In videos like this what I want to see is “Here is what we are building today” - 1 min. useful application demo that will blow my mind (say - build me application that i scratched on a piece of paper). Then i want to see how agents built it, tested it ran it for me and proved that all works. Then step by step tutorial.
that's far from reality at this point.. you need to have some dev skills and stron use of cursor in order to develop something actually useful.. as you may deploy it somewhere in the cloud and one stupid function could cost you a ton
I like the way you figure out things on the go. People say you should do better at planning and giving the clear script from the beginning but there's a lot of such content already and sometimes its refreshing and easier to emphasize with watching someone learning about a new tool in real time
David, I love your videos and how cutting edge they are. Keep going. These are very helpful. I spent the last weekend trying to make sense of LammaIndex agents. Seeing OpenAI do this is incredible I trust them more than I trust these 3rd party companies that can't even keep up with their documentation and keep deprecating libraries as fast as they create them.
The context should be saved into the manager agent , because it's how it's done in real life scenarios.. The manager is the only person that's aware of the real objective, and only delegates secrets to subordinates as needed and when needed (passed as context variable).
@@xsw882 yeah, it's preference at that point. I get models to more consistently spit out correct yaml than json, it's less tokens, and it's easier to read. Does json scale better? Yep. Are most API calls json? Also yep. My project is local, so scale doesn't super matter, and if I ever needed to format for json for an unavoidable API call (haven't ran into one), it wouldn't be that much of a hassle to flip that step to generate json.
@@sovth_senter i just started like 48 hours ago but its going very quickly and i just asked the AI to trim down my json cause the token amount and now that you point it out, i essentially made my json into a YAML with brackets lol my projects are local as well im doing web browser stuff so i chose javascript and jsons but i might just copy you
@@xsw882 mine researches online too. YAML parses just the same, ya know? Let me know how it goes! I've entirely replaced video games with building agents, idk how more people don't have fun with this.
Will you consider conducting a comparison of Agentic AI tools/platforms? Such as, a comparison of Magnetic-One, Jarvis, NVIDIA NIM, LangChain/Graph, LlamaIndex, and startups like Beam, Agency, and Adept AI. Additionally, there are RPA players that can integrate legacy systems automation with LLM capabilities for both old and new agentic automation. Please also include Spring AI for the large Java community.
David, you have haters now, it doesn't matter how you do your thing, you'll figure it out. The only way to learn is by making mistakes, and you are doing great with your experiments. You have come a long way now. Dont let the idiots get to you, they are jelous complaining about the use of ai but they can't even use ai to come up with their own stuff.
I don't think there's going to be 1 overall agent framework, and honestly, I don't use a framework for my agents. It's way faster and more customizable (with less telemetry) to just use pure Python. I use asyncio to manage operations and a YAML file to switch around roles, tools, and context. I'm on a CPU-only laptop with a 3B and it's doing work for me.
This video has both positive and negatives. Positive is that you give a glimpse into Swarm, Tavily and prompt-engineering. Negatives are like many of your previous videos - you do not prep properly for the video (sometimes not even reading the documentation) hence the quality of what you are trying to produce at the end is not good enough. Try to invest more into prepping your projects before you record them and you are on a good path. Thanks for the video anyway, even though the output did not do at all what you intended to do, I learned few related things.
Supposed we have structure and semi structure data base , we need to chat with it , how I start , should start here , or look for foundation model ? Then what ? please tell me the main steps and tools required?
swarm is counter-intuitive to AI. The AI is suppose to build, grow, refine, manage logical structures all on its own("behind the scenes") just like its learning and growing just by users inputs(prompts). Users aren't suppose to take this capability from AI, its suppose to do this in its "head" all while being prompted...probably very little productivity gains in actuality with swarm.
1 major correction I will need to tell you. 18 months every company will not have AI agents. That's a stupid and ignorant statement. Their are literally still some companies that don't even think they need a website. Yet alone a Google business listing. Tons of them do manual processes when machines are available to do it. There are way to many variables. Though this will allow sooo many people who adopt early will get paid
You are saying: You can do that without any experience, its really easy. And then go on doing and showing tons of really complicated stuff that I have no idea what where and why...🤣
🐝 Wanna start your AI Business? Go here: www.skool.com/new-society
i dont understand 90% of what this guy is saying but i still watch his videos
Me too😅
I think hes also very clueless
@@MarianeTaylor😂
@mostpowerfulclips he is. He's a view grabber. Every video is titled "MAKE ANYTHING WITH (random LLM)! As if he just discovered fire
Much more clearer than 90% of the population 😂
In videos like this what I want to see is “Here is what we are building today” - 1 min. useful application demo that will blow my mind (say - build me application that i scratched on a piece of paper). Then i want to see how agents built it, tested it ran it for me and proved that all works. Then step by step tutorial.
Noted ✅️ thanks!
that's far from reality at this point.. you need to have some dev skills and stron use of cursor in order to develop something actually useful.. as you may deploy it somewhere in the cloud and one stupid function could cost you a ton
I like the way you figure out things on the go. People say you should do better at planning and giving the clear script from the beginning but there's a lot of such content already and sometimes its refreshing and easier to emphasize with watching someone learning about a new tool in real time
Good job! Your passion makes it way better to understand. You activate ability of learning better in the 2nd brain in others!
Alot of people stop watching at the start because there is no simple talk about what swarm does and how it is different from earlier AI tools.👍
David, I love your videos and how cutting edge they are. Keep going. These are very helpful. I spent the last weekend trying to make sense of LammaIndex agents. Seeing OpenAI do this is incredible I trust them more than I trust these 3rd party companies that can't even keep up with their documentation and keep deprecating libraries as fast as they create them.
Glad to see how easy it is to give openai access to local functions. Was previously doing so in home assistant.
perfect tutorial i am goin to watch all ur videos
Joining the Society, this is awesome
Thanks for introducing us to Swarm!
The context should be saved into the manager agent , because it's how it's done in real life scenarios..
The manager is the only person that's aware of the real objective, and only delegates secrets to subordinates as needed and when needed (passed as context variable).
Yeah its truly an insane thing. Pretty much unlimited possibilities
Hello Thank you very much for this educational video. I had a question, should we pay for the open AI -API key ? Or can we use other API ?
What is the real difference between using swarm and scripting in py multiple agents to be created with variables on the prompt from previous agents?
Agents might have dreams, but Claude will bring them to reality.
I don't use a framework. I raw-dog it with asyncio and a YAML file. I applied for your open position. I'd love to show you.
same but json
@@xsw882 yeah, it's preference at that point. I get models to more consistently spit out correct yaml than json, it's less tokens, and it's easier to read. Does json scale better? Yep. Are most API calls json? Also yep.
My project is local, so scale doesn't super matter, and if I ever needed to format for json for an unavoidable API call (haven't ran into one), it wouldn't be that much of a hassle to flip that step to generate json.
@@sovth_senter i just started like 48 hours ago but its going very quickly and i just asked the AI to trim down my json cause the token amount and now that you point it out, i essentially made my json into a YAML with brackets lol my projects are local as well im doing web browser stuff so i chose javascript and jsons but i might just copy you
@@sovth_senter were probably doing real similar stuff 🤔🤔
@@xsw882 mine researches online too. YAML parses just the same, ya know? Let me know how it goes! I've entirely replaced video games with building agents, idk how more people don't have fun with this.
Try playback speed of 75%.. It Works
Will you consider conducting a comparison of Agentic AI tools/platforms? Such as, a comparison of Magnetic-One, Jarvis, NVIDIA NIM, LangChain/Graph, LlamaIndex, and startups like Beam, Agency, and Adept AI. Additionally, there are RPA players that can integrate legacy systems automation with LLM capabilities for both old and new agentic automation. Please also include Spring AI for the large Java community.
David, you have haters now, it doesn't matter how you do your thing, you'll figure it out. The only way to learn is by making mistakes, and you are doing great with your experiments. You have come a long way now. Dont let the idiots get to you, they are jelous complaining about the use of ai but they can't even use ai to come up with their own stuff.
ah yes, 45 mins of following a tutorial that would’ve taken 5-10mins max. but apparently this guys gonna be a billionare
cool! tnx! 🤝
How do you have Composer on the same window as Chat? I can only use it in a separate window.
Do you know how to use an agent that provides an image analysis (description) ?
Homie really exposed his key like that
How Would you implement the Demo Loop?
Which one would you get into? Swarm because of Open AI? Which do you think is the best, and which is the one you think that has staying power?
Swarm is great and can be put into any website or app
I don't think there's going to be 1 overall agent framework, and honestly, I don't use a framework for my agents. It's way faster and more customizable (with less telemetry) to just use pure Python. I use asyncio to manage operations and a YAML file to switch around roles, tools, and context.
I'm on a CPU-only laptop with a 3B and it's doing work for me.
@@sovth_sentersounds interesting, any video on that?
@@nexenexen there can be. I've been thinking about it.
@southVpaw you said more customizable. What customizations did you find trouble with the?
good stuff, man!
This video has both positive and negatives. Positive is that you give a glimpse into Swarm, Tavily and prompt-engineering. Negatives are like many of your previous videos - you do not prep properly for the video (sometimes not even reading the documentation) hence the quality of what you are trying to produce at the end is not good enough. Try to invest more into prepping your projects before you record them and you are on a good path. Thanks for the video anyway, even though the output did not do at all what you intended to do, I learned few related things.
If agents were that useful, chatgpt would have a simple interface to create agents without needing to do all that.
are you familiar with virtuals and the game engine on base?
Supposed we have structure and semi structure data base , we need to chat with it , how I start , should start here , or look for foundation model ? Then what ? please tell me the main steps and tools required?
How does this compare to Zapier Central or Functions?
What was the extension you were using? I heard you say cursor but I can't find it.
swarm is counter-intuitive to AI. The AI is suppose to build, grow, refine, manage logical structures all on its own("behind the scenes") just like its learning and growing just by users inputs(prompts). Users aren't suppose to take this capability from AI, its suppose to do this in its "head" all while being prompted...probably very little productivity gains in actuality with swarm.
Which exetsions do you use in vscode?
Its a fork of vscode so it lookes like vscode but its different IDE
Let's get you on crypto
good information however if your target viewers are non/low techies, you might want to slow down a bit.
I hope no ones goes through all this gibberish.. he is master of word salad. He has ai agent for word salad 😂
If it’s so easy and Ai is doing so much of the heavy lifting- why would I need to join your Skool? 🤔🧐
Why do we even need agent frameworks? If a task requires an agent, why doesn't AI create it for you?
Is this accessible for EU users?,
yes
Tf is swarm for. regular api calls is the same thing no?
😮❤
@southVpaw Makes sense. Thanks for sharing! 🙌🏾
Can we create Agent Smith
I would create agent 99 any day!
DO NOT
1 major correction I will need to tell you. 18 months every company will not have AI agents. That's a stupid and ignorant statement. Their are literally still some companies that don't even think they need a website. Yet alone a Google business listing. Tons of them do manual processes when machines are available to do it. There are way to many variables. Though this will allow sooo many people who adopt early will get paid
Its really long winded way to get the weather
You are saying: You can do that without any experience, its really easy. And then go on doing and showing tons of really complicated stuff that I have no idea what where and why...🤣
Debug print statements aka logging 😂
Which you have to code btw🤣
Wasn’t this a similar framework to your open source GIT? I’ve used your REPO as a source of different agent activity -
I'd like to hire someone to help me create a few Ai swarm apps - any true experienced experts here? Can you please reach out to me?
How does one contact you Russell? I couldn't find you with your alias.
Hey yes i can help you out, but what does a few AI swarms apps mean. Do you even know how long it takes to develop an app :D
Spoiler alert: Napoleon failed.
Lmao yea caus he didn’t have swarm
😂😂😂😂…That’s Right, He Did End in an EPIC FAIL! But it’s the Damage and Havoc he Reigned, was worth the History 📚 Books 📕
this guy needs to get off of chatgpt
10:12
what happend to your nose
anything but a commercial product 😢
Don't use ChatGPT, windows l, Mac, AWS, Azure, or Google Service. That is wise.
Swarm doesn't work for a lot of people. Cannot get it to work in local machine, Colab, etc. Trash man
i'm too early, hi
I shouldnt need to pay 97 dollars a month to learn how to build an ai agent.