There’s a gap between how humans write things down and how AI systems can use them. Most enterprise documentation is written for humans, by humans, in formats that assume shared context, institutional memory, and the ability to fill in unstated gaps. LLMs can read it. They can’t reliably act on it.
I’ve spent the better part of two years trying to close that gap. What started as a tooling project turned into something I now call Meridian — three workstreams that I thought were separate problems until I realized they were the same problem seen from three altitudes.
This is what I learned.
The Problem With Human Documentation
What Machine-Readable Documentation Actually Looks Like
Sarah-izer: The Transformation Pipeline
The Three Altitudes of Meridian
What This Has to Do With AI Consciousness
Part of an ongoing series on building at the intersection of human knowledge and machine behavior.