What is an AI Agent?
Table of Contents
Agentic Systems can be divided into Workflows and Agents
- Workflows are structured systems that orchestrate Large Language Models (LLMs) and tools through predetermined, scripted pathways. The execution follows predefined logic and decision trees.
- Agents, in contrast, are autonomous systems where LLMs make real-time decisions about their own processes and tool selection. They maintain dynamic control over task execution, adapting their approach based on context and requirements.
An agent is like an autonomous entity that operates on your behalf, leveraging AI technologies to perform tasks and interact with the world.
AI Agents can make decisions and take actions on their own to achieve a goal without being told exactly what to do at every step.
Agentic AI is a system that is using one or more agents . It is a similar term.
| Type | Reactive | Tool Use | Reasoning | Planning | Proactivity |
| RAG Chatbot | ✅ | ❌ | ❌ | ❌ | ❌ |
| Tool-Augmented Chatbot | ✅ | ✅ | ❌ | ❌ | ❌ |
| Agentic AI | ✅ | ✅ | ✅ | ✅ | ✅ |
For any autonomous Agent the reliability drops. With Langraph the Reliability increases significantly
Typical Langchain Application ( Its a chain of things happening one after another )
LangGraph Application
Stateful workflows represented as graphs.
| Feature | LangChain | LangGraph |
| Purpose | Toolkit to build LLM apps (chains, tools, agents) | Framework to manage complex workflows with state |
| Style | Linear or reactive chains | Graph-based, supports loops, retries, memory |
| Best Use Case | Simple chatbots, RAG apps, tool usage | Multi-step workflows, agents with memory, conditional paths |
| State Handling | Stateless or partially stateful | Fully stateful; remembers and transitions based on logic |
| Example Use | “Book a flight” using a flight API | “Plan a vacation” (ask budget → choose flights → book hotel → loop if error) |
Part-1 : Building a very basic AI Agent using Langgraph ( Windows )
YT Reference : How to make Jupyter notebook up and running in VSCODE
Create a folder ( Open cmd )
- mkdir langgraph && cd langgraph
- code .
Creating VE ( Open Terminal, ctrl + ~)
- python -m venv venv
- source venv/Scripts/activate ( for git bash )
Install necessary plugins using uv or pip
- pip install langgraph python-dotenv ipykernel notebook
ipython will be installed as a dependency to ipykernel, so no need to install separately ( pip install ipython ❌ )
Create : New Jupyter Notebook
- ctrl + shift + p
- create: New Jupyter Notebook
- Add Python Environment (as given in the Reference Video Above)
Create Class
from typing import TypedDict
class PortfolioState(TypedDict):
amount_usd:float
total_usd:float
total_inr:float
Define Functions
def calculate_tot(state:PortfolioState) -> PortfolioState:
state['total_usd'] = state['amount_usd']*1.08
return state
def convert_to_inr(state:PortfolioState) -> PortfolioState:
state['total_inr'] = state['total_usd']*85
return state
Build the Graph
from langgraph.graph import StateGraph, START, END
builder = StateGraph(PortfolioState)
builder.add_node("calculate_total_node", calculate_tot)
builder.add_node("convert_to_inr", convert_to_inr)
builder.add_edge(START, "calculate_total_node")
builder.add_edge("calculate_total_node", "convert_to_inr”)
builder.add_edge("convert_to_inr", END)
graph = builder.compile()
Generate the Flow Diagram
from IPython.display import Image, display
display(Image(graph.get_graph().draw_mermaid_png()))
