My boss has been using Claude Code to review MRs. He claims it's a quality add, perhaps it is, but I suspect it's more a shaming tool and he probably doesn't have time to properly review all his MRs. Whatever the actual intent, it worked, as more of us have been using the tool. I've used it a few times to review local changes in the ~1k LoC range. Some random musings and ramblings:
How much can you trust the dollar cost in /usage? I used up a third of my daily tokens(sorry, forgot to record actual amount) and it was ~$5. Yet, others at my work are looking at implementing as part of CI and cost(among other things) is a problem. $5 for even one true positive is a steal.
It's hit or miss, it will find genuine bugs but it will also point out issues that other code(in which it has access to) will easily refute.
Commit summaries can be impressive, almost worth $5 on that alone.
I'm being generally positive because I'm familiar with the code it's analyzing, I just spent hours/days working with the code. Thus, it's easy for me to understand its findings. I don't understand how people are using this to write feature-size code and have that same amount of familiarity with the results. It seems analogous to asking someone else to review your code. You cannot expect them to take the time to become as intimate with your code as you(exeptions, of course).
A bit ironic that Claude charged me tokens to add token, context, and cost usage to the status line...and fucked it up.
www.rtk-ai.app
