What is an AI Agent? (Explained Simply)

Key points

• A chatbot answers. An agent acts.
• An agent can use tools (search, calculator, APIs) on its own.
• An agent works in a loop until the goal is done.

The one-line definition

An AI Agent is a program that uses an LLM (like Claude or GPT) as its "brain" to decide what to do, then uses tools to actually do it, repeating until a goal is achieved.

Note: Brain (LLM) + Hands (tools) + a Loop = Agent.

Chatbot vs Agent (the key difference)

CHATBOT (just talks) AGENT (talks AND acts) ──────────────────── ──────────────────────

You: book a flight You: book a flight │ │ ▼ ▼ ┌─────────┐ ┌─────────┐ │ LLM │ │ LLM │ ◄────┐ └─────────┘ └────┬────┘ │ │ │ │ loop ▼ ▼ │ "Sure! Here is how ┌─────────────┐ │ you can book a │ USE A TOOL │──┘ flight..." (just │ search/book │ words, no action) └─────────────┘ │ ▼ ✅ Flight booked!

Every agent has 3 parts

• The Brain (LLM) — The language model that thinks and decides the next step. Example: Claude / GPT-4 reading the goal and saying "I should search first."
• The Tools (Hands) — Functions the agent can call to affect the real world. Example: web_search(), send_email(), run_code(), query_database()
• The Loop (Habit) — It keeps thinking → acting → observing until the job is done. Example: Think → call tool → read result → think again → finish

The basic agent loop (memorise this picture)

┌──────────────────────────────────────────┐ │ │ ▼ │ ┌──────────┐ ┌──────────┐ ┌────────────┐ │ │ THINK │──► │ ACT │──► │ OBSERVE │──┘ │ (decide) │ │ (use a │ │ (read the │ │ │ │ tool) │ │ result) │ └──────────┘ └──────────┘ └────────────┘ ▲ │ │ goal not done yet? │ └────────────────────────────────┘ │ goal done ▼ ✅ FINISH

A tiny code example (don't worry, read it like English)

Here is the simplest possible agent loop. Notice it is literally just a while loop around "ask the LLM what to do next".

goal = "Find today's weather in Mumbai"
history = []

while True:
    decision = llm(goal, history)        # 1. THINK
    if decision.is_final_answer:
        print(decision.answer)            # done!
        break
    result = run_tool(decision.tool,      # 2. ACT
                      decision.args)
    history.append(result)                # 3. OBSERVE

▶ Try it: a tiny agent (no real LLM, runs in your browser)

Edit the goal string and press Run. Real Python runs right here in your browser.

# This is a fake agent so you can FEEL the Think -> Act -> Observe loop.
# Change the goal below and hit Run!

tools = {
    "weather": lambda city: f"{city}: 31C, sunny",
    "shout":   lambda text: text.upper(),
}

def fake_brain(goal):           # pretend the LLM decided this
    if "weather" in goal:
        return ("weather", "Mumbai")
    return ("shout", goal)

goal = "what is the weather in Mumbai?"
tool, arg = fake_brain(goal)    # 1. THINK
result = tools[tool](arg)       # 2. ACT
print("Tool chosen :", tool)
print("Observation :", result) # 3. OBSERVE

▶ Free playground: write any Python you like

This is a real Python interpreter (Pyodide). Experiment freely!

# Your sandbox. Type anything and Run.
print("Hello from real Python in your browser!")
print("2 to the power 10 =", 2 ** 10)
for i in range(1, 4):
    print("counting", i)

When do you actually need an agent?

Beginner misunderstandings

Remember these 3 lines

• Chatbot = answers. Agent = answers AND acts.
• Agent = Brain (LLM) + Tools + Loop.
• The LLM decides; your code executes. Always.

Key takeaways

• An AI agent uses an LLM as a brain to decide actions and tools to perform them.
• It runs a Think → Act → Observe loop until the goal is reached.
• Use agents only when the task is multi-step and needs live actions/data.
• The LLM never runs tools itself — it requests them and your code executes.

Frequently Asked Questions

Related topics

• Goal Setting & Monitoring
• Learning & Adaptation
• Memory Management Pattern
• Multi-Agent Collaboration
• Prompt Chaining Pattern

Browse all Agentic AI Patterns topics →