IDUG NA 2026

AI agents are moving beyond chatbots and code suggestions. The next frontier is agentic AI — systems that autonomously plan multi-step tasks, invoke tools, handle branching logic, and execute end-to-end workflows with minimal human intervention. This session shows how that capability applies directly to Db2 for z/OS operations today, using tooling already available in your environment.
The architectural foundation is simpler than it appears: a well-documented CLI is already an agent interface. The Zowe CLI ecosystem — including its Db2 plugin and the richer vendor-supplied extensions published to the Zowe conformance catalog — gives an AI agent a complete, auditable means to act on the mainframe. The agent does not wait for step-by-step instructions; it reasons over the task, selects the right tool from its available plugin set, chains the necessary operations, and executes.
This session presents a layered agent architecture purpose-built for Db2 workflows in production environments:
Execution layer: A base skill encoding the Zowe CLI interface — translating agent intent into exact, logged commands with safe defaults enforcedDomain layer: A Db2-specific reasoning layer encoding your object model, naming conventions, environment topology, and operational runbooks — the institutional knowledge that typically lives in a senior DBA's headGuardrail layer: Explicit constraints on what the agent executes autonomously versus what requires human sign-off, scoped by environment, subsystem, or operation classWe will demonstrate this using vendor-supplied database management plugins available in the Zowe ecosystem. By exposing operations like DDL generation, schema comparison across subsystems, validation, deployment, and object migration through a CLI interface, an agent can autonomously execute workflows that previously required manual expert coordination — generating DDL for a target object, comparing it against a reference subsystem, validating dependencies, and staging the deployment artifact as a single orchestrated task.
We will also show how the compounding architecture scales across your full mainframe stack. Institutional knowledge encoded once — for Db2, for your SCM tool, for your CICS or MQ environment — becomes a reusable agent capability. Each domain layer builds on the same execution foundation, and the agent chains them together as the workflow demands.
Attendees will leave with a concrete architecture for deploying agentic AI on Db2 z/OS, practical guardrail patterns that make autonomous execution safe in production, and a clear understanding of how the depth of the Zowe plugin ecosystem — open-source and vendor-supplied alike — becomes a direct multiplier for what an AI agent can accomplish.