Directly Responsible Individuals (DRI)

Simon Willison went hunting for a clean definition of "Directly Responsible Individual" and landed on the GitLab handbook's version, which traces the phrase back to Apple. The idea is simple on paper: one person is ultimately accountable for whether a specific project, initiative, or task succeeds or fails. No committee, no shared blame. One name.

Willison's real interest isn't corporate org charts, though. It's what happens to that concept once LLM-powered agents start doing real work inside companies. As these systems take on tasks that used to require a human owner, someone has to decide whether the agent itself can be the DRI, or whether that title has to stay with a person.

His answer is firm. An agent can execute, plan, even make decisions, but it can't be held accountable in the way the DRI framework demands. Accountability requires the possibility of consequences, and a model weighted file doesn't feel anything when a project tanks. Willison frames this as a uniquely human trait, not a technical limitation that better engineering will eventually solve.

He backs this up with a line from an IBM training slide dated 1979, which stated flatly that a computer can never be held accountable and therefore must never make a management decision. Nearly half a century later, with agents drafting code, managing pipelines, and making judgment calls that used to sit with junior staff, that IBM slide reads less like an old-fashioned warning and more like a rule a lot of teams are quietly ignoring.