Unbiased Review Langraph
I’m sure there’ll be lots of discussion whether I’m right or wrong in my assessment. I welcome you to the comments section to provide your point of view. We’ll also explore the potential real world applications of this a a agents.
How could businesses use language of agents to improve their operations? What industries could benefit from Autogen’s approach? And how might Kuya agents change the game when it comes to customer service? So if you’re thinking about integrating AI agents into your business, this video will give you the knowledge you need to make an informed decision even if you’re not a developer.
So, whether you’re a tech geek like me, a business owner looking to stay ahead of the curve or you’re someone who is curious about the future of AI, you wouldn’t want to miss this deep dive into the world of AI agents. But before we get into the specifics of each framework, let us take a step back and understand why multi agent collaboration is so important.
Andrew Ng pointed out at the recent Ascent Conference. Given a complex task like writing software, a multi agent approach would break down the task into subtasks to be executed by different roles. Such such as software engineer, product manager, designer, and quality engineer. With different agents accomplishing different subtasks. Now, you might be thinking, wait a minute, if we’re just making multiple calls to the same large language model, why bother with multiple agents? Well, as Andrew Ng explains, there are several compelling reasons. 1st, it works way better than just using LLM calls.
Many teams are getting good
results with this method. And ablation studies shown here on the screen shows that multiple agents give superior performance to a single agent no matter what LLM you’re using. Number 2. Even though some LLMs can accept very long input contacts, their ability to truly understand long, complex inputs is mixed.
An agentic workflow in which the LLM is prompted to focus on one thing at a time can give much better performance. And number 3, perhaps most importantly, the multi agent design pattern gives us as developers a framework for breaking down complex tasks into subtasks, which allows us to create a much more complex task and being able to debug it much easier. With that context in mind, let’s dive into our first framework, AutoGen.
The oldest and most mature framework of the 3, AutoGen supports multi agent systems and streaming output, making it a versatile choice for complex projects. When it comes to customization, Origin allows you to update agent system messages, giving you the flexibility to tailor your agents to your needs.
One of the standout features of Origin is a containerized code execution, which provides an extra layer of safety. Basically, you can protect your system from any potentially harmful code that l l m can accidentally develop. Skynet has become self aware. Additionally, Autogen’s feedback cycle enables agents to solve issues autonomously, saving you time and effort in the long run.
Mastering Agents: LangGraph Vs Autogen Vs Crew AI – Galileo
When it comes to randomness, probably one of the biggest weaknesses of AutoGen is hard to fine tune the outcome of the AutoGen MultiGen application. Also, there’s still problems with infinite loops where unless you put fairly conservative number of max iterations, you will run out of your OpenAI budget very fast. Now, let’s talk about the user experience.
AutoGen Studio offers a UI layer, but it can be a bit slow and not very intuitive. Kind of like navigating a maze blindfolded. The computation is pretty decent, but the studio might leave newcomers scratching their heads. It’s a little bit confusing. Coil quality is another area where AutoGen could improve with some verbosity and self recursive functions that might make your code look like a tangled mess of spaghetti. But wait, there is more. Let’s check out Lang Graph.
But before we move on, if you’re finding this information valuable, let us know in the comments. Love hearing about your experiences with this AI agents. Next up, we have a Langraph, the newest framework of the bunch. This training new technology uses direct to cachely graphs as a foundation for its agent applications. Langraph provides a good mental model for users.
By defining notes and specifying agents can be a bit verbose, like filling out a text form for every little thing. On the bright side though, Langraph has a good documentation with with clear examples, so you won’t feel like you lost in a foreign country without a map. What is also amazing is that the code quality is also cleaner and better organized compared to AutoGen, making it easier to navigate and maintain.
Language of examples focus on web browsing and scraping, customer service, info gathering, code assistance. Also, the systems that you could build with Langraph are not only that. Basically, you could have multiple agents working in collaborative approach or a supervisional approach, and you could actually have different level systems. Like, you have a manager, you have a director, you have CEO.
It’s innovative because with Langraph, you could build anything that AutoGen can do, but, also, you could build much more. So we’re a big fan of Langraph at AiSiMP. We think it’s probably one of the best technologies that came out in 2024, and we’re excited to use it for our own projects.
Now get ready for the CrewAI. They’re doing something completely different. Last but not least, we have a krui AI, the middle child of the group. It’s like the Goldilocks of agentic frameworks. Not too old, not too new, but just right. CrewAI boasts an intermediate level of maturity and hierarchical agent structure.
Although it lacks native support for dynamic planning. Customization, Crew AI is a breeze, thanks to support for agent and task definitions. Also, Crew AI is built on Langchain, which makes it compatible with Langsmith, which as we saw with Langraph, is invaluable tool when it comes to debugging and optimizing your agents. The documentation is stop touch.
It was clear examples, core concepts, nicely explained. We have how to guides from installing, getting started, creating custom tools, trip planner, create Instagram post. This one is was very interesting example, which I will link the video of 1 of the creators. I forgot his name. He has a very cool video about how to use llama 3 with crew AI to create Instagram posts.
Very advanced, to be honest with you. As I mentioned before, integration is another area where crew AI shines. It can be integrated with other systems like Langraph and works with both local and global LLMs, making it a versatile choice for AI engineers and entrepreneurs whether you wanna host your own solution within the company or you wanna have it available via API to the rest of the world.
In conclusion, choosing the right agentic framework depends on your specific needs and priorities. Whether you value maturity, ease of use, or unique features, there is a framework out there for you. As Andrew Ng notes, the output quality of multi agent collaboration can be hard to predict. Especially, when allowing agents to interact freely and providing them with multiple tools.
However, the more mature patterns of reflection and tool use are much more production ready systems. If this video has helped you understand AI agents better, give it a thumbs up and let us know which approach you’re most excited to try out. And if you’re interested in learning more, be sure to check out the papers recommended by Andrew Ng in the description below.
Until next time, happy coding, and may your AI agents be ever in your favor.
Hi, this is a comment.
To get started with moderating, editing, and deleting comments, please visit the Comments screen in the dashboard.
Commenter avatars come from Gravatar.