{
  "id": 20341070,
  "title": "Taxonomy of multi-agent design patterns, anti-patterns, and lessons from manufacturing",
  "url": "https://urgent.news/2026/11/16/taxonomy-of-multi-agent-design-patterns-anti-patterns-and-lessons",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-16T21:53:46.000Z",
  "source": {
    "name": "AWS Machine Learning",
    "slug": "aws-machine-learning",
    "url": "https://aws.amazon.com/blogs/machine-learning/taxonomy-of-multi-agent-design-patterns-anti-patterns-and-lessons-from-manufacturing/"
  },
  "original_language": "en",
  "account": "Enterprise teams are increasingly developing multi-agent systems, but often face failure due to insufficient distributed systems experience. Studies show that approximately 88% of AI pilots never reach production. Stanford HAI research demonstrates that when coordination overhead outweighs the benefit, two models collaborating can perform worse than when operating individually. This highlights a clear pattern: teams add complexity at a faster rate than they develop understanding of the system.\n\nTen steps that are each 95% reliable only result in an overall end-to-end reliability of 59.9%, while 20 such steps lead to 35.8% reliability. This illustrates the risk that each dependent handoff introduces, as each step multiplies the potential for failure. In this article, we explore how to determine whether a use case requires multiple agents, select an appropriate coordination pattern, and identify costly anti-patterns early on. We also examine how factors like takt time, safety boundaries, edge data, and shift cycles influence the architecture.\n\nCoordination patterns in multi-agent systems can be categorized into two tiers based on who controls the flow. Tier 1 coordination patterns involve predefined control flow, while Tier 2 patterns allow the AI model to determine the flow at runtime.\n\nTier 1 patterns include deterministic control flow, where the path is defined by the team. The Sequential Pipeline pattern implements linear execution where each node's output feeds into the next, maintaining predictable execution timing. A typical example is a document processing ETL chain (extract - transform - validate). Another Tier 1 pattern is the Conditional Workflow, where logic-based edges select the next node based on conditions. This pattern enables branching and sending each case down a different path. A retry mechanism up to N times can be enforced to prevent unbounded iteration, as seen in a pass/fail workflow that escalates to a human if the maximum number of retries is reached. The Conditional Router pattern classifies the input and directs it to the appropriate processing path, making decisions at the entry point. This approach is ideal for intent-driven dispatch, quality-gate decisions, customer inquiry routing, and anomaly triage.\n\nTier 2 autonomous patterns rely on the AI model to determine the control flow at runtime, offering flexibility for work with undefined steps. Plan-based orchestration is a common autonomous pattern, where a single orchestrator agent acts as a project manager. It maintains an explicit task plan with assignments and dependencies, dynamically assigns work to stateless worker agents, and monitors progress. If a worker fails, only that worker is retried, leaving the other agents unaffected. This approach reduces the risk introduced by each dependent handoff compared to Tier 1 patterns.\n\nIn summary, when developing multi-agent systems, it is crucial to carefully consider the coordination pattern based on the use case requirements. Starting with Tier 1 patterns and progressively moving to Tier 2 autonomous patterns as needed can help mitigate risks and improve overall reliability. Understanding the impact of coordination overhead and potential anti-patterns is essential for successful multi-agent system implementation.",
  "summary": "Multi-agent coordination is a distributed systems problem, not a model problem. This post gives a field-tested taxonomy of multi-agent design patterns and anti-patterns, a framework for deciding when multiple agents are warranted, and how to build cost-efficient, reliable systems on AWS.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}