{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreigjlwpch7gdq552tn5panjr2imkq76llu2cel5utr7kbww7e3rrii",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mk2yoo343j72"
  },
  "path": "/t/optimizing-agentic-architecture-strategies-for-reducing-high-token-costs-in-multi-intent-workflows/1379120#post_2",
  "publishedAt": "2026-04-22T07:34:38.000Z",
  "site": "https://community.openai.com",
  "textContent": "As for a optimizing “agent per tool call”. Experience of our lab showed that if you can accomplish the action without LLM call after routing - do it. This makes workflow more reliable and predictable.\n\nWe basically use another LLM(agent) call only in case we need semantically extract parameters necessary to perform action",
  "title": "Optimizing Agentic Architecture: Strategies for Reducing High Token Costs in Multi-Intent Workflows"
}