Using Factory.ai
I spent most of yesterday building a pretty simple website called Prompts
https://prompts.iwonderdesigns.com
The idea came from my wife. She and a few photographer friends run a small creative exercise together. Each week, one of them sends out three words, and they take turns doing this week by week. Everyone then spends the week capturing photos inspired by those words, and at the end they compare results. Same prompt, totally different interpretations. It is a simple structure, but it creates a surprising amount of focus and creative energy.
As soon as she explained it, I thought this would be fun to open up to a wider audience. It also felt like a perfect excuse to lean hard on an AI coding tool and see how far I could push it.
So I used Factory.ai.
If you have not heard of it, Factory is a coding focused LLM that I have had really good luck with. It tends to produce cleaner, more intentional code than most “generate me an app” tools I have tried, especially when you give it enough context to understand why you are building something, not just what you want.
How I prompted it
In my initial prompt, I described:
- What the app should do along with a general look and feel
- The creative motivation behind it
- The kinds of tradeoffs I cared about, like simplicity, extensibility, and not over engineering
Factory responded by asking around 15 clarifying questions, which I actually appreciated. Once I answered those, it generated a full working site.
What Factory handled
Out of the box, it gave me:
- A clean, sensible database schema using Prisma
- Image storage wired up to Cloudflare R2
- NextAuth with Google sign in already configured
It also did a bunch of boring but important setup that tools often skip:
- Linting rules
- Basic unit test scaffolding
- A reasonable project structure
What I still did manually
There were a few things I expected to do myself:
- Create a local Postgres database and a production one using Supabase’s free tier
- Set up a Google OAuth client
- Create an R2 bucket and configure public access
- Spin up a Vercel project for hosting
Extra features I added afterward
Once the foundation was in place, I layered on some UI and feature improvements:
- User profile pages
- Submission history
- A better lightbox experience
- Support for image only, text only, or combined submissions
- A much nicer text editor using TipTap
- Proper documentation, mostly for future me
The entire project, from idea to live site, cost $12.
- $12.00 for Factory.ai
- $0.00 for Supabase
- $0.00 for Cloudflare R2
- $0.00 for Vercel
- $0.00 for Google OAuth
Total ongoing cost: effectively nothing
Final thoughts
This experience reinforced something I have been feeling for a while. For greenfield projects, prototypes, and ideas you want to get out of your head and into the world quickly, tools like this dramatically flatten the slope.
Factory handled foundational work that usually burns a lot of setup time, and it did it thoughtfully. I still made architectural decisions, added features, and shaped the product. The difference was how quickly I got from idea to something real.
Honestly, this was the most fun I have had building something in a while.