We prove a characterization of first-order string-to-string transduction via $\lambda$-terms typed in non-commutative affine logic that compute with Church encoding, extending the analogous known characterization of star-free languages. We show that every first-order transduction can be computed by a $\lambda$-term using a known Krohn-Rhodes-style decomposition lemma. The converse direction is given by compiling $\lambda$-terms into two-way reversible planar transducers. The soundness of this translation involves showing that the transition functions of those transducers live in a monoidal closed category of diagrams in which we can interpret purely affine $\lambda$-terms. One challenge is that the unit of the tensor of the category in question is not a terminal object. As a result, our interpretation does not identify $\beta$-equivalent terms, but it does turn $\beta$-reductions into inequalities in a poset-enrichment of the category of diagrams.