When you’re building a product, there are always more things you could build than you have time to build.

This is especially true when you’re building with AI. It’s easy to spend a week building something that has nothing to do with the problem you’re actually trying to solve.

I felt the same way about infrastructure.

HealthFirstAI needed a database, authentication, file storage, backups and security. I could build each of these myself, or I could use something that already did most of it.

So I picked Supabase.

The important part wasn’t that Supabase was the best possible infrastructure. It was that it was good enough.

PostgreSQL was a particularly good fit because hospital data has a lot of relationships. Patients have admissions. Admissions have encounters. Encounters have documents and clinical events. I didn’t want to spend time inventing a different way to represent something a relational database already handles well.

There is a temptation to build the infrastructure yourself because it feels more serious.

I think that’s usually a mistake.

If the infrastructure isn’t the problem you’re trying to solve, it should take up as little of your attention as possible.

Supabase lets me do that.

I’d rather spend my time figuring out how a doctor works than figuring out how to run a database.