Zig isn't ready for AI because AI isn't ready for Zig. There is a bootstrapping problem. Quality training data is limited because it is new, and many humans don't yet have the experience needed to be good at it.
Since Zig is newer, developers are often learning the language alongside the AI. A novice developer might accept an AI's bad allocator usage or improper async handling simply because they lack the expertise to audit it properly. This creates a situation that causes AI to be less effective when using it for Zig development when compared to C and C++ development efficacy, even if they review proposed changes prior to submission of the pull request. Citing Zigs experience as indicative of a global condemnation of AIs usefulness in a software development context represents a complete failure to look at the big picture and place the facts in context. One can use an AI to explore this further.
After thinking a bit about this I asked Gemini (free version) "I am trying to determine if AI generated zig code would likely be of poorer quality than C code due to a few factors. One is that zig is new so many developers may not be very skilled with it, and another is that there is limited training data as compared to C or C++." It had a lot of insightful commentary on my point. In fact there was far too much to copy and paste here, but anyone interested can do so if they want to explore my point (made primarily in the first paragraph of this post) further.