The authors manipulated repetitions and/or changes of abstract response rules and the specific stimulus- response (S-R) associations used under these rules. Experiments 1 and 2, assessing trial-to-trial priming effects, showed that repetition of complete S-R couplings produced only benefits when the rule also repeated (i.e., rule-S-R conjunctions) but costs when identical S-R couplings repeated while rules changed. In Experiments 3 and 4, the authors manipulated amount of experience with specific rule-S-R conjunctions and demonstrated integration between rules and S-R couplings in terms of cumulative practice effects. However, unlike short-term priming effects, cumulative practice supported generalization of experience with specific S-R couplings across rule boundaries (Experiment 4). Results are discussed in terms of constraints on models of hierarchical control and in terms of qualitatively different ways in which people profit from very recent experiences (i.e., all-or-none access to working memory representations) versus cumulative experience (i.e., similarity-based retrieval from long-term memory).