Agents, Agents and more Agents 🤖🤖🤖

We will have a quick look at the new Agents framework released by Mistral AI. Agents are everywhere!

LINKS:

💻 RAG Beyond Basics Course:

Let's Connect:

Signup for Newsletter, localgpt:

00:00 Introduction: The Buzz Around Sam Altman's Strawberry Picture
00:09 Emerging Trends in Agentic Frameworks
01:08 OpenAI's Structured Outputs: A Game Changer
02:54 Mistral's New Agentic Workflows
03:27 Building Custom Agents with Mistral
07:03 Deploying and Testing Your Agent
10:54 Advanced Multi-Agent Workflows
12:14 The Future of Agentic Systems

All Interesting Videos:

[jameehuang]

The AI templates help maintain consistency in tone and style across my social channels. It’s made brand building so much easier.

[MrAhsan]

Please a demo project where multiple agents have some roles assigned to them. For example, You can mimick how financial analysts from wall street make analysis about stocks or stuff like that or Different agents working together in a News Channel setting talking about situation in different countries and commenting on them.

May be these are some dumb suggestions. If you have something better in mind, please do that.

One more thing, I am learning Langgraph so It'd be very helpful if you use that in your upcoming video(s).

[joydeepbhattacharjee]

Soon we will have backend frameworks that can serve very personalized apps that are generated on the fly for individual users.

[vitalis]

What’s the different between using agents and have the LLM create various roles, especially for summary? Is it mainly because of the context window tokens?

[stephquiz]

Have you done something where teh agent can use my api keys say from aws or azure and do tasks for me? actual tasks where i just chat what i need like what are memebrs of this group and then it generates me the list without giving me code ot run on my own?

[wryltxw]

do you think it is better to use langgraph or use mistral agentic framework directly? what could be the tradeoffs.

[ShyrikRyss]

This is still not an AI agent, but a set of chats that are simply connected to each other (the output of one is passed to the input of another). And such a method is very susceptible to any errors (an error in the response of one of the chats will lead to incorrect outputs of subsequent chats when generating the final response), despite the precisely specified formulation of tasks and input / output.

The same implementation can easily be done for free in the ComfyUI workflow locally on your machine by connecting several LLM generators (connecting their outputs to the input of the next generator) and assigning them tasks.

Read what an AI Agent should actually be able to do. In short, an AI agent should understand any of your requests and qualitatively and continuously search for answers to them and perform calculations, and not act according to a given pattern, as in your example (sequential operation of chats).

This claim is not to you, the author of the video, but to those who create these "agent" systems and call them agents. This is the LLM chat system, but not the agents in AI terminology.