DEVOURED
Context graphs and decision traces to the rescue
A December 2025 Foundation Capital paper introduces "context graphs" and "decision traces" as a crucial missing provenance layer for enterprise RAG and AI agent workflows.
What: A December 2025 paper from Silicon Valley VC firm Foundation Capital, titled “AI’s trillion-dollar opportunity,” introduced the concept of a "context graph." This knowledge graph is designed to capture "decision traces," which record the full context, reasoning, and causal relationships behind critical business decisions to provide provenance for AI agents and RAG systems.
Why it matters: This signals a growing industry focus on the need for explainable AI and robust operational memory in enterprise environments. Generic AI models struggle with specific business contexts, highlighting that effective enterprise AI requires understanding not just data, but also the historical *how* and *why* of decisions to prevent AI from hallucinating.
Takeaway: If working on enterprise AI or agent systems, explore the concepts of context graphs and decision traces to consider how to log and structure operational memory for better AI reasoning and explainability.
Deep dive
- A December 2025 paper by Foundation Capital, "AI’s trillion-dollar opportunity," introduced the concept of "context graphs" and "decision traces."
- A context graph is a knowledge graph designed to capture "decision traces," which record the observable reasoning and causal relationships behind past business decisions.
- This approach is seen as a missing provenance layer for retrieval-augmented generation (RAG) and AI agent workflows in enterprises.
- It complements existing enterprise systems (like ERPs, CRMs) by storing operational memory: what evidence mattered, which relationships were relevant, policies applied, and exceptions made.
- The article argues that AI needs all three types of human memory: episodic (decision traces), semantic (facts/schemas), and procedural (skills/how-to), to prevent hallucinations.
- Graphs are crucial because enterprise context is fundamentally based on connections and relationships, which vector search struggles to capture.
- A context graph can act as a "graph of graphs," guiding AI agents to the correct underlying data sources (e.g., accounting database for accounting questions).
- GraphRAG, which improves retrieval from graphs for AI systems, is an essential part of this emerging ecosystem.
- The market is evolving quickly, and enterprise AI teams should explore innovations like context graphs without setting architectures in stone yet.
Decoder
- Context graph: A knowledge graph designed to capture and map the full context, reasoning, and causal relationships behind critical business decisions and organizational data.
- Decision traces: Records of the observable reasoning behind how decisions were made, including rule application, exceptions, conflict resolution, approvals, and governing precedents.
- GraphRAG (Retrieval-Augmented Generation in graphs): A technique that uses graph databases to store and retrieve structured knowledge, improving the ability of AI systems to access and utilize this knowledge for generating more accurate and contextually relevant responses.
Original article
Full article content is not available for inline reading.
Explore all 58 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi