chapter three

3 Building a multi-agent crew

 

This chapter covers

  • Crew project setup
  • Specialized agents and task design
  • Knowledge sources in crews
  • Short-term, long-term, and entity memory
  • Delegation and process types

After implementing a single agent in chapter two, we’ll now explore the most important capability of building agentic applications with CrewAI: assembling multiple highly specialized agents into a crew and having them collaborate on a real, multi-step workflow. This matters because most production use cases, such as marketing pipelines, research assistants, sales ops, content generation, and internal tooling, rarely consist of a single prompt. They’re sequences: gather context, analyze, decide, produce, enrich, publish, and so on. A single agent can do parts of that, but a crew lets you split the work into roles, enforce order, and reuse tools and knowledge across the whole workflow.

This chapter is for you if you want to automate a process that today still requires a human to coordinate the steps manually. Think of a workflow where someone needs to understand the business, identify relevant competitors, select valuable keywords, and turn those insights into high-quality blog posts. By the end of this chapter, you'll have a crew that handles all of this automatically.

3.1 Initializing the project

3.2 Creating the keyword researcher

3.3 Creating the topic researcher

3.4 Creating the blog writer

3.5 Loading knowledge

3.6 Assembling the crew

3.7 Understanding process types

3.8 Summary