What I learned after letting AI build too much of my apps

I’ve been using AI pretty heavily while building my latest project, and one thing surprised me: the code can look completely fine while the architecture is already going in the wrong direction.

The worst example was multi user data. Everything worked perfectly during testing because I was basically the only user. It was only when thinking through multiple users that I realized how easily AI can make assumptions like “there’s only one current user” or forget ownership checks.

Since then, I’ve started giving AI more context before asking it to code:

• who can access the data

• where the data comes from

• what must never be changed

• how errors should be handled

• what happens when multiple users do the same thing

It feels less like “prompting for code” and more like giving a junior developer the context they’re missing.

For those using AI to build real products, what’s one mistake you learned to prevent early?