Six things that will be wrong with it
None of these mean you did it badly. They are what an AI-built prototype looks like at the point it starts working, and knowing them by name is the difference between a build somebody can pick up and one nobody can. Every one of them belongs in your what's-left field rather than in a bug report.
- It works on your machine
- Your laptop has your environment, your keys, your one test account. That is not evidence it works.
- Secrets in the browser
- If an API key is in code the browser downloads, it is public. This is the most common serious mistake in AI-built apps, and it is invisible until it is expensive.
- No real authentication
- A login screen is not authentication. Most first builds have a hardcoded user and a screen that looks like a gate.
- The database lets anyone read anything
- Managed databases often start permissive so you can move fast. That default is not meant to survive contact with strangers.
- Numbers that are confidently wrong
- A total that sums the wrong rows still renders beautifully. Check one figure by hand against the data before you trust any of them.
- Cost that scales badly
- One model call per item is fine for forty items and ruinous for forty thousand. Work out what it would cost at real volume.
When you have something
Post it. Screens, one line on what it does, and what is still missing — that last part is the whole point, and it is what lets somebody pick your build up instead of starting again.