Building Effective Agents

Anthropic Engineering Blog post distilling lessons from working with dozens of teams building LLM agents across industries. The central thesis: the most successful implementations use simple, composable patterns rather than complex frameworks. The article introduces a taxonomy of agentic systems, distinguishing workflows (LLMs orchestrated through predefined code paths) from agents (LLMs that dynamically direct their own processes and tool usage).

Five workflow patterns are presented in order of increasing complexity: prompt chaining, routing, parallelization, orchestrator-workers, and evaluator-optimizer. Each is described with architecture, use cases, and examples. The foundational building block is the augmented LLM — a model enhanced with retrieval, tools, and memory.

Three core principles for agent design: simplicity, transparency (showing planning steps), and careful crafting of the agent-computer interface (ACI) through tool documentation and testing. The ACI concept parallels HCI — investing the same effort in designing interfaces for models as for humans. Practical tool engineering advice includes avoiding formats that require counting or escaping, using absolute paths over relative ones, and iterative testing of tool definitions.

Two production domains are highlighted: customer support (natural conversation flow + tool integration + measurable success criteria) and coding agents (verifiable via automated tests + structured problem space).

Key connections

Directly extends the Managed Agents Architecture by the same organization — that post covers infrastructure, this one covers design patterns
The brain-hands split in Brain-Hands Decoupling is the infrastructure counterpart to the augmented LLM concept here
The orchestrator-workers pattern maps to the “many brains, many hands” idea in Managed Agents
The ACI concept connects to tool design patterns in Hermes Agent and OpenClaw

Building Effective Agents

Key connections

Ingestion manifest