# Multi-agent system A system in which multiple agents exchange work, state, or decisions under an explicit coordination and authority model. technical-term · Multi-agent & protocols · emerging · Reviewed 2026-08-09 ## Definition A multi-agent system assigns different roles or tasks to multiple model-driven agents and coordinates their communication, shared state, tools, and dispositions. More agents can create specialization or independent challenge, but they also add correlated error, identity, concurrency, security, and accountability risks. Orchestration frameworks are proliferating; dependable coordination, assurance, and governance patterns remain immature. ## Why it matters Multiplicity is an architecture choice, not an automatic capability or quality gain. Independence requires separation of model lineage or deterministic method, context, ownership, and decision authority—not merely different agent names. ## System anatomy - **Roles:** Bounded responsibilities and permitted actions for each agent. - **Coordination:** Routing, delegation, handoffs, conflict resolution, and termination. - **Shared state:** Concurrency-safe facts, artifacts, and task status. - **Authority model:** Which agents can propose, execute, block, or only advise. ## Important distinctions - **Parallel model calls:** Independent calls are not a multi-agent system unless state or decisions are coordinated. - **Independent verification:** A second agent is independent only when its method, lineage or tooling, ownership, scope, and block authority are separated. ## Implementation signals - Use the fewest roles that create measurable value - Give every agent a distinct identity and least privilege - Define ownership of shared state and disagreement - Keep human release authority separate and named ## Failure modes - Agents endlessly delegate to one another - Several agents reproduce the same model blind spot - Shared memory becomes a race condition or prompt-injection channel ## Related knowledge - [Orchestrator or supervisor agent](https://aisdlc.ai/agentic-engineering/orchestrator-supervisor) — A coordinating component that decomposes work, routes tasks, monitors progress, and resolves or escalates outcomes across agents or tools. - [Agent handoff](https://aisdlc.ai/agentic-engineering/agent-handoff) — A bounded transfer of task responsibility, state, evidence, and permitted actions from one agent or human role to another. - [Independent verifier systems](https://aisdlc.ai/agentic-engineering/multi-agent-verification) — A separation-of-judgment architecture in which builder agents, evaluator agents, deterministic checks, domain experts, and authorization authorities challenge different failure surfaces. - [Agent identity & delegated authority](https://aisdlc.ai/agentic-engineering/agent-identity) — The identity and authorization discipline that treats an enterprise agent as a non-human principal with attributable, purpose-bound, time-bound permissions. ## Sources and further study - [DeepLearning.AI — Agentic AI](https://www.deeplearning.ai/courses/agentic-ai/) — A practitioner course organizing agentic systems around reflection, tool use, planning, multi-agent workflows, evaluation, and optimization. **Use:** Official guidance. **Limitation:** Course taxonomy is instructional rather than normative; it does not itself define enterprise authorization or assurance requirements. - [A2A Project — Agent2Agent Protocol v1.0](https://a2a-protocol.org/latest/announcing-1.0/) — A protocol milestone for interoperable communication and collaboration between agents. **Use:** Standard or protocol. **Limitation:** A specification defines an interface or control pattern; conformance alone does not establish authorization, security, or fitness for a use case. - [Anthropic — Harness design for long-running application development](https://www.anthropic.com/engineering/harness-design-long-running-apps) — A planner, generator, and evaluator architecture for multi-hour autonomous application development. **Use:** First-party case study. **Limitation:** This first-party account documents one organization, product, or implementation context and should not be generalized without local evidence. - [NIST NCCoE — Identity and Authority for Software Agents concept paper](https://www.nccoe.nist.gov/sites/default/files/2026-02/accelerating-the-adoption-of-software-and-ai-agent-identity-and-authorization-concept-paper.pdf) — A concept paper exploring how established identity and authorization practices can apply to software and AI agents. **Use:** Official guidance. **Limitation:** Guidance describes recommended practice; citation does not prove that a control is implemented or effective in a particular environment. --- This library synthesizes cited research, standards, official documentation, and clearly attributed practitioner perspectives. Maturity describes the state of a concept—not vendor endorsement, production readiness, or permission to deploy. Benchmarks and demonstrations do not replace use-case evaluation, governed controls, independent verification, or named human release authority.