<p>LangChain published an important text about how agentic systems are structured β and why most frameworks on the market confuse key concepts.</p>
<p>They criticized a recent guide from OpenAI: it offers a convenient but oversimplified abstraction of an agent, hiding the main thing β control over the context that goes into the LLM. Without it, it's impossible to build a reliable system.</p>
<p>LangChain offers something different: LangGraph β a framework with a clear separation of workflows and agents, a declarative API for structure, and imperative/functional API inside. Instead of black boxes β explicit nodes and connections, normal code, access to memory, and built-in support for human-in-the-loop.</p>
<p>The key idea: real agentic systems are not just one agent, but a mix of workflows and agents. And it is the control over the transfer of context into the LLM that makes the framework suitable for production.</p>
<p>π Source: <a href="https://blog.langchain.dev/how-to-think-about-agent-frameworks/">https://blog.langchain.dev/how-to-think-about-agent-frameworks/</a></p>;
<p>π₯ Bonus (Agent Framework comparison): <a href="https://docs.google.com/spreadsheets/d/1B37VxTBuGLeTSPVWtz7UMsCdtXrqV5hCjWkbHN8tfAo/edit?gid=0#gid=0">https://docs.google.com/spreadsheets/d/1B37VxTBuGLeTSPVWtz7UMsCdtXrqV5hCjWkbHN8tfAo/edit?gid=0#gid=0</a></p>;
<p>#langgraph #langchain #agent #openai #ai</p>
<p>βββββββββ<br>ΠΡΡΠ»ΠΈ Π Π²Π°ΡΠ΅Π²Π°<br>βββββββββ</p>
