What Microsoft announced
The November 17 entry covers Azure SQL Database. Database Agent now produces AI-generated recommendations for detected database issues. Those recommendations explain why the issue matters, the likely cause and the suggested next step. According to the changelog entry reproducing the announcement, they draw on the available evidence and on Microsoft-authored scenario knowledge. They separate observed evidence from AI-generated conclusions. They also flag uncertainty when evidence is incomplete, stale or conflicting. Recommendations are shown for your review before any state-changing action.
The agents live inside Fabric Database Hub, so the context matters. Database Hub is Microsoft's single view of a database estate. It covers SQL Server, Azure SQL, PostgreSQL, Azure Cosmos DB and SQL database in Fabric. The agents were first teased at the September FabCon and SQLCon event in Barcelona. Microsoft described them there as AI-powered specialist agents that continuously observe and analyze database signals. A WindowsForum report on the Database Hub general availability notes that Database Hub is generally available, while Copilot and Database Agents remain in preview.
What "contextual Copilot" actually means
The Microsoft Learn documentation draws a line that the headline blurs.
- Core detection and investigation for Azure SQL Database runs without a large language model. It doesn't consume Copilot capacity.
- Contextual AI is optional, and it covers only two scenarios. Those are high CPU and long-running query issues. It uses an LLM and the Copilot capacity your administrator has assigned.
- The AI assistance doesn't change your database. It helps you explore the evidence.
For high-CPU and long-running-query issues, the interface offers three prompts:
- Recommended action opens the Copilot pane with the issue as context.
- Explain why shows an explanation inline.
- Analyze slow query gives inline suggestions for a captured query.
Issue types covered
For Azure SQL Database, Database Agents can detect these scenarios:
- High CPU
- High memory
- High I/O
- Long-running query
- Blocking query
- Missing index
- Dropping unused or duplicate indexes
- Query plan regression
Don't assume every item on that list gets LLM-generated analysis. Only high CPU and long-running query are documented with contextual AI.
Who can use it
Supported: Azure SQL databases, including serverless and Hyperscale.
Not supported:
- Azure SQL elastic pools
- One-vCore serverless databases
- Azure SQL Managed Instance
- SQL Server on Azure Virtual Machines
- SQL Server enabled by Azure Arc
The last two exclusions matter for Windows-centric shops, because many SQL Server estates run on VMs or Arc-enabled servers. Those estates are outside this preview.
Prerequisites from Microsoft Learn:
- Sign in to Fabric with an identity that has access to the Azure SQL databases you want to monitor.
- Use a supported database. If it's serverless, make sure it's awake so it can be evaluated.
- Keep the database in the supported region associated with your Fabric tenant home region. Cross-region enrollment isn't supported.
- Enable performance monitoring for Microsoft SQL.
- Hold database read permission to view issues.
- Hold Azure RBAC write permission on every selected database resource when saving an agent.
Turning it on
In Fabric, open Database Hub and select Manage agents. Then choose Database agent settings. There you pick databases, review agent policies and save the configuration. Visual Studio Code support for Azure SQL is listed as coming soon.
Saving checks your Azure write permission on every selected database. If any one fails, the whole save fails. The documented fix is to remove the resources you lack permission for and save the rest.
Timing and status
- Enabling performance monitoring can take about 15 minutes to propagate before a database can be enrolled.
- After you enable a database or scenario, a first issue can take about 15 minutes to appear. It may appear sooner.
- If nothing shows up, wait at least 15 minutes. Then confirm the source condition is still active, check that the matching policy option is enabled, and refresh the list.
Issue statuses in Database Hub:
| Status | Meaning |
|---|---|
| Active | The condition needs attention |
| Investigating | The agent is gathering or analyzing evidence |
| Investigation completed | Guidance is available, and the response is manual |
| Failed | The investigation didn't complete |
| Closed | The issue is no longer active and stays in history |
A "Failed" status doesn't mean a database change failed, because the agent never executes changes. Investigations retry automatically up to three times for retriable errors. If a condition returns after closure, a new issue is created and the old one stays in history.
Analysis: why the guardrails matter
Microsoft is being careful here, and that's sensible. The documentation says the feature doesn't grant extra access. It doesn't bypass approvals or change management, and it doesn't replace production validation. Closing an issue doesn't prove the workload recovered.
Two practical cautions follow:
- Captured evidence is a snapshot. It describes the condition when it was detected. Recheck the live database before acting.
- This is a preview. Microsoft's Azure Updates definitions describe preview as available for non-production use and testing. Supported scenarios, configurations and regions can change.
A reasonable evaluation plan is to enroll a non-critical Azure SQL database. Let the agent flag a high-CPU or slow-query event. Compare its recommendation with what your DBA would conclude. Treat any suggested change as a ticket, not a button press.
Bottom line
Database Agent with contextual Copilot adds guided, evidence-based triage to Azure SQL Database in Fabric Database Hub. The AI layer is optional and covers two scenarios. The feature is limited by database type and region, and it never makes changes for you. Teams on Managed Instance, elastic pools, SQL Server VMs or Arc-enabled servers will have to wait.
References
- (In preview) Public Preview: Database Agent for SQL with contextual Copilot Azure Updates · 2026-11-17T22:43:01Z
- Public Preview: Database Agent for SQL with contextual Copilot — Cloud Changelog cloudchangelog.com
- Database Agents (Preview) - Microsoft Fabric | Microsoft Learn learn.microsoft.com