World's Fair 2026
Intrinsic + Extrinsic + Learned Knowledge — Pablo Castro, Distinguished Engineer & CVP, Microsoft
Overview
This talk explores the three categories of knowledge that power AI agents: intrinsic, extrinsic, and learned. Intrinsic knowledge, embedded within models through training data, has been a primary driver of recent AI advancements, particularly in coding. Extrinsic knowledge, accessed through retrieval-augmented generation (RAG) patterns, allows agents to ground themselves in organizational data and external information. Learned knowledge emerges from agents' ongoing work, enabling continuous improvement and the capture of unique organizational capabilities.
Who should watch
- AI Engineers
- Product Managers
- Builders of AI agents and systems
- Those interested in grounding AI agents with data
- Individuals looking to optimize agent performance through learning loops
Key takeaways
- Intrinsic knowledge, inherent in models from training data, has fueled significant progress, especially in coding agent capabilities.
- Extrinsic knowledge, accessed via sophisticated retrieval systems, is crucial for grounding agents in organizational context and external data.
- Microsoft Foundry offers a platform for integrating various models and managing knowledge bases, supporting both simple and complex retrieval needs.
- Retrieval systems have evolved from basic vector search to complex, combined methods that improve performance in real-world scenarios.
- Agentic retrieval allows systems to reflect on data and determine if information needs are met, enhancing performance in complex cases.
- Learned knowledge is generated through agent optimization, creating learning loops that capture and leverage unique organizational insights.
- The agent optimizer in Foundry facilitates evaluating, generating, and deploying improved agent configurations based on performance data.
- Microsoft IQ provides a suite of capabilities, including Work IQ, Fabric IQ, and Web IQ, to connect agents to diverse organizational and public data sources.
Notable quotes
*Intrinsic knowledge in models has been a key inflection point, powering experiences even before large language models became widely known.*
*The evolution from simple vector search to sophisticated retrieval systems, often combining multiple methods, yields better results.*
*Learned knowledge, captured through agents performing work and optimizing processes, can effectively embed what's unique about an organization.*
Unofficial community note. Prefer the recording for nuance.