I believe, however, that to do this is tantamount to abandoning [local] act consequentialism. For to embrace this alternative standard is simply to claim that rules should be evaluated directly in terms of the goodness of their consequences... Under the proposed revised standard, after all, the fact that a given rule produces right acts does no real work whatsoever. The rule is evaluated, rather, simply in terms of the goodness of its results.
Kagan seems to be falsely assuming that internalized rules can only affect outcomes by means of the downstream actions they give rise to. But this is a mistake. If an evil demon threatens death and destruction unless we internalize some rule R (but he doesn't care about what actions we perform), then this makes R a damn good rule to internalize. If R is something arbitrary -- say a rule of spelling -- then it may not affect our downstream propensity to perform beneficial actions at all (unless you think the world is deprived of value when Americans leave the 'u' out of 'colour'; I'm sometimes of this opinion). So local AC, as defined above, would fail to recognize this as the best rule to be internalized. Rather than collapsing into global consequentialism, it is a distinct -- and distinctly inferior -- view.
P.S. Kagan's confusion is revealed by his emphasis on 'right acts' in the penultimate quoted sentence. What's distinctive about local AC, in contrast to global consequentialism, is not that the former assesses rules in terms of the rightness of acts (as if consequentialists thought that was something importantly different from 'goodness'). It's that local AC assesses internalized rules in terms of the [rightness or goodness of] acts thereby produced, rather than in terms of all good consequences of said internalized rules.