How do you check the quality of code that AI wrote?
A short learning video. A renovation company owner and a software engineer check her booking app, built with AI.
What this video shows
Nora had her booking app built with an AI tool in two weeks. It works in the demo. Sam shows her that you do not need to read code to judge it: you ask for proof. A demo shows the screens; the risks sit underneath. Then one real bug: two customers book the same slot at the same moment, and both get a yes. Sam walks through six checks (tests that can fail, no keys in the code, customer data locked, a second reviewer, a real-customer run on the live app, and someone watching the errors after launch), fixes the bug and adds a test so it never comes back. The full list of twelve questions is in the article.
If you want these checks run on your app, or an app built with them from day one, see how I can help or tell me about it.
Transcript
0:03 Nora: How do you check the quality of code that AI wrote?
0:08 Sam: You do not need to read the code. You ask for proof.
Sam: Tests that run. No keys in the code. Customer data locked. A second reviewer. And a real user, on the live app.
0:20 Nora: An AI tool built my booking app in two weeks. Customers book a site visit on their phone.
Nora: It works. But is it good?
0:28 Sam: A demo shows the screens. The risks sit underneath.
Sam: Tests. Rules about who sees which data. Passwords and keys. Outside code. Error logs.
Sam: AI code now runs almost every time. But in a big study this year, only just over half of it passed the security tests.
0:48 Sam: Look. Two customers book the same slot, at the same moment.
Nora: And both of them got a yes!
0:54 Sam: Check one: tests. Small programs that try the app again and again, after every change.
Sam: And break something on purpose. A good test turns red.
1:05 Sam: Check two: passwords and keys. They must never sit in the code. This one does.
Nora: So anyone who sees the code can use it?
Sam: Yes. So it moves to a locked setting.
1:16 Sam: Check three: your customers’ data. Can one customer ever see another customer’s booking?
Sam: This is the most common hole in apps built with AI. One rule missing, and the whole database is open.
1:30 Sam: Check four: a second reviewer. Not the AI that wrote it. A fresh one, with no memory of writing it.
Sam: And then a person.
1:39 Sam: Check five: use it like a real customer. On the live app, on a phone.
Nora: I book. I change the time. I cancel.
1:47 Sam: In my team, every change goes down the same line.
Sam: The builder shows proof. Machines run the tests and scans. A different agent reviews. A tester uses the live app. And a person says yes.
2:02 Sam: And here is the fix for the double booking. The slot locks, and a new test makes sure it never comes back.
2:09 Sam: Check six: after launch, someone watches the errors. And the backups get a test restore.
Sam: If something breaks at night, you know before your customers do.
2:20 Nora: Now I don’t just hope my app works. I have proof.
2:24 Sam: Good code made by AI is code you can prove. Tested, locked, and checked by someone else.
Have an app idea, a stuck app, or a daily job you want an AI agent to do?
Write three lines. I reply within one working day with a plain answer.
