A (long) welcome from First1000

This is a newsletter where I dive deep into my journey to get my first 1,000 customers for lazyweb.com. Lazyweb is a one-person, AI-native business. AI writes the code, designs the apps, builds internal tools, and runs all my data pipelines for building an Internet archive for mobile apps.

I have been at Duolingo since 2022, starting my journey on the growth retention team (think streak, notifications, etc.). My timing could not have been better; a few months into the job, we got connected to the folks at OpenAI who were working on a very secret project where Duolingo would be one of 5 partners (that project was GPT-4, and ChatGPT didn’t exist).

The most obvious use case was Duolingo notifications—our biggest growth lever—and, lucky me, I owned it. I ran probably the first GPT experiment in the world… at least that I know of, given how panicked OpenAI was when I told them that we wanted to use this in production (but wouldn’t tell anyone, since GPT-4 wouldn’t launch until 9 months later).


From there on, I moved to monetization, where I spent most of my time. Monetization was my calling. I worked on our New Year’s promo, owned the purchasing experience, and most of our subscription hooks. At some point, I owned surfaces my team was responsible for that generated 50–60% of the company’s revenue. It was a huge responsibility and quite a ride. I made the company probably something like $50–100m in incremental revenue and then went on to work on Ads.

Ads were interesting because they were our biggest bookings business that we didn’t have a team to own. They were really important for our margins, which became more and more important as we: 1) went past the ZIRP era and public investors cared more about bottom lines than growth; 2) AI costs started creeping up as we integrated AI into every part of our business, putting pressure on our bottom line.

Everyone else at the company, CEO included, hated ads because of the poor experience and brand misalignment of advertisers coming from programmatic auctions. It was the one part of the Duolingo user experience we had very little control over, and if you ever used Duolingo, you could tell how much we sweat the small details and cared about making users feel delighted and motivated using our app. Ads were the opposite of that.

I am currently working on ads and it has been a journey. I built a new modular ads system that helped us use AI to bring our characters to life. It was really fun working with animators, illustrators, and writers who could have worked at Pixar but, by some miracle, chose to work at Duolingo. We had fun; we created bits and developed our characters’ backstories through ads (my favorite was how we took our bear, which was barely visible, and made him into this creature that roasts humans for making dumb decisions)

Modular ads (“Mads”) made it so in 5 minutes you can generate an ad for Super Duolingo with our characters saying stuff and place them in different environments (see example). Before, it took us a quarter to produce an ad, and it was way less effective. That turned into one of our biggest monetization levers; we produced hundreds of ads and built machine learning systems to deliver the right ad to the right person with the right message at the right time. It was beautiful 😍.

From there, we hired an OG ads sales person who ran ads for Buzzfeed, built infrastructure to allow other companies like Universal and Marriot to use our characters to advertise on Duolingo. It took this new ad format, that helped us built our brand outside the app, and sold it directly to external partner. This new format help us attract the right partners and co-develop IP with them, beat every benchmark we could find on performance ads and gave Duolingo control over the thing we hated the most…ads. And of course, made us a ton of money.

It has been a ride, a beautiful one…

But throughout that journey, I saw Duolingo and the world around us change with AI. Since I saw the first GPT-4 demo before the world took notice of OpenAI, it was extremely obvious that this thing would change how we generate content and enable new capabilities to do things like Mads (with text-to-speech, script generation, and programmable animations). It was not obvious that it would change every aspect of what it means to build a product.

At Duolingo, we found the best use of AI was doing infeasible things that would have taken us years to do manually but that we have very strong expertise, taste, and opinions on (like generating new courses to teach every language from every language).

Personally, as a PM, I found it most helpful for helping me come up with new experiment ideas, doing deep research to understand best practices so I don’t have to fall through all the pitfalls when ramping up on new projects or domains—like moving from growth to monetization, or all of a sudden owning our third-party ads and being in charge of growing a business that is very different—or even optimizing our paywall, which is the highest-ROI thing you can do in a consumer subscription business if your goal is to increase revenue without impacting engagement or DAUs.

What I did overtime was I started auditing all the companies I looked up to, who were competitors, or had a very similar problem to us (e.g fitness apps) Overtime, that list ended up being around 200 companies.

I screenshotted their flows and put them into ChatGPT and asked it questions like “What is the best CTA for a paywall based on all the flows I injected into your memory?” Soon afterward, I started updating these flows (with the help of my awesome team) so we kept track of what experiments these companies ran and launched over the last year or half a year. Then the GPT got smarter; it would detect the differences and surface themes of changes and trends in the industry.

And I thought… oh man, what if I could do this for all the top 1,000 apps on the App Store? What if… what if I could just index all the flows for all the apps that matter, feed them into a GPT, and ask it any question about any product, industry, or domain?

Well, it was one of those problems that was infeasible and would have taken me years to do manually but that I have very strong expertise, taste, and opinions on. Sounds like exactly the kind of problem AI is primed to solve at every step of the journey.

That is how lazyweb.com was born.

First100 is a newsletter about my journey using AI exclusively to design, build, and market lazyweb.com to get its first 1,000 customers. It’s just me and a team of AIs, a lot of ideas, and an ever-evolving world with a tech breakthrough every other week.

If I can promise you anything, it’s that this will be a fun journey, and I hope it sparks a few ideas as I document my journey with absolutely no filter.

Join me for the ride 🚀, you can also find me on the X where I spend a ridiculous amount of time.