I think I found a problem worth building for, but I keep talking myself out of it

This is me thinking out loud, so bear with me.

A couple weeks ago I shipped a change I did not really understand. The AI wrote it, I skimmed it, tests passed, it merged. Then it broke, and opening the file I felt this out of body thing of not recognizing my own work. My name was on the commit. No memory of the decision.

I laughed it off. Then I noticed I do this constantly now.

Here is the shape of it. For years the slow part of building was writing code, so we threw everything at that. Autocomplete, Copilot, agents that write more in a minute than I read in ten. Writing is basically free now. But reading was always the real bottleneck, and nobody made that faster. So there is a growing pile of code that works, passes every test, and is understood by no one. People are calling it comprehension debt. It is technical debt that does not even look messy. You find out it is there when something breaks and the whole team reads the source together for the first time.

The reason I think there might be a business here: nobody measures it. We track how fast we ship a dozen ways. None of it measures whether anyone still understands what got shipped.

And here is where I talk myself out of it. Maybe it is not painful enough to pay for. Maybe it feels like surveillance the moment you productize it. Maybe the answer is just "prompt better." Maybe agents get so good nobody cares whether a human understands the code at all.

Then I open my editor, look at code I supposedly wrote yesterday, feel nothing, and think: no, this is real.

If you build with AI in the loop, does this itch you too, or have I lost the plot?