Why not use the agent system itself to optimize agent creation? Start with an agent in charge of creating and overseeing other agents as needed to optimize towards a specific goal. 🖖😸👍
@@WhatsAI First of all you are advocating for the use of a framework right away, you should NOT start by looking at agentic system at all. Try your best to find the simplest workflow as much as you, otherwise you are putting youself in a postion where you will be forced to trade-off complexity, latency and cost for a better task performance thinking AI agent will do the job but in fact employing a real person may give you near results with less cost, latency and more simplicity. The recent hype of AI Agents tend to focus more on the bright side of having an agentic systems running in your company or business (exactly like what you are doing in this video), but completely ignoring the fact that agents are still in testing stage and the training data in the used model is playing a massive impact on how good or bad your agent is performing. Same agents architecture if used with any of the claudes, GPTs, geminis, crew, deepseek ..etc surely will have a different result/performance than if used with private LLM or VLM or whatever model you are using as a business. So we end up with complex and expensive architecture and not even doing the job right if you are not clear on the foundations of that whole hype. So if you are working in specific field that isn't as common as others, there probably not much data out there enough to train the LLM with enough data to allow your agent to do its job right, hence adopting an agent in your field at the moment will come with a costly price. I see this is compeltley ignored in most of the youtube videos and I guess this means the whole hype should be advocating about building a sophosticated agents/workflows instead of making it like oh here is just few lines of python code and here you go; you got yourself an agent running. An advice from a less experienced person in the AI field, try advocating more towards automation and workflows, you would be surprised how most of the jobs will be done right away .. do you know what is workflow ... it\s simply calling the APIs of the LLMs directly instead of sinking deep in another complex layer of frameworks.
Why not use the agent system itself to optimize agent creation? Start with an agent in charge of creating and overseeing other agents as needed to optimize towards a specific goal.
🖖😸👍
modelise it and make it real
You are advocating this in a very very very WRONG way
Oh really, how so?!
@@WhatsAI
First of all you are advocating for the use of a framework right away, you should NOT start by looking at agentic system at all. Try your best to find the simplest workflow as much as you, otherwise you are putting youself in a postion where you will be forced to trade-off complexity, latency and cost for a better task performance thinking AI agent will do the job but in fact employing a real person may give you near results with less cost, latency and more simplicity.
The recent hype of AI Agents tend to focus more on the bright side of having an agentic systems running in your company or business (exactly like what you are doing in this video), but completely ignoring the fact that agents are still in testing stage and the training data in the used model is playing a massive impact on how good or bad your agent is performing. Same agents architecture if used with any of the claudes, GPTs, geminis, crew, deepseek ..etc surely will have a different result/performance than if used with private LLM or VLM or whatever model you are using as a business. So we end up with complex and expensive architecture and not even doing the job right if you are not clear on the foundations of that whole hype.
So if you are working in specific field that isn't as common as others, there probably not much data out there enough to train the LLM with enough data to allow your agent to do its job right, hence adopting an agent in your field at the moment will come with a costly price. I see this is compeltley ignored in most of the youtube videos and I guess this means the whole hype should be advocating about building a sophosticated agents/workflows instead of making it like oh here is just few lines of python code and here you go; you got yourself an agent running.
An advice from a less experienced person in the AI field, try advocating more towards automation and workflows, you would be surprised how most of the jobs will be done right away .. do you know what is workflow ... it\s simply calling the APIs of the LLMs directly instead of sinking deep in another complex layer of frameworks.