What was announced
The announcement came from Microsoft Azure's product marketing lead, with Oracle executives as co-authors. It splits into three audiences:
- IT operations: Azure Workload Center for Oracle, in limited preview.
- End users: an Oracle AI Database plug-in for Excel, described as available today.
- Developers: Oracle AI Database as a data source for agents built in Microsoft Foundry, using managed Model Context Protocol (MCP) servers and the A2A protocol.
Oracle AI Database@Azure underpins all three. Oracle and Microsoft say it spans 33 Azure regions, which they describe as more than any other hyperscaler. They don't list the regions or say how or when the comparison was made.
IT operations: Azure Workload Center for Oracle
The preview introduces a Workload Virtual Instance (WVI). A WVI groups a workload's architecture, resources and dependencies into one object, so admins don't manage each component separately. The WVI connects with Azure Monitor Health Models and the Azure Copilot "Run Oracle" agent. Together they tie monitoring signals from the application, database and infrastructure layers into one picture.
Microsoft's broader Ignite materials describe the same capability in similar terms. The Book of News says Azure Workload Center extends agentic operations to Oracle and SAP for workload health monitoring and configuration validation. That page calls it a private preview. Microsoft's Azure roundup says Azure Workload Center is now in limited preview. The two documents use different preview labels. Treat the program as invitation-only until Microsoft publishes eligibility details.
The worked example
Microsoft's scenario starts with business users reporting a slow order-processing application. This is an illustration, not a documented procedure.
- The admin signs in with their existing Microsoft Entra identity.
- Workload Center shows workload health across infrastructure, database and application layers, and Copilot summarizes where to focus.
- The admin uses a natural-language prompt to start an investigation by several agents. Each agent's Oracle data access is limited by its identity profile.
- Database calls go through the managed MCP server, and every action is logged for audit.
- The agents combine their findings into a single operational conclusion.
The post doesn't define "identity profile." It also doesn't give audit-log retention, the exact Entra configuration, or the Oracle workload types the preview supports. It doesn't say whether every step above works in the limited preview. Admins shouldn't plan around this flow until Microsoft documents it.
Access controls
The post says Oracle AI Database@Azure supports Microsoft Entra authentication and Oracle Deep Data Security. The stated goal is that agents stay within users' existing permissions. That is a design claim. No authorization matrix or deployment guidance was supplied. Security teams should test what an agent can actually reach under their own role and identity setup before trusting it.
End users: the Excel plug-in
The new Oracle AI Database plug-in for Excel is described as available today. It connects Excel directly to Oracle Database and pulls data into a worksheet. Users can then ask Copilot to reason over that data without exporting it first.
The announcement leaves out several details a deployment would need:
- supported Excel editions
- supported Oracle Database versions
- authentication prerequisites
- licensing
- tenant availability
Check each of these before promising the plug-in to finance or planning teams.
The pitch goes beyond the spreadsheet. Microsoft and Oracle say Copilot can combine Oracle transaction data with Work IQ context to help explain margin changes, demand shifts, reconciliation problems and supply-chain disruptions. Copilot is said to identify exceptions, draft an explanation, and prepare a "governed action plan for review." The post doesn't define "governed." The wording describes assistance and human review. It doesn't describe Copilot executing business actions by itself.
Developers: Foundry, MCP and A2A
Developers can build agents against Oracle AI Database through Foundry with managed MCP servers and A2A. According to the post, MCP access covers Oracle AI Database running in OCI or as Oracle AI Database@Azure. It works through approved SQL and PL/SQL operations, tools and predefined reports. Application roles decide which tools each user or group can use. The data stays in place instead of being copied out to the AI.
The two example applications are:
- A sales-intelligence app with separate agents for account and order data, cross-sell analysis and account health. It returns a brief through an API.
- A customer-service agent grounded through Foundry IQ. It uses Oracle-backed memory and similar past interactions to answer questions with citations.
The post offers no setup tutorial and no list of available tools. It names no supported SDKs or versions for the Oracle integration.
What Foundry's own documentation says
Microsoft's Foundry documentation covers the A2A tool in general. It doesn't cover the Oracle integration specifically. It says to use the generally available a2a tool type for new integrations. The older preview tool targets A2A protocol 0.3 and stays in preview for existing integrations. Foundry's A2A tool uses version 1.0 and is generally available. That status applies to the A2A tool, not to every Oracle workflow in this announcement.
The documentation also carries a warning that matters for anyone wiring in third-party agents. It tells developers to treat provider-supplied agent descriptions as untrusted input and to enforce permissions and approvals independently of model-generated decisions. The Oracle MCP route is Microsoft-and-Oracle-sanctioned. Even so, the same least-privilege thinking applies.
Latency and FedRAMP claims
Proximity Placement Groups (PPGs). The post says PPGs let customers place their database closer to applications. It reports roughly 50% lower VM-to-Exadata network round-trip latency compared with deployments without PPG alignment. Treat this as Oracle's own figure. The post gives no methodology, baseline numbers or workload description.
FedRAMP Class D. The post says certification is "coming soon" for US Government and public-sector workloads. That means it is not certified today. Government customers shouldn't read this as current authorization. I couldn't retrieve Oracle's own FedRAMP page to confirm more specific timing, so I'm not repeating any date.
Context and analysis
- Oracle isn't the only database story at Ignite. Microsoft's database team also announced AI-assisted Oracle application conversion for Azure Database for PostgreSQL. Microsoft is pushing both routes: run Oracle on Azure, or move off it. Which fits depends on how entrenched your Oracle estate is.
- Oracle is treated as a first-class data source. The pattern is the same one Microsoft uses elsewhere. Governed access sits in the middle, and agents are never given raw database credentials.
- The practical risk is permissions. Agents that act on a user's behalf are only as safe as the role model beneath them. Administrators should review Entra roles, Oracle application roles and MCP tool lists together.
What to do now
- Ask your Microsoft or Oracle account team about limited-preview eligibility for Azure Workload Center for Oracle.
- Before rolling out the Excel plug-in, confirm Excel and Oracle Database version support, licensing, and how users authenticate.
- Pilot Foundry agents with narrow Oracle roles and approved tools only. Review the audit trail before widening access.
- Don't use the 50% latency claim or the FedRAMP "coming soon" language in compliance or architecture decisions. Wait for documented numbers and certification.
Microsoft and Oracle have described a coherent direction, with identity, audit and tool-level governance as the selling points. The documentation for deploying it is still thin. Until it arrives, the claims remain vendor descriptions.
Update: Additional details (November 18, 2026)
Microsoft’s Azure Updates listing now labels Azure Workload Center for Oracle and the Azure Copilot Run Oracle Agent as a Limited Preview. The service is offered to select customers for non-production use and testing, confirming that it is not a broadly available preview or production-ready release.
That clarification resolves the conflicting preview terminology in earlier Ignite materials. Organizations should treat access as controlled and test-only until Microsoft publishes broader availability, supported configurations, and production-use guidance.
References
- (In preview) Limited Preview: Azure Workload Center for Oracle and Azure Copilot Run Oracle Agent Azure Updates · 2026-11-17T22:46:30Z
- Microsoft and Oracle turn data into insights and actions - Microsoft Azure Microsoft Azure · 2026-11-17T18:35:00+00:00
- Connect to an A2A agent endpoint from Foundry Agent Service - Microsoft Foundry | Microsoft Learn learn.microsoft.com