How Governed AI Agents Work: Skills, Context, and Guardrails

Description

What actually makes an AI agent trustworthy enough to use with real business data? It comes down to architecture: skills, tools, context, and guardrails working together.

In this session, we break down how governed AI agents are built - combining skills, tools, and business context to reason, act, and stay fully auditable at every step.

The conversation covers:

• Skills as building blocks: how skills act as recipes that define which analytical tasks an agent can perform, and how customers control which ones are turned on.
• Tools and execution: how skills connect to tools that carry out actions—like creating a metric or scheduling an export through governed APIs.
• The context layer: how the Analytics Catalog, AI Knowledge, and AI Memory give agents the business context they need to reason accurately.
• Why agents never touch raw data: how the LLM only generates a query definition, while a deterministic query engine handles execution, keeping results accurate and self-correcting.
• Full observability: how every token in and out of the LLM is captured for complete auditability.
• Self-improving agents: how specialized agents like the Semantic Quality Agent and AI Memory Agent continuously monitor and improve the system.

Watch this session to see the architecture that makes AI agents both powerful and governed - without compromising accuracy or control.