Verification and repair

Rationale grounded in precedent

A rationale grounded in precedent is an explanation of an autonomous action that points at a rule or preference the person previously set, rather than at the agent’s decision process.

Control sits with the agent

When to use it

Any action taken without asking, particularly one triggered by an external event the person did not see.

What it prevents

Valid autonomous actions being read as bugs, which stops people forming an accurate model of what the agent will do next.

Why this happenedacted without asking
Moved the Acme renewal to next Tuesday.
BecauseYou asked for renewals never to land on a Friday.
Set by youIn the standing brief for account work

Where it stands

Partly, in agents that cite a stored preference when they act. Rarely done well.

Sources

Related patterns: Autonomy dial · Action audit and time-limited undo

All agent design patterns

Review an agent feature with us30 minor send a message