There is an old philosophical problem. Hume argued that memory does not just record the past; it constitutes the self. What you remember shapes what you perceive, what you decide, and what you ignore.
We tend to believe that the era we live in is unique. For the most part, it is not because people are people, and their hopes and fears remain the same over the course of evolution. But, in some ways, they are unique, and this is where we find the small gap of what matters.
Recently, the news was full of stories about Anthropic running pilots inside Fortune 100 companies, where agents monitor internal chats, documents, and meetings and begin surfacing patterns that influence who gets hired, what gets funded, and how risk gets assessed. We do the same thing locally at Intradiegetic, although not at the same scale, but with A LOT more impact.
The agent reads the stream of organizational life and acts on it; indeed, it is reading what I am writing right now as it is saved in the general writing corpus and ingested. The impact: constant comparison to other activities going on in the company; what has been written in the past and is being worked on right now; and what belongs only to us at Intradiegetic. It is local AI, as a partner, not a service.
Institutional memory has always encoded the priorities, biases, and blind spots of the people who built it. The difference now is that this memory is being processed at machine speed, with recommendations attached.
That creates a structural problem that most firms are not ready to answer. When an agent surfaces a pattern in real time, who decides what a pattern worth surfacing looks like? Where does the accountability architecture live? These are governance questions dressed in technical clothes.
The data these agents learn from is not neutral. It reflects the organization as it was, not necessarily as it should be. A system trained on your internal communications will learn your politics, your silences, and your defaults. The question worth asking is whether you want those defaults encoded into the decision layer.
The real shift is in how we think about knowledge: do we expect to file it, or is it to be read back in real time?
That is a fundamentally different relationship with organizational memory. And the design challenge is not how to make the agents faster or cheaper. It is about building systems that can surface what the organization does not already know it believes.
Hume would be awed. He might also remind us that what we choose to remember is already a decision.
Leave a Reply
You must be logged in to post a comment.