DEV Community

Kashaf Abdullah
Kashaf Abdullah

Posted on

Software with a Brain: Inside Agent-Based Intelligent Systems

#ai #agents #agentbase

What is an Agent Based Intelligent System?

"Agent Based Intelligent Systems are computer systems where autonomous software programs (agents) work together to solve complex problems by sensing their environment, making decisions, and taking actions."

Real Life Analogy šŸŒŸ

Think of it like a restaurant kitchen:

  • Each chef (agent) works independently
  • They sense orders (environment)
  • Communicate with each other
  • Coordinate to prepare meals
  • Adapt when problems arise

Part 2: Core Concepts with Examples

1. What is an Agent?

Definition: An agent is a computer program that can perceive its environment through sensors and act upon that environment through actuators.

Key Properties of Agents:

  • Autonomy: Operates without human intervention
  • Reactivity: Responds to environment changes
  • Proactiveness: Takes initiative to achieve goals
  • Social ability: Communicates with other agents

2. Environment Types Explained

A. Fully Observable vs Partially Observable

Fully Observable (Agent sees everything):

  • Chess Game: All pieces visible
  • TicTacToe: Complete board visible
  • Calculator: All inputs visible

Partially Observable (Agent has limited info):

  • Poker Game: Can't see opponent's cards
  • Self-Driving Car: Can't see around corners
  • Stock Market: Don't know others' strategies

Blog Tip: Most real-world problems are partially observable, making them perfect for agent-based solutions.

B. Deterministic vs Stochastic

Deterministic (Predictable outcomes):

  • Calculator: 2+2 always = 4
  • Vending Machine: Insert money ā†’ get snack
  • Traffic Light: Red ā†’ Stop, Green ā†’ Go

Stochastic (Random/Uncertain outcomes):

  • Weather Prediction: 70% chance of rain
  • Dice Roll: Random number 1-6
  • Stock Market: Can't predict exactly

C. Static vs Dynamic

Static (Environment doesn't change):

  • Crossword Puzzle: Same until solved
  • Sudoku: Numbers stay fixed
  • Written Exam: Questions don't change

Dynamic (Environment constantly changes):

  • Traffic System: Cars keep moving
  • Live Sports: Game situations change
  • Social Media Feed: New posts appear

D. Discrete vs Continuous

Discrete (Limited states):

  • Chess: 64 squares, finite positions
  • ON/OFF Switch: Only 2 states
  • ATM Menu: Limited options

Continuous (Infinite states):

  • Drone Flight: Infinite positions
  • Car Speed: 0 to 200+ km/h
  • Room Temperature: Any value possible

3. The PEAS Model (Agent Design Framework)

PEAS = Performance, Environment, Actuators, Sensors

Example 1: Amazon Delivery Drone

  • Performance: Deliver package in under 30 min
  • Environment: Airspace, weather, obstacles
  • Actuators: Propellers, camera, gripper
  • Sensors: GPS, camera, wind sensor

Example 2: ChatGPT

  • Performance: Provide accurate, helpful responses
  • Environment: User conversation context
  • Actuators: Text generation, API calls
  • Sensors: User input text, history

Example 3: Tesla Self-Driving Car

  • Performance: Safe navigation, no accidents
  • Environment: Roads, traffic, pedestrians
  • Actuators: Steering, brakes, accelerator
  • Sensors: Cameras, radar, LiDAR

Part 3: Types of Agents (Detailed with Examples)

1. Simple Reflex Agents

"Act based on current situation only"

IF [condition] THEN [action]

Examples:

  • Roomba Vacuum: "If dirt detected ā†’ clean"
  • Smoke Detector: "If smoke ā†’ alarm"
  • Basic Traffic Light: "If timer expired ā†’ change"

Code Example:

class SimpleReflexAgent:
    def act(self, sensor_input):
        if sensor_input == "dirt":
            return "clean"
        elif sensor_input == "wall":
            return "turn"
        else:
            return "move_forward"
Enter fullscreen mode Exit fullscreen mode

Limitation: No memory, no learning

2. Model Based Agents

"Keep track of the world state"

Examples:

  • GPS Navigation: Remembers your route history
  • Smart Home System: Knows when you usually return
  • Game AI: Remembers your past moves

Real Scenario:

Your smart thermostat:

  • Remembers you come home at 6 PM
  • Knows outside temperature
  • Adjusts accordingly
  • Learns your preferences

3. Goal Based Agents

"Achieve specific objectives"

Examples:

  • Delivery Robot: Goal = deliver package
  • Chess AI: Goal = checkmate opponent
  • Trading Bot: Goal = maximize profit

How It Works:

Goal: Reach Destination in 30 mins

Options:

  1. Highway ā†’ 25 mins āœ“ Achieves goal
  2. Local road ā†’ 40 mins āœ— Fails goal
  3. Shortcut ā†’ 20 mins āœ“ Achieves goal

Agent picks best option for goal achievement

4. Utility Based Agents

"Maximize happiness/satisfaction"

Examples:

  • Google Maps: Shows fastest AND fuel efficient route
  • Amazon: Suggests products you'll likely buy
  • Netflix: Recommends shows you'll enjoy most

Utility Calculation:

Route Options:
ā”œā”€ā”€ Route A: 20 mins, ā‚¹50 fuel (Utility = 80/100)
ā”œā”€ā”€ Route B: 15 mins, ā‚¹100 fuel (Utility = 75/100)
ā””ā”€ā”€ Route C: 25 mins, ā‚¹30 fuel (Utility = 90/100) āœ“ BEST

Agent picks highest utility = Route C
Enter fullscreen mode Exit fullscreen mode

5. Learning Agents

"Improve from experience"

Examples:

  • ChatGPT: Gets better with more conversations
  • Spotify: Learns your music taste
  • AlphaGo: Learns from millions of games

Learning Process:

  1. Try action ā†’ Get result
  2. Remember outcome
  3. Adjust strategy
  4. Try again
  5. Improve over time

Part 4: Multi-Agent Systems (MAS)

What is MAS?

"Multiple agents working together or competing to solve complex problems"

Real-World MAS Examples:

1. Uber/Ola System

ā”œā”€ā”€ Rider Agent ā†’ Wants cheap ride
ā”œā”€ā”€ Driver Agent ā†’ Wants high fare
ā”œā”€ā”€ System Agent ā†’ Matches both
ā””ā”€ā”€ Payment Agent ā†’ Handles transaction
Enter fullscreen mode Exit fullscreen mode

2. Amazon Warehouse

ā”œā”€ā”€ Picking Robots ā†’ Get products
ā”œā”€ā”€ Packing Robots ā†’ Package items
ā”œā”€ā”€ Labeling Robots ā†’ Print labels
ā””ā”€ā”€ Shipping Robots ā†’ Load trucks
Enter fullscreen mode Exit fullscreen mode

3. Smart City Traffic

ā”œā”€ā”€ Traffic Light Agent 1 ā†’ Main Street
ā”œā”€ā”€ Traffic Light Agent 2 ā†’ Side Street
ā”œā”€ā”€ Emergency Agent ā†’ Ambulance priority
ā””ā”€ā”€ Monitoring Agent ā†’ Oversee everything
Enter fullscreen mode Exit fullscreen mode

Agent Communication

How Agents Talk:

Message Types:

  1. INFORM ā†’ "Temperature is 30Ā°C"
  2. REQUEST ā†’ "Please move left"
  3. QUERY ā†’ "What's your position?"
  4. AGREE ā†’ "Okay, moving left"
  5. REFUSE ā†’ "Can't move, blocked"

Communication Protocol Example (FIPA-ACL):

<message>
  <sender>traffic_light_1</sender>
  <receiver>traffic_light_2</receiver>
  <performative>inform</performative>
  <content>heavy_traffic_detected</content>
</message>
Enter fullscreen mode Exit fullscreen mode

Coordination Mechanisms

1. Cooperation (Working Together)

Swarm Robots Cleaning:

ā”œā”€ā”€ Robot 1: Cleans left side
ā”œā”€ā”€ Robot 2: Cleans right side
ā”œā”€ā”€ Robot 3: Empties dust bins
ā””ā”€ā”€ All share map information
Enter fullscreen mode Exit fullscreen mode

2. Competition (Rivalry)

E-commerce Bidding:

ā”œā”€ā”€ Buyer Agent 1: Bid ā‚¹1000
ā”œā”€ā”€ Buyer Agent 2: Bid ā‚¹1500
ā”œā”€ā”€ Seller Agent: Wants highest price
ā””ā”€ā”€ Auctioneer Agent: Manages bids
Enter fullscreen mode Exit fullscreen mode

3. Negotiation (Reaching Agreement)

Salary Negotiation Bot:

ā”œā”€ā”€ Employee Agent: Wants ā‚¹80,000
ā”œā”€ā”€ Company Agent: Offers ā‚¹60,000
ā”œā”€ā”€ HR Agent: Suggests ā‚¹70,000
ā””ā”€ā”€ Both agree on ā‚¹70,000
Enter fullscreen mode Exit fullscreen mode

The Future: Where Are We Headed?

Next 5 Years 

ā”œā”€ā”€ šŸ¤– Personal AI Agents
ā”‚   Your own digital twin managing your life
ā”‚
ā”œā”€ā”€ šŸ˜ļø Smart Cities
ā”‚   Thousands of agents coordinating traffic,
ā”‚   energy, waste, and emergencies
ā”‚
ā”œā”€ā”€ šŸ§¬ Healthcare Swarms
ā”‚   Nano-agents monitoring your body,
ā”‚   detecting diseases early
ā”‚
ā”œā”€ā”€ šŸ’¼ Autonomous Companies
ā”‚   Businesses run entirely by agent
ā”‚   negotiations and transactions
Enter fullscreen mode Exit fullscreen mode

Next 10 Years 

ā”œā”€ā”€ šŸŒ Global Agent Internet
ā”‚   Agents from different systems,
ā”‚   countries, and companies talking globally
ā”‚
ā”œā”€ā”€ šŸ§  Human-Agent Teams
ā”‚   Humans and AI agents working as equals
ā”‚
ā”œā”€ā”€ šŸŖ Space Exploration
ā”‚   Agent swarms exploring Mars and beyond
Enter fullscreen mode Exit fullscreen mode

Conclusion

Agent-Based Intelligent Systems are how we'll build the intelligent, automated, and adaptive world of tomorrow by creating software that thinks, works, and collaborates like living organisms.

Whether it's your Roomba vacuum, Google Maps, or self-driving cars ā€“ agents are already everywhere. And they're only getting smarter.

Written by Kashaf Abdullah

Software Engineer | MERN Stack | Web Development

Top comments (1)

Collapse
 
mamoor_ahmad profile image
Mamoor Ahmad •

This is a really nice intro to agent-based systems ā€” the restaurant kitchen analogy at the start actually made it click for me right away.

I especially liked the PEAS breakdown with real examples like ChatGPT and Tesla. Most articles just throw definitions at you but this one actually makes you go "oh wait, I already use these every day."

One thing I'd love to see in a follow-up post, how do these agents actually coordinate in practice when things go wrong?
Like what happens when two agents in a MAS disagree or conflict with each other?
That's where it gets really interesting and messy. šŸ˜…

Also the "Next 10 Years" section got me thinking we're honestly not that far from agent swarms running parts of our daily lives. Kind of exciting and terrifying at the same time.

Great write-up Kashaf, keep them coming šŸ‘