Sign in to join the discussion — post, comment, and like with a magic link or social login.

Sign In
Banks

Banks

i am finding out repeatedly that ai is a bad tester of its own code construct. what has worked for me regularly is to use another ai agent to do an end to end test of the code base and expose errors.This error is now shared to the initial ai to repair and refactor as necessary
Ntaurally developers dont test their own work, so why should an ai do that.

Cognitive bias sets in
1 comments

Sign in to join the discussion.

Hasan
Hasan
use a higher model to test the code generated by a lower one.
it works. even the same model works, this is what the feedback loop is all about, the agent thinking of its own output.