Type: url
Original: https://www.anthropic.com/research/building-effective-agents
Status: generated
Summary
Anthropic’s account of what worked across dozens of customer teams building LLM agents, whose headline finding is negative: the most successful implementations used simple, composable patterns rather than complex frameworks or specialized libraries. It groups everything under “agentic systems” and separates workflows (LLMs and tools orchestrated through predefined code paths) from agents (the LLM directing its own process and tool usage), then inventories five workflow patterns — prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer — plus the autonomous agent, built up from the augmented LLM as foundational building block. Its standing advice is to start with LLM APIs directly, find the simplest solution possible, and add complexity only when it demonstrably improves outcomes; a note records that much of the tooling landscape it describes has changed since December 2024.