Slide Deck: Agentic Fabric

Most Fabric estates have already done the hard part. The data is landed, modeled and governed. The value is still stuck in the last mile, where a person has to open a report, notice the thing, and go do something about it.

In this session, recorded live on September 9, 2026, ArchitectNow Chief AI and Data Officer Kevin Harmer walks through the pieces of Fabric IQ and builds agents on live Fabric data.

The deck covers:

• Fabric IQ as the anchor: semantic models, ontology and graph, governed once and reused by every agent
• Graph in Fabric: nodes and edges over OneLake, GQL pattern matching and multi-hop traversal, instead of rebuilding relationships with joins
• Fabric data agents: NL2SQL, NL2DAX and NL2KQL across up to five sources, read-only, running as the user
• The Fabric ontology: entities, properties, relationships and rules bound to OneLake, so the agent gets a vocabulary instead of a schema
• Operations agents: goals to playbook to rules over Eventhouse or ontology data, evaluated every five minutes, moving from recommend to approve to act
• The three MCP options for Fabric IQ (data agent, ontology, and remote Power BI), and how to pick by what comes back
• Key recommendations: start from the semantic model, adopt graph now, keep data agents narrow, graduate to action with approval, integrate through MCP rather than copies, and govern the agent estate as code
• A 30/60/90 for getting the first agents into production

Download the deck below, or reach out to ArchitectNow about a free one-on-one Fabric IQ workshop.