I'm kind of disappointed, since the "infrastructure" I feel is "more important" involves how you prevent LLMs from turning your codebase and product into spaghetti... as opposed to the infrastructure for making it become spaghetti faster and cheaper.
Left to their own devices, AI can add a lot of extra code you don’t need, overbuild features and functionality, and create a lot of unintended consequences.
If you want to protect your codebase, robust review and carefully deploying changes is essential as you rightfully point out.
Ironically, that’s also what our product does (ai code analysis that runs your code), but as you can see we wrote the post from our experience with infrastructure benefits in mind.