Meta Unveils 'Organizational Second Brain' AI Agent to Preserve Expert Knowledge

Meta has built a 'second brain' system based on AI agents to preserve and share organizational expertise. Unlike typical domain-specific agents, this system integrates two layers—structured knowledge architecture and self-improvement loops—to sustain expertise.

The core innovation is compiling expert feedback into verified updates without model retraining. This transforms one-off corrections into the organization's cumulative institutional memory, making it scalable to any enterprise domain based on searchable text.

Architecture and Operating Principles

The system consists of four interdependent layers.

• Knowledge System: Offline processes analyze source documents and refine them into structured knowledge files containing organizational interpretations and constraints. This addresses the inefficiency of re-deriving information from sources during every inference.
• Reasoning Layer: Includes explicit procedures that mimic the actual thought processes of domain experts.
• Evaluation Framework: Gates all changes to ensure quality.
• Improvement Loop: Incorporates expert feedback into the knowledge and reasoning layers to continuously enhance performance.

Practical Value and Scalability

This system helps Meta's SMEs (Subject Matter Experts) reduce time spent on repetitive inquiries, allowing them to focus on complex tasks requiring judgment. It particularly contributes to maintaining evaluation consistency and reducing risk in high-stakes domains such as compliance. While borrowing concepts similar to Andrej Karpathy's LLM Wiki or Google's Open Knowledge Format, it strengthens citation fidelity and institutional consistency, managing over 200 files under a strict classification system.