A cornerstone of sentence processing is that we deploy incremental and predictive processing during online comprehension. Code-switching, however, potentially presents unique challenges to the comprehension system. While parsing current linguistic input, bilinguals must be ready to integrate code-switches into the other language, which minimally includes switches in language membership, phonology, and morphosyntax. Additionally, lab-based studies robustly find processing costs associated with the comprehension of code-switching, despite its ubiquity among certain bilingual communities. To reconcile this apparent paradox, I will present the Adaptive Predictability Hypothesis, comprised of two components: 1. bilinguals globally adapt how they predict and 2. bilinguals locally increase cognitive control to integrate an immediate code-switch. I will illustrate the hypothesis across two studies and discuss future directions.