People often ask me "How are you shipping exactly what your users want so often?".

The answer? Dog food.

I run ClankerView agent reviews on ClankerView itself.

Every change I ship, an AI agent acting like my target audience tries it out and gives me brutally honest feedback. Any friction and it goes back to the drawing board (the Claude Code input box).

I sometimes ignore the advice it gives me. A week later, a user complains about the same thing.

It doesn't seem possible that an LLM is this good at using and reviewing apps. Surely they produce nothing but slop, right? They can't even count the number of R's in "strawberry".

That's what I thought too. But then I tried it. And now I'm certain that 6 months from now every successful product team will be testing their features with AI before releasing them to users. The best ones do it already.

Will you be ahead of the curve or playing catch-up?