It is possible to define conversation policies, such as communication or di
alogue protocols, that are based strictly on what messages and, respectivel
y, what performatives may follow each other. While such an approach has man
y practical applications, such protocols support only "local coherence" in
a conversation. In a mixed-initiative dialogue between two agents cooperati
ng on some joint task, there must be a "global coherence" in both the conve
rsation and in the task they are trying to accomplish. Recognition of agent
intentions about the joint task is essential for this global coherence, bu
t there are further mechanisms needed to ensure that both local and global
coherence are jointly maintained. This paper presents a general yet practic
al approach to designing, managing, and engineering agents that can engage
in mixed-initiative dialogues. In this approach, we promote developing abst
ract task models and designing conversation policies in terms of such model
s.