How to Build a Multi Agent AI System
ฝัง
- เผยแพร่เมื่อ 9 ก.พ. 2025
- Want to learn more about AI agents and assistants? Register for Virtual Agents Day here → ibm.biz/BdaAVa
Want to play with the technology yourself? Explore our interactive demo → ibm.biz/BdK8sd
WatsonxCrewAI Github → ibm.biz/BdK8ic
Ever wondered how to automate tasks with specialized AI Agents using Large Language Models? Nicholas Renotte shows you how to create a team of agents that can work together to complete tasks using AI prompting techniques. From creating tasks to generating keynote speeches, Nick covers it all in 15 minutes or less.
AI news moves fast. Sign up for a monthly newsletter for AI updates from IBM → ibm.biz/BdK8sD
#largelanguagemodels #AIprompting #aiagents
Yes, please, would love a series with agents performing more complex tasks across multiple platforms.
What other tasks are you looking to learn about?
How to fix a faucet from hanging over
A first task that be make the tasks needed to solve a certain project, then make the agents interact with their competences in each task and solve with the best solution possible, with multiple iterations like using set based design and a value evaluation of alternatives. Great content by the way.
@jgayoso do you work in AI agents as well???
This is an amazing presentation: entertaining , enlightening and helpful. Looking forward to all the sequels. Thank you!
From Kenya, and am impressed with the tutorial, just trying to self learn AI. Soon I hope to build an AI agent.
Very clean use of prompting.
That’s great, man! Create a crew. I’m 15, though. I don’t understand why you need to rush into this. I don’t see the urgency in your needs. You can explain the same in 30 minutes without rushing or putting pressure on yourself. To be honest, I feel like everyone wants to show how quickly things can be done, but speed isn’t the goal. Just relax and enjoy what we have.
Two reasons. 1. It's a demonstration, key noting ease of use and functionality. And 2. Because he's a badass and he can. Meaning, once you get to a level of mastery of anything, you function at a higher level.
I get it, modern world is too rushy... But here he states 15 minutes as a goal, actually emphasis the easy of use of this framework, yet remains understandable. I enjoyed this demo and think i'll give a try.
Wow. Idk coding but interesting to see your advanced syntax in action.
Next episode use Watson Discovery as a search tool. Would be awesome!
Nice!! Please continue this series with more complex cases.
Seein nick here is incredible.
Hey, the model used for the first agent is not supported by IBM any more. I tried use another, but I face a error about LiteLLM. Any tip?
yeah , cool work IBM. Multipitle AI agents in 15 minutes!
Is it possible to take these agents to another platforms ???
🙌great explainer, will def be trying this out
Lo uu yogur x😊😅😊😊😊😊😊
Excellent stuff
Amazing!! Thanks for sharing!
Hey everyone! 👋
I’m building a multi-agent setup on my own cloud (GCP) in a BYOC model. I’m considering deploying each agent as a separate Cloud Run service for scalability.
Would this approach be effective, or is it something LangGraph doesn’t recommend? Is it necessary to deploy agents independently for scalability, or are there better practices for managing multiple agents efficiently?
Looking forward to your thoughts! 🚀
awesome man, learnt real AI programming from it🎉
Hi you mentioned cost control. Can you estimate how much would one run of those two agents cost, using 500 tokens as maximum? Many thanks. Awesome video.
Ayyyyyy so glad to see you here
Thanks for this video. I really appreciated it. Please proceed with the serie of multi agent videos :)
Is there a course from IBM to build local Agents???
Great video 🎉
Do you work on AI Agents???
Great video. Many thanks for uploading this.
Hey Nicholas, I was missing you ;-) Keep It Rocking!
Hey Nicholas. Great Video. You are awsm as always😎
How does task2 know to use the output of task1 as its context? (The crewAI documentation mentions a Task context parameter, which you don't provide.) -- update: I answered my own question: "the output of one task is automatically relayed into the next one"
That was not clear to me before you said it.
This was my confusion as well. Thanks for providing the answer.
Can we take the AI agents to other platforms as well ???
The development of multi-agent AI systems is fascinating! Look at other platforms with enhanced features for more complete tools that make this procedure easier. AIDevelopment #AIAgents
is it possible that the user gets information, which agent provided which answere, so he can see, which model is responsible for the answere?
This is my first time hearing anything like this it's so fascinating and I want to learn more what should I do I know absolutely nothing about this but I think it's coding? Help me please anyone
Hey I know about this. I know that it's using the python programming language, and I know that there's probably some libraries and API's somewhere that make it work.
I ran into issues with pydantic when I tried to initially test.
great workflow:)
Nice job, v impressive how much he can do so quickly!
the wording "react prompting" isn't confusing at all.
Thank you.
Nice. Will be interesting to see building AI Agents Routing. Differents LLMs integrated and executing better models to specific issue.
Very cool, got it installed and running. Do the tokens control how much it can output ? I did one with 3 agents and it seemed to cut of the end of the output
Excellent work. Please put more use cases.
IBM is impossible to work with. I tried settting up an account like 20 times, talked to customer service and and they never set me up. I’m sure I can do this in GCP or AWS just the same. I also used Watson in a commercial setting and there was an outage like once a month. I’ll never work with IBM ever again.
Do you work in building the AI agents ???
is crewAI free?
So basically IBM Watson uses CrewAI ?
No, you can use different agent frameworks.
All the things he has done , is it free ? If not, then how can i make ai agent app with free sources ?
Is current AI technology capable of an agent that asks questions that I am interested in and express opinions that I agree with for the purpose of online conversations
Thanks for sharing this, How reliable these results are in real ?
Great!
how do u get Watsonx api, serper api key? are these open source or paid?
Would this be considered a cooperative multi agent system or a competitive multi agent system ?
Make it a series. Please
API Key ?? I dont like it. Hope there is possibility to run this locally.
What do you think about Zero agent?
do I need to add permissions in the cloud I keep getting error invoking WatsonLLM instance
Wow. So powerful.
You just glossed over the api key part though and that is the biggest entry barrier to using these services. Credit card required. Multiple accounts must be registered. I appreciate the video but man-- can you walkthrough these parts as well?
CrewAI is helping build the pathway to Cognitive Organizations!
Things moving so fast that even the Llama 3 used in this video is outdated. My days.
It's amazing how multi-agent systems can enhance efficiency and innovation in various fields. Have you tried building any multi-agent AI systems yourself? I'd love to hear more about your experiences!"
Hi Nicholas,
Awesome tutorial, but I'm encountering some issues. The model ibm-mistralai/merlinite-7b is deprecated, so I switched to a different one. I'm not sure if it's related, but I can't pass the information to the tool. I'm getting this error below. Can you help? Is anyone else facing the same issue? Thanks!
> Entering new CrewAgentExecutor chain...
I need to find promising research in the field of quantum computing. I will start by searching the internet for recent advancements in quantum computing.
Action: Search the internet
Action Input: {"search_query": "recent advancements in quantum computing"}
Failed to convert text into a pydantic model due to the following error: 1 validation error for CrewPydanticOutputParser
pydantic_object
subclass of BaseModel expected (type=type_error.subclass; expected_class=BaseModel)
Also a series with agents performing more complex tasks would be great!
Have @Wonderful 248th !/ / thanks
yo
King
Really bad. Trying to sell your product without even explaining what is an agent. Poor by İBM
I mean he said a phrase about agent, let's not forget that...
Great
but i guess, this is not free?
Llama3 is free
I follow you, my friend, and you are a very excellent person
Therefore, I hope to communicate with you as soon as possible, or via WhatsApp. I want to talk to you, brother.❤❤❤❤❤
Man you approach is totally WRONG. That's not how you teach people about anything. May be you need to learn first before you get to a point of teaching.
Fast talking videos with not good explanations
more
I hate autogen/crewai tbh. lacking in imagination.
Are u working on AI agents ???
Agents just suck. They are not good at all. You can simply ask chatgpt to do all of this instead of writing so many codes
Please don’t understand why you explain that you have to npm install crew ai … again your videos are always leaving initial installation information so videos are not helpful and fast talking is not helpful for layman and confusing for new beginners … all of videos never 👎🏾 give the setup properly
pip install crewai
simple as that 💀
@@GleniLame he could also just google or chatgpt it lol, definition of lazy