Multi-Agent Orchestration & Conversations using Autogen, CrewAI and LangGraph
Understanding with examples and code explanations
We are into the 2nd year of the Generative AI thunderstorm and it is still as strong as ever. The newest concept to add onto this never-ending innovation saga is Multi-Agent Orchestration which is quickly gaining traction in the developer community. We will be discussing MAO in my 151st blog post !
My debut book →
LangChain in your Pocket: Beginner’s Guide to Building Generative AI Applications is out now !
In this post, we will be discussing Multi-Agent Orchestration at length and implementation for some popular packages like Autogen, CrewAI and LangGraph
What is Multi-Agent Orchestration?
Different usecases
When to use Multi-Agent Orchestration?
Autogen (basics, group discussion, local LLM integration)
CrewAI (AI Tech team, Team blog writing)
LangGraph (basics, Debate application, Interview App, Movie scripting, Interview Panel)
So let’s get started
What is Multi-Agent Orchestration?
Multi-Agent Orchestration (MAO) refers to multi AI agents with different capabilities working together to solve a problem.
Some key features to consider:
The AI-Agents engage in a step-by-step discussion to reach a final solution.
The AI-Agents communicate in a conversational manner, resembling human interaction
The agent switching is autonomously managed by LLM.
Users can assign different roles to the AI-Agents within the team.
This actually is very exciting to play with and can be very handy to solve a number of complex usecases.
Usecases
Software Development : Using MAO, one can create multiple Tech AI roles like PM, Programmer, Code Reviewer and let an entire product lifecycle managed by this team. I’ve discussed this in one of the usecases.
Decision Making: A dummy round-table conference debating over a topic can be conducted where different Agents have different roles to take business decision.
Simulations : An entirely fake scenario with different agent playing different roles.
Movie Scripting on a given theme, Optimizing Logistics operation, etc.
Now, as we are well aware of what is Multi-Agent Orchestration, next we will discuss a few important packages and a handful of demos on some major applications:
Autogen
The most popular of the lot, Autogen by Microsoft is a MAO package which has expertise in task which require code execution and generation. You can find more about Autogen in the below beginners usecase explanation
Now, as you’re now well versed with the UserProxy and Assistant Agent being used in autogen (as discussed in the above example), in the next example, you can explore how to enable group discussion between different agents using Autogen
Though more compliant with OpenAI, it does support other APIs and even local LLMs using Ollama & LiteLLM as well
CrewAI
Though Autogen being more popular, I personally feel CrewAI is the easiest to use and the best if you’re not into programming. You can check a baseline example how I created two agents with different expertise to write a technical blog below.
Next up, you can even create an entire Technical team with a few lines of code using CrewAI with ease
LangGraph
LangGraph is an extension of the famous LangChain package, which is a lot more complex than both autogen and CrewAI but gives way more flexibility and customization compared to other. Trust me, if you’re good at programming, you’re gonna love it. Also, it is not just limited to MAO but can be extended to many other LLM based usecases. If you are new to LangGraph, you can check the below tutorial
I even developed a debate app using LangGraph which creates two AI-Agents depending upon the topic that debates on a given topic
There are other packages like ChatDev, OpenDevin, etc as well which I haven’t covered but are equally good and popular. Do check them out as well.
Not just this, you can even perform complex usecases like 1–1 interviews and even and entire interview incorporating multiple rounds and interviewees
And even movie scripting
See you soon !!
