# IBM watsonx Orchestrate CURRENT_STATUS: UNKNOWN OBSERVATION: SPEC_ONLY FRESHNESS: UNKNOWN PAGE_TYPE: reference CANONICAL_URL: /tools/ea-144-ibm-watsonx-orchestrate CLAIM_IDS: none EVIDENCE_IDS: EVD-EA-144-URL-F49DBAED789D, EVD-EA-144-URL-3CA0E547871D, EA-144-R11-RESEARCH, EA-144-R11-WEB-READBACK UNKNOWN_IDS: EA-144-UNKNOWN-01, EA-144-UNKNOWN-02, EA-144-UNKNOWN-CURRENTNESS OPS_IDS: none ## Summary 「IBM watsonx Orchestrateの公式資料は、複数agentの発見・接続・統合・評価・統治を行うcontrol planeと、ADK／LangGraph import／AI Gateway／human approval等の経路を案内している。機能存在は、契約・plan・region・RBAC・互換性・業務品質・SLAの保証ではない。」 ## Current judgment IBM official product/developer materials describe watsonx Orchestrate as an agent-management/control-plane product with agent discovery/connection/orchestration, owners/dependencies/activity/cost visibility, access/policy/guardrails, third-party agent environments, ADK, LangGraph import, tool/API integration, human approval, evaluation/regression testing, and multi-provider AI Gateway capabilities. ## Scope provider mapping / organization overlay / interoperability / governance / security / audit / cost / operability。 ## Confirmed - ESTABLISHED: IBM公式ページは、watsonx Orchestrateをagent管理・control planeとして、agent discovery/connection/orchestration、owners/dependencies/activity/cost visibility、access/policy/guardrails、third-party environment、ADK、LangGraph import、tool/API、human approval、evaluation/regression、multi-provider AI Gatewayを製品要素として記載している。 - NOT_ESTABLISHED: “enterprise-grade”“production”の製品表現は、全plan/region/tenantでの有効化、import互換、policyの完全阻止、評価の妥当性、SLA・ROIを証明しない。ページの機能存在と利用組織の契約・権限を同一視しない。 IBM official product/developer materials describe watsonx Orchestrate as an agent-management/control-plane product with agent discovery/connection/orchestration, owners/dependencies/activity/cost visibility, access/policy/guardrails, third-party agent environments, ADK, LangGraph import, tool/API integration, human approval, evaluation/regression testing, and multi-provider AI Gateway capabilities. NOT_ESTABLISHED: Feature and marketing statements do not establish enablement for every plan, region, or tenant; third-party import parity; complete policy prevention; evaluation validity; SLA; ROI; production quality; or organization-specific entitlement. ## UNKNOWN - EA-144-UNKNOWN-01: current product/API release、各plan/region、third-party import parity、AI Gateway routing policy、approval/audit schema、data isolation、pricing、実業務品質・ROI・障害復旧。 - EA-144-UNKNOWN-02: current product/API release; plan/region availability; third-party import parity; AI Gateway routing policy; approval/audit schema; data isolation; pricing; production quality; ROI; incident recovery. - EA-144-UNKNOWN-CURRENTNESS: 外部根拠の再確認日がcanonical body/control-planeで確定していないため、currentnessはUNKNOWNとして保持する。日付を推測補完しない。 ## Evidence - EVD-EA-144-URL-F49DBAED789D (RESEARCH): source_class=official product documentation; exact_title=IBM watsonx Orchestrate; author/org=IBM; canonical_url= — https://www.ibm.com/products/watsonx-orchestrate - EVD-EA-144-URL-3CA0E547871D (RESEARCH): source_class=official product/developer documentation; exact_title=AI agent development and deployment; author/org=IBM; canonical_url= — https://www.ibm.com/products/watsonx-orchestrate/developers - EA-144-R11-RESEARCH (RESEARCH): EA-144 監査済みresearch / canonical revision binding - EA-144-R11-WEB-READBACK (IMPL): EA-144 Web body identity / canonical readback ## Failure / limits third-party agent importのversion/API差、AI Gateway routingの誤選択、human approvalの誤承認・race、policy設定漏れ、activity/cost trace欠損、connector side effectが残る。IBMのページ上の機能・case/marketing値は独立再現benchmarkではない。 ## Failure / limits IBM公式の“enterprise-grade”“production”表現を独立保証へ変換しない。agent catalog・control planeの存在からユーザー組織の全agent接続可否を推測しない。 UNIT_CLOSE_STATUS: INDIVIDUAL_BOUNDED_CLOSE / EVIDENCE_RECORDED / UNKNOWN_PRESERVED --- ## Failure / limits Third-party agent import can vary by version/API. AI Gateway routing can be misconfigured. Human approval can be incorrect or race with execution. Policy configuration can be incomplete. Activity/cost traces can be incomplete. Connector side effects remain possible. IBM product/case/marketing statements are not independently reproduced benchmarks. ## Practice IBM契約・plan・region・trial/production、ADK版、LangGraph互換範囲、AI Gatewayとmodel provider、connector/API credentials・scope、RBAC、data isolation、OTel/Langfuse、quota・billing。 ## Practice IBM official product/developer docs + PRODUCT_VOLATILE。Docs記載の機能存在と導入経路を保存した段階。IBM tenantの設定・認証・外部書込み・課金・本番migrationは未実施。 ## Practice 公式Docsはevaluation/regression testing機能を記載するが、どのevaluator・dataset・thresholdで、どの業務品質・安全率・costを達成したかは本Packetで確立していない。UNKNOWN。 ## Practice Security=central policy・RBAC・approval・connector scopeをagent/tool/model別に設定。Privacy=import agent、gateway、memory/trace、外部providerのデータ境界。Audit=owner・dependency・version・activity・approval・policy decision・costを相関。Cost=control plane、model、connector、evaluation、storage、retryを含む。Operability=import compatibility、rollback、policy fail-closed、provider outage、trace completenessを検証。 ## Practice IBM provider mappingとorganization overlayへ、control plane・agent import・gateway・approval・evaluation・policyを別claimで渡す。次回はpricing/plan表、region、release、ADK/LangGraph import matrix、RBAC/audit/data isolation、公開incident、実業務evalを補う。 ## Practice IBM contract/plan/region/trial-production state; ADK version; LangGraph compatibility scope; AI Gateway/model-provider support; connector/API credentials and scopes; RBAC; data isolation; OpenTelemetry/Langfuse integration; quota and billing. ## Practice Security: central policy, RBAC, approval, and connector scope must be evaluated per agent/tool/model. Privacy: imported agents, gateway traffic, memory/trace data, and external providers create distinct data boundaries. Audit: correlate owner, dependency, version, activity, approval, policy decision, and cost. Cost: include control plane, model, connector, evaluation, storage, and retry. Operability: verify import compatibility, rollback, policy fail-closed behavior, provider outage handling, and trace completeness. ## Practice IBM documentation describes evaluation/regression testing capability, but this packet does not establish evaluator/dataset/threshold choices or achieved business quality, safety rate, or cost. UNKNOWN. ## WHY_THIS_MATTERS_TO_UPDATE 複数agentを発見・接続・統合・評価・統治するcontrol planeと、ADK・LangGraph import・AI Gateway・human approvalの製品境界をIBM provider mappingへ更新する。 ## SCOPE_ROUTING provider mapping / organization overlay / interoperability / governance / security / audit / cost / operability。 ## CLAIM_OR_RESEARCH_QUESTION IBM公式ページがwatsonx Orchestrateのcontrol plane、agent開発経路、multi-provider、approval、evaluation、governanceを何と記載しているか。 ## EVIDENCE_TABLE - source_class=official product documentation; exact_title=IBM watsonx Orchestrate; author/org=IBM; canonical_url=https://www.ibm.com/products/watsonx-orchestrate; publication/update/access_date=current product page accessed 2026-09-13 JST; version/release/tag/commit/API revision=current product page（製品version未固定）; supporting_locator=Overview / Features / Discover more AI agents / Govern / Mitigate risk（build/deploy/orchestrate/manage/governするagent management platform、agents/owners/dependencies/activity/cost visibility、access/policy/guardrails、third-party agent environment接続とBedrock agentの取り込み、PII等のrisk policyの記述）; evidence_label=SOURCE_REPORT+PRODUCT_VOLATILE; relation=support+limit - source_class=official product/developer documentation; exact_title=AI agent development and deployment; author/org=IBM; canonical_url=https://www.ibm.com/products/watsonx-orchestrate/developers; publication/update/access_date=current page accessed 2026-09-13 JST; version/release/tag/commit/API revision=current product docs; supporting_locator=Build new agents or bring your own / Everything your agents need in production（ADKによるlow-code/pro-code、LangGraph agent import、API/tool integration、human-in-the-loop approvals、evaluation/regression testing、OpenTelemetry/Langfuse、AI Gatewayによるmulti-provider model routingの記述）; evidence_label=VERIFIED_SPEC+PRODUCT_VOLATILE; relation=support+context ## FINDINGS - ESTABLISHED: IBM公式ページは、watsonx Orchestrateをagent管理・control planeとして、agent discovery/connection/orchestration、owners/dependencies/activity/cost visibility、access/policy/guardrails、third-party environment、ADK、LangGraph import、tool/API、human approval、evaluation/regression、multi-provider AI Gatewayを製品要素として記載している。 - NOT_ESTABLISHED: “enterprise-grade”“production”の製品表現は、全plan/region/tenantでの有効化、import互換、policyの完全阻止、評価の妥当性、SLA・ROIを証明しない。ページの機能存在と利用組織の契約・権限を同一視しない。 ## LIMITS / COUNTEREVIDENCE / FAILURE ENVELOPE third-party agent importのversion/API差、AI Gateway routingの誤選択、human approvalの誤承認・race、policy設定漏れ、activity/cost trace欠損、connector side effectが残る。IBMのページ上の機能・case/marketing値は独立再現benchmarkではない。 ## AVAILABILITY_GATES IBM契約・plan・region・trial/production、ADK版、LangGraph互換範囲、AI Gatewayとmodel provider、connector/API credentials・scope、RBAC、data isolation、OTel/Langfuse、quota・billing。 ## IMPLEMENTATION_STATE IBM official product/developer docs + PRODUCT_VOLATILE。Docs記載の機能存在と導入経路を保存した段階。IBM tenantの設定・認証・外部書込み・課金・本番migrationは未実施。 ## BENCHMARK_OR_EVAL 公式Docsはevaluation/regression testing機能を記載するが、どのevaluator・dataset・thresholdで、どの業務品質・安全率・costを達成したかは本Packetで確立していない。UNKNOWN。 ## SECURITY / PRIVACY / AUDIT / COST / OPERABILITY Security=central policy・RBAC・approval・connector scopeをagent/tool/model別に設定。Privacy=import agent、gateway、memory/trace、外部providerのデータ境界。Audit=owner・dependency・version・activity・approval・policy decision・costを相関。Cost=control plane、model、connector、evaluation、storage、retryを含む。Operability=import compatibility、rollback、policy fail-closed、provider outage、trace completenessを検証。 ## SAFE_WORDING_FOR_FUTURE_BOOK 「IBM watsonx Orchestrateの公式資料は、複数agentの発見・接続・統合・評価・統治を行うcontrol planeと、ADK／LangGraph import／AI Gateway／human approval等の経路を案内している。機能存在は、契約・plan・region・RBAC・互換性・業務品質・SLAの保証ではない。」 ## UPDATE_WORK_HANDOFF IBM provider mappingとorganization overlayへ、control plane・agent import・gateway・approval・evaluation・policyを別claimで渡す。次回はpricing/plan表、region、release、ADK/LangGraph import matrix、RBAC/audit/data isolation、公開incident、実業務evalを補う。 ## OPEN_UNKNOWN current product/API release、各plan/region、third-party import parity、AI Gateway routing policy、approval/audit schema、data isolation、pricing、実業務品質・ROI・障害復旧。 ## STOP_CONDITION IBM公式の“enterprise-grade”“production”表現を独立保証へ変換しない。agent catalog・control planeの存在からユーザー組織の全agent接続可否を推測しない。 UNIT_CLOSE_STATUS: INDIVIDUAL_BOUNDED_CLOSE / EVIDENCE_RECORDED / UNKNOWN_PRESERVED --- ## PUBLIC_CLAIM IBM official product/developer materials describe watsonx Orchestrate as an agent-management/control-plane product with agent discovery/connection/orchestration, ADK, LangGraph import, tool/API integration, human approval, evaluation/regression, policy/guardrail controls, and multi-provider AI Gateway capabilities. These product statements do not establish availability for every plan/region/tenant, import parity, policy completeness, SLA, ROI, or production quality. ## ESTABLISHED IBM official product/developer materials describe watsonx Orchestrate as an agent-management/control-plane product with agent discovery/connection/orchestration, owners/dependencies/activity/cost visibility, access/policy/guardrails, third-party agent environments, ADK, LangGraph import, tool/API integration, human approval, evaluation/regression testing, and multi-provider AI Gateway capabilities. NOT_ESTABLISHED: Feature and marketing statements do not establish enablement for every plan, region, or tenant; third-party import parity; complete policy prevention; evaluation validity; SLA; ROI; production quality; or organization-specific entitlement. ## LIMITS_FAILURE_ENVELOPE Third-party agent import can vary by version/API. AI Gateway routing can be misconfigured. Human approval can be incorrect or race with execution. Policy configuration can be incomplete. Activity/cost traces can be incomplete. Connector side effects remain possible. IBM product/case/marketing statements are not independently reproduced benchmarks. ## AVAILABILITY_GATES IBM contract/plan/region/trial-production state; ADK version; LangGraph compatibility scope; AI Gateway/model-provider support; connector/API credentials and scopes; RBAC; data isolation; OpenTelemetry/Langfuse integration; quota and billing. ## SECURITY_PRIVACY_AUDIT_COST_OPERABILITY Security: central policy, RBAC, approval, and connector scope must be evaluated per agent/tool/model. Privacy: imported agents, gateway traffic, memory/trace data, and external providers create distinct data boundaries. Audit: correlate owner, dependency, version, activity, approval, policy decision, and cost. Cost: include control plane, model, connector, evaluation, storage, and retry. Operability: verify import compatibility, rollback, policy fail-closed behavior, provider outage handling, and trace completeness. ## BENCHMARK_EVAL IBM documentation describes evaluation/regression testing capability, but this packet does not establish evaluator/dataset/threshold choices or achieved business quality, safety rate, or cost. UNKNOWN. ## OPEN_UNKNOWN current product/API release; plan/region availability; third-party import parity; AI Gateway routing policy; approval/audit schema; data isolation; pricing; production quality; ROI; incident recovery. NEXT: /agent/start.json STOP: This is a knowledge projection, not an execution order.