Select AI agent teams can now be managed like code

Oracle's in-database AI agents got the "boring" features that make them production-ready. These Autonomous AI Database updates for Select AI agent teams landed between July and September 2026.

What's new

• Export and import: export a team as a CLOB or to Object Storage or a directory file, and import it back from JSON (the Open Agent Specification format). Secrets, ACLs and privileges are not exported.
• GET_DEFINITION returns executable PL/SQL for supported objects (including DBMS_CLOUD_AI and DBMS_CLOUD_AI_AGENT objects), so you can compare and recreate them across environments.
• Share a team across schemas: DBMS_CLOUD_AI_AGENT.GRANT_TEAM_ACCESS lets another user or role run it by its owner-qualified name.
• A managed A2A server exposes published teams, so external agents can discover and call them over standard HTTPS.

Why it matters: the hard part of agents isn't the demo. It's versioning, access control and promotion between environments. Treat agent teams like any other schema object: script them, commit them, and deploy them through your pipeline.

Sources: Autonomous AI Database: what's new, Portable Select AI agent teams, A2A server docs.