Knowledge systems · LinkedIn adaptation
What should an agent remember as a fact?
A domain model can help an agent distinguish observations, interpretations, and confirmed facts.
The important design choice in agent memory often comes before storage. An agent has to represent what it knows, where it learned it, and how certain it should be. Ravi used a small health-record prototype to demonstrate the difference between a reported fact and a hypothesis drawn from evidence.
Give evidence and interpretation their own place
Consider a support assistant. A customer reports a service interruption. Logs show a deployment at about the same time. An outage caused by that deployment is a hypothesis. If the agent records all three statements with the same status, a future answer can quietly turn that hypothesis into history.
Write down the distinctions your workflow needs
Start with three fields: what was observed, where it came from, and who has confirmed an explanation. Test whether the agent can update the record when new evidence arrives and explain why its answer changed. The shape of the record matters whether it lives in a file, a database, or a graph.