• Coordination overhead in multi-agent AI often exceeds task cost, yet necessity unclear. • Monotonicity: a task is coordination‑required iff new info can invalidate prior conclusions. • Classic interdependence taxonomy aligns with monotonicity, yielding a decision rule. • Coordination Tax quantifies avoidable coordination overhead for organizational tasks, reducing overhead. • Simulations confirm predictions; 74% of 65 enterprise workflows are monotonic. • O*NET analysis shows 42% monotonic, implying 24‑57% coordination spending unnecessary.
Article Summaries:
- Researchers in multi‑agent systems have shown that coordination is only required when a task is non‑monotonic-i.e., when new information can invalidate earlier conclusions. By mapping a classic taxonomy of organizational interdependence onto this monotonicity criterion, they derived a decision rule and a metric called the Coordination Tax that estimates avoidable coordination overhead. Simulations confirm the theory, and a survey of 65 enterprise workflows (74 % monotonic) and 13,417 occupational tasks from O*NET (42 % monotonic) suggests that 24-57 % of coordination effort may be unnecessary for correctness.
Sources: