AT-301b · Module 1

Responsibility Matrices

4 min read

A role without boundaries is a liability. At 20 agents, undefined boundaries create 190 potential overlap pairs — that is n*(n-1)/2 collision surfaces. Each collision means duplicated work, contradictory outputs, or worse, both agents modifying the same artifact simultaneously. The RACI matrix adapted for agent teams solves this: every task gets exactly one Responsible agent, one Accountable lead, zero or more Consulted specialists, and zero or more Informed observers.

The methodology is straightforward. List every task your team performs. For each task, assign exactly one R. If two agents share R on the same task, you have a design flaw — not a feature. HUNTER finds leads. CLOSER works them. LEDGER tracks pipeline. No overlap. No ambiguity. The matrix is the contract.

Do This

  • Assign exactly one Responsible agent per task — zero ambiguity
  • Document negative space: what each agent explicitly does NOT do
  • Review the matrix weekly for drift — roles expand silently

Avoid This

  • Share responsibility between two agents — "they both handle it" means neither owns it
  • Define roles by capability instead of task — agents will optimize for breadth
  • Assume the matrix stays accurate without auditing — entropy is real