Authentic Developer Education with Dylan Fox from AssemblyAI

Jack: Hi everyone, this is Jack from scaling DevTools. The show that investigates how DevTools go from zero to one. I'm joined today by Dylan, the founder of Assembly AI, a speech to text API. , thank you so much for joining us, Dylan,

Dylan: Yeah, thanks for having me on excited to be.

Jack: could you tell us a little bit about Assembly AI?

Dylan: Yeah, definitely. So, in brief, we are building an API that, developers and product teams can send, audio files to, and then our APIs can automatically transcribe and. We say, understand those, audio and, sometimes video files, basically running them through our AI models that can do, things like automatic transcription, content, moderation, topic detection, and summarization. There's a few other things that can be, applied as well to that audio video data. So we have product teams, building features and, products, on top of audio and video data leveraging. Our APIs and our AI models.

Jack: I've seen that you've done a lot of very interesting AI content. Could you talk a little bit about your content strategy?

Dylan: Yeah. Yeah, sure. So, we've been working on the company for a while and I think when we think about. Our content strategy. It's really about wanting to try to make sure that developers know about assembly before they even have a need to use us. Because we just want to be a tool that they think about when they're trying to solve a problem that could be solved with our API in the future.

And that could be a week in the future or years in the future. and I think Twilio is a. Example of a company that did this really well in the early days. And I have a lot of friends at Twilio. I had even at one point applied to be a developer evangelist at Twilio, like years and years ago, and ended up taking a different job, but, uh, what they did a great job of is making sure that, developers just knew about Twilio before you even had a, a need.

And I, for myself as a developer, I find out about cool developer tools all the time. And that's through content or just through word of mouth, like we were talking about earlier. And even if I don't have a need, sometimes it resonates with me. And then I just kind of bookmark it in my brain. And then it might be months or years later down the road.

And I go back and use that tool because the need arises. I have so many examples of this I could give, but, that's how we think about our content strategy. I think, there's definitely some content you can create as a developer tool that targets high. and a high intent audience, so to speak where it's like, okay, people are for us looking to find an API that can automatically transcribe audio and video content. So there's a lot of content we can create to target that high intent, so to speak traffic, but really the bigger goal is especially for us because we're in a new. Market where a lot of people don't even know that you can do what you can do with our APIs. And it's such a new industry and new field that for us, it's about just generating awareness that we exist and that there are APIs for this that you can use now. And hoping that inspires a lot of folks to build things with our APIs. Which is, I think exactly what are very similar to what Twilio and a lot of other companies have done really well. So that's how we think about it, at a high level.

Jack: And how do you balance the goal of promoting, assembly AI with actually creating really authentic, useful pieces of content?

Dylan: I think that we try to take the position of like, Hey, if we're just creating really good content, for the developer community about interesting things, it shouldn't even matter if it's about assembly or if it's talking about assembly, if it just lives on our website, then, hopefully because of that, developers will be exposed to assembly and learn about us, you know, and the way I kind of think about it is like, some of the best news outlets that I. Follow or read, I'm a fan of the economist, for example, and I think the economist's brand is so popular because they have good content. , not because they're like logo or their distribution strategy. It's that could be totally off here. Cause I don't know anything about their go to market strategy, but I am a fan of the economist because their content is. So I recommend it to friends and I, follow it as a news outlet because they've kind of earned my trust as, Hey, this is good content and so I think we think similarly where our goal is, Hey, let's just put out. Really interesting content for the developer community, not even with the goal of trying to get people, to sign up to our API, just, with the goal of putting out good content for the developer community.

That's educational that is genuine and authentic. And our hope is that, as a result of doing that, developers will come to know assembly as a, trusted. Life developer tool because they see that we're putting out high quality content and that we're competent and that we're, a good team of engineers.

And as a result, we'll become interested in our, product and service. And I think there's a lot that goes into that. Like for us, we know that our, like there's a certain type of developer we target. Like for example, front end developers are probably not very interested in. API. so if we were to put out content about, popular new front end libraries, even if that, content did really well, those developers might not be that interested in assembly as a API or company, because it's what we do is very D. From front end development. So what we tend to focus on are developers or product people that are interested in ML or data science, and that are professional software developers. And so you'll see like a lot of the content we put out is around machine learning models. So like when Dolly two came. Which is a really popular model from open AI, for text to image generation, we put out a deep dive video and article on like how that model actually works under the hood. And we're actually open sourcing pretty soon, a version of, Google's imagine model that, folks can just run and train and, , we do that with, a specific developer in mind, but for us, it's just about, Hey, we wanna try to put out. Content that is, genuinely like good content that we would wanna read ourselves that , if we're scrolling through Reddit or hacker news, we'd wanna read. And if we're doing that for the right type of developer, then hopefully they click around to our homepage and they're like, Hey, this company looks cool. This is interesting. And they remember it or they talk to their friends about it and then could be months or years later. That person signs up or a portion of those people sign up and use the API or recommend it for a project internally at their company and we have a lot of examples of this already, but that's at a very high level. How we, think about it.

Jack: That's really interesting. So you're very fixed on who you are creating content for, but within that, it could be very kind of whimsical, like fun, interesting piece of content, as long as it's targeted to your audience.

Dylan: Yeah, exactly. For example, like Stripe is such a big brand in the developer community now and, where they can create content about their products or just about Stripe, how they do things. And that is very popular because people care about Stripe, the hacker news community. is like very passionate about Stripe.

And there's a few companies that, are lucky to be in that type of position. But we're very conscious that where we are today, like people don't really care about assembly. You know, like we're a new company, we care about it, and we're trying to grow community of developers that care about assembly.

But if we're honest ourselves, we're so early that like, if all of our content's about assembly or how to do X with assembly that just won't resonate with people because I, as a developer, wouldn't care about that. If I saw something about that on Reddit or hacker news, I wouldn't really care, or what I would be interested about as a developer, Yeah, how a new neural network architecture works how to build like a, tutorial for how to build a speech recognition model and pie torch or something. , and as a result, then I get exposed to the assembly brand and company, and that's how we try to keep people. Engaged. So I think that's where, in my opinion, some developer tool companies, get it wrong. All their contents about themselves. And unless you're like a strike, I don't think that really works. And there's not even very popular developer tool companies. It's like very rare to be a Stripe where you could just put out content about yourselves and that goes viral. It. Tesla Stripe. There's like not many companies like that. It's pretty rare.

So we, we don't tend to even like to write about ourselves that much. Like if you go to our YouTube channel, we don't really, even say who we are, it's. Go right into the content. And it's about, , like one of our recent videos was around, like how to deploy ML models with fast API and Docker. And we, put out like a pie torch crash course, and we don't have like a two minute intro about assembly in the beginning. We just put out good content and hope that, will expose. Us as a company, two developers. And, if we keep doing that, then we'll build that, awareness with developers and hopefully in the future, they have a need and they come back and use our API.

Jack: One of the challenges that I can see that I'm wondering about is for instance, like your do two content, it seems like it's maybe not quite like the research paper level, but it's like quite close to that kind of level in terms of complexity and ha there can't be too many people.

In the world that are capable of like writing this kind of content, how are you able to produce that? I'm sure you could write it, but then also I imagine you've got a lot of other things as the founder that you need to be doing.

Dylan: It's hard. It's yeah, no, it's definitely hard. And something that we're, we're still trying to figure out how we scale that, if you're a really good ML person, you. Probably don't like the amount of people that are really skilled ML or deep learning folks that, wanna be in what basically is like a marketing role, is rare.

It's something that we're actively trying to figure out how we scale up. But yeah, I mean, it has to be really good content because otherwise, , no one's gonna care is, is what we are really honest about with ourselves. It's very much like a quality over quantity type thing. So I do think though, there are folks out there that there there's a lot of content creators, in the ML space, especially now with YouTube and TikTok. That we're in like that era, there are a decent amount of ML, [00:11:00] influencers, so to speak. And so, we found that as a good area of, opportunity to go to find folks. But it is a challenge for sure. It is a challenge for sure. I wish I had a answer or a, solution, but we're still trying to figure it.

Jack: Yeah, that's still really, really interesting. Kind of going slightly different topic, but I saw you went through, YC. What was it like at the beginning and getting your first users love to hear a little bit about that?

Dylan: We're an AI company and so, We moved a lot slower in the beginning because we had to train models and that took weeks and you'd get feedback and then you'd have to go train a new model and that would take weeks and YC is only 12 weeks. So, we're more like a hardware company where our ability to iterate was a lot slower because we had to go train these compute intense AI models. And iterate on them while we were looking for product market fit. Whereas, uh, other companies that, were just building, you know, crowd apps, right. It's a lot easier to go get feedback iterate over the weekend, go back to users with an updated UI or feature set. So for us, we launched on hacker news. and that brought in a lot of developers that were interested in the API. we were fortunate to be a member of YC. So, you know, there's a ton of YC companies and there's like internal launch forums that you can, use to launch to, YC alumni basically. And so we did. And then it took a while. But we just kind of kept at it and kept building that word of mouth. And we're still doing that today, you know? So like, we still have a ton of, like a long way to go, but in, in the early days it was really word of mouth, hacker news, product hunt. And then just trying to be scrappy and, find ways to get in front of folks. And a lot of like outreach to potential, startups that could use our API, getting involved with hackathon. So it was a combination of a lot of different things, but none of it worked overnight. It took, I mean, we started the company. Five years ago, like I was a single founder, started the company pretty much, right when we got into YC in the summer of 2017. And it really was only like, you know, in the past two years that stuff's really started to, take off. And within those two years in the last like year in particular where things have really started to take off at a higher pace, like we've raised. Now it's like 65 million for the adventure funding for the company.

But you know, like 90% of that funding has been in the last six months. and we've been around for five years. So it just takes sometimes a long time. And I think that's one thing that, . Folks have to keep in mind that it can just take a while. You have to be honest with all right. Are, are we really just hitting a dead end and we need to like, pivot or, maybe this, like, isn't a great idea. Or do we have a lot of conviction that there's product market fit for what we're doing? And we just have to keep after it. It's a really tough question. but for us it did take a while.

Jack: That's amazing. I have a lot of respect for solo founders, especially going through YC must have been very hardcore.

Dylan: was not fun. not fun. Yeah. It kind of happened by accident. Like I just applied to YC almost as like a, as like an exercise, just to try to articulate what we were doing. And then, and submitted the application late. I was actually like in Europe at the time and got an email that they wanted to interview me for YC.

So I had to fly back from Europe, do the interview and then YC started like the week later. So I was very unprepared and, I had like, basically just started working on the company. I quit my job a couple months prior. So it was like, all right, we're doing this thing. , it was pretty intense. I would not recommend that to other

Jack: But you did it and here you are. So.

Dylan: Yeah,yeah, yeah, exactly.

Jack: Dylan, thanks so much for joining us. I think that's all we've got time for. Where can people learn more about you and about assembly AI?

Dylan: The best place is really on Twitter. So if you just look us up on Twitter assembly, AI. You can also go to our website assembly, ai.com. That's really the best place to see what we're up to and see some of the developer content we're putting out.

Jack: Amazing. Thanks so much. That was so interesting.

I know you're probably very busy, so really appreciate your time. Thank you so much.

Dylan: You too. Thanks for having me on.

Jack: Thank you for listening. We'll be back soon.