🧪Lab Session: Build a Real-World Multi-Agent AI Crew With CrewAI

Introducing 🧪Lab Sessions. This session is a paid session and will be structured as follows:
First 20 minutes

• Demonstrating the multi-agent crew we are going to build
• Providing you with the prompts, configs, and step-by-step workflow. Only Available To Attendees.

Remaining Session:

• You will build, and we will be on standby to help when you get stuck.
• You will leave with having built the crew.

What You Will Need For This Lab: ( But we can help you set it up too)

• Python 3.10 or higher installed
• VS Code (or your editor of choice) with a terminal
• CrewAI installed: pip install crewai crewai-tools
• An LLM API key (OpenAI or Google Gemini both work)
• Watch The CrewAI Fundamentals Session In The Resources If You Did Not Attend

Learning Outcomes
1. Define agents with role, goal, and backstory
Learn to configure a CrewAI agent with a role, goal, backstory, and LLM, so you skip writing system prompts by hand.
2. Write tasks and orchestrate them with a crew
Learn to set a task description, agent, and expected output, then run the crew sequential, concurrent, or hierarchical.
3. Plug in real tools like the Serper search API
Learn to attach the Serper Google search API to an agent so its answers come from live data, not just training data.
4. Let the agent write its own custom function
Learn to set allow_code_execution so the agent writes and runs its own Python when no pre-made tool fits the problem.
5. Run agent-written code safely in Docker
Learn to set code_execution_mode to safe, plus a timeout and max_retry_limit, so agent code never touches your machine.
Why This Topic Matters
Real work rarely fits the tools you already have. A client asks for a number nobody tracks, or a report that pulls from three places at once, and you are back doing it by hand at 11pm. A crew handles that: agents that search live data, hand work to each other, and write their own Python in Docker when no tool exists. Build one and you stop being limited by what a single chat window can do.