Non-Determinism Doesn't Matter: Why LLM Output Is Just a Means, Not the End

I used to see non-deterministic LLM output as a fatal flaw for software development. Then I realized: the result must be exact, but the path to it doesn't have to be. LLMs are actors in a system, not code. By turning their probabilistic output into deterministic tools, we can harness randomness today to reduce randomness tomorrow—and use fewer tokens while doing it.