Agents as a Service 支える Data / Context AI Readable な情報 ワークフロー アクション状態の管理 評価・承認プロセス, etc. 人間主体、AI は補助 Agent 主体、人間は承認 なんか大変そう〜! 参考: Loop Engineering / O'Reilly Agents as a Service / Sierra 3
Rule / Policy Workflow Knowledge Memory Glossary Response phrasing PURPOSE A goal the agent knows how to pursue: dispute a charge, file a claim, book a flight. Each journey has a trigger and an outcome. A way for the agent to interact with external systems: pull an itinerary, check coverage, process a refund. Guardrails and business logic expressed in natural language (e.g., “Premium cardholders waive foreign transaction fees.”). Step-by-step guidance for situations that require a specific sequence, like regulated intake or multi-step verification. Help center articles, product docs, FAQs, and internal policies the agent can access on demand. The customer’s history: past conversations, preferences, and prior issues. The terminology your business uses: product names, plan tiers, and internal jargon. Brand voice and tone. - アプリケーション実装とデータ構造が密 - 新しい Data Management の予感 参考: Context Engineering / Sierra Ontology System / Palantir 4