While fault-tolerance requires relatively mild assumptions about the nature of the errors, the overhead associated with coherent and non-Markovian errors can be orders of magnitude larger than the overhead associated with purely stochastic Markovian errors. One proposal, known as Pauli frame randomization, addresses this challenge by randomizing the circuits so that the errors are rendered incoherent, while the computation remains unaffected. Similarly, randomization can suppress couplings to slow degrees of freedom associated with non-Markovian evolution. Here we demonstrate the implementation of circuit randomization in a superconducting circuit system, exploiting a flexible programming and control infrastructure to achieve this with low effort. We use high-accuracy gate-set tomography to demonstrate that without randomization the natural errors experienced by our experiment have coherent character, and that with randomization these errors are rendered incoherent. We also demonstrate that randomization suppresses signatures of non-Markovianity evolution to statistically insignificant levels. This demonstrates how noise models can be shaped into more benign forms for improved performance.