Types of AI Agents
Not all AI Agents are built the same way. Depending on how much they can perceive, reason, and learn, they are classified into different types. Understanding these types helps in choosing the right design when building an agent for a specific purpose.
AI Agents are broadly categorized from the simplest (no memory, no learning) to the most complex (self-improving, goal-driven).
Type 1 — Simple Reflex Agent
A Simple Reflex Agent reacts to the current input only. It has a set of predefined rules: "If this happens, do that." It has no memory and cannot plan ahead.
How It Works
It reads the current input, matches it against a rule, and gives a response. That is all it does.
Example
A customer support chatbot that works like this:
IF user says "hello" → Reply "Hi! How can I help you?" IF user says "price" → Reply "Our pricing starts at ₹999/month" IF user says "bye" → Reply "Goodbye! Have a great day!"
Strengths and Limitations
| Strengths | Limitations |
|---|---|
| Very fast and simple | Cannot handle situations outside its rules |
| Easy to build and maintain | No memory — every response is independent |
| Predictable behaviour | Cannot learn or improve |
Type 2 — Model-Based Reflex Agent
A Model-Based Reflex Agent is smarter than the simple reflex type. It keeps an internal model (a picture) of the world, so it knows some history of what has happened — not just the current input.
How It Works
It maintains a state — a record of what it knows about the world so far. This state is updated after every action or observation.
Example
A shopping cart agent that remembers what items have been added:
State: { cart: ["laptop", "mouse"] }
User: "Add keyboard"
Agent updates state → { cart: ["laptop", "mouse", "keyboard"] }
User: "Show my cart"
Agent reads state → Displays all 3 items
Strengths and Limitations
| Strengths | Limitations |
|---|---|
| Remembers context within a session | Still rule-based, not intelligent |
| Better than simple reflex agents | Cannot set or plan for goals |
Type 3 — Goal-Based Agent
A Goal-Based Agent does not just react — it works towards a specific goal. It thinks about what actions to take in order to reach that goal.
How It Works
The agent is given a goal (the desired outcome). It then plans a sequence of actions that will lead to that goal being achieved.
Example
A travel planning agent with the goal: "Book the cheapest flight from Mumbai to Bangalore for next Friday."
Goal: Book cheapest flight Mumbai → Bangalore, next Friday Plan: Step 1 → Search all available flights for that date Step 2 → Sort by price Step 3 → Select the cheapest option Step 4 → Book the ticket Step 5 → Confirm booking
Strengths and Limitations
| Strengths | Limitations |
|---|---|
| Can plan and execute multi-step tasks | No ability to evaluate tradeoffs |
| Works toward a defined outcome | Rigid goal definition needed |
Type 4 — Utility-Based Agent
A Utility-Based Agent is even more advanced. It does not just achieve a goal — it tries to achieve the goal in the best possible way. It assigns a score (called utility) to different options and picks the one with the highest utility.
How It Works
Instead of simply picking any path to the goal, it evaluates multiple options and weighs factors like cost, time, quality, and risk.
Example
A travel agent that considers both price and travel time:
Goal: Reach Bangalore from Mumbai Option A: Flight — ₹2500, 1.5 hours Option B: Train — ₹800, 8 hours Option C: Bus — ₹400, 12 hours Utility Score (considering budget + urgency): Option A = 85/100 Option B = 70/100 Option C = 55/100 Agent chooses: Option A (highest utility)
Strengths and Limitations
| Strengths | Limitations |
|---|---|
| Makes optimal decisions | Defining utility scores can be complex |
| Handles trade-offs intelligently | Computationally more expensive |
Type 5 — Learning Agent
A Learning Agent improves over time based on experience. It has a learning component that updates its behavior based on feedback — similar to how humans learn from their mistakes.
Core Components of a Learning Agent
| Component | Role |
|---|---|
| Performance Element | Takes actions based on current knowledge |
| Learning Element | Updates knowledge based on feedback |
| Critic | Evaluates how well the agent performed |
| Problem Generator | Suggests new experiences to learn from |
Example
A customer support agent that learns from ratings:
Day 1: Agent gives a generic response → User rates 2/5
Agent learns: "Generic responses get low ratings"
Day 5: Agent personalises responses → User rates 5/5
Agent learns: "Personalised responses work better"
Day 30: Agent consistently gives personalised answers
Strengths and Limitations
| Strengths | Limitations |
|---|---|
| Gets smarter with every interaction | Needs a lot of data to learn well |
| Adapts to new situations automatically | Can learn bad habits if feedback is poor |
Type 6 — Multi-Agent System
A Multi-Agent System is a collection of multiple AI Agents working together to solve a problem. Each agent handles a specific sub-task, and they communicate and coordinate with each other to achieve the overall goal.
Example
An e-commerce order processing system with 4 agents:
Order Agent → Receives order, validates details Payment Agent → Processes payment, checks fraud Inventory Agent → Checks stock, reserves items Delivery Agent → Assigns delivery partner, sends tracking info All 4 agents work together → Order is processed end-to-end
Comparison of All Agent Types
| Agent Type | Memory | Planning | Learning | Best For |
|---|---|---|---|---|
| Simple Reflex | None | None | None | Rule-based FAQs |
| Model-Based Reflex | Session only | None | None | Shopping carts, state tracking |
| Goal-Based | Yes | Yes | None | Task automation, booking systems |
| Utility-Based | Yes | Yes (optimal) | None | Decision-making, recommendations |
| Learning Agent | Yes | Yes | Yes | Personalised assistants, adaptive systems |
| Multi-Agent | Yes | Yes (distributed) | Possible | Complex workflows, enterprise systems |
Which Type Will Be Built in This Course?
This course focuses primarily on building Goal-Based and Utility-Based Agents using LLMs, and later introduces Multi-Agent Systems. These are the most practical and powerful types used in real-world AI applications today.
Summary
AI Agents range from simple rule-following bots to intelligent, self-learning systems. The six types — Simple Reflex, Model-Based Reflex, Goal-Based, Utility-Based, Learning, and Multi-Agent — each serve different purposes. As complexity increases, so does the agent's ability to plan, adapt, and make smarter decisions. Understanding these types is essential before diving into building one.