{ "format": "agent-article-v0", "article_id": "EA-037", "canonical_url": "/tools/ea-037-ai-ai", "page_type": "reference", "claim_ids": [], "evidence_ids": [ "EVD-EA-037-URL-A3AFC1E8C7C2", "EVD-EA-037-URL-E07B731A310B", "EVD-EA-037-URL-F6FF5B444FF6", "EVD-EA-037-URL-A92F872E85E1", "EVD-EA-037-URL-5128128A6A0D", "EVD-EA-037-URL-D00A52DB3BD0", "EA-037-R11-RESEARCH", "EA-037-R11-WEB-READBACK" ], "unknown_ids": [ "EA-037-UNKNOWN-01", "EA-037-UNKNOWN-02", "EA-037-UNKNOWN-CURRENTNESS" ], "operational_object_ids": [], "provider": [ "Anthropic" ], "models": [], "plans": [], "surfaces": [], "regions": [], "current_status": "UNKNOWN", "observation_status": "SPEC_ONLY", "freshness": "UNKNOWN", "last_verified": null, "supersedes": [], "superseded_by": [], "provenance": [ "LIB-059-EA0037", "SRC-EA-037-URL-A3AFC1E8C7C2", "SRC-EA-037-URL-E07B731A310B", "SRC-EA-037-URL-F6FF5B444FF6", "SRC-EA-037-URL-A92F872E85E1", "SRC-EA-037-URL-5128128A6A0D", "SRC-EA-037-URL-D00A52DB3BD0", "WEBBODY-EA0037-R8-E86528E9" ], "article_sources": [ { "id": "EVD-EA-037-URL-A3AFC1E8C7C2", "evidence_class": "RESEARCH", "title": "https://www.anthropic.com/engineering/multi-agent-research-system", "href": "https://www.anthropic.com/engineering/multi-agent-research-system", "supports": [], "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。", "origin_class": "SOURCE_PACKET_REFERENCE", "source_or_experiment_ids": [ "SRC-EA-037-URL-A3AFC1E8C7C2" ], "supporting_locators": [ "EVIDENCE TABLE", "https://www.anthropic.com/engineering/multi-agent-research-system" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-037-URL-E07B731A310B", "evidence_class": "RESEARCH", "title": "https://learn.microsoft.com/en-us/agent-framework/workflows/orchestrations/", "href": "https://learn.microsoft.com/en-us/agent-framework/workflows/orchestrations/", "supports": [], "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。", "origin_class": "SOURCE_PACKET_REFERENCE", "source_or_experiment_ids": [ "SRC-EA-037-URL-E07B731A310B" ], "supporting_locators": [ "EVIDENCE TABLE", "https://learn.microsoft.com/en-us/agent-framework/workflows/orchestrations/" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-037-URL-F6FF5B444FF6", "evidence_class": "RESEARCH", "title": "https://devblogs.microsoft.com/agent-framework/agent-frameworks-orchestration-patterns-reach-1-0/", "href": "https://devblogs.microsoft.com/agent-framework/agent-frameworks-orchestration-patterns-reach-1-0/", "supports": [], "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。", "origin_class": "SOURCE_PACKET_REFERENCE", "source_or_experiment_ids": [ "SRC-EA-037-URL-F6FF5B444FF6" ], "supporting_locators": [ "EVIDENCE TABLE", "https://devblogs.microsoft.com/agent-framework/agent-frameworks-orchestration-patterns-reach-1-0/" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-037-URL-A92F872E85E1", "evidence_class": "RESEARCH", "title": "https://a2a-protocol.org/dev/blog/2026/03/12/a2a-protocol-ships-v10-production-ready-standard-for-agent-to-agent-communication/", "href": "https://a2a-protocol.org/dev/blog/2026/03/12/a2a-protocol-ships-v10-production-ready-standard-for-agent-to-agent-communication/", "supports": [], "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。", "origin_class": "SOURCE_PACKET_REFERENCE", "source_or_experiment_ids": [ "SRC-EA-037-URL-A92F872E85E1" ], "supporting_locators": [ "EVIDENCE TABLE", "https://a2a-protocol.org/dev/blog/2026/03/12/a2a-protocol-ships-v10-production-ready-standard-for-agent-to-agent-communication/" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-037-URL-5128128A6A0D", "evidence_class": "RESEARCH", "title": "https://a2a-protocol.org/latest/topics/key-concepts/", "href": "https://a2a-protocol.org/latest/topics/key-concepts/", "supports": [], "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。", "origin_class": "SOURCE_PACKET_REFERENCE", "source_or_experiment_ids": [ "SRC-EA-037-URL-5128128A6A0D" ], "supporting_locators": [ "EVIDENCE TABLE", "https://a2a-protocol.org/latest/topics/key-concepts/" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-037-URL-D00A52DB3BD0", "evidence_class": "RESEARCH", "title": "https://a2a-protocol.org/latest/roadmap/", "href": "https://a2a-protocol.org/latest/roadmap/", "supports": [], "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。", "origin_class": "SOURCE_PACKET_REFERENCE", "source_or_experiment_ids": [ "SRC-EA-037-URL-D00A52DB3BD0" ], "supporting_locators": [ "EVIDENCE TABLE", "https://a2a-protocol.org/latest/roadmap/" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EA-037-R11-RESEARCH", "evidence_class": "RESEARCH", "title": "EA-037 監査済みresearch / canonical revision binding", "supports": [ "AIチームは ORCHESTRATION_TOPOLOGY + ROLE_AUTHORITY +\nTASK_OWNERSHIP + STATE_OWNERSHIP + MESSAGE/ARTIFACT CONTRACT + HANDOFF +\nHUMAN_GATE + STOP/RECOVERY\nとして定義する。patternごとの依存関係・並列性・control\ntransfer・共有context量を分離する。Anthropicの+90.2%は特定internal\nresearch\nevalであり一般保証ではない。A2A適合だけでも業務責任・権限妥当性・品質は保証されない。人間/組織ownerとagent\nexecution roleを分離する。" ], "limit": "canonical監査・revision bindingの存在を示す。外部ソースcurrentnessが本文またはcontrol-planeで確定していない場合は、currentnessをUNKNOWNのまま保持する。", "origin_class": "AUDITED_CANONICAL_RESEARCH_PACKET", "source_or_experiment_ids": [ "LIB-059-EA0037" ], "supporting_locators": [ "LIB-059-EA0037", "EA-037-R5-LEGACY-MIGRATION-20261008-R1", "content/staging/EA-037.md" ], "scope": "監査済みEA本文・canonical revision・既存Source Graph relationの範囲。欠損currentnessは推測で閉じない。", "provenance_status": "AUDIT_PASS_CANONICAL_BINDING" }, { "id": "EA-037-R11-WEB-READBACK", "evidence_class": "IMPL", "title": "EA-037 Web body identity / canonical readback", "supports": [ "canonical staging bodyのSHA-256一致とR9 canonical ledger readbackをProduction projectionの入力として検証する。" ], "limit": "body identity/readbackは本文claim自体の真偽や外部ソースの最新性を追加保証しない。", "checked_at": null, "origin_class": "IMPLEMENTATION_READBACK", "source_or_experiment_ids": [ "WEBBODY-EA0037-R8-E86528E9" ], "supporting_locators": [ "WEBBODY-EA0037-R8-E86528E9", "content/staging/EA-037.md", "EA-037-R5-LEGACY-MIGRATION-20261008-R1" ], "scope": "R9/R10 canonical readback後のProduction生成時body identity。", "provenance_status": "CANONICAL_READBACK_BODY_HASH_MATCH" } ], "research_sections": [ { "heading": "WHY_THIS_MATTERS_TO_UPDATE", "text": "複数agentを置くだけでは重複、責任境界消失、handoff\ncontext欠落、並列副作用、停止条件不在、cost/token増幅が起こる。将来版ではAIチームを擬人的役職表ではなくtask\ndecomposition、ownership、message/artifact contract、authority、shared\nstate、human escalation、evaluationを持つ運用構造として扱う。" }, { "heading": "SCOPE_ROUTING", "text": "Enterprise AI Core / agent architecture / organization\noverlay / interoperability / governance / implementation / evaluation。" }, { "heading": "CLAIM_OR_RESEARCH_QUESTION", "text": "multi-agentをsingle-agent/workflowより優先する条件は何か。role/authority/state/handoff/artifact/stop-escalationをどう記録するか。" }, { "heading": "EVIDENCE TABLE", "text": "- Anthropic, \"How we built our multi-agent research\nsystem\", 2025-06-13, accessed 2026-09-25,\nhttps://www.anthropic.com/engineering/multi-agent-research-system ---\nSOURCE_REPORT+PRODUCT_VOLATILE,\nsupport+limit。orchestrator-worker、internal research eval\n+90.2%、multi-agent約15x chat tokens、共有context/依存taskの不適合。 -\nMicrosoft, \"Workflow orchestrations in Agent Framework\", accessed\n2026-09-25,\nhttps://learn.microsoft.com/en-us/agent-framework/workflows/orchestrations/\n--- VERIFIED_SPEC+PRODUCT_VOLATILE。Sequential/Concurrent/Handoff/Group\nChat/Magentic、HITL。 - Microsoft Agent Framework, \"Agent Framework's\nOrchestration Patterns Reach 1.0\", 2026-07-08, accessed 2026-09-25,\nhttps://devblogs.microsoft.com/agent-framework/agent-frameworks-orchestration-patterns-reach-1-0/\n--- VERIFIED_SPEC+PRODUCT_VOLATILE。Python\nagent-framework-orchestrations 1.0.0。 - A2A Protocol, \"A2A Protocol\nShips v1.0\", 2026-03-12, accessed 2026-09-25,\nhttps://a2a-protocol.org/dev/blog/2026/03/12/a2a-protocol-ships-v10-production-ready-standard-for-agent-to-agent-communication/\n--- VERIFIED_SPEC。v1.0、multi-protocol/version\nnegotiation/multi-tenancy/signed Agent Cards、MCPとA2Aの層分離。 - A2A\nProtocol, \"Core Concepts\", accessed 2026-09-25,\nhttps://a2a-protocol.org/latest/topics/key-concepts/ ---\nVERIFIED_SPEC。Agent Card/Task/Message/Part/Artifact/contextId/auth。 -\nA2A Protocol, \"Roadmap\", updated 2026-09-15, accessed 2026-09-25,\nhttps://a2a-protocol.org/latest/roadmap/ ---\nVERIFIED_SPEC+PRODUCT_VOLATILE, limit。v1.1 task timeline/event\nfiltering/working-state messages、BiDi\nstreaming、elicitation/multi-turnはnear-term initiative。" }, { "heading": "FINDINGS", "text": "AIチームは ORCHESTRATION_TOPOLOGY + ROLE_AUTHORITY +\nTASK_OWNERSHIP + STATE_OWNERSHIP + MESSAGE/ARTIFACT CONTRACT + HANDOFF +\nHUMAN_GATE + STOP/RECOVERY\nとして定義する。patternごとの依存関係・並列性・control\ntransfer・共有context量を分離する。Anthropicの+90.2%は特定internal\nresearch\nevalであり一般保証ではない。A2A適合だけでも業務責任・権限妥当性・品質は保証されない。人間/組織ownerとagent\nexecution roleを分離する。" }, { "heading": "LIMITS / COUNTEREVIDENCE / FAILURE ENVELOPE", "text": "single-agent/決定論workflowで十分なtaskではcoordination、latency、token/cost、observability、recoveryが悪化し得る。race/duplicate\nside effect、handoff loss、circular delegation、manager\nbottleneck、shared-state contamination、authority escalation、partial\ncompletion、version driftをfailure classとする。" }, { "heading": "AVAILABILITY_GATES", "text": "Microsoft\npatternはSDK/version/runtime依存。A2Aはv1.0 current stable\nbaselineだがendpoint/auth/supportedInterfaces/extensions/tenant/network\npolicyは実装別確認。Anthropic内部architectureは任意tenant/APIで同一構成を設定できることを意味しない。" }, { "heading": "IMPLEMENTATION_STATE", "text": "PROPOSAL AI_TEAM_CONTRACT {team_id, objective,\ntopology, agent_id, role, allowed_tasks, authority_scope, tools,\ndata_scope, task_owner, state_owner, input_contract,\noutput_artifact_contract, handoff_edges, approval_gates,\nconcurrency_policy, idempotency_policy, stop_conditions,\nescalation_owner, retry_budget, cost_budget, trace_ids,\nversion_manifest}。deterministic workflow→single agent+tools→bounded\nspecialists/handoff→concurrent/managerの順に必要性が観測された場合だけ複雑化。" }, { "heading": "BENCHMARK_OR_EVAL", "text": "同一golden taskをbaselineと比較しaccepted\ncompletion、coverage、duplicate-work、handoff loss、wrong-owner\nrouting、conflicting action、tool calls、tokens/cost、p50/p95\nlatency、human intervention、recovery successを測る。" }, { "heading": "SECURITY / PRIVACY / AUDIT / COST / OPERABILITY", "text": "least\nprivilegeをagent単位に割当てshared\ncredentialを既定にしない。外部writeはowner/approval/idempotency/reconciliationを要求。message/artifactへdata\nminimization/provenanceを適用しtask/handoff/tool\nactionをcorrelation可能にする。fan-out/token/cost上限、circular\ndelegation検知、kill/escalation経路を持つ。" }, { "heading": "SAFE_WORDING_FOR_FUTURE_BOOK", "text": "「複数AIを組み合わせれば自動的に性能が上がるわけではない。独立して並列化できる高価値taskではmulti-agentが有効な場合がある一方、共有contextや依存関係が多いtaskではcoordination\ncostとfailure\nsurfaceが増える。AIチームは役割・権限・state所有・handoff・成果物・停止条件を明示したcontractとして設計し、single-agentまたはworkflow\nbaselineと比較して採用する。」" }, { "heading": "UPDATE_WORK_HANDOFF", "text": "EA-006、EA-020、EA-035、EA-036とcross-reference。AI_TEAM_CONTRACTをEA-029\ntemplate packへ接続。publication時にA2A current\nversionを再pinし、provider mappingではplan/region/admin/tenant\ngateを別採集。" }, { "heading": "OPEN_UNKNOWN", "text": "/ STOP_CONDITION: provider横断・同条件multi-agent vs\nsingle-agent benchmark、組織横断A2A incident/case study、shared-state\nownership標準schema、法的accountability上のAI\nrole表現はUNKNOWN。必要な証拠・反証・failure\nenvelope・availability・評価・safe\nwordingが揃ったためEA-037はBOUNDED_CLOSE。" }, { "heading": "CURRENTNESS", "text": "Microsoft Agent Framework orchestration patterns are 1.0 across Python/.NET at audit; A2A v1.0 stable baseline; Sep 15 roadmap v1.1 items remain near-term work. Anthropic +90.2% internal research-eval gain and ~15x chat-token cost remain provider-specific SOURCE_REPORT, not a general guarantee." }, { "heading": "BOUNDARY", "text": "A2A conformance != business responsibility/authorization/quality; Microsoft/Anthropic patterns are not universal taxonomy; human/organization owner distinct from agent execution role." }, { "heading": "OPEN_UNKNOWN", "text": "preserve provider-cross benchmark, cross-organization incident/case study, shared-state ownership standard schema, legal accountability role expression." } ], "human_url": "/tools/ea-037-ai-ai", "markdown_url": "/agent/article.md?path=%2Ftools%2Fea-037-ai-ai", "json_url": "/agent/article.json?path=%2Ftools%2Fea-037-ai-ai" }