When a model run is expensive, what is your default move?

Options

• I would rather burn tokens on several parallel tries than spend an hour tuning one prompt — speed beats elegance when the clock is loud.
• I slow down and craft one tight prompt with examples and constraints; brute force feels like paying laziness forward.
• It depends on whether I am exploring or shipping: explore cheap and messy, ship with a reviewed chain.
• Brute force is fine for prototypes, but it teaches you nothing about the system you are building.
• I run a quick spam pass, then distill what worked into a reusable prompt or small tool.
• Cost is not the issue — wrong answers in production are; I cap retries and add checks.
• I treat the model like a junior: give context once, ask for a plan, then execute in steps.
• Honestly I avoid expensive models for most tasks and only escalate when stuck.
• Teams should standardize prompts in repos so individuals are not reinventing precision every day.