is you wanna make sure that things are interconnected, that it's really easy for anyone from any particular point or from the beginning to get to that individual destination.

Mhmm.

This is a nice way of, like, a really, like, engineer way of saying that you can now get answers directly to any questions that you have instead of needing to worry about finding the right paragraph or the right keyword or the right answer from a page. Yeah. It's all built from content when the doc so the doc still needs to exist. Yeah. Right?

That's so good. Yeah. Yeah. So you kind of don't even need to, like, have the perfect experience for everyone. It's like Yeah. Because they'll be able to choose their own kind of adventure there

Exactly.

If you've got Yeah. All the information. Yeah. So you're kind of just feeding Yeah. Chat gbt or whatever.

Exactly. Sense. I think this is the first thing that is starting to already change the dynamic of technical content. I mean, as it already has with, like, chat gbt, you know, with Stack Overflow Yeah. You know, Copilot and all those great things, the the way it's kind of, like, infiltrated into, like, the docs world for any future company is, like, now more and more users are just getting the answers directly to what they're looking for Yeah.

you know,

this kind of rebirth of, like, this, you know, prior to, you know, like, Google, there was just all these index in like, you know, these individual pages back when search engines were just, like, a bunch of links. True. You know? And now you have this, like, magical thing where you can get to the answer directly.

That's so true. That's a really cool way to think about it.

Yeah. Yeah. Like, does that replace fundamentally, like, what Wikipedia still is? Yeah. No. But it enhances, like, the the way that ability for users to actually get to the answer directly. Mhmm. And I think that's just the way that, like, LMs have unlocked this kind of new, like, you know, ability that people haven't been able to do before.

Do you think it changes, like, the role of docs in a sense then in in not not just in, like, filling it out, but, like, if someone does come to you, is it gonna be less is there gonna be less there gonna be less people coming to your docs Yeah. In a sense, But then maybe the ones that come are, like, expecting Yeah. More or

Yeah. I I think there's still gonna be the 2 roles to it Mhmm. Where it's, like, it'll never be, like, everyone knows, like, they'll come in and they'll just simply ask chat and get an answer directly. Because I think there's another part of Docs that's super useful which is to even know what questions to ask Yeah. And to know what the product does.

Yeah. Right?

There's still gonna be textbooks. Yeah. You need to know what you don't know yet. Mhmm.

And I

think there's still is gonna be the serving the role in that. I think what AI does is it enhances like this, like, kind of like this kind of like optionality depending on the use case. Because the use case of docs as often is is you just wanna get to an answer to fix this, like, terminal, like, you know, error message that you're getting. Right? That's one big unlock, that has never been done before.

So do you think there's any is there now, like, a point of having, like, how to do this in, you know, 15 different languages Yeah. And, like Yeah. How to do this slightly different use case and this slightly different use case?

Yeah.

Or, like, do we not even because we could probably just generate those now.

Yep. Do

we need to is it worth generating something that can be generated by ChargeGpc when they need it anyway?

Yes.

Yes. So

so on the kind of the note on, like, the tutorial stuff Yeah. I think you described it perfectly. I I think we at Middlify are kind of experimenting with this right now Yeah. Which is, like, could you fundamentally be like, okay. Like, you're in Java.

I used to. Yeah.

Right? It's gotten to a good enough threshold where it is actually trustworthy Mhmm. Like, the vast, vast, vast majority of times. And then the time that's actually wrong is not because of the model, it's purely because of, like, what content you even have within your docs. Which actually then changed the paradigm of, like, how you even think about the responses here.

Yeah.

What does that mean? It means Missing. You the docs aren't there. You know? Like, or maybe even sometimes even worse, like, the product should have it, though.

Like, you

know, like, this is not there in the product. And that changes the way people are actually writing docs now. And so we're starting to see that a lot of our customers are literally taking that feedback, reading it, like, every single week or every like, the technical writers

have been professional. You're capturing that information so that they can see it.

And then they're looking at this, and they're like, these are the questions, these are responses. These responses here, not getting answers. Yeah. These responses here It's

a killer feature.

Yeah. Yeah. Yeah. Use middle of mine. So You

got me on that one. That's great.

And or, like, just like these answers are just wrong.

You

know? Like, go go go fix them. Like, I can't think there's a better way to just highlight exactly how you can how you can do that better. And now that the models are there, you're actually able to unlock this change in behavior.

Yeah.

And so from, like, what we've seen, it's like those two main things from chat to changing the way people are editing the docs is just the beginning of what I believe will be a long tail of many different ways that kind of, like, you know, writing the docs to reading and interpreting as a user is gonna be different.

Yeah. That's that's so cool, actually. It's like you're getting this, like, constant, like, feedback of, like, what people are trying to do.

Yeah. Exactly. That's incredible. Yeah.

Okay. Han, that was extremely interesting,

and I

think I just got sold all the notify.

So,

if you have one takeaway for people listening, what would it be?

Oh, man. I'm not I'm not I'm not a I'm not a I'm not a wise I'm not a wise guy. I don't have the best advice to give. How about I give a cow joke instead?

Great.

A cow joke? That's okay. Yeah. Why do cows have hooves instead of feet?

Why? Well, why do cows have hooves instead of feet?

It's because they lack toes.

That is horrendous. Thank you.

I appreciate I appreciate the pity laugh. Thank you. Yeah.

Brilliant. Okay. Yeah. Well, Han, thank you so much for coming. Where can people learn more about Mindify and about you?

Honestly, just just, just, just look us on minnify.com, I guess. But yeah.

Brilliant.

Thank you so much for having me on, Jack.

Thanks, Tom, for coming. Yeah. And we timed that perfectly because my roommate has just arrived.

Oh, perfect. Awesome. That was great. Thank you. Yeah. Really well done.