We cut PR wait time and running costs by rethinking CI as a system, from the infrastructure underneath it to how work gets scheduled and tests get parallelized.