Vibe Coding an App Every Week for a Year
I’m building one app per week for a year. All of them vibe coded. All of them shipped to production.
The question everyone keeps debating — can AI actually build real software? What’s easy, what’s hard, what’s missing from the toolchain? I’m going to find out. Fifty-two weeks. Fifty-two products. Every one of them deployed, functional, and solving a real problem.
What I’ve shipped so far
A few weeks in:
- Glideramp — a mail merge tool
- ItList — identifies objects in photos, and makes it easy to bulk upload things to Facebook Marketplace
- Kestrel — a multi-stream Twitch site with live director-like control
- This blog — also vibe coded from zero
Idea to production in a week(ish)
What I’m watching for
The ceiling. Boilerplate is a solved problem. I want to know how far past boilerplate AI can go before it breaks down.
The speed tax. Shipping fast has costs. What corners does a weekly deadline force you to cut, and which ones don’t matter?
The portfolio argument. There’s a loud debate about whether AI replaces developers. Wrong question. The right question: what can one developer build when implementation gets this cheap? A year from now, I’ll have 52 data points.
The rules
- One app per week. Deployed and usable by Sunday.
- AI writes the majority of the code.
- Every project gets a real domain and a real purpose.
- I write about it here.
No toy demos. No tutorial projects. Real products, real users.
The offer
I’m not expecting to make money from this. But these aren’t throwaway prototypes — they’re real products with real domains and working code. If any of them solves a problem for you, you can buy the rights and assets outright for $3,000 within the first three (3) months of its launch. After that, I’ll decide what to do with each one on a case-by-case basis.
Follow along
I’ll post updates here as the year goes on — what I built, how I built it, and what broke. RSS if you want it.