Anthropic

{
  "article_type": "research",
  "authors": [],
  "chars": 688,
  "content": "## When (and when not) to use agents\n\nWhen building applications with LLMs, we recommend finding the simplest solution possible, and only increasing complexity when needed. This might mean not building agentic systems at all. Agentic systems often trade latency and cost for better task performance, and you should consider when this tradeoff makes sense.\n\nWhen more complexity is warranted, workflows offer predictability and consistency for well-defined tasks, whereas agents are the better option when flexibility and model-driven decision-making are needed at scale. For many applications, however, optimizing single LLM calls with retrieval and in-context examples is usually enough.",
  "observed_at_ms": 1792517470655,
  "outline": [
    {
      "anchor": "what-are-agents",
      "level": 2,
      "text": "What are agents?"
    },
    {
      "anchor": "when-and-when-not-to-use-agents",
      "level": 2,
      "text": "When (and when not) to use agents"
    },
    {
      "anchor": "when-and-how-to-use-frameworks",
      "level": 2,
      "text": "When and how to use frameworks"
    },
    {
      "anchor": "building-blocks-workflows-and-agents",
      "level": 2,
      "text": "Building blocks, workflows, and agents"
    },
    {
      "anchor": "building-block-the-augmented-llm",
      "level": 3,
      "text": "Building block: The augmented LLM"
    },
    {
      "anchor": "workflow-prompt-chaining",
      "level": 3,
      "text": "Workflow: Prompt chaining"
    },
    {
      "anchor": "workflow-routing",
      "level": 3,
      "text": "Workflow: Routing"
    },
    {
      "anchor": "workflow-parallelization",
      "level": 3,
      "text": "Workflow: Parallelization"
    },
    {
      "anchor": "workflow-orchestrator-workers",
      "level": 3,
      "text": "Workflow: Orchestrator-workers"
    },
    {
      "anchor": "workflow-evaluator-optimizer",
      "level": 3,
      "text": "Workflow: Evaluator-optimizer"
    },
    {
      "anchor": "agents",
      "level": 3,
      "text": "Agents"
    },
    {
      "anchor": "combining-and-customizing-these-patterns",
      "level": 2,
      "text": "Combining and customizing these patterns"
    },
    {
      "anchor": "summary",
      "level": 2,
      "text": "Summary"
    },
    {
      "anchor": "appendix-1-agents-in-practice",
      "level": 2,
      "text": "Appendix 1: Agents in practice"
    },
    {
      "anchor": "a-customer-support",
      "level": 3,
      "text": "A. Customer support"
    },
    {
      "anchor": "b-coding-agents",
      "level": 3,
      "text": "B. Coding agents"
    },
    {
      "anchor": "appendix-2-prompt-engineering-your-tools",
      "level": 2,
      "text": "Appendix 2: Prompt engineering your tools"
    },
    {
      "anchor": "acknowledgments",
      "level": 2,
      "text": "Acknowledgments"
    }
  ],
  "published_on": "2024-12-19",
  "section": "When (and when not) to use agents",
  "slug": "building-effective-agents",
  "summary": "We've worked with dozens of teams building LLM agents across industries. Consistently, the most successful implementations use simple, composable patterns rather than complex frameworks.",
  "tags": [
    "Agents",
    "Tool use",
    "Claude Developer Platform",
    "Coding"
  ],
  "teams": [
    "Engineering"
  ],
  "title": "Building effective agents",
  "truncated": false,
  "updated_at": "2026-10-09T06:34:08Z",
  "url": "https://www.anthropic.com/research/building-effective-agents"
}