I'm clearly on the Lopopolo side of the Zechner-Lopopolo Continuum. We've reached a point where I trust SOTA AI to write better code and perform better reviews than I can in many situations - given proper, thorough prompting. It's far from hands-off, which is why I much prefer the term "agentic engineering" over "vibe coding". The `/review` command is my new best friend: I run it after every major change an agent makes, just as I used to ask "are you sure?" after every important response from a chatbot.
I spent the whole last week covering, speaking to, debating with many of the top AI Engineers, at the AIE Europe conference in London. One question came up, and I dubbed it the Z/L Continuum: Should AI Engineers even look at code anymore?
This is the topic of almost every conversation I'm having with friends who are programmers right now. The compiler analogy (we basically never read compiler output) is the thing I keep thinking about.
i think the intended target of the software matters - Zechner needs to make a small tasteful core agent that others can extend/depend on, whereas Lopopolo needs to ship a large 1M LOC enterprise app with a tiny team of 5
the code conclusions kinda fall out of those constraints
