*𝓣𝓲𝓶𝓮𝓼𝓽𝓪𝓶𝓹𝓼 𝓫𝔂 𝓘𝓷𝓽𝓮𝓵𝓵𝓮𝓬𝓽𝓒𝓸𝓻𝓷𝓮𝓻* 0:00 - CrewAI Key Updates 0:56 - Creating a New Project with CrewAI's CLI 3:30 - Exploring YAML Configs for Agents and Tasks 4:53 - Example Project: Blog Post Creator 7:29 - Planning Steps: New Feature Overview 10:30 - Training Your Crew: Improving Output Quality 16:20 - Testing and Evaluating Crew Performance 21:10 - CrewAI's New Features and Future Potential
🎯 Key points for quick navigation: 00:00:00 *🤖 CrewAI Updates Overview* - Introduction to recent updates in CrewAI, - Features like CLI added for easier creation of projects, - Enhanced usability for non-coders with YAML structure. 00:02:25 *🏗️ Setting Up CrewAI Projects* - Use of command line to create new crews and related config files, - Facilitation for non-coders to contribute through config files, - Overview of different agent roles and tasks setup. 00:05:26 *🔍 Tools and Environment Setup* - Details about agent tools and environment variable integration, - Using different models for varied purposes, - Setting up planning flags and enabling detailed logging. 00:07:50 *🚀 Planning Steps Feature* - Introduction to new planning steps feature in CrewAI, - Description of its utility and implementation, - Example of step-by-step execution logs. 00:10:21 *🎓 Training Process and Feedback* - Overview of training methodology for improving prompt responses, - Feedback mechanism for refining output quality, - Example walkthrough of using the training component. 00:14:24 *📊 Enhancing Output with Training* - Iterative feedback improving content structure and style, - Creating consistent output across different inputs, - Saving training data to guide future tasks. 00:16:54 *🛠️ Testing and Evaluation* - Process of conducting tests using multiple iterations, - Performance scores for output evaluation, - Considerations for model selection in testing. 20:57 *🚀 CrewAI Enhancements and Performance* - Discusses improvements in CrewAI, emphasizing the benefits of new training strategies, - Highlights the importance of additional planning features for better results, - Mentions ongoing development and reliability concerns, suggesting LangGraph for more reliable needs. 21:58 *💡 User-Friendly Features and Future Plans* - Describes the ease of use for non-coders with CrewAI's new features, - Plans to explore more agent ideas and improvements in future content, - Invites viewer engagement through comments and encourages subscribing for future updates. Made with HARPA AI
Ahh I have pulled it apart. lol Actually wondered if I should make a seperate video about all the tricks they are doing. It is a set of dicts with the human in the loop guidance and example changes. Thanks Mike.
@@samwitteveenaiI’d love a more deep dive, including all those nice tricks. Those deeper dives also help with concepts when doing more low level things in LangGraph. My favorite tool for learning all of these is PraisonAI. It uses LangGraph, CrewAI, AutoGen, and more.
If you're trying to push scale. I'm doing around 4 billion tokens every week. Input and output total ... I highly recommend utilizing Langchain chain or llammer index looking at the memory agent . Also utilizing prompt caching in between the agents. We cut our bill by 86%
Utilizing Crew AI is great for building out an MVP, but it's not user-friendly enough to start manually changing some deeper configurations. I still think it's a great framework, it's just too attractive for when you have to get real work done with volume.
21:52 I already heard what you said at the end of the video, that rather than being a matter of preference it is more a matter of how precisely we want the agents to work and how easy it is to create them, if we need precision we will surely use LangGraph
Awesome! Is it possible to convert and run Crew AI agents in web UI, either with Streamlit or React/Next.js? It would be great to have an admin area to manage all Crews as well
I have actually been playing with making a LangGraph version of this. Let me see how it goes, I like the flow engineering parts and some cool stuff in the paper as a whole.
@@samwitteveenai Yes, would love to see the comparison between LangGraph and CrewAI/autoGen on specific case. love to use high-level framework but struggle with the reliability issue
Not a big fan of CrewAI framework just yet. When you can mix non AI tasks ( not dependent on LLM ) coupled with improvements in speed will give it a shot again.
totally understand not liking CrewAI, I gotta be fair though and say it has gotten better and they have some nice ideas in there now with these updates, but still you give up a lot of control for the ease of use.
crewAI creator here, great video, super well explained!
*𝓣𝓲𝓶𝓮𝓼𝓽𝓪𝓶𝓹𝓼 𝓫𝔂 𝓘𝓷𝓽𝓮𝓵𝓵𝓮𝓬𝓽𝓒𝓸𝓻𝓷𝓮𝓻*
0:00 - CrewAI Key Updates
0:56 - Creating a New Project with CrewAI's CLI
3:30 - Exploring YAML Configs for Agents and Tasks
4:53 - Example Project: Blog Post Creator
7:29 - Planning Steps: New Feature Overview
10:30 - Training Your Crew: Improving Output Quality
16:20 - Testing and Evaluating Crew Performance
21:10 - CrewAI's New Features and Future Potential
is it an automatic service?
Sam, that‘s great. Thx for the summary and the insight. Life is too short to keep up with every framework. 🎉
Great video again, again and again. Thank you so much for not beating around the bush and only sharing useful information
🎯 Key points for quick navigation:
00:00:00 *🤖 CrewAI Updates Overview*
- Introduction to recent updates in CrewAI,
- Features like CLI added for easier creation of projects,
- Enhanced usability for non-coders with YAML structure.
00:02:25 *🏗️ Setting Up CrewAI Projects*
- Use of command line to create new crews and related config files,
- Facilitation for non-coders to contribute through config files,
- Overview of different agent roles and tasks setup.
00:05:26 *🔍 Tools and Environment Setup*
- Details about agent tools and environment variable integration,
- Using different models for varied purposes,
- Setting up planning flags and enabling detailed logging.
00:07:50 *🚀 Planning Steps Feature*
- Introduction to new planning steps feature in CrewAI,
- Description of its utility and implementation,
- Example of step-by-step execution logs.
00:10:21 *🎓 Training Process and Feedback*
- Overview of training methodology for improving prompt responses,
- Feedback mechanism for refining output quality,
- Example walkthrough of using the training component.
00:14:24 *📊 Enhancing Output with Training*
- Iterative feedback improving content structure and style,
- Creating consistent output across different inputs,
- Saving training data to guide future tasks.
00:16:54 *🛠️ Testing and Evaluation*
- Process of conducting tests using multiple iterations,
- Performance scores for output evaluation,
- Considerations for model selection in testing.
20:57 *🚀 CrewAI Enhancements and Performance*
- Discusses improvements in CrewAI, emphasizing the benefits of new training strategies,
- Highlights the importance of additional planning features for better results,
- Mentions ongoing development and reliability concerns, suggesting LangGraph for more reliable needs.
21:58 *💡 User-Friendly Features and Future Plans*
- Describes the ease of use for non-coders with CrewAI's new features,
- Plans to explore more agent ideas and improvements in future content,
- Invites viewer engagement through comments and encourages subscribing for future updates.
Made with HARPA AI
Great video again, thanks Sam. Keen to see what is inside that pickle file.
Ahh I have pulled it apart. lol Actually wondered if I should make a seperate video about all the tricks they are doing. It is a set of dicts with the human in the loop guidance and example changes. Thanks Mike.
@@samwitteveenaiI’d love a more deep dive, including all those nice tricks. Those deeper dives also help with concepts when doing more low level things in LangGraph. My favorite tool for learning all of these is PraisonAI. It uses LangGraph, CrewAI, AutoGen, and more.
Can we disable telemetry yet? Last time I tried CrewAI, it didn't offer a way to disable it.
Hi Sam, I was struggling a lot with this last week. So I gave up... but now I am back2business :-)
Sam your videos are so helpful as always
The thing that worries me is the cost of going about a task this way, and the longevity of the framework when using it in production.
If you're trying to push scale. I'm doing around 4 billion tokens every week. Input and output total ...
I highly recommend utilizing Langchain chain or llammer index looking at the memory agent .
Also utilizing prompt caching in between the agents.
We cut our bill by 86%
Utilizing Crew AI is great for building out an MVP, but it's not user-friendly enough to start manually changing some deeper configurations.
I still think it's a great framework, it's just too attractive for when you have to get real work done with volume.
After reduction of LMMs cost I will return to play with CrewAI - thanks for update:)
lol this was partly my thinking as well. Still burnt a crazy amount of tokens
why dont you use local llms like Llama3.1?
Do you still prefer LangGraph over CrewAI?
21:52 I already heard what you said at the end of the video, that rather than being a matter of preference it is more a matter of how precisely we want the agents to work and how easy it is to create them, if we need precision we will surely use LangGraph
Hi sam, can you give tutorial about chatgpt free fine-tuning?
Awesome! Is it possible to convert and run Crew AI agents in web UI, either with Streamlit or React/Next.js? It would be great to have an admin area to manage all Crews as well
Sam you are a beast thank you! Would you be willing provide tutorial on how to setup and run the sakani ai scientist on github
I have actually been playing with making a LangGraph version of this. Let me see how it goes, I like the flow engineering parts and some cool stuff in the paper as a whole.
@@samwitteveenai Yes, would love to see the comparison between LangGraph and CrewAI/autoGen on specific case. love to use high-level framework but struggle with the reliability issue
🙏Thanks 🙏Thank 🙏Thanks
Very helpful videos about crewai
Please make video about NL2SQL & MYSQL TOOLS FROM CREWAI
Not a big fan of CrewAI framework just yet. When you can mix non AI tasks ( not dependent on LLM ) coupled with improvements in speed will give it a shot again.
totally understand not liking CrewAI, I gotta be fair though and say it has gotten better and they have some nice ideas in there now with these updates, but still you give up a lot of control for the ease of use.
Understood and thank you for the reply 🙏
Why not use it within a framework like Langgraph?
jrue can sleep better at night. he will retire happier than dame
Yes more on yhis