AI Agent without Jargon Chapter 1 – What is an AI Agent?

Reading Time: 5 minutes

Disclaimer: 

No jargon is used or harmed in writing this tutorial.

This is a “no-jargon” attempt to explain AI Agents. In case you find any jargon, please report it to me, and I’ll take corrective measures 🙂

Chapter 1 – What is an AI Agent?

Character 1 – Jigyaasu Tayal, VP Technology at a mid-size technology company

Experience: 18 years in software architecture & engineering leadership

Character 2 – Saral

Scene: 

Jigyaasu has just walked out of a high-stakes C-suite meeting. 

(Jigyaasu enters his office with a notepad. He looks panicked.)

(He stares at a whiteboard that says in very large font: “AI Agent?”)

Jigyaasu (takes out his phone, and starts mimicking leadership):
” Mr. Jigyaasu, we want AI agents integrated into our product!’ 

As if it’s like sprinkling coriander on a salad!”

Jigyaasu (sighs):
“No YouTube videos don’t help… all I get is fancy diagrams and zero clarity.”

(The door opens. A casually confident human walks in.)

Jigyaasu does not notice him. He is still on the phone

Saral (smiling):
“Sorry I’m late. I heard I’m being considered for the job?”

Jigyaasu (confused):
“I’m not hiring anyone…”

(Back to his phone)
 

Jigyaasu:
“They want an AI agent that can talk to customers, classify tickets, predict churn, wash their car… and what not?”

Saral:
“Actually, I can’t wash the cars. But we can talk about the rest of the activities

Jigyaasu (irritated):
“I told you, I don’t want you.”

Saral (grinning):
“You do. You just don’t know it yet.”

(Jigyaasu stares, does not know what to say.)

Saral: So you don’t know the difference between an LLM and an AI Agent, do you?

Jigyaasu (defensive):
“Of course I know!”

Saral:
hmmm.. I got it, let me tell you the difference between LLMs and AI Agents

Concept Explanation 1

Let’s break it down.
ChatGPT can help you write a reply to an email, right?

but for that, you have to:

• Copy that email
  
• Paste it in ChatGPT
  
• Ask it that – respond in a polite and professional tone
  
• Copy the reply
  
• Paste it into Gmail
  
• And then send
  

Now imagine ChatGPT has access to Gmail.
You simply say: “Reply to the email from Nikhilesh.”
The agent searches your inbox, finds Nikhilesh’s email, writes a reply, and saves it in the draft folder

The only thing you need to do now is – check the email.

And if it’s good, then just press the send button.

Jigyaasu:
“Wait, so, you are saying, AI Agent can read my email…” 

 Saral:
“That’s correct.”
 

Jigyaasu:
“And generate a reply?”
Saral:
“Yes.”

 Jigyaasu (talking to himself):
“That’s Creepy… but looks productive.”

But who are you? What’s your name?

AI – my name is Saral and I am an AI

Jigyaasu  (looks further confused) – what? Why have you come here?

AI – To help you. I noticed that you are struggling with Agentic AI, so I thought, ‘Let me help you with that.’

Jigyaasu  – Seriously??

Saral – Yes, I can make you learn Agentic AI in simple language… 

Jigyaasu  (still suspicious) – ok, so you are an AI, then answer my questions –  how do you do what you do? How do you acquire so much intelligence? How do you learn so many things? Will you take away people’s jobs? How….

Saral – Hold on hold on…

Until now, humans have been making me learn; it seems it’s my turn to make them learn about AI. 

Jigyaasu seems interested now

Concept Explanation 2

Let us continue. So the main difference between an LLM and an AI Agent is that
LLMs generate content for you.
AI Agents act on your behalf.

Originally, ChatGPT worked in isolation—if it knew, it answered. 

If not, well…it hallucinated or said – I can’t assist with that.

Now? 

With agentic AI, ChatGPT can access external tools and actually get the job done for you.

Jigyaasu (trying to sound smart):
“Ah, so you mean RAG, right?”
Saral:
“No. RAG is different.” 

But I am glad you pointed this out because this is a major misconception.

Concept Explanation 3

RAG lets LLMs pull info from a vector database.

While AI Agents can access anything—Google Search, Wikipedia, Maps, databases, even your email.

AI agents are basically

LLM +
Reasoning +
Access to external applications +
Self-reflection

They:

1. Break a problem into steps
   
2. Decide which external applications they need to access
   
3. Once they get the answer, they self-reflect on their answers. If they think they do not have the right answer, they try again
   

Jigyaasu:
“Sounds like LLM went through a corporate training, and came back with a 3-step action plan.”

Saral (laughs):

That’s a good one! 

Concept Explanation 4:

Another misconception many people have is that assigning a role to an LLM makes it an AI agent; that’s not the case
True AI Agents combine LLMs with code and tools via APIs.

They decide if they have to use a tool. If yes, then which tool to use, when to use it, and how.

Jigyaasu:
“So I don’t want have to tell you which tool to use?”

 Saral:
“Correct. 

I decide. I execute. I double-check.”

 Jigyaasu:
“And I just… clap from the sidelines?”
Saral:
“Exactly. 

Or take credit. That works too.”

Concept Explanation 5:

Let’s go deeper.

An AI Agent has 3 key components:

1. Tasks – What needs to be done
   
2. Tools – APIs or external apps to perform tasks
   
3. Memory – For remembering, reflecting, and improving
   

You define the task.
You describe the tool.
The Agent decides if and how to use it.

The tool could be:

• Google Search
  
• A database
  
• A Python interpreter
  
• Any specific website
  

Jigyaasu:
“This is getting exciting. Can we build an actual AI agent?”
Saral:
“Of course. Let’s build an AI Search Agent.”

Concept Explanation 6:

Let’s build a basic AI Search Agent.

So, first, what we are doing we are using Gemini 2.5 Flash, and we are asking, Who is the president of the US?

We know that- LLMs are pre-trained models, which means they have a knowledge cut-off. So, you can see, the answer that the model generated was – 

The current president of the United States is Joe Biden.

This is outdated information because there is a knowledge cut-off.

Jigyaasu:
“So you’re that one friend who gives advice based on 2021 data?”
Saral:
“Yeah. I still think NFTs are hot.”

 “So how do we fix this? 

Jigyaasu:
Let me give you a newspaper subscription?”

Saral:
“Well close. Or give me access to Google Search.”

Jigyaasu:
This is what I meant

Concept Explanation 7

We’ll access Google search using SERP API key. You can get the SERP API key for free.

Then we define the tool name, which is SERP API, and then we create an agent. 

We pass the tool, we pass the LLM, and then you can see we have done verbose equals to true, 

Verbose equals to true actually gives us the information about how the model is thinking, how the model has come up with the final answer. m

And now, let us ask the same question: Who is the president of the US?

Now, see the model is thinking – I need to find out who the current president of the United States is.

So, the action It has decided is to search 

But what is the action input? 

It has decided that it will search – “who is the president of the united states”.

So, when it searched, it got The search results mention President Trump. I need to confirm if he is still the president.. 

Because it is not sure, it again decided to search, and this time the action input was – current president of the united states. 

this time the agent got the confirmation that The search results indicate that Donald Trump is the current president of the United States

And then the agent provided the final answer – Donald Trump

Jigyaasu (leaning back, impressed):
“Very nice…so.. you think, search, and then double-check – exactly what my mom does when I say I have locked the door.”

Saral (smiles):

“It seems your mom is smarter than you “

Jigyaasu (grinning):
“I started this session talking to a 2021 chatbot…and now I’ve got a Sherlock Holmes with an API key that can get things done for me”

Saral (smiling):
“Another human, fine-tuned for the AI era.”