A&A’s mistake is to assume that the parts and wholes of cognitive systems are fixed by how they sometimes function. That is, they assume that because Haugeland himself sometimes functions as a self-contained cognitive system, he must always function as such. The case of the road to San Jose is designed as a counter-example to precisely that inference. A&A, therefore, are themselves guilty of a fallacy, that of assuming that because something sometimes functions in a certain way, it must always function in that way.
I don't think we assume this. In fact, that assumption is inconsistent with our idea that extended cognition is possible. So, in principle, you might give Otto some sort of ultra-fancy neural prosthetic (unlike the humble notebook) that really does function just like a brain region. That could be a case of extended cognition.
Instead, we assume that just because something causally contributes to successful performance does not suffice to establish it as a constitutive element of a cognitive process. The sun contributes to one's reading a newspaper outside, but is not a cognitive processor, right?
Of course, A&M may have some other conditions that distinguish mere causation from constitution, but let's have those on the table.