<p>Clark (<i>Philosophical Psychology</i> 19(3):291–307, 2006) proposes that a standard challenge to the hypothesis of extended cognition can be avoided in the case of linguistically structured cognition, because the role played by our public manipulation of linguistic artifacts is irreducible to the role played by the brain’s operations over internal representations. I demonstrate that Clark’s argument relies on a view of the brain’s cognitive architecture to which he no longer subscribes. I argue that on Clark’s later view of the brain as engaged in ‘predictive processing’, his earlier defense of extended cognition from this challenge is no longer an effective strategy. I explore the implications of this for Clark’s attempts to reconcile his previous arguments for extended cognition with his characterization of the predictive-processing brain.</p>