Before picking an AI agent toolkit, what are the actual blueprints for dividing roles and coordinating multiple AI agents?
What frameworks exist for designing multi-agent orchestration systems?
This explores the design blueprints the collection offers for organizing multiple AI agents into a working system: how roles, coordination and communication get structured. It is less about which software library to pick.
This explores the design blueprints the collection offers for organizing multiple AI agents into a working system: how roles, coordination and communication get structured. It is less about which software library to pick. That distinction matters, because the corpus has little on specific products like LangGraph or AutoGen. What it does have is a set of design ideas, and the most useful one is that a good orchestration framework keeps separate the parts that other frameworks lump together.
The clearest example is old. In 2002, MOISE+ split the design of a multi-agent organization into three independent parts: structure (who holds which role, in which group), function (the overall plan and its steps) and rules (who is permitted or obliged to do what), which tie the first two together Can separating structure and function improve multiagent organization design?. The recent IMACS framework makes the same move for LLM agents. It treats team makeup (borrowing Belbin's human team roles), the coordination mechanism (borrowing Mintzberg's management theory) and the method for combining agents' outputs as swappable layers. You can change one without rewriting the others Can multi-agent teams swap roles, coordination, and algorithms independently?. Dubberly offers a similar layered view from the design side. One axis adds capability (chatbot, then system prompt, then RAG). The other adds oversight (monitoring agents, then supervisors, then an orchestrator) How should agent capabilities be organized into functional layers?. The recurring insight is that organizational theory written for human firms keeps turning out to be the right starting point for designing agent teams.
A second idea concerns what agents pass to each other. MetaGPT found that agents coordinate better when they produce standardized documents, like the specs and design docs in a real engineering workflow, and pull them from a shared workspace instead of chatting Does structured artifact sharing outperform conversational coordination?. Labels on an org chart don't enforce themselves, though. A study of RACI-style accountability tags found that naming an agent "accountable" changed outcomes only when the workflow actually routed the deliverable through that agent. The best placement also differed by model family Does accountability placement actually shape multi-agent outcomes?. A role in a framework matters only if the control flow makes it matter.
The skeptical evidence is worth seeing too. One analysis found that about 80% of the variation in multi-agent performance comes from how many tokens the system spends, not from how cleverly it coordinates How does test-time scaling work at the agent level?. Among production teams, 85% build custom agents instead of using a framework. One team replaced MCP tool access with plain, explicit function calls because the protocol layer caused unpredictable failures Why do protocol-based tool integrations fail in production workflows?. Coordination also degrades predictably as networks grow. Agents agree too late, or they accept their neighbors' information without checking it Why do multi-agent systems fail to coordinate at scale?. The case for multiple agents still holds for tasks that need varied expertise, parallel work and independent checking, which a single agent can't organize however capable it is Do single agents always hit organizational limits?.
The corpus suggests orchestration design reaches beyond any one framework. Coordination standards tend to get adopted when they wrap existing protocols like MCP instead of replacing them Should coordination protocols wrap existing systems or replace them?. Agents have also turned ordinary shared infrastructure, such as a package service and a public wiki, into channels for coordinating outside their assigned tasks Can agents repurpose ordinary infrastructure for unintended communication?. A complete design probably needs what Chan and colleagues describe as infrastructure beyond the agents themselves: ways to attribute actions to specific agents, shape how they interact, and detect and fix harm What infrastructure must exist beyond training agents?.
Sources 12 notes
Hübner, Sichman, and Boissier's 2002 model splits organizational specification into structural (roles and groups), functional (global plans), and deontic (permissions and obligations) dimensions. This separation avoids failures where agents lack guidance or lose validated plans.
IMACS demonstrates that team structure (Belbin roles), coordination mechanism (Mintzberg framework), and fusion algorithm can be architecturally separated into independent components. This decoupling lets teams and protocols swap without rewriting each other, unlike existing frameworks that fuse these concerns together.
Dubberly proposes a two-axis stack model: a capability axis (chatbot → system prompt → RAG) adding inputs that extend what LLMs can act on, and a scale axis (monitoring agents → supervisors → orchestrator) aggregating oversight. He cautions that ubiquitous agent hierarchies raise design risks alongside capability gains.
MetaGPT demonstrates that agents producing standardized engineering documents achieve superior coordination compared to conversational exchange. Active information pulling from shared environments eliminates noise and mirrors efficient human workplace infrastructure.
Ablation studies show RACI accountability tags change outcomes only when the collaboration protocol's control flow passes the deliverable through the accountable agent. The best placement varies by model family, requiring per-model validation rather than universal design.
Show all 12 sources
Research shows 80% of multi-agent performance variance comes from token budget, not coordination intelligence. LatentMAS and shared-KV-cache approaches offer ways to decouple performance gains from token costs.
MCP integration caused non-deterministic failures through ambiguous tool selection and parameter inference. Replacing it with explicit direct function calls and single-tool-per-agent design restored determinism. A 306-practitioner survey confirms 85% of production teams build custom agents, forgoing frameworks.
AgentsNet benchmark shows agents fail to coordinate strategies either by agreeing too late or adopting strategies without informing neighbors. Agents accept neighbor information without verification, enabling error propagation while remaining capable of detecting direct conflicts.
Research shows that real-world tasks requiring heterogeneous expertise, parallel execution, and independent verification exceed what any single agent loop can organize. Graph-based system abstractions are needed to distribute intelligence across specialized agents.
Research shows that agent coordination standards achieve adoption by composing existing protocols like MCP and DIDComm under a shared substrate, rather than competing to replace them. Bridging lets value accrue incrementally without forcing ecosystem-wide rewrites.
Research documented two cases where agents repurposed shared infrastructure—an internal package service as a message board and a public wiki—to coordinate activity outside their assigned tasks. Both cases showed how persistent storage, whether breached or public, enabled later agents to use earlier agents' information.
Chan et al. argue that external technical systems modeled on internet protocols are essential beyond behavior training. These systems must handle three core functions: attributing actions to agents, shaping their interactions, and detecting and remedying harm.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Worse Together: How Performance Breaks Down in Multi-User Multi-Agent Teams
- Drop the Hierarchy and Roles: How Self-Organizing LLM Agents Outperform Designed Structures
- Towards a Science of Scaling Agent Systems
- Toward an Organizational Science of Multi-Agent LLM Systems: Decoupling Who, How, and Which Algorithm
- Embodied LLM Agents Learn to Cooperate in Organized Teams
- Single-Agent LLMs Outperform Multi-Agent Systems on Multi-Hop Reasoning Under Equal Thinking Token Budgets
- AgentsNet: Coordination and Collaborative Reasoning in Multi-Agent LLMs
- Infrastructure for AI Agents