Authentic Developer Education with Dylan Fox from AssemblyAI
AssemblyAI's YouTube channel doesn't open with a company intro — it just starts the tutorial. Dylan Fox's reasoning: outside of a handful of companies like Stripe, content that's actually about your own product almost never resonates, no matter how good your product is.
Dylan Fox is founder and CEO of AssemblyAI, a speech-to-text and audio understanding API used by product teams to add transcription, summarization, and content moderation to their own products.
We get into why he wants developers to know AssemblyAI exists long before they ever have a need for it, why the company's deep technical content, like a breakdown of how DALL-E 2 works, deliberately avoids mentioning the product at all, and what it was like being a solo, unprepared founder rushing back from Europe for a YC interview.
The Q&A below is experimental — extracted from the episode transcript with AI. If you spot a mistake, let us know at jack@scalingdevtools.com.
What does AssemblyAI actually do?
We build an API that developers and product teams send audio and video files to, which then runs through our AI models to do things like automatic transcription, content moderation, topic detection, and summarization. Product teams use our APIs to build features on top of audio and video data without needing their own AI models.
What's the actual thinking behind AssemblyAI's content strategy?
The goal is that developers know about AssemblyAI long before they ever have an actual need for us — we want to be the tool that comes to mind when someone hits a problem we could solve, whether that's a week or years in the future. Twilio is a great example of a company that nailed this early on — I actually applied to be a developer evangelist there years ago and took a different job instead. As a developer myself, I discover cool tools constantly through content or word of mouth, and even without an immediate need, something will resonate and get bookmarked mentally, only to get used months or years later when the need actually shows up. There's definitely content worth creating for high-intent searchers actively looking for a transcription API, but the bigger goal, especially in a category this new, where a lot of people don't even realize this kind of API exists, is pure awareness — inspiring people to realize this is something they can build with at all.
How do you balance wanting to promote AssemblyAI with creating genuinely useful, authentic content?
Our position is that if we're creating genuinely good content for the developer community, it shouldn't matter whether it's directly about AssemblyAI at all — just having it live on our site is enough exposure. I think of it like The Economist, a publication I genuinely like — its brand is strong because the content itself is good, not because of some clever distribution trick. I recommend it to friends because it's earned my trust through quality. We aim for the same thing: put out genuinely educational, authentic content for developers, without the explicit goal of getting anyone to sign up, and trust that developers will come to see AssemblyAI as competent and trustworthy as a byproduct of consistently doing that well.
How do you make sure that content is actually reaching the right developers, given how broad 'developer content' could be?
We're very deliberate about who we're targeting — front-end developers, for instance, are probably not very interested in our API, so content about popular front-end libraries, even if it performed well, wouldn't bring us the right audience. We focus on developers and product people interested in ML and data science. When DALL-E 2 came out, we put out a deep-dive video and article on how the model actually works under the hood, and we're about to open-source a version of Google's Imagen model people can run and train themselves. The bar is simple: would we ourselves want to read this while scrolling Reddit or Hacker News? If we hit that bar for the right audience, some portion of those people will eventually click through, remember us, or mention us to a colleague — sometimes months or years before they actually sign up.
Is there a risk in making content that's genuinely fun or whimsical, as long as it's aimed at the right audience?
Not at all, as long as it's targeted correctly. Stripe is a great example of a company big enough in the developer community that content directly about Stripe itself performs well — Hacker News is genuinely passionate about them specifically. But we're honest with ourselves that at our stage, nobody inherently cares about AssemblyAI yet. If all our content were about AssemblyAI or 'how to do X with AssemblyAI,' it just wouldn't resonate — I wouldn't personally care about that if I saw it on Reddit. What I would care about is how a new neural network architecture works, or a PyTorch tutorial for building a speech recognition model — and through that, I'd naturally get exposed to the AssemblyAI brand. I think that's where a lot of dev tool companies get it wrong: making all their content about themselves, which only really works if you're a company as established as Stripe or Tesla, and those cases are genuinely rare. Our YouTube channel doesn't even open with an intro about who we are — we go straight into the content, whether that's deploying ML models with FastAPI and Docker or a PyTorch crash course, and let exposure to the brand happen as a side effect.
Producing content close to research-paper depth, like the DALL-E 2 breakdown, seems genuinely hard to scale. How do you actually manage that as a founder with a hundred other things to do?
It's genuinely hard, and something we're still figuring out how to scale. There just aren't that many highly skilled ML and deep learning people who also want to work in what's effectively a content or marketing-adjacent role — that's a real bottleneck. We're honest with ourselves that the content has to be genuinely excellent, or nobody will care; it's a quality-over-quantity approach by necessity. There is a real and growing community of ML content creators and influencers on YouTube and TikTok now, and we've found that a good pool of people to look toward, but it remains a real, unsolved challenge for us.
What was it actually like going through YC and getting your first users, given AssemblyAI is fundamentally an AI company rather than a typical CRUD app?
We moved much slower than a typical YC company because training models takes weeks, and you'd get feedback, retrain, and wait weeks again — much closer to a hardware company's iteration cycle than a typical software startup that can ship a UI update over a weekend, especially within YC's tight 12-week batch. We launched on Hacker News, which brought in a lot of genuinely interested developers, and used YC's internal alumni launch channels too. But none of it worked overnight — it took real, sustained word of mouth, Product Hunt, scrappy outreach to startups that could plausibly use our API, and getting involved with hackathons.
How long did it actually take before things really started working?
We started the company five years ago, and it's really only been in the last two years, especially the last year, that things picked up meaningfully. We've now raised $65 million in venture funding, but around 90% of that came in the last six months alone, despite the company being five years old. It genuinely can take a long time, and you have to be honest with yourself along the way about whether you've genuinely hit a dead end and need to pivot, or whether you have real conviction in product-market fit and just need to keep pushing through — that's a genuinely hard question to answer honestly in the moment.
You went through YC as a solo founder. What was that experience actually like?
It kind of happened by accident — I applied almost as an exercise, just to force myself to articulate what we were building, and submitted the application late. I was actually in Europe at the time when I got the email inviting me to interview, so I had to fly back for it, and the batch started the following week. I was very unprepared, having quit my job only a couple months earlier, and it was genuinely intense — not something I'd necessarily recommend to others in the same rushed way.