AI School · Level 4 · Lesson 4

The stack: GitHub, Vercel, Supabase and Claude Code

Léelo en español →

To have a site of your own — your pharmacy's, an internal tool, a dashboard with your own sales — you need four pieces, and only one of them is AI. It is worth understanding what each one does before touching any of them, because most of the mess comes from asking one to do another one's job.

PieceWhat it doesWithout it…
GitHub Holds the code and all of its history. Every change is dated and can be undone. A change that breaks something cannot be rolled back. It is the piece people skip most and the one that costs most.
Vercel Takes what is in GitHub and serves it on the internet, with its domain and its certificate. You have the code and nobody sees it.
Supabase The database and the user accounts. What has to be kept and who is allowed to read it. The site works but remembers nothing: not an order, not a customer, not an account.
Claude Code
or Codex
Writes the code, inside your own project. You write it, or you commission it.
You do not need to know how to program. You need to know what to ask for and how to check it. That is what changes: the work is no longer writing, it is deciding and reviewing. And three decisions are never delegated — what data you keep, who can read it, and what the page claims. None of the three is a technical question.

Why this order

GitHub first, always, even if the project is two files. With an agent writing, the question is not whether it will break something: it is whether, on the day it does, you can go back. Without history the answer is no — and there is no warning, because an overwritten file looks exactly like one that never existed.

Then Vercel, connected to GitHub. From that point on every change you push publishes itself, within a minute. That is exactly the good part and exactly the trap: there is no "right, now publish" button. What you pushed is live.

Supabase comes in when there is something to remember. And Claude Code or Codex as soon as the repository exists — not before, because its context is the project.

What it costs

All four have a free tier, and for a pharmacy site with normal traffic the free tier is enough. I am deliberately not putting the actual figures here: they change every few months and a lesson with stale numbers is worse than one with none. What does not change is the shape:

And a warning before you start. If patient or staff data is going to live on that site — a record, a payslip, a phone number — stop reading and go back to the first lesson. Setting up the infrastructure is the easy part; deciding what is allowed to live inside it is the part with consequences, and it does not get fixed afterwards.

← Back to the school Next: coding with Claude Code or Codex →