Agentic AI

Traditional artificial intelligence (AI) systems rely heavily on predefined rules and human intervention. They typically react to inputs or execute preset instructions. In contrast, agentic AI systems operate with a high degree of autonomy, performing tasks with minimal supervision. Unlike traditional reactive systems, agentic AI is proactive, leveraging machine learning (ML) models to mimic human-like decision making in real time [1].

LangGraph is a low-level framework developed by the creators of LangChain, designed to facilitate the implementation of agentic AI through graph-based workflow orchestration. While LangChain provides high-level abstractions for building LLM-powered applications, LangGraph focuses on the construction of stateful and agentic workflows using a graph-based execution model [2, 3]. In the rest, we first explore the graph-based architecture of LangGraph through code examples adapted from [4]. Thereafter, we develop several simple agentic AI systems using LangGraph, following designs also provided by [4].

LangGraph - Basics

In this section, we implement a series of LangGraph-based projects to develop a deeper understanding of its fundamental concepts. A graph in LangGraph consists of three primary components: a start state, one or more nodes, and an end state. These components are interconnected through edges, which define the transitions between them. A node within the graph can be designed to accept either a single input (e.g., a single value) or multiple inputs (e.g., a list of values). We begin by constructing a simple graph with a single node that processes a single input, then extend it to handle multiple inputs. Subsequently, we increase the number of nodes to create more complex workflows. All these initial experiments are conducted under deterministic transitions. Next, we explore conditional transitions, where edges are traversed based on specific criteria. Finally, we conclude the section by examining cyclic graphs (loops), which enable iterative workflows.

• Single-Node Graph

• Sequential Graph

• Conditional Graph

• Looping Graph

LangGraph - Agents

One of the primary applications of LangGraph is the development of agentic AI systems designed to minimize human intervention by leveraging one or more autonomous agents. These agents integrate large language models (LLMs) into the system and employ appropriate tools to accomplish tasks with minimal supervision. In the remainder of this section, we present several scenarios that demonstrate the construction of such systems. We begin with a simple chatbot that engages in basic conversations with the user. We then extend this chatbot by integrating memory, enabling it to retain and utilize conversational context. Next, we improve the agentic system by incorporating one or more tools, implemented as functions, to achieve specific objectives. Finally, we develop an agentic system capable of retrieving information from designated resources and files.

• Bot

• Reasoning and Acting (ReAct) Agent

• Retrieval-Augmented Generation (RAG) Agent

References

[1] C. Stryker, “What is agentic ai?” International Business Machines (IBM), accessed: 2025, https://www.ibm.com/think/topics/agentic-ai.

[2] LangChain, https://www.langchain.com/.

[3] LangGraph, https://www.langchain.com/langgraph.

[4] freeCodeCamp.org, https://youtu.be/jGg_1h0qzaM?si=69DsFmR2TMN259HC.