Reviewing 150k loc line-by-line is not practical. I think we need a new set of tools to facilitate such tasks.

Otherwise, we will be spending most of our time monitoring AI's code writing process.