Building a truth layer for persistent agent memory

Over the past year, the main thing blocking me from automating real workflows with AI agents hasn't been intelligence. It's trust in the state they accumulate. I wrote about why it keeps failing and the system I'm building in response: Neotoma.