Assign exactly one agent as accountable for each decision…
When designing RACI-style accountability for cognitive agents, assign exactly one agent as 'accountable' (final decision authority) for each contested decision or resource, allowing other agents to be consulted or informed but not to hold veto power.
When This Fires
- When Define non-overlapping agent scopes — scope collisions… (non-overlapping scopes) can't fully eliminate agent overlap for a specific decision
- When the same decision triggers input from multiple agents with different preferences
- When designing the priority hierarchy (Resolve inter-agent conflicts with documented priority…) for contested decisions
- When internal conflict repeatedly arises around the same decision type
Common Failure Mode
Shared accountability: "Both my health agent and social agent get a vote on food decisions." Shared accountability is no accountability — it produces the same deliberation-at-decision-time that agents were designed to eliminate. One agent must own the decision. Others can provide input (consulted) or be notified (informed), but the accountable agent decides.
The Protocol
(1) For each decision type that triggers multiple agents, designate one agent as Accountable: this agent has final authority. (2) Designate other relevant agents as Consulted (provide input before the decision) or Informed (notified after the decision). (3) The accountable agent considers consulted agents' input but is not required to follow it. (4) Document the RACI assignment for each contested decision type. (5) When a contested decision arises, the accountability assignment resolves it without real-time deliberation: the accountable agent's recommendation wins. (6) Review assignments quarterly — the right accountable agent for food decisions at one life stage may not be the right one at another.