Enterprise AI agents often fail at the point where knowledge must become controlled action. Retrieval-augmented generation can ground an agent in the correct documentation, but it does not guarantee that the agent will follow required steps, perform validations, respect approval boundaries, or apply operational policies consistently.
This session presents an enterprise architecture that combines Microsoft Foundry IQ with procedural memory in Foundry Agent Service. Foundry IQ provides permission-aware, citation-backed access to enterprise knowledge across runbooks, policies, architecture documents, and incident records. Procedural memory complements this knowledge layer by extracting reusable execution patterns from audited agent trajectories—capturing when a procedure applies, which actions are required, and which controls must be enforced.
Using an IT incident-management scenario, we will compare a knowledge-grounded agent with one enhanced by procedural memory. The demonstration will show how the agent progresses from retrieving a runbook to executing it consistently while preserving human approval for sensitive actions.
We will also address production considerations including memory isolation, retention and TTL, auditability, access control, observability, and protection against stale or unsafe memories. Attendees will leave with a practical pattern for building governed agents that can transform enterprise knowledge into reliable, repeatable workflows.