Building reliable agentic systems, which encompass both workflows and agents, requires ensuring that large language models (LLMs) have the appropriate context at each step, including the control over content and execution of relevant tasks. Agentic frameworks, like LangGraph, provide orchestration by combining declarative and imperative APIs, with agent abstractions simplifying initial development but potentially complicating context management. The blog critiques OpenAI's approach for conflating declarative frameworks with agent abstractions, emphasizing the need for frameworks that offer flexibility, reliability, and explicit control over LLM context. The text further discusses the importance of distinguishing between workflows and agents, noting that most production systems integrate both to balance predictability and flexibility. LangGraph is highlighted as a versatile orchestration framework that supports both workflows and agents with features like persistence, fault tolerance, and human-in-the-loop interactions, essential for production-ready agentic systems.