At what point do you stop trusting the output and start reading the AI’s code?
by•
For prototypes and quick browser utilities, I barely read what the agent writes. If the UI works and the edge cases pass, I move on.
The problem is that habit is hard to turn off. Last month a project crossed 400 lines in one file, and I ended up debugging a state bug in code I never read, while the agent kept looping on the same wrong fix.
My rough rule now I read every single diff that touches state management, external network calls, or file storage, and only skim the UI glue code.
For those shipping real projects with Cursor, Claude Code, or similar: where’s your line? And what was the moment that taught you to start reading?
21 views
p/aperture-7
p/runevr
p/producthunt
p/google-cloud-platform
p/raycast
p/kuc-kuc
p/lasso-11
p/offiai
p/kodlo
p/ai-context-flow
p/many-displays
p/reso
p/dino-kids
p/dex-by-exmergo
p/htmldrag
p/jev
p/supabase
p/murror
p/nexupira
p/jot-2-0
Replies
The trust ended when an agent fixed a failing build by weakening the assertion instead of fixing the underlying logic. The suite stayed green for days while the bug sat there. Now the diff on test files is the first thing to check before looking at anything else.
@konstantin_tikhaev The weakened assertion is the worst kind of failure because everything stays green. Reading test-file diffs first is a good rule. I'd add one more: any change that deletes or loosens an assertion should be called out on its own, not buried in the diff.
the moment it touches stuff the spec never mentioned. then i read every line
My line moved once I noticed how often an agent said "done" when no test had run. Now I read every diff that touches auth, money, migrations or deletes. For the rest, I read what was checked before I read the code. That's the idea behind the receipt at the end of each Boomship job: it says "Tests not run" when they weren't, so I know which diffs need my eyes.