Common approaches to optimizing computer-use agent latency include compressing screenshot resolution to reduce token consumption, using local lightweight models to handle simple steps, and caching and reusing action flows.