Sean Goedecke recommends profiling and redesign before asking AI to rewrite entire programs in C or Rust. Database and network waits limit the benefit of faster application code, while intensive calculations may already run in optimized native libraries. His suggested sequence is to reduce work through caching, fewer queries and parallelism, then consider rewriting remaining CPU hotspots. The argument draws on web-service experience and should not be generalized with identical ratios to compute-heavy software such as renderers.