Intro to Agents

If you’re reading this, you’ve probably heard people say:

“2025 will be the year of AI Agents.”

But what are AI agents, and why is everyone talking about them? How are they different—or better—than the Large Language Models (LLMs) we already use?

Why LLMs Are Powerful (But Limited)

Since 2020, LLMs like ChatGPT have changed how we work. They’ve doubled our productivity by doing tasks like:

• Writing code
• Drafting emails
• Creating templates
• And much more

But here’s the catch: LLMs can’t take real-world actions. They can only give you text or images as replies. Let’s break this down.

What’s an “Action”?

Tools Make LLMs Powerful Agents

What if LLMs could use tools, like humans use hammers or calculators? That’s exactly what AI agents do!

Real-Life Example: Booking Flights

Imagine asking ChatGPT: “What’s the total cost of the cheapest round-trip flights from New Delhi to Bali?” Today, it can’t answer this—it only knows data up to 2023.

But if we give it two tools:

Now, here’s how ChatGPT becomes an AI agent:

1. It uses the Google Flights tool to find all flight prices for your dates.
2. Picks the cheapest options for both legs of the trip.
3. Adds the two prices using the calculator tool and gives you the total cost.

What Tools Can AI Agents Use? (Spoiler: No Limits!)

You might wonder: "How many tools can we connect to an AI agent? Is there a limit?"

The answer is exciting: there's no limit. If a tool can help the LLM solve a problem, it can be added. The more tools, the smarter the agent becomes.

How Do AI Agents Actually Work? (Step-by-Step)

Let's break down the process of how an AI agent uses tools to solve problems. Imagine it like a chef in a kitchen—thinking, choosing tools, and combining ingredients to serve a dish:

Your Questions, Answered

get_flight_prices(
    departure_city="DEL",  # Delhi
    arrival_city="DPS",    # Bali
    dates="2024-11-15 to 2024-11-22"
)

def get_google_flights(
    departure_code: str,  # IATA code (e.g., "DEL" for Delhi)
    arrival_code: str,    # IATA code (e.g., "DPS" for Bali)
    departure_date: str,  # Format: "YYYY-MM-DD"
    return_date: str,     # Format: "YYYY-MM-DD"
    cabin_class: str = "economy"
    ) -> list[dict]:
        """
    Fetches flight prices from Google Flights API.
    Returns list of flights sorted by price.
    """
    response = requests.get(
        "https://api.google.com/flights",
        params={
            "departure_code": departure_code,
            "arrival_code": arrival_code,
            "departure_date": departure_date,
            "return_date": return_date,
            "cabin_class": cabin_class
        }
    )
    # Process response  
    flights = response.json()["flights"]  
    return sorted(flights, key=lambda x: x["price"])  # Sort by cheapest

get_google_flights(  
    departure_code="DEL",  
    arrival_code="DPS",  
    departure_date="2024-11-15",  
    return_date="2024-11-22"  
)

[  
    {  
        "airline": "Air India",  
        "price": 18500,  
        "departure_time": "08:30",  
        "flight_number": "AI-876"  
    },  
    {  
        "airline": "IndiGo",  
        "price": 20300,  
        "departure_time": "14:15",  
        "flight_number": "6E-132"  
    }  
]