Agentic AI and How It Works

What is an AI Agent?

• A Large Language Model (LLM) is powerful at understanding and generating natural language.

• But it is:

  • Trained on past data.

  • Trained on large public data.

  • Stateless and isolated — it cannot access real-time or private data, connect to external systems, update databases, or call APIs on its own.

• Think of an LLM as a brain without a body — it can think, reason, and plan, but it can’t act.

So we create Agents:

An AI agent is a software system that uses artificial intelligence to autonomously perform tasks and achieve goals, often without direct human intervention.

• The agent uses the LLM to plan or interpret, and external tools to take action (e.g., calling APIs, querying databases, interacting with users).

• This makes the system capable of doing real-world tasks like:

  • Getting the weather

  • Searching the web

  • Sending an email

  • Writing to a database

Example: Why LLM Alone Isn’t Enough

• If you ask an LLM:
  "What's the weather in Bangalore right now?"
  It will fail to answer accurately because it was trained on past data.

• But if you:

  1. Fetch the live weather using a function, and

  2. Insert it into the system prompt, like:

     "User asked for weather in Bangalore. The current weather is 28°C, cloudy."

Then the LLM can respond intelligently, e.g.,

"It’s currently 28°C and cloudy in Bangalore. Do you want to pack an umbrella?"

But what are Tools ?

• A tool is a function or API the agent can use to complete a specific task.

• You can build tools for:

  • Weather API calls

  • Database reads/writes

  • File uploads

  • Google Search

  • Email sending

  • Code execution

How It Works:

1. You define a function in your codebase (e.g., getWeather(city)).

2. You describe this function to the LLM using a prompt or function schema (like in OpenAI’s function calling).

3. The agent decides when to use that tool based on the user query.

LLM + Tools = Agent

Real-Life Agent Frameworks

Some libraries and frameworks help you build agents easily:

• LangChain – Lets you define tools, memory, agent behaviors

• OpenAI Function Calling / Assistants API

• Auto-GPT / BabyAGI – Autonomous agents that loop until a goal is met

• Semantic Kernel – Microsoft’s framework for agents

Note: This article is my understanding from the GenAI cohort from chaicode website. #chaicode #chaiaurcode