What AI agents are

0:01 [Music] hi I'm Corey St p a cloud advocate here at Microsoft part of our AI Cloud advocacy team and I have the pleasure of delivering to you lesson number 17 of our generative AI for beginners course covering AI agents in this lesson we're going to cover a few things first we're going to answer the question of what exactly are AI agents what are AI agent Frameworks how do they sort of work and what are the difference between them and when should you actually use an AI agent when building generative AI applications the learning goals of this this lesson is going to be explaining how AI agents work understanding the difference between some of the popular Frameworks out there and then how to design properly working with AI agents so what exactly are AI agents this is a very open definition of what AI agents are uh it means a lot to different people when they're sort of building with applications but to

1:08 standardize this or to keep it really General we're going to understand that to for an AI AG to be defined you need to have one a large language model two some sort of state and then lastly tools so with a large language model essentially what I like to say is it's kind of this Choose Your Own Adventure game the large language model is trying to reason or decide on which tools to actually use um when working or interacting with users and then we have this idea of state which is really this the sort of the context of the conversation interaction with users or even other large language model in terms of the past what it's been received and the present of what it's sort of planning to do next then it's also maintaining those particular results uh so that the user is getting the task they want to have completed or that information that they need and then when we talk about tools that could be any sort of external system whether it's a database uh an API uh even another large language model or maybe even just some code whether that's python function things like that that are used being utilized by the large language model or

2:19 being interpreted by responses that the large language model should in the next process execute the function so now that we have a very general definition let's actually look at how different AI agent Frameworks sort of implement this idea of large language model State and tools so in the case of Lang Chan who also has uh available features we have