In early 2025, I moved from creating code in ChatGPT to Replit, and after a few months I changed to a different setup.
Early experimentation
For early experimentation, I use v0. While I have many apps on the radar, I settled on v0 for this because I already use it to host some of my projects.
Apart from that, my frustrating early experiences are no longer happening. It’s a solid AI app builder platform now.
The option to choose from a rich template gallery and design systems is also a plus.
For longer-running projects
If I move a project from early experimentation or prototyping to working on it more long-term, I move over to Cursor.
While Cursor’s target audience in the beginning was more developer-focused, its recent updates put building with AI front and center.
Database
Supabase. No question about it. It comes with a generous free plan and is a solid option, not only for storing your data, but so much more. I recently had to add Google OAuth to a project, and it was all supported out of the box.
Workflow
As an AI product builder, not a dev, I move slow. I don’t work with 5 agents in parallel. I carefully prompt, I use plan mode in Cursor a lot. I review, I often prompt to get things explained in simpler terms. And I use Google and Claude a lot on the side.
Linear is the app I use for managing my projects. It’s connected to Cursor and GitHub (and supports many more services) and allows you to push work to your agents directly from tasks.
What am I building?
Internal apps, apps that we use with clients. Dashboards that show social media data or other analytics. Apps that speed up content creation, and little productivity helpers like browser extensions.
The biggest public project yet is LinCal, a calendar app for Linear. For this project, I onboarded a dev who now works with me on it.
Conclusion
The stack will keep shifting. What stays constant is the approach. LinCal didn’t happen because of the tools — it happened because of how I used them.
