The Renaissance Technologist

I was a Product Manager at Intuit for more than two years. 1 year before ChatGPT launched, and 1 year after. Then, I randomly started an AI consulting business that turned me into a Forward Deployed Engineer. Recently, I started a podcast, and the more episodes I record, the more I keep hearing what I've experienced firsthand: that traditional roles on a software development team are converging into a generalist "Builder" role (someone needs to come up with a cooler name and I think that should be me. Let's call these people "renaissance technologists"). I'll write a more philosophical post on this soon, but first, here's a story of how I vibe-coded an app to 38,000 users in a week (this is a stream of consciousness talking to my computer, so bear with me on the writing style. I didn’t get a chance to polish it. I need every minute I can get building shit right now).

When my business partner and I first started our consulting business, we were helping retailers like Louis Vuitton, Dior, etc do virtual try-ons by fine-tuning models and building workflows before the Nano Banana and ChatGPT image models came out. Since we were working with 3D modeling and image diffusion models pretty heavily, we had a lot of experience with it.

As a side project, I vibe coded an app that could take a 2D image and convert it into a 3D model. We started by finding state-of-the-art open source image-to-3D models, like Trellis by Microsoft and Hunyuan 3D by Tencent. I figured out how to run those models serverlessly on serverless GPUs from Modal.com. I built a quick landing page optimized for SEO (we basically just copied a Framer template that we really liked). Then I started building the actual application.

I knew I wanted to use Supabase for auth, and I knew we were using Modal for the model deployment (I got that from a hint from a friend) and I just started vibe coding. I did some research on other image-to-3D model websites, got a general vibe for how they were architected, and then just started building.

I knew I wanted somewhere for users to upload multiple angles of an object they wanted to turn into a 3D model. And I knew I wanted users to be able to test multiple models at the same time for a specific object, just because Trellis works well on some objects whereas Hunyuan works better on other objects (like jewelry), at least at that time.

A lot of the work I ended up doing was debugging because I didn't have a ton of experience with Docker. At the time, when we were prompting the models to help me build Docker files, that's where I was getting a lot of errors. But after a few days of prompting the model and learning a little more about Docker, I got it to work pretty well.

After getting some startup credits from Modal.com, we were ready to go. The cost really wasn't a huge factor, especially because we were running it serverlessly. It wasn't like we were just paying for the server to be running all the time. We just subsidized the rest of the server costs with the revenue of our consulting business and kept prices low for users to make it easy for users to try us out.

Then we just started posting about it on Reddit, Product Hunt, and whatever. We ended up getting 3,000 users within a month, and then it just kept going up from there because people were posting about it. It went viral on the Chinese side of Twitter, and people were posting about it there. We ended up getting 17,000 users within 3 months and now we’re at 38,000. And it's still running to this day.

We would message users every once in a while, and we would get a lot of feedback from the support emails about what people were using it for. We got a lot of interest from interior designers who were uploading images of objects they wanted to put inside 3D models of rooms. We got a lot of interest from people who were trying to create jewelry, so they would upload images of jewelry that inspired them and they wanted to be able to play around with that. And so we let them do that. That gave me ideas to add new features.

We also got a lot of interest from animators who wanted to put in character reels (basically a 2D strip of a character from multiple angles), and then turn that into a 3D model. We realized that we should probably turn that into a separate section of the application so people knew that was functionality we could handle. So we implemented that.

And so it was all very ad hoc in terms of how we built the product, but I think that's how product is going to be now, especially at startups. We get feedback from users, and we would just vibe code a new feature, test it, get some feedback, and iterate. Maybe you'll have to have a more formal process at bigger companies, but at startups, that's basically what we did and I’m pretty sure that’s how it’s going to be done from now on (it’s no coincidence that all the newest YC companies say that 90%+ of their code is generated by AI).

This is all possible. Yes, I know, I'm technical. I have a CS background, and I’m more technical than your average person. But this is all doable. The models are only getting better by the day, and as long as you're willing to learn, you can do this now, which is pretty awesome. Previously I would get hung up on debugging, but models are so good at debugging now that that's not really an issue anymore.

PMs and managers didn’t code because it wasn’t a high-leverage activity before, but at least for now, it is. The combination of my PM skills plus actually developing something in real time, getting feedback, and iterating on that to build a product is becoming a VERY valuable skill. This is how products are going to be built going forward, and I’m pretty excited to see it and watch the transition happen in real time.

Right now the only people doing this are AI-Native, the first adopters: YC startups, generalists like me. But pretty soon this is going to be everyone. This is just the way that product is going to be done.

More and more, as I do more podcasts talking to engineering leaders, product leaders, and start-up founders, I’m getting more and more convinced that there is not going to be a delineation between a software engineer and a product person. You’re going to need basically just one person who can do everything. And if you can’t get technical enough to do this, you’re not going to be useful.