Not all theorists believe in the psychological reality of rules. Those who don't may still use production systems merely as convenient programming languages or notations to express the processes producing some behaviour. This is like writing your model in Pascal: it doesn't commit you to believing that memories are stored as 32 bit words, or that the brain is full of machine code. There's an article by Richard Young Production Systems for Modelling Human Cognition in Expert Systems in the Microelectronic Age edited by Michie (E.U.P. 1980; PSY KH:M 58), in which he identifies three levels of theoretical commitment. This, the least committed to PSs, he calls the language level.
The next level down is the rule level. This is the level of Young's own work. We believe in the psychological reality of rules, and perhaps in some aspects of the conflict resolution. But we don't (necessarily) extend this commitment to details of the architecture.
Finally, there's the immediate processor level, where the PS architecture is - as described in the previous section - a model of human cognitive architecture. Most of Newell's work has been at that level. It's interesting to note though that Newell's level of committment appears to have changed between the Psych. Rev. article quoted below and his PS work.