Prompt file imported from IvonneAkerBP/model-redisigner-repo (
.github/prompts/optimize-ai-cost.prompt.md). Copyright stays with the author.
Review what I am about to run, or just ran, for AI cost. Tell me how to do it cheaper without losing quality.
Check, in order:
- Model fit — is the model heavier than the task needs? Name a lighter category (Lightweight / Versatile / Powerful) if so.
- Extended capabilities — is the 1M-token context window or higher reasoning on when the task does not need it?
- Input size — is stable context (instructions, tool definitions, reference material) being re-sent instead of cached or front-loaded?
- Output size — is the output uncapped or verbose when a terse, structured answer would do?
- Session hygiene — is unrelated history, unused tools/MCP servers, or large/generated files inflating the context? Should this be a new chat or a
/fork? - (API only) batching — is non-urgent bulk work going through a real-time call instead of a batch?
Return three things: the levers to apply now (concrete, in priority order); the single measurement that will confirm it worked (status dashboard %, Cache Explorer hit-rate, or the API usage block); and any claim you could not ground. Do not invent prices — if a number is needed, point me to the live pricing page.