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.