Building a Reliable Multi-Agent Competitor Intelligence System with CrewAI

Single-agent LLMs hallucinate, miss context, or produce shallow analysis. Multi-agent systems fix this by giving specialized roles, opposing incentives (especially a dedicated critic), and layered reviews — exactly what the arXiv paper 2601.14351 demonstrated with >90 % internal error interception in real financial tasks.

In this tutorial you’ll build a production-ready competitor intelligence crew that takes any product/company and outputs a professional, fact-checked strategic report. Perfect for product managers, strategy teams, and developers who need trustworthy AI output.

What You’ll Build

Input: “Cursor AI code editor” + competitors Output: A polished Markdown competitive intelligence report (executive summary, SWOT tables, pricing comparison, risks, recommendations, sources).

Step 1: Setup (5 minutes)

# Recommended: use uv
uv pip install crewai crewai[tools]

# Or classic
pip install crewai crewai-toolsCode language: PHP (php)

Create a .env file:

OPENAI_API_KEY=sk-...
SERPER_API_KEY=...   # free tier at serper.devCode language: PHP (php)

Step 2: Complete Runnable Code (main.py)

import os
from dotenv import load_dotenv
from crewai import Agent, Task, Crew, Process
from crewai_tools import SerperDevTool, ScrapeWebsiteTool
from langchain_openai import ChatOpenAI

load_dotenv()

# Tools
search_tool = SerperDevTool()
scrape_tool = ScrapeWebsiteTool()

# LLMs (you can mix models for diversity & cost)
research_llm = ChatOpenAI(model="gpt-4o-mini", temperature=0.3)
critic_llm   = ChatOpenAI(model="gpt-4o", temperature=0.1)   # stronger reasoning
writer_llm   = ChatOpenAI(model="gpt-4o-mini", temperature=0.7)

# ====================== AGENTS ======================
researcher = Agent(
    role="Senior Market Researcher",
    goal="Gather comprehensive, up-to-date, evidence-backed information on competitors",
    backstory="You are an extremely thorough researcher who always verifies sources "
              "and digs into product pages, news, reviews, pricing, and recent launches.",
    tools=[search_tool, scrape_tool],
    llm=research_llm,
    verbose=True,
    allow_delegation=False,
)

analyst = Agent(
    role="Strategic Competitive Analyst",
    goal="Turn raw research into deep strategic insights, differentiators, and risks",
    backstory="You think like a top-tier management consultant (McKinsey/Bain level). "
              "You excel at identifying patterns, opportunities, and threats.",
    llm=research_llm,
    verbose=True,
    allow_delegation=False,
)

critic = Agent(
    role="Skeptical Quality Critic",
    goal="Find every factual error, hallucination, bias, logical gap, or missing perspective",
    backstory="You are a ruthless but constructive auditor. Your only job is to "
              "protect the team from bad decisions based on flawed analysis.",
    llm=critic_llm,
    verbose=True,
    allow_delegation=False,
)

writer = Agent(
    role="Executive Report Writer",
    goal="Produce clear, professional, visually structured competitive intelligence reports",
    backstory="You write for C-level executives. Your reports are concise, "
              "action-oriented, and use excellent Markdown formatting and tables.",
    llm=writer_llm,
    verbose=True,
    allow_delegation=False,
)

# ====================== TASKS ======================
research_task = Task(
    description="Research {product} and its main competitors ({competitors}). "
                "Cover features, pricing, recent news, customer sentiment, and market positioning.",
    expected_output="Detailed research summary in Markdown with sources and quotes.",
    agent=researcher,
)

analysis_task = Task(
    description="Perform deep competitive analysis: SWOT, differentiation, risks, "
                "and strategic recommendations based on the research.",
    expected_output="Structured strategic analysis with tables and clear insights.",
    agent=analyst,
    context=[research_task],
)

critique_task = Task(
    description="""Critically review the entire analysis. Flag any:
- Factual inaccuracies or outdated information
- Logical inconsistencies or biases
- Missing important angles
If everything is solid, reply exactly 'CRITIQUE PASSED'.
Otherwise reply 'REVISIONS NEEDED:' followed by detailed fix list.""",
    expected_output="Critique result + revision instructions if needed",
    agent=critic,
    context=[research_task, analysis_task],
)

report_task = Task(
    description="Write the final executive competitor intelligence report. "
                "Incorporate all previous work and any critic revisions. "
                "Include executive summary, tables, and actionable recommendations.",
    expected_output="Complete professional Markdown report ready for executives.",
    agent=writer,
    context=[research_task, analysis_task, critique_task],
    output_file="output/competitor_report.md",
)

# ====================== CREW ======================
crew = Crew(
    agents=[researcher, analyst, critic, writer],
    tasks=[research_task, analysis_task, critique_task, report_task],
    process=Process.hierarchical,          # manager coordinates
    manager_llm=ChatOpenAI(model="gpt-4o", temperature=0.1),
    verbose=2,
    memory=True,          # unified memory system
    cache=True,
)

# ====================== RUN ======================
if __name__ == "__main__":
    result = crew.kickoff(inputs={
        "product": "Cursor AI code editor",
        "competitors": "GitHub Copilot, Windsurf, Replit Agent"
    })
    
    print("\n✅ Report generated!")
    print("📁 Saved to: output/competitor_report.md")Code language: Python (python)

Mastering CrewAI Flows: Building Hierarchical Multi-Agent Systems | by Jishnu Ghosh | Medium

Step 3: Run It

mkdir output
python main.pyCode language: CSS (css)

You’ll get a high-quality report with far fewer hallucinations thanks to the dedicated critic.

Going Further: Automatic Retry Loop with CrewAI Flows

For true “Team of Rivals” reliability, wrap the crew in a Flow that automatically loops back to the analyst if the critic rejects the output (max 3 attempts).

See the official self_evaluation_loop_flow example in the CrewAI repo for the pattern — it’s only ~30 extra lines and gives production-grade self-correction.

---

Resources

• CrewAI docs: https://docs.crewai.com
• Official examples repo (including stock_analysis and self_evaluation_loop_flow): https://github.com/crewAIInc/crewAI-examples
• Original paper that inspired this: https://arxiv.org/abs/2601.14351

---

🚀 Ready? Start your free trial on today

---

• Discord – Share your thoughts
• GitHub Repo – Code of blog articles ready to start
• Follow Us on X @regolo_ai
• Open discussion on our Subreddit Community

---

Built with ❤️ by the Regolo team. Questions? regolo.ai/contact or chat with us on Discord