It is generally agreed that there are two kinds of if-then rules: logic-based rules, in which if-then has a logical semantics, and reactive rules, in which if-then represents change of state without a logical semantics. I will argue that reactive rules have an implicit logical semantics, as goals that need to be satisfied by generating a model that makes the goals true. The logical semantics of reactive rules can be made explicit by making change of state explicit, and by understanding the rules as meaning that if some conditions are true at a time, then some actions are performed at a future time. This is the approach taken by the modal logic MetateM. The same approach can be used with a non-modal logic, such as the situation calculus or event calculus. However, all these logics use frame axioms to reason that if a fact is true in a state, then it remains true in the next state, unless it is terminated by the change of state. Reasoning by means of frame axioms is intolerably inefficient, compared with destructive change of state in conventional reactive systems. I will argue that this inefficiency can be avoided in logic-based systems by using destructive change of state, with frame axioms becoming an emergent property. But giving a logical semantics to reactive rules leaves open the relationship of reactive rules with ordinary logic-based rules. I will argue that ordinary logic-based rules can be understood as beliefs that help an agent satisfy its goals.