A screen is not a product

A good-looking screen is not a product. I learned that while building my first AI product, and I learned it slowly, because screens are the easiest thing to judge. You open them, they look right, and it feels like progress.

The loop is the product

A product is what happens between a person arriving and a person getting the result they came for. If someone signs up, sets up their account, starts their first task and gets stuck at any step, you have a demo. The demo can be beautiful. It is still a demo.

So I changed how I review work. I no longer ask whether a page looks done. I ask whether a new user can complete the whole loop without me standing next to them.

Fake output is worse than an honest error

The product depends on outside AI providers. When one is not configured, it is tempting to show sample output so the screen looks alive. I now think that is the wrong call. A fake result tells the user something false. An honest message that says the provider is not configured tells them exactly where things stand.

Fake output is worse than an honest “provider not configured” state.

What I check now

• Can a new user get from sign-up to a real result on their own?
• Does every failure say what failed and what to do next?
• Is anything on screen pretending to work when it does not?

I am the product owner on this, not the engineer. My job is to keep asking these questions until the answer is yes. I am still learning how to do that well, and I will keep writing about it as I go.