Agree is considered one of the fundamental operations of Minimalist Syntax, responsible for establishing long-distance dependencies (e.g. agreement, wh-movement). Various models of Agree have been suggested over the last few decades, and the operation has become increasingly flexible and parameterized (e.g. with articulated probes, separation of interaction and satisfaction) to account for complex data. However, as we strengthen Agree, what extra typological predictions do we make? Do they coincide with observed data, or do we face significant overgeneration? In the latter case, can we identify the individual modifications to Agree that result in a mechanism that is “too strong”, in some way? I’ll present my (very ongoing) work, which includes a face-value implementation in Python of Chomsky 2000/2001’s version of Agree. I should come with some interesting patterns that it generates, and am hoping to frame things more as a discussion than a full presentation. Looking forward to hearing people’s thoughts!