AI Agent Orchestration Platform  ·  View 18 of 32  ·  4 · Runtime

Multi-Agent Coordination

Six coordination patterns, and the claim that all six are workflow definitions rather than different platforms.

Editable source SVG draw.io All views
Trigger Decompose Delegate Execute Converge Supervisor and worker Single goal Supervisor splits Fan-out to workers Isolated contexts Supervisor merges Planner and executor Open task Plan as DAG Steps to queue Executor per step Replan on failure Hierarchical teams Programme goal Team leads Depth limit 3 Team sub-runs Roll up by level Peer collaboration Shared brief No fixed owner Message passing Shared scratchpad Quorum or timeout Debate and critic Draft answer Critic prompt Round cap 3 Revise and rescore Best scored wins Specialist routing Ambiguous intent Capability match Registry lookup One specialist Return or reroute Multi-Agent Coordination — Six Patterns on One Runtime Application we own Interface / broker Decision point Queue / topic Security / platform Data store All six run on the same orchestrator and the same worker image. The pattern is a workflow definition, not a different platform. v 1.0 · owner Data & AI Global Practice · date 2026-08

Decisions

  • The pattern is expressed in the workflow definition; the orchestrator and the worker image are the same for all six
  • Context is isolated by default and shared only where the workflow declares it, because shared context is how one agent's error becomes every agent's premise
  • Agent-to-agent communication is explicit message passing through the orchestrator, not direct calls — otherwise there is no trace and no policy check

Coordination safety

  • Delegation depth capped at 3 and debate rounds capped at 3, both from the workflow spec
  • A coordination timeout on every fan-in; a missing worker result converges with a partial answer rather than hanging
  • Deadlock prevention by construction: the delegation graph is acyclic and validated at publish, and dynamic delegation may not revisit an ancestor

Risks

  • Cost grows superlinearly with agents; the per-run token budget is enforced across the whole tree, not per agent
  • Dynamic delegation makes cost and latency hard to predict before the run; those workflows carry a lower default budget ceiling
  • Peer collaboration with no fixed owner is the hardest pattern to debug; it is Phase 2 and gated on the trace viewer being good enough
◀  Execute an Agent All views Durable Execution and Replay  ▶
Solution Architecture v1.0 · Data & AI Global Practice · 2026-08 · 32 views · Microsoft Azure18-multi-agent-patterns