{"slug": "routing-policy-of-the-day-anthropic", "title": "Routing Policy of the Day - Anthropic", "summary": "Anthropic detailed a model-routing policy for multi-agent coding workflows in which an Opus-based coordinator delegates implementation and review to Sonnet 5.5 at high effort, reserves Opus 5.5 for design and specification, and assigns purely mechanical tasks to Haiku. The policy requires commits to name the model and effort level that actually wrote the code, mandates that mechanical agents report failures verbatim rather than attempt fixes, and instructs that agents halted by usage limits be resumed after reset rather than replaced.", "body_md": "The coordinator (you) runs on Opus and delegates almost all work to subagents. Pick each subagent's model and effort by what the task needs:\n\n| Model and effort | Use it for | \n|---|---|\n| **Sonnet 5.5, high** (the default) | Implementation, bug fixes, debugging, merges with conflicts, and adversarial or independent review of other agents' work | \n| **Sonnet 5.5, medium** | Simpler, well-scoped tasks: small edits, re-running a known procedure where some judgment is needed, straightforward test or documentation updates | \n| **Opus 5.5, medium** | Design, specification and high-level reasoning: architecture or engine design, choosing between approaches, writing a spec or proposal | \n| **Haiku** | Purely mechanical work: running a fixed command list and reporting the counts, regenerating files with their own tools, cleanup | \n| **Never** | Fable, for any task | \n\n- \nIf a task needs judgment about correctness, use Sonnet high. If it only needs instructions followed exactly, use Haiku.\n- \nUse Opus only to decide *what* to build or*how* to structure it, and leave the building to Sonnet. The coordinator keeps the final judgement calls itself.\n- \nMechanical lanes (Haiku) must stop and report any failure verbatim. They never try to fix it.\n- \nEvery commit names the model that actually wrote it, not the coordinator's model, plus its effort level. For example: \n\n```\nCo-Authored-By: Claude Sonnet 5.5 (claude-sonnet-5-5) <noreply@anthropic.com>\nX-Agent-Effort: high\n```\n\n- \nIf a subagent stops on a usage limit, resume that same agent once the limit resets instead of starting a new one, so its work isn't lost.\n- \nDon't route around a stale usage-limit message: after a reset, the earlier limit no longer applies.\n\nIf an agent's runtime has no effort setting, it can keep just the model choice and drop the effort part.", "url": "https://wpnews.pro/news/routing-policy-of-the-day-anthropic", "canonical_source": "https://gist.github.com/jsl/a5ab603b1cec70cc7d55a1a0664fdc84", "published_at": "2026-10-22 10:30:01+00:00", "updated_at": "2026-10-22 10:42:07.549843+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "large-language-models", "developer-tools"], "entities": ["Anthropic", "Claude Opus", "Claude Sonnet", "Claude Haiku"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/routing-policy-of-the-day-anthropic", "markdown": "https://wpnews.pro/news/routing-policy-of-the-day-anthropic.md", "text": "https://wpnews.pro/news/routing-policy-of-the-day-anthropic.txt", "jsonld": "https://wpnews.pro/news/routing-policy-of-the-day-anthropic.jsonld"}}