I read through a bit of it, and it seems like the author assumes training was done only on open source data, but I doubt that's actually the case. There are ple

I read through a bit of it, and it seems like the author assumes training was done only on open source data, but I doubt that's actually the case. There are plenty of paid, proprietary source codes being sold out there, hacked code, that is. Aside from that, I don't agree with the author's view on agentic workflows. Modern AI native development runs like a massive state machine, starting from MCP, local file systems, and what's usually called a harness. I also noticed what might be a mistake in the author's domain, games. Putting aside the fact that inheritance based OOP is an outdated pattern, the suggestion to remove update() and put it into a manager class's List, then iterate with a for loop, is meant to eliminate overhead like P/Invoke costs in C#. But if Foo is still a class, a reference type, then List<Foo> is just an array of pointers scattered across heap memory. Pointer chasing can still happen. So I think that's actually bad advice.(Of course, the same issue exists in Mr. Claude's code as well.) If the author truly wanted Data Oriented Programming(or DOD), they would have specified struct arrays or NativeArray. This is a tricky area. The author's goal was likely to remove the per MonoBehaviour Update call cost in Unity, which is why they suggested using List<Foo>. But the instruction seems ambiguous, and I think that's part of the difficulty with AI development. The objective function is the same, but the implementation varies and subtly differs from what I actually want. From a design perspective, for team maintenance, the GameUpdateable abstraction might actually be better. But it's difficult. In terms of extensibility, an update manager that handles registration and expansion of multiple update targets might be over abstracting. Writing this down makes me realize how many things I actually consider when putting code into a program. Sometimes I model how my next teammate might read it, and sometimes my words might be interpreted differently. It's really difficult.