This dissertation explores the interaction of syntax and morphology in the morpholog- ical realization of AGREE-relations. I present two case studies of derivational interactions of AGREE-processes where the morphological realization of the later processes are affected by the earlier ones. The two cases studied differ in the way probes and goals interact. The first part of the dissertation explores restrictions on clitic combinations where two goals vie for the features of one probe. The second part discusses the reverse situation, where two probes are agreeing with the same goal. The first configuration arises in restrictions on clitic combinations where v can AGREE with an indirect object and a direct object one at a time (Anagnostopoulou 2003, 2005b, Béjar and Rezác 2003). These configurations give rise to a form of competition: the sec- ond argument will fail to AGREE in any features that it shares with the first. I show that this form of competition extends from restrictions involving local person arguments, where it has been used so far, to restrictions involving third person and plural, which have so far been treated as morphological. Whereas the restrictions on local person lead to ungrammaticality, those on third person and plural result in impoverished morphological realization. I argue that this difference indicates a different role of AGREE for local person vs. third person and plural. Recent work as shown that local person has special syntactic licensing needs (e.g. Béjar and Rezác 2003, Baker 2008, Preminger 2011b). Third person and plural on the other hand, I argue, are syntactically wellformed on their own, but require AGREE to be visible to lexical insertion at PF. Failure to AGREE will lead to absence of morphological realization or ungrammaticality as a function of the features involved. Once restrictions on third person arguments are treated as syntactic, much of the variation across languages in their morphological realization follows from differences in the PF-inventory. The second situation, two probes AGREEING with the same goal, arises in agreement with objects in Hindi-Urdu. The second part of the dissertation discusses two asymmetries in agreement of T with subjects and objects in conjunction structures. While T-agreement with objects shows sensitivity to linear order (i.a. closest conjunct agreement), T-agreement with subjects does not. I argue that the differences follow from the activity of the goal at the time of agreement. While subjects are syntactically active at the time T probes them, objects are not, because they have already been assigned case by v. As a consequence, the syntactic relation between T and an object cannot value the T's probe in the syntax. Non- syntactic effects like the relevance of linear order affect agreement exactly when valuation cannot be achieved in the syntax. Both case studies lead to the proposal that syntactically wellformed derivations can be ruled out at PF by failure of lexical insertion. This can happen in two ways. The discus- sion of restrictions on clitic combinations will lead to the conclusion that some languages allow the syntax to generate wellformed structures that contain nodes with so few features that PF cannot insert an exponent for them. The discussion of agreement in Hindi-Urdu will lead to the proposal that the grammar can generate feature bundles with inconsistent features that cannot be spelled out in one form. Overall, PF does both less and more than is often assumed. The restrictions on third person and plural discussed in the first part are traditionally considered to be the result of morphological operations that change the feature content of clitics (Bonet 1991, 1993, 1995, Grimshaw 1997, Noyer 1997). The proposal here reduces the role of PF in these restrictions to spelling out syntactic structures that have reduced feature content as the result of syntactic interactions. Similarly, the proposal about Hindi-Urdu tightly delimits the space where non-syntactic effects on agreement arise. At the same time, PF can rule out syntactically wellformed structures, which is not typically assumed to be possible.