What is agentic AI in Glean AI? - Glean Tutorial

- [Instructor] I'm going to teach you the exciting concept of agent AI and I'm going to walk you through demos to show you how to build agents in Glean AI. So we are going to go to Agents here and we are going to create some agents in the next few lessons. First, I want you to understand: What are agents? So you see this link says How agents work? That takes you to full documentation and lessons on how agents actually work. I'm going to give you a short summary. Agent AI is basically automating AI using the reasoning power of large language models and allowing the agent to perform complex tasks by breaking it down into simple tasks. So the starting point of the automation is a trigger. A trigger is an event that initiates the execution of an agent within the system. It can be a condition in which the agent will start or it could be an event. The second thing is a set of steps. This is what we will build as we build agents. The steps are made into actions or flows. Action could be something as simple as getting information from a document, or it could be more complex about taking action, creating content, example, responding to a customer service request. Let's take a step back and understand what makes up an agent AI ecosystem. So we talked about trigger. That's what starts the automation. But what is an agent AI? So an agent AI is about automating using a large language model. That means there is at least one LLM. In fact, in Glean, you can choose what LLM the agent is going to work with. It could be more than one LLM. The second thing is it is going to have access to some documentation and it is going to take some action. So, we will define those actions, and there is something called chain of thought reasoning. And so the action doesn't have to be a simple action, could be a flow or a branch based on some condition, and an agent interacting with other multiple agents called multi-agent or subagent system. So we are going to define actions and we are going to define flow, and you'll learn all this as we go through our demo. So there's trigger to start the agent, there's the actions and the flow that we will define based on what we want the automation to be about. These are the steps we will go through. One important thing in agents is that it is using the reasoning power of the large language models. So, what is reasoning? Reasoning is the ability of the large language model to make the agent self-reflect. So if the chat bot built using Glean agent is going to give an answer and the answer is incorrect, then typically in the chat bot, it'll forget and continue to keep giving the same answer. In this case, because the agent is expected to automate towards some goals, it'll use its own reasoning power, look back, and say, "Did I give the right answer? Was the customer happy? Or how do I improve my performance in the future?" So, to make that happen, as the agent executes every step, it actually stores it in the memory. And so that is called memory, and we will also learn about memory. If you want to learn more, you can click, and again, Glean has given beautiful documentation for you to understand all the concepts. So coming back, we are going to be creating agents, we are going to set triggers, steps, flows, and create some automation. And I will show you some beautiful demos and you'll be able to follow along and try them on your own.