Your code has versions. Your data has nothing.

The worst documented losses in AI building aren't code; code has versions in half a dozen places. They're data: the user accounts, orders, and uploads in your database, which one confident "fix" from an AI tool can destroy with no history and no way back. Your app's code and its data live completely separate lives, and most people only protect the first one.

What lives where

• Code: the app itself. Recoverable from your AI tool's history, GitHub sync, or your project's home. Losing it is painful, not fatal.
• Data: everything your users and you put into the app, almost always in Supabase (or similar). It has exactly as many copies as you've made. Usually: one.

Back up your Supabase data by hand, today

1. Open your Supabase project → Database → Backups. Paid plans include daily backups; check yours actually shows recent ones rather than assuming.
2. On any plan, take a manual export now: Database → and use the export options, or ask your AI: "Write me a script that exports every table in my Supabase database to CSV files using my connection details, so I have a local copy."
3. If users upload files, those live in Storage, separately from tables; download those buckets too.
4. Put the export somewhere that isn't the same laptop the project lives on.

The rule for risky moments

Before letting any AI restructure your database ("migrate", "clean up", "fix the schema"), take a fresh backup first, every time. The disasters in this category are all one skipped backup away from being non-events.

Automating it

Manual backups work until the week you forget. Plyo backs up your project's database nightly once connected: encrypted snapshots you can download any time, with a "what's inside" summary in plain English, and a backup-before-risky-changes tool your AI calls automatically. Restoring takes a fresh snapshot of the current state first, so even a restore can be undone.