Why multiple agents solve complex problems in Agentic AI - Experiment to Prove It

Experiment - Why multiple agents solve complex problems

Problem:You want to understand why using multiple agents working together can solve complex problems better than a single agent.

Current Metrics:A single agent solves a problem with 70% success rate but struggles with tasks requiring diverse skills or parallel work.

Issue:The single agent cannot handle complex tasks efficiently because it lacks specialization and parallel processing.

Your Task

Show that multiple agents collaborating improve problem-solving success rate to at least 90%, especially on complex tasks.

You must keep the total number of agents reasonable (e.g., 3 to 5 agents).

Each agent should specialize in a different subtask.

Agents communicate only through simple messages.

Hint 1

Hint 2

Hint 3

Hint 4

Solution

Agentic AI

import random

class Agent:
    def __init__(self, name, skill):
        self.name = name
        self.skill = skill

    def solve(self, task):
        # Agent solves task if skill matches task type
        if task == self.skill:
            success = random.random() < 0.98  # 98% success if skill matches
        else:
            success = random.random() < 0.5   # 50% success otherwise
        print(f"{self.name} working on {task}: {'Success' if success else 'Fail'}")
        return success

class Coordinator:
    def __init__(self, agents):
        self.agents = agents

    def solve_complex_problem(self, subtasks):
        results = []
        for task in subtasks:
            # Find agent with matching skill
            agent = next((a for a in self.agents if a.skill == task), None)
            if agent:
                result = agent.solve(task)
            else:
                # If no specialized agent, pick random agent
                agent = random.choice(self.agents)
                result = agent.solve(task)
            results.append(result)
        overall_success = all(results)
        print(f"Overall problem solved: {'Yes' if overall_success else 'No'}")
        return overall_success

# Define agents with different skills
agents = [
    Agent('Agent A', 'math'),
    Agent('Agent B', 'language'),
    Agent('Agent C', 'vision')
]

coordinator = Coordinator(agents)

# Complex problem requires math, language, and vision subtasks
subtasks = ['math', 'language', 'vision']

# Run multiple trials to estimate success rate
trials = 1000
successes = sum(coordinator.solve_complex_problem(subtasks) for _ in range(trials))
print(f"Success rate with multiple agents: {successes/trials*100:.2f}%")

# Compare with single agent trying all subtasks
single_agent = Agent('Solo Agent', 'math')
single_successes = 0
for _ in range(trials):
    results = [single_agent.solve(task) for task in subtasks]
    if all(results):
        single_successes += 1
print(f"Success rate with single agent: {single_successes/trials*100:.2f}%")

Created multiple agents each specialized in a different skill.

Added a coordinator to assign subtasks to the right agent.

Simulated success rates showing specialized agents perform better on their tasks.

Compared multi-agent success rate to single agent handling all tasks.

Results Interpretation

Single Agent Success Rate: ~25%

Multiple Agents Success Rate: ~92%

This shows that dividing a complex problem into parts and letting specialized agents handle each part greatly improves success.

Using multiple agents with different skills allows parallel work and specialization, which helps solve complex problems more effectively than a single general agent.

Bonus Experiment

Try adding communication between agents to share partial results and improve overall success.

💡 Hint

Implement a simple message passing system where agents can request help or share intermediate outputs before final coordination.

Practice

1. Why do multiple agents working together solve complex problems better than a single agent?

easy

A. Because agents do not communicate and work independently without sharing.

B. Because one agent can do all the work alone without help.

C. Because they divide the work and share knowledge to find solutions faster.

D. Because multiple agents always produce the same results as one agent.

Solution

1. Step 1: Understand agent collaboration

   Multiple agents split a big problem into smaller parts and work on them separately.

2. Step 2: Recognize knowledge sharing

   Agents share what they learn, combining their results for a better solution.

3. Final Answer:

   Because they divide the work and share knowledge to find solutions faster. -> Option C

4. Quick Check:

   Multiple agents collaborate = better solutions [OK]

Hint: Think teamwork: many hands make light work [OK]

Common Mistakes:

• Assuming one agent can solve everything alone
• Ignoring the benefit of sharing knowledge
• Thinking agents work without communication

2. Which of the following is the correct way to describe multiple agents working together?

easy

A. Agents divide tasks and communicate their findings.

B. Agents compete to solve the same task alone.

C. Agents work independently without sharing any information.

D. Agents ignore each other and solve unrelated problems.

Solution

1. Step 1: Identify correct teamwork behavior

   Multiple agents divide tasks and share results to solve complex problems.

2. Step 2: Eliminate incorrect options

   Options A, B, and D describe no communication or competition, which is not teamwork.

3. Final Answer:

   Agents divide tasks and communicate their findings. -> Option A

4. Quick Check:

   Task division + communication = teamwork [OK]

Hint: Look for teamwork and communication keywords [OK]

Common Mistakes:

• Choosing options that say agents work alone
• Confusing competition with collaboration
• Ignoring the need for communication

3. Consider this Python-like pseudocode for two agents working on parts of a problem:

agent1_result = 5
agent2_result = 7
combined_result = agent1_result + agent2_result
print(combined_result)

What will be the output?

medium

A. 57

B. 12

C. Error

D. None

Solution

1. Step 1: Understand variable values

   agent1_result is 5 and agent2_result is 7, both numbers.

2. Step 2: Calculate combined_result

   Adding 5 + 7 equals 12, so print outputs 12.

3. Final Answer:

   12 -> Option B

4. Quick Check:

   5 + 7 = 12 [OK]

Hint: Add numbers, not strings, to get sum [OK]

Common Mistakes:

• Treating numbers as strings and concatenating
• Expecting an error from simple addition
• Ignoring the print output

4. This code tries to combine results from two agents but has an error:

agent1 = 10
agent2 = 20
combined = agent1 + agent2_result
print(combined)

What is the error and how to fix it?

medium

A. Variable 'agent2_result' is undefined; change to 'agent2'.

B. Syntax error due to missing colon.

C. Cannot add integers; convert to strings first.

D. Print statement is missing parentheses.

Solution

1. Step 1: Identify variable names

   Code uses 'agent2_result' but only 'agent2' is defined.

2. Step 2: Fix variable name

   Replace 'agent2_result' with 'agent2' to fix the NameError.

3. Final Answer:

   Variable 'agent2_result' is undefined; change to 'agent2'. -> Option A

4. Quick Check:

   Correct variable names avoid errors [OK]

Hint: Check variable names carefully for typos [OK]

Common Mistakes:

• Assuming syntax error without checking variables
• Thinking addition of integers causes error
• Ignoring exact error message

5. In a system with three agents solving parts of a complex task, agent A finds data patterns, agent B cleans data, and agent C builds a model. Why is this multi-agent approach better than one agent doing all steps?

hard

A. Because one agent would do all steps faster without errors.

B. Because splitting tasks causes confusion and slows down work.

C. Because agents do not need to share results to succeed.

D. Because each agent specializes, speeding up the process and improving quality.

Solution

1. Step 1: Understand specialization benefits

   Each agent focuses on one task, becoming better and faster at it.

2. Step 2: Recognize teamwork advantage

   Sharing results lets agents build on each other's work for a better final model.

3. Step 3: Compare with single agent approach

   One agent doing all tasks may be slower and less effective due to multitasking.

4. Final Answer:

   Because each agent specializes, speeding up the process and improving quality. -> Option D

5. Quick Check:

   Specialization + teamwork = better results [OK]

Hint: Think specialists working together beat one multitasker [OK]

Common Mistakes:

• Believing one agent is always faster
• Ignoring the need for communication
• Thinking splitting tasks causes delays