Biological systems must sense and adapt to changes in their environment. Molecular networks capable of such adaptation belong to two well-known classes, feed-forward and feedback structures, but the fundamental limitations and tradeoffs of these two classes remain unknown. Here we study the advantages and limitations of the feedforward class using three-node circuits representative of these architectures. The feedforward model we investigate displays a tradeoff between the sensitivity of the response (its peak response) and its precision (its error in its return to steady-state). We suggest two ways in which this tradeoff can be alleviated: (1) by introducing a nonlinearity in the production of a specific node in the network, or (2) by adding a feedback loop to the input. We present analytical and numerical examples to support our findings.