Author(s): Rick Hightower

Originally published on Towards AI.

Part 1: The throwaway orchestration glue you keep rewriting every time two LLM calls have to cooperate is exactly the thing CrewAI was built to delete.

You have done this before. Two model calls need to cooperate: one drafts, one reviews. So you write the glue. You pass the first output into the second prompt, parse the response, add a retry when the reviewer’s JSON comes back malformed, and thread through a third call when the review says “redo it.” Somewhere around the fourth if branch, you realize you have hand-built a small, brittle workflow engine that exists only to make two LLM calls behave like coworkers. The next project, you build it again.

After the introduction, the article explains what CrewAI replaces: custom loops for tool-use and multi-step cooperation, bespoke orchestration code, and tight coupling to a single model provider. It introduces the core primitives—Agent, Task, and Crew—and contrasts them with higher-level orchestration concepts like Flows. Using a “buggy-shop” example repository, it clarifies how the framework’s loop runs internally via kickoff(), how expected_output works as acceptance criteria, and why the simplest “hello crew” example is the right starting point. The post also covers practical setup (installing the CLI, setting API keys, and using verbose traces), then previews the real project shape generated by the CLI (YAML configs plus a small Python wiring class). Overall, it emphasizes the mindset shift: describe the team and the work, not the control flow.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI