This post is part of an ongoing series examining the forces reshaping how organizations establish truth, make decisions, and maintain control in an AI-driven world — from the economics of intelligence to the architecture of what comes next.
The Faster Filing System Problem
Foundation Capital's trillion-dollar thesis on context graphs is correct about the problem and wrong about the solution. They're right that AI needs context: temporal, spatial, semantic, provenance, and confidence dimensions [10–12]. But their vision is fundamentally static: a map of known relationships that AI can query.
That's not intelligence, that's a faster filing system.
Consider their example: an LLM with a context graph queries "What is our refund policy for this specific customer?" and traverses nodes to find policy documents, customer history, temporal validity, and exception cases [13, 14]. Better than pure LLM retrieval, yes. But it assumes someone already mapped those relationships, that relevant context is documented, that relationships are static, and that you know what questions to ask.
This works for structured workflows where relationships are explicit. But the vast majority of organizational knowledge is informal, context-dependent, undocumented understanding that lives in people's heads, in hallway conversations, in the "we've always done it this way" decisions that never get written down.
The Real Problem: Hidden Relationships
Real intelligence isn't querying known relationships, it's discovering hidden ones — patterns that emerge from data that were never explicitly encoded. It's understanding why a decision was made even when the reason was never documented. It's capturing the informal context that makes an experienced engineer say "that won't work" before running a single test.
Foundation Capital's static context graphs fail on four dimensions:
1. No Discovery Mechanism — They can't find relationships you didn't know to encode, identify emergent patterns across disparate data, or discover why certain decisions consistently produce better outcomes.
2. No Tacit Knowledge Capture — Organizational knowledge lives in people, not documents. Decisions are informed by informal context: market sentiment, team dynamics, unstated assumptions. Critical "why" reasoning rarely gets documented in queryable form.
3. No Temporal Evolution — Relationships change as systems evolve, but static graphs don't adapt. What worked last quarter may not work this quarter, and the graph doesn't know why.
4. No Self-Organization — They require human ontology design and maintenance. They break when systems change or new domains emerge. They cannot handle unstructured, real-world data.
The trillion-dollar opportunity isn't in building better maps of what we already know. It's in building dynamic autonomous graphs that discover what we don't know we need to know and capture the tacit knowledge that makes those discoveries actionable.
Next up: what dynamic autonomous graphs actually look like in practice — the four capabilities that separate systems that query reality from systems that discover it.
Works Referenced
Foundation Capital. "Context Graphs: AI's Trillion-Dollar Opportunity." foundationcapital.com, 2024.
"What Are Context Graphs." simple.ai, 2024.
"How Context Graphs Help AI Understand and Automate Real Work." intellectyx.ai, 2024.
"Context Graphs." usefluency.com, 2024.