From Kun: "AI can write so much code that if you review every single line, you become the bottleneck. So I don't even review the first-pass code from the agent. I use a tool I built called No Mistakes.
I did a lot of prompt engineering to get the agents to scrutinize the change very hard, so any edge case, bug, or logical error gets caught. It's a very high recall phase.
When I first built it, I ran parallel testing. I let the agents review the change, and I reviewed it myself, to see how often I caught something they didn't. I used that to iterate on the prompts and the workflow.
Eventually, I got to a point where I find myself never catching anything the agents don't catch."
Full interview on @PeterYangYT!
