Microsoft is putting AI agents into database administration, migration, and development tools, moving the technology from conversational assistance towards systems that can investigate problems and carry out multi-step work across enterprise data estates.

The company used its SQLCon and Fabric conference in Barcelona to introduce Database Hub in Microsoft Fabric as a central view of SQL Server, Azure SQL, PostgreSQL, Azure Cosmos DB, and SQL databases in Fabric. The public-preview environment brings inventory, performance, health, security, risk, and optimisation signals together, while database agents can interpret those signals and recommend actions.

Microsoft is also extending agent-driven workflows into migration. A preview of Skills for SQL migration to Azure can handle assessment, migration, and validation steps, while GitHub Copilot Agent Mode in SQL Server Management Studio is now generally available for tasks including dependency analysis, execution-plan investigation, DDL generation, and reviewing database changes.

The collection of announcements amounts to a more operational form of enterprise AI than adding a chatbot beside an existing administration console. Database tools sit close to production systems, which means useful automation can reduce repetitive diagnostic work, but an incorrect action can also affect applications, transactions, performance, and data availability.

Agents move into the control plane

Database Hub is designed to give administrators a single view across engines and environments rather than forcing them to inspect resources separately. Microsoft says findings are grouped around security, performance, and optimisation, allowing administrators to move from an estate-wide signal into a particular resource and then invoke an agent to help understand or address the issue.

The SQL and PostgreSQL agents are beginning in different interfaces, with SQL appearing through Database Hub and PostgreSQL through Visual Studio Code, while Microsoft plans to extend the model across its management surfaces. The agents combine database context with tools and connectors so they can do more than produce generic text about a problem.

That direction changes the governance problem. A model summarising an alert has limited ability to damage a production database, whereas an agent capable of proposing or executing optimisation steps has to operate inside permissions, approvals, auditing, and change-control processes that enterprises already apply to administrators and automation scripts.

Microsoft says human judgement remains part of the model and that actions are intended to respect existing permissions and operational guardrails. How those controls behave in live deployments will be more consequential than whether an agent can write a competent query because database environments contain years of application assumptions and dependencies that may not be apparent from one telemetry signal.

AI meets ordinary database engineering

The company is pairing the agent push with less glamorous engineering controls. SQL Projects and Schema Compare are now generally available in SQL Server Management Studio, allowing database changes to be compared and placed into source-control workflows. Microsoft has also made SQL Formatter generally available in SSMS and Visual Studio Code.

Those additions are relevant precisely because more database code may be machine-generated. Organisations still need a repeatable process for examining what changed, testing it, committing it, and moving it between environments, regardless of whether the first version was written by a database administrator or an AI system.

Microsoft is also adding vector indexes to operational SQL databases and extending Data API Builder with vector and JSON support plus REST, GraphQL, and Model Context Protocol endpoints. That gives applications and agents more direct routes into business data without requiring every organisation to create a separate vector database for AI workloads.

The architectural appeal is straightforward: operational data stays under an existing SQL security model while AI applications gain search and retrieval capabilities closer to the source. The practical risk is that exposing more production information to agents makes identity, authorisation, metadata governance, and query controls more important rather than less.

Sovereignty is another strand of the Barcelona announcements. SQL Server on Azure Local is now generally available in connected and disconnected configurations, while Azure Arc-enabled SQL Server has expanded into additional European regions and SQL Server on Azure virtual machines is available in Microsoft’s Bleu sovereign-cloud environment in France.

Taken together, Microsoft is trying to make its database platform span two enterprise pressures that can pull in different directions: organisations want more automation and AI access to operational information, but many also want tighter control over where data runs and how infrastructure is managed.

Database administration is a useful test of whether agentic AI can survive contact with production technology. The work contains repetitive diagnosis and configuration tasks that lend themselves to automation, although databases are also systems where small mistakes can propagate quickly into customer-facing applications and business operations.