LangChain
Building Effective Agents with LangGraph
LangChain takes the patterns from Anthropic's Building Effective Agents post and implements every one of them from scratch in LangGraph: the augmented LLM, prompt chaining, parallelization, routing, orchestrator and workers, evaluator and optimizer, and finally a real agent looping over tools. The useful part is the boundary it draws. Workflows are predefined code paths and you should reach for them first; an agent is what you build when the sequence of steps genuinely cannot be known ahead of time.