Member-only story
Building Autonomous Multi-Agent Systems with CrewAI
What‘s Multi-Agent Autonomous System and How to build one with CrewAI and LangChain?

Motivation
In fact, we are not unfamiliar with these concepts; we know them from movies. A person commands their AI, and the AI carries out these commands by using various tools. This is the path we are on today with the rise of AI systems. The era is gradually changing. In the past, people couldn’t undertake a task alone and would need a team. Without a team, they would either run out of energy after a while or hit the limits of their abilities. In the end, successful projects come from teams made up of individuals with different skills.
Teamwork makes the dream work.
However, these days a new technology has started to make a name for itself. We can call it the next phase of AI before AGI: ‘Agents.’ So, what are these agents? Before diving into the code, let’s talk a bit about the structure of multi-agent systems
How does it work?
To put the equation simply: Multi Agent Systems = AGENTs + TOOLs + TASKs
It’s a system where multiple agents are equipped with various tasks and tools.
Agent
We are familiar with role-playing games, where your character has a role, like a warrior, for example. Throughout the game, you put yourself in their place, aiming to complete the game by finishing the quests that shape their backstory from one adventure to the next. Similarly, researchers have discovered that large language models (LLMs) can be motivated to perform tasks optimally when given roles, backstories, and objectives. This allows us to motivate LLMs to carry out various tasks with just a few simple prompts.
Agents essentially break down the assigned tasks into simple steps and then execute those steps by ‘thinking’ — yes, thinking — through them in sequence. This enables us to create an agent that not only performs steps thoughtfully but also consults other agents with different areas of expertise, rather than relying on a single LLM to input prompts and receive outputs.