Accumulating evidence in the cognitive and linguistic sciences suggests that people are often near-optimal actors, being exquisitely tuned to the world around them. In contrast, I describe a range of observations indicates that when producing language, speakers are notably suboptimal and insensitive to many important features of their linguistic expressions and communicative environments. For example, speakers produce words based on factors other than what they mean; they sometimes choose descriptions that ignore what their addressees do and do not know and that violate their own communicative goals; and they are largely insensitive to the linguistic ambiguity of their utterances. These insensitivities arise at least partly because speakers are responsive to their own cognitive needs: They choose words and sentence structures that are readily accessible, and choose descriptions referring to features that draw their attention. I argue that speakers' productions show sensitivity to their own needs like this because producing language is hard -- especially, it's harder than understanding language. As such, it is not speakers who are optimally tuned to their environment, but speakers and hearers together, each making up for the challenges of the other, who exhibit a division of labor for communicative success.