Skip to content

AGT-06 · Multi-agent collaboration: choose, build and operate topologies

Teams add more agents when one would do, then struggle with agents passing work back and forth, overwriting shared state and multiplying cost. Incidents are hard to trace because no one can see which agent did what.

  • Who reads it: Solution architects and GenAI engineers designing multi-agent systems and DevOps engineers running them
  • Outcome: The simplest topology that works, with hop limits and traceable handoffs
  • Pack owner: James (catalog owner role: Agent Lead)

Master catalog entry

  • Catalog name: Multi-Agent Collaboration Templates (P2)
  • Why it matters: Enables the complex workflow use cases that justify large programmes
  • What it removes: Handoff, shared state and conflict resolution patterns pre-built
  • Stack: LangGraph, AutoGen patterns, Semantic Kernel agents
  • Maturity target: Level 2 — supervisor, sequential, parallel and debate topologies

Playbooks

ID Story Title Kind Complexity Urgency Status Owner Planned publish
PB-AGT-06-01 AGT-06-S1 Choose a multi-agent topology: sequential, parallel, supervisor, handoff or debate Decision guide High 2 draft Sivaraj 2026-12-14
PB-AGT-06-02 AGT-06-S2 Build a supervisor with workers and hop limits in LangGraph How-to Medium 3 draft Sanjay Sarvesh 2026-12-18
PB-AGT-06-03 AGT-06-S3 Build group chat, handoff and Magentic orchestrations in Microsoft Agent Framework How-to High 2 draft James 2026-12-09
PB-AGT-06-04 AGT-06-S4 Investigate a multi-agent incident: ping-pong, conflicting writes or cost spike Runbook High 5 draft James 2026-11-24