An agent reviewing its own output has the same blind spots, the same training-data gaps, and a strong bias toward declaring its work "good". But even having a different model review the code is flawed and not good enough, especially for those whose code can't be wrong, like in regulated industries.
I remember having to spend much more time on the existing search engines trying to find what I was looking for, even though there were much less sites.
Since Google, sifting through all the data has become easier, although I worry about how it's also become influenced by our search history and navigational habits.
This in turn, lands up limiting what we are actually exposed to and might reduce the probability of us discovery new things outside of our comfort zone.
Do we go for something like DuckDuckGo, that isn't influenced by our history, or do we use google to present us with what we want?