https://www.anthropic.com/engineering/building-effective-agents

この標準文書は未翻訳です。以下に英語の原文を表示します。

この情報源から得られる知見

  • Use only to determine appropriate autonomy and orchestration levels: simple composable workflow patterns before complex agent frameworks; increase autonomy only when task complexity and tool diversity genuinely require it.
  • Map workflow/agent architecture choices to visible user controls: approval boundaries, trace visibility, stop, retry, and escalation paths must be proportionate to autonomy level.
  • Distinguish predefined workflow steps from agent decisions; users must be able to understand, interrupt, and recover from both.

AwesomeDSへの反映先

実装・成果物

直接リンクされた成果物はありません。

注意事項と模倣防止

Transfer orchestration level selection and visible-control proportionality principles; this source does not govern visual design, token use, or component behaviour.

証拠レベル: first-party-guidance · 最終確認 2026-07-16 · 確認周期 90d