DEV Community

Anirudh
Anirudh

Posted on

How much of your AI-written code do you actually understand? (2-min survey)

I'm a student doing a small research project on how people deal with code their AI tools write.

Something I keep seeing around me: Claude Code or Cursor writes a function, it works, it ships, and a week later nobody can explain it. Some people call this "comprehension debt."

I'm trying to figure out if that's a real problem for most people or not. Also curious what everyone does while the agent is running.

If you use AI coding tools, 2 minutes would help a lot: https://forms.gle/h32W8AyLHaSaFF8u5

No signup, nothing to buy. I'll post a summary here once enough people respond. Happy to just discuss in the comments too.

Top comments (1)

Collapse
 
jes_p_d4f61f7aa2ec405a418 profile image
Dev_PJS •

I work with agents all day and my honest answer is: I understand the code at the level of the record, not the line. I stopped pretending I'd re-read everything. What I keep instead is a check I can actually run — the run ends with what was changed, what was verified, and what decision got made where nobody specified one. If I can't explain a function, that's a signal it needs a smaller scope or a test, not that I need to study it. Comprehension debt is real, but for me the interest payment isn't confusion — it's the first time you have to change the thing and discover nobody, human or agent, can say why it's shaped that way.