Summary
Cole Medin tackles a problem that most AI second-brain guides skip entirely: memory decay. After more than six months building and running his own AI second brain, Medin explains why these systems inevitably accumulate stale, contradictory, or outright incorrect information — and how to engineer a solution that detects and repairs it automatically.
The video walks through the typical three-layer architecture shared by most second brains (core memory documents, daily logs, and a knowledge graph of entities and concepts) and shows why each layer creates its own contradiction risks. Using a fictional example of an agency owner named Dana whose client retainer changed from $4,000 to $9,500 per month across multiple conversations, Medin demonstrates how agents can silently pull from outdated records and produce wrong answers with apparent confidence.
The practical fix centers on building a reconciliation layer that runs scheduled audits across the knowledge base, identifies documents that speak to the same entity in conflicting ways, and resolves or flags the discrepancies. Medin also demonstrates Granola — a meeting notes tool with an MCP connector — as a way to feed structured, dated information into the second brain to reduce the staleness problem at the source. The video includes live agent interactions and is aimed at builders who already have a second brain running and want to make it reliably accurate over time.
📺 Source: Cole Medin · Published August 07, 2026
🏷️ Format: Tutorial Demo
