Why can't AI maintain its own code?

I wonder how the size of the repositories/lines of code to maintain impact the team size. Good Coding Practices always advocated composability.

And the reason was to keep the complexity low.

Theoretically speaking, if you have component A, another component B then you can have A(B), B(A), A(), B(), A(A), B(B) .... you get the idea.

And this is the power of composability. Once you have figured out A and B, you can chain them in any which way to create multiple permutations of your code, and THAT is how you build fast.

Not only figuring out the right A and B is a hard problem, but AI is so good at spewing code that for every new feature it will write its own very specific component called A1(if you can call it a component).

Now you may ask what's the harm with that. The problem is that A1 is slightly different from A2 which is slightly different from A3 and as you keep "building", you are eventually going to create more components, with even more permutations which makes the whole thing hard to maintain.

If not structured correctly, AI which itself struggles with context, is going to create even more code for itself then it has to then maintain, which just increases the probability of a bug ridden software.

So back to the team size, with AI producing tons and tons of code, and not being able to maintain it, will we eventually revert to having the same team sizes as we currently do but just in a different format?

Just a thought as I keep hearing how AI is going to reduce the product triad sizes.

But we'll see how this plays out.