We decreased our LLM costs with Opus | Hacker News

A Hacker News discussion highlights an LLM routing pattern that uses cheap Haiku for triage and reserves Opus for unresolved cases.

The setup aims to cut inference costs by avoiding premium-model calls for issues already known or easy to resolve.

Commenters debated whether the approach is truly about reasoning, RAG, or just clever prompt and context management.

The thread also discussed how to build similar routing and harness systems in Claude Code and other agent workflows.