The Three IQs: Ground Your Agents in Knowledge, Data, and Work | LIVE171
Marco Casalaina and Ayca Bas explain how to build grounded, enterprise-ready agents by delegating context to three “IQ” systems: Foundry IQ for knowledge, Fabric IQ for data, and Work IQ for human/work context, with a live demo showing how this approach avoids brittle, hand-wired pipelines.
Overview
The session argues that agents shouldn’t try to “hold” all context inside the model prompt. Instead, they should delegate to reusable intelligence layers that provide the right context at the right time.
The three IQs covered are:
- Foundry IQ (knowledge grounding): grounding an agent in curated knowledge so it can answer and act with the right enterprise context.
- Fabric IQ (data grounding): grounding an agent in organizational data (analytics/data estate) so it can reason over and use data as part of its work.
- Work IQ (human/work context grounding): grounding an agent in the context of work (people, tasks, communications) so it can operate in a way that matches how work actually happens.
Session structure (from chapters)
Discussion topic introduced: the IQ framing
- The presenters introduce the “IQ” concept as a way to think about where an agent gets context from.
The “Four IQs” list
- The session references a broader set of IQs (including Web IQ) and then focuses on the three highlighted in the title: Foundry, Fabric, and Work.
Collaboration background
- The presenters provide background on the project collaboration and how the IQ approach is intended to be reused across agent experiences.
Improving agent performance with refined instructions
- They discuss improving agent behavior by tightening and refining instructions, rather than relying on ever-growing prompts.
Live demo: agent behavior and next steps
- A live demo shows the same agent in action and how the approach supports iterative improvement.
Demo: agent checking email and sending replies
- The demo includes an agent checking its own email and sending replies, illustrating “work context” grounding.
Independent identities for agents
- The presenters call out the value of agents having independent identities for both security and functionality.
Multi-IQ interactions
- The session transitions to showing interactions between multiple IQ systems working together.
IQ Series modules
- They point viewers to the broader IQ Series modules for further learning: https://aka.ms/iq-series