Create an agent using templates - Glean Tutorial

- [Instructor] Now, we are going to learn how to create an agent using readily-provided templates. So I went to app.glean.com, I clicked on Agents, and next I'm going to click on Create agent. And we've already learned how to use a prompt to tell AI to generate an agent. So now, we are going to use the many templates provided. And as you can see, I clicked on engineering, and it is showing me many different templates. You can read through them and you can try any of them. There's HR templates, there's IT template. I want you to think about: What is agent AI? What is it made up of? What does it do for you? Every time you create an agent, even if it is a simple click or a simple prompt, I want you to be aware of your responsibility of what you're automating, what documents is it accessing, and how is it doing this, and how is it serving your customers? So, look at this. This is a template for IT. This is not the template we are going to use to create the agent, but I want you to read this. It says create a knowledge base article from a template. So this is to check if a ticket can be answered by an existing knowledge base article. So what is agent AI? Agent AI is automating AI using the reasoning power of large language models to break a complex task into multiple simple tasks to run it, to execute it. That is the power of an agent. So here, this agent is going to check a given ticket number exists in a knowledge base, which means it is going to search company articles to find out whether the ticket exists. In this case, it is going to look at a support database to see whether the information exists. So, I'm going to look at a support template because I want to go back to the same example of what we did last time. We created a simple ticket timeline agent. And so I want to do the same thing using this template. It says create a detailed timeline of events for a support ticket. I click, and the agent is ready. So anything that is marked in a blue outline has to be customized for this specific use case. So the first step, do you remember what invokes an agent is a trigger. It can run manually or it can run on a schedule. So if you say Allow the agent to run on a schedule, then it'll run on a schedule. Otherwise, it is going to run manually. And in this case, because we used a template, it has already pre-populated this field to say it is a text file and it is going to be provided with the ticket URL. And it is given the description as the URL of the customer support ticket to summarize. It's going to generate a timeline of the support. Again, all decisions have been made for us. It's almost like the AI use case, but this is ready-made templates, and it is going to say, "Okay, this is how it is going to provide the response. In what format is it going to provide the response?" And it is reading this document. That's how it's going to provide it. And if you simply press Publish, the agent is published. If you press Review, it'll show you what it'll look like to run this agent. So the ticket number has to be provided, and that is going to be a URL of the support ticket because it's not just a ticket number. It is going to take the ticket number and it is going to look at the entire conversation of the support ticket because it's going to summarize all the conversations we've had with the customer, all the different interaction in the support call, and what is in the log to give us a timeline, which is a chronological summary. Now, if you did that and you run, you can get the agent to work. So now, we have created another agent with a template. I want to show you one more. So we looked at examples from engineering and HR. We actually created an agent from support. I want to go to Sales because I want to show you a working agent and end-to-end run it for you to see what the experience is like. So, in this case, think of a sales situation. It is actually saying, you know, "What are the prospect's outreach email? Can we create a personalized outbound message? Can we look at a list of all open action items from yesterday so we can plan a schedule for today?" Or you just look at a particular customer account and look at an existing customer database to understand a snapshot of that customer account. A lot of different things are possible for the agent to summarize for us. I'm going to choose this Draft a competitive brief. So, the agent has to go look at internal documents where you have information about competition. It'll also augment this from the public web. So I clicked that and it created the agent in draft for me. It says draft because now I get to change it and customize it if I want it. So the first step is the trigger, and I'm going to let it run manually. It's created the search field, and it says it's going to search for all internal mention of the competitor. And then, there is another action where it says, "Formulate web search queries to further research the same competitor," and then it drafts a one-page response. And in this response, you can actually see whether it is factual. Again, it has chosen it to be factual. It can be factual or it can be creative. We learned about this. Creative just means that it's a expressive, diverse response and you can see that it is using the large language model GPT-4.1 as default. And we are ready. So let me go and say Publish. And the agent is published. If I do Preview, it'll actually give me an option to run this agent. So I'm going to test this out with a company name. Since I'm working here on LinkedIn Learning course, I'm going to safely say, "How about if we find LinkedIn as the company to analyze?" And I don't have access to any internal documents from LinkedIn, so this company's search of internal mentions is not going to give me any results. I've said, "Go find web search queries," so that's what it's going to do. Editors, you can speed this and just show that it is thinking, moving fast, and then it's getting to the results. It is actually showing insights from web research and has a lot of information on the product offerings and target audience. And it's doing a decent competitive analysis. But this is all from public documents it has found from the internet. And it even analyzes and summarizes that on how much is the revenue and how many users, reference how many users and everything else. And it says limitations is no internal data or proprietary insight was available, because in this case, imagine this is a sales scenario of analyzing a competitor. I just chose LinkedIn as the company. And so if you have internal document of your sales database where you interacted with the client and where somebody else got a business, all of that will be inside your internal documents. And this can do that for you for this... Rephrase, for the sake of this demo, I just wanted to run a competitive analysis from public data set and just show you, and it was intelligent enough to say, "Hey, I did not get any internal data or proprietary insights. All insights are from the public web source." So you get the idea of what is possible when you create an agent using a template. So now, try this for your own use cases, for your own department, and make sure it has access to all the documents you need so that it can give you a really functional agent that you can use.