Like the letters in the alphabet forming words, reusing components of a heterogeneous mixture is an efficient strategy for assembling a large number of target structures. Examples range from synthetic DNA origami to proteins self-assembling into complexes. The standard self-assembly paradigm views target structures as free-energy minima of a mixture. While this is an appealing picture, at high speed structures may be kinetically trapped in local minima, reducing self-assembly accuracy. How then can high speed, high accuracy, and combinatorial usage of components coexist? We propose to reconcile these three concepts not by avoiding kinetic traps, but by exploiting them to encode target structures. This can be achieved by sculpting the kinetic pathways of the mixture, instead of its free-energy landscape. We formalize these ideas in a minimal toy model, for which we analytically estimate the encoding capacity and kinetic characteristics, in agreement with simulations. Our results may be generalized to other soft-matter systems capable of computation, such as liquid mixtures or elastic networks, and pave the way for high-dimensional information processing far from equilibrium.