AI works well when it understands the context in which it is operating. Yet much of that context has never been written down.

Systems carry years of decisions, exceptions and compromises that may look like mistakes to anyone—or any AI—without access to their history. The obvious correction can easily break something that was done deliberately.

Before CMS Connect 26 in Montréal, Andrei Kalamkarov wrote about the memory he was building for AI. His conversations at the conference expanded that thinking: perhaps this is not only about helping an AI work with one system, but about preserving institutional memory more broadly.