System of Record

A company might hold a customer’s address in its CRM, its billing tool, a data warehouse and a dozen spreadsheets. Only one of those is the system of record for that field: the place where changes are made first and the copy everyone else defers to. Salesforce for accounts, Workday for employees and an ERP for orders are common examples.

For AI agents, systems of record matter twice. They are the trustworthy source an agent should read when it needs facts, which is why they show up in discussions of the context layer. And they are where an agent’s actions land: closing a ticket or updating a deal is a write to a system of record, usually through an API or a Model Context Protocol server. That second role is why permissions and audit trails on those writes need the same care as for a human user.

In episode 61 at 8:12, LlamaIndex’s Jerry Liu lists connectors to tools like Confluence or Salesforce as providers of context, and says the idea of a system of record will stick around because agents need a way to store, act and operate on data.

Neo4j’s Sudhir Hasbe draws a finer line in episode 55 at 9:15 and at 22:58: operational databases are the system of record for the present, and he describes a CRM such as Salesforce as a system of record for decisions. The episode was sponsored by Galileo; the read is at 1:37.