Why we put memory in files instead of vectors.
A long-running argument about how to structure an agent's memory, and the operational reasons we landed on flat MCP files instead of an embedding store.
Practical writing about building agentic systems for real businesses. Lessons from deployments, methodology refinements, architectural decisions, and the occasional opinion.
A long-running argument about how to structure an agent's memory, and the operational reasons we landed on flat MCP files instead of an embedding store.
Most agents will tell you they finished a task. A working agent verifies first, then reports.
What changed between deployment one and deployment three, and what stopped changing.
One email per new post. Nothing else. Unsubscribe in one click.