Connect the Microsoft estate to accountable AI work.

An original candidate architecture

1. Experience and integrationExisting business channels → API Management, application adapters and Logic Apps where suitable.
2. Context and meaningExisting records → AI Search and Storage where appropriate; Purview supports data-governance work. Owners maintain the semantic model.
3. Reasoning and workflowFoundry supported models and Agent Service; Logic Apps for configured business integration workflows.
4. Authority and effectConfigured Entra identity and resource permissions, human-decision application and authoritative backend checks.
5. Evidence and operationFoundry tracing and evaluation capabilities, connected Azure Monitor services and customer-owned decision and outcome records.

This is a proposed CAE assembly rather than a Microsoft reference diagram or a tested integration. The provider is not a partner or endorser of this initiative.

What documentation supports

Foundry Agent Service documents managed agents, models, tools, identity and observability capabilities. Features have distinct release stages; the overview identifies its agent optimiser as preview. This map does not require that preview feature.

Microsoft’s landing-zone reference architecture shows how Foundry can connect to supporting search, storage, application, secrets and monitoring services. It is a reference for adaptation, not evidence that the customer environment has those controls.

API Management manages APIs. Logic Apps supports integration workflows across cloud and other systems. Purview supports data governance. The CAE interpretation connects these categories to the discovery model; it does not equate them with AWS or Google Cloud services.

Controls and deployment choices still to make

• Verify current product names, regional availability, model availability, quotas and feature release status.
• Confirm required Azure subscriptions, service tiers and any connected product licences. Existing Microsoft licences are not assumed to cover the architecture.
• Implement user and agent permissions, source-level retrieval restrictions, safe tool adapters and backend transaction checks.
• Build the human-review flow, exact-action binding, rejection, timeout and escalation; identity alone is not approval.
• Evaluate output quality, privacy, logging, incident handling and tested recovery.
• Include model, search, storage, networking, integration, support and human-review costs, with portability and exit effort.

No Azure subscription, customer record, model or integration is provisioned by this site. A legal compliance, security or performance outcome requires separate evidence. Compare all capabilities and customer-owned controls.