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.