Sure, AI makes writing code easy, but code review remains a bottleneck. The individual coders are finding ways to use AI to improve their workflows, but as long as organizational culture remains stuck in the old ways of reviewing code, the overall throughput will not improve significantly. That's been my personal experience at Amazon. I get significant pushback when I submit a CR more than a few pages long. You need to have high degrees of trust within a team for this to work at scale, and that's very rarely the case in my group.
Stuck in the old ways of reviewing code? In other words, what you're suggesting should be the new way is to simply accept the LLM's code salad and push it to production as soon as the pipeline lights up green?
