Introduction to AI Agents

Hello and welcome to this session on Introduction to Agents. My name is Janette Roush, and I'm the Chief AI Officer for Brand USA. Today, we're going to be talking about AI agents. And AI agents are one of the hottest topics right now when it comes to how AI is evolving as a technology and what it means for work specifically.

One of the phrases that you're going to hear me say a lot during this presentation is agentic AI or agentic tools or agentic systems. Agentic is a made-up word. It sounds fancier than just agent, but it means the same exact thing. Somebody was trying to make a regular word sound fancier and so they made it into agentic, and so if you hear me say the word agentic throughout this presentation, I'm just talking about agents.

So I want to start with a definition. What is an AI agent? An AI agent is goal-oriented software that's able to plan and execute multi-step tasks in order to achieve a specific objective. So the way that I like to think about this is a chatbot that you just ask questions and it responds, that is not an agent. But the minute that that software is able to take actions, is able to make decisions without you, and is able to work through complex workflows—even if you're not there, even if you're not prompting it every single time—that is when it becomes an agent.

And so the simplest way to think about agents is that they can plan and take action without constant human input, and that is the main distinction between just a regular chatbot and an AI agent.

So what I've done is I have tried to group or categorize agents that are particularly relevant for DMOs into four main buckets. I call them the operator agents, researcher agents, builder agents, and automator agents. And as I go through each of these four different types, what you're going to see is there's going to be actual tools that you can use that fall into each of these different categories.

And so the first category is operator agents. Operator agents are agents that are essentially using the mouse for you. They're doing browser tasks. They're automating things that you would do yourself using your mouse on your computer. And so for DMOs, this might be automated lead generation, scraping a particular website to pull down information, being able to extract data from a website in order to reformat it or put it somewhere else.

And then the actual tools that are examples of operators, there's one called Browse.ai, there's another one called Manus.im. Both of these are tools that essentially are using the internet as if they were a human being, clicking through things and performing tasks for you.

And so what I want to do real quick is I'm going to show you just a demo using a different tool, which is ChatGPT, and I'm going to use it to demonstrate what an operator agent might do for a DMO.

So let me jump over to ChatGPT. This is using the 01 model, which is their reasoning model. And so I'm going to give it a really specific prompt, a really specific task, and I want to see if it can actually complete this task for me using operator agent functionality.

And so I'm saying, please go to usa.gov, which is the official website of the US government. It has a listing of all federal holidays. I want you to use your web browsing capabilities. I want you to go to that site and compile a comprehensive list of all federal holidays with detailed information, and then once you've compiled that information, output it in a CSV format so that I can download that data and then I can either import it into Excel or I can import it into a CRM or whatever tool I'm using.

And so as I hit enter on this, what ChatGPT is going to do is it's thinking first, it's reasoning through what steps it needs to take in order to answer this particular task, this particular prompt. And then once it's finished reasoning, it's going to start taking actions. And the actions that it's taking are all browser-based, meaning it's going to go to the website, it's going to read through the website, it's going to find the information, and then it's going to compile it and respond back.

And so this is what it means to be an operator agent. And while it's doing that in the background, what I'm going to do is I'm going to show you just one other example of what an operator agent could be used for. So on this particular website, I'm on Manus.im, and this is a tool that has a number of different what they call AI workers that they've already pre-trained for specific tasks.

And so you can have an AI worker who's going to generate creative ads. There is one that's going to do competitive price tracking. And so this is a really interesting one for DMOs because what it can do is you can tell it, "Here are three different competitors in my competitive set. I want you to monitor their websites and I want you to track any changes in the pricing for hotel rooms." And it will go to those websites, it will look at the pricing for different hotel rooms or different products that are being offered by those competitors, and then it will compile them in a list and will send you regular updates every time that the pricing changes.

And so for doing competitive analysis, this is a really, really powerful use case. It's being able to essentially automate the collection of data from competitive sets in order to do regular analysis and make sure that you are staying competitive with everybody else who's in your particular market or your competitive set.

And so that's the first example. That's what we mean by an operator agent.

But then if I jump back over to ChatGPT, you can see that ChatGPT has finished this task. It has gone to usa.gov, it has compiled the list of all federal holidays for 2025, and it has provided them right here in this list. And then it has also converted that into a CSV format, which is a file format that's able to be imported into other tools. And so I can download this, I can import it into Excel, I can import it into a Google Sheet, or I can import it into whatever tools I'm using in order to make sure that our marketing calendar is matching up against federal holidays, that we're not scheduling events on federal holidays, or that we are scheduling particular campaigns to launch around specific holidays.

And so that is what it means to be an operator agent. And we're going to come back to this later, but there is one really important caveat with operator agents, which is they still hallucinate.

And so I want to show you one other example. Let me just pull this up real quick.

So I asked ChatGPT, I gave it this prompt. I said, "Imagine you are a lead generation specialist for a destination marketing organization. Please use your web browsing capabilities to identify 20 corporate meeting planners who work at Fortune 500 companies in the pharmaceutical industry. For each planner, provide the name of their company, their LinkedIn URL, and a brief explanation of why their company might be interested in hosting a corporate event in a U.S. destination."

So this is essentially a task that a salesperson at a DMO would be doing. They would be researching and trying to find leads, trying to find new companies that they could be reaching out to in order to book their corporate events or their association conferences at their destination.

And so ChatGPT ran through this list, and it came back with 20 different leads. And it gave me the person's name, the company name, and their LinkedIn URL. And then it gave me a brief explanation of why their company might be interested in hosting a corporate event.

So that seems great. It seems like it accomplished the task. I could hand this to my sales team and they could start reaching out to these people. But what I did is I went through each of these, and I clicked on the LinkedIn URLs to see, does this person actually exist? And I found multiple examples where the person did not exist. There was a LinkedIn profile that ChatGPT created, but it did not match any actual LinkedIn profile.

And so this one right here, Alicia Grant, who supposedly works at Eli Lilly, when I clicked on this LinkedIn profile, it took me to a profile for somebody completely different. It was not Alicia Grant, and it was not somebody who worked at Eli Lilly. And so what ChatGPT did is it hallucinated this person. It made them up. It created a plausible-sounding name, a plausible-sounding company, and a plausible-sounding URL, and then it responded as if that person was real, even though they're not.

And so this is one of the most important things to understand, is that even though agents can do incredible, powerful, impressive things, they still have this fundamental issue of hallucination where they will make things up, and those made-up things will sound really convincing and they will look really real, and so you've got to be on guard when you're using these tools to make sure that you're not trusting them blindly, that you are actually verifying the information that they give you.

Now, the solution to this—and we're going to come back to this later, but the solution to this is to connect the agents to a source of truth, to a database that has verified, correct information. And the way that you do that is through a technology called the Model Context Protocol, or MCP, which is something that we're going to cover in another session.

But just to give you a brief overview, MCP is a way for AI to connect to authoritative databases and pull information that is verified and correct in real time. And so instead of the AI just making things up, the AI is able to query a database and get back information that is accurate. And so for things like accessible travel, where you need to make sure that the ramp that you're recommending actually exists, you can use MCP to connect to an authoritative database that has verified information about accessible travel routes. Or if you're trying to find the capacity of a venue, instead of the AI just guessing at what the capacity might be, you can use MCP to query a database that has the actual capacity listed for that particular venue.

And so MCP is the solution to this hallucination problem, and it's going to be critical as we move forward with using these agents in high-stakes scenarios where we need to make sure that the information that the agents are providing is accurate and correct.

All right, so that is operator agents. That is the first category.

The second category is researcher agents. And so researcher agents are agents that are able to perform really deep analysis and synthesis of information. And so unlike a simple search tool where you just type in a keyword and it brings back a list of links, a researcher agent is able to go through dozens or even hundreds of different sources, analyze them, synthesize the information, and then compile a report that gives you strategic insights.

And so for DMOs, this means you could have a researcher agent that analyzes all of your competitors at once, that looks at what they're doing, looks at their marketing strategies, looks at their pricing, looks at their campaigns, and then synthesizes that information into a single report that you can use for your strategic planning. Or you could use a researcher agent to do market research, to look at trends in the travel and tourism industry, to look at what travelers are saying on social media, and then compile that information into insights that you can use for your marketing.

And so researcher agents are really powerful for strategic intelligence, for understanding the market, for understanding your competition, and for making data-driven decisions.

All right, the third category is builder agents. And so builder agents are agents that can create digital products from just natural language prompts. And so instead of having to hire a developer or having to learn how to code yourself, you can just describe what you want in plain English, and the builder agent will create it for you.

And so tools like Lovable.ai, for example, are able to generate fully functional websites from just a text prompt. You can say, "I want a website that has a homepage, an about page, a contact form," and then Lovable will generate that entire website for you, with code, fully functional, ready to deploy.

Another example is Claude Artifacts, which is a feature inside of Claude, and it's able to create interactive applications, data visualizations, games, tools—all sorts of different digital products—from just a conversational prompt.

And so for tourism marketers, this means you can do rapid prototyping. You can test out ideas really quickly without needing to have a whole engineering team behind you. You can build interactive tools for your stakeholders, you can build simple applications that help with internal workflows, and you can do all of that without needing to know how to code.

Let me actually show you an example of this. So I'm going to jump over to Lovable.ai, and I'm just going to show you really quickly how this works.

So Lovable is a tool where you just describe what you want, and it builds a website for you. And so I'm going to give it a prompt. I'm going to say, "Create a simple registration website for a tourism conference. It should have a hero section with an image, an about section, a registration form that collects name, email, and company, and a footer with contact information."

And so I'm going to hit enter on this, and Lovable is going to start building this website for me. And as it's building it, what you're going to see is it's writing the code in real time, it's generating the design, and then it's going to give me a live preview of what this website looks like.

And so this is happening in real time. This is not a template. This is not something that was pre-built. This is being generated from scratch based on the prompt that I just gave it. And so within just a few seconds, I now have a fully functional website that I can deploy, that I can share with people, that I can use for an actual event.

And so that is the power of builder agents. You can create digital products incredibly quickly without needing to have technical skills.

All right, and then the fourth and final category is automator agents. And so automator agents are agents that orchestrate workflows across multiple different platforms. And so these are the agents that are connecting your CRM to your email marketing tool to your calendar to your website, and they're automating multi-step processes that would normally require a lot of manual work.

And so tools like N8N, Agent.ai, Google Gemini Gems—these are all examples of automator agents. They're able to take a workflow that you describe, and then they set up all of the connections and all of the automation so that it runs without you having to do anything.

And so for DMOs, this might mean automatically updating your CRM whenever somebody fills out a form on your website. Or it might mean automatically sending follow-up emails after a sales call. Or it might mean automatically generating reports and sending them to your stakeholders on a regular schedule.

And so automator agents are really the infrastructure layer for AI-powered operations. They're what allow you to connect all of these different tools and systems together and have them work seamlessly without you having to manually do all of that work.

Let me show you an example of Agent.ai, which is one of these automator agents.

So Agent.ai is a platform where you can build agents that perform specific tasks. And one of the examples that they have is a sales intelligence agent. And so what this agent does is you give it a LinkedIn profile URL, and then it goes out, it analyzes that LinkedIn profile, it looks at the person's work history, their connections, their posts, and then it compiles a report that gives you insights about that person that you can use for sales outreach.

And so let me just show you how this works. I'm going to give it a LinkedIn profile. I'm going to give it my own LinkedIn profile just as an example. And so I'm going to paste that in, and then I'm going to hit "run agent," and Agent.ai is going to start doing its work.

And so what it's doing is it's going to that LinkedIn profile, it's analyzing all of the information, and then it's going to compile a report for me. And this report is going to include things like what are the person's key skills, what are they interested in, what kind of content do they post about, what are some potential conversation starters that you could use if you were going to reach out to them.

And so this is incredibly useful for sales teams because instead of having to manually research every single lead, you can just give the agent a list of LinkedIn profiles, and it will do all of that research for you and give you a report that you can use to personalize your outreach.

And so that is automator agents. They're orchestrating all of these different workflows, they're connecting to different platforms, and they're doing the work for you so that you can focus on the strategic, high-value tasks.

All right, so those are the four categories: operator agents, researcher agents, builder agents, and automator agents. And understanding these four categories is really helpful because it allows you to think about, "Okay, what are the different use cases for AI agents in my organization? What are the types of tasks that I could be automating or augmenting with these different types of agents?"

Now, I want to talk a little bit about what the future looks like. And so one of the things that's happening right now is we're moving from a world where you have individual agents that do specific tasks to a world where you have swarms of agents that work together.

And so imagine a scenario where you have a DMO manager who is directing a whole team of AI agents. One agent is tracking sentiment on social media. Another agent is drafting emails to stakeholders. Another agent is generating custom economic impact reports for different audiences. And all of these agents are working together, they're coordinating, and they're accomplishing complex tasks without the human having to do all of that manual work.

And so the role of the human in this future is not to do the tasks themselves, but to be the strategic director of these agents. You're the one who's setting the goals, you're the one who's making the decisions, you're the one who's prioritizing what needs to get done, and then the agents are the ones who are doing the actual execution.

And so this is a really exciting future because it means that humans can focus on the things that humans are best at—strategy, relationships, creativity, critical thinking—and the agents can handle all of the repetitive, time-consuming tasks that take up so much of our time right now.

Now, coming back to this idea of the source of truth and MCP, this is going to be absolutely critical as we move into this future of agent swarms. Because if you have a whole swarm of agents working together and they're all hallucinating, they're all making things up, then the output is going to be completely unreliable. And so we need to make sure that these agents are connected to authoritative, verified sources of truth so that the information they're providing is accurate.

And so the Model Context Protocol is this technology that allows AI to connect to databases and query them in real time. And so for tourism, this could mean things like verifying accessible travel routes, checking venue capacities in real time, making sure that attraction operating hours are accurate—all of these things that are really critical for travel planning and where hallucinations could cause real problems for real people.

And so MCP is the solution, and it's going to be essential as we move forward with using AI agents in high-stakes scenarios.

All right, so just to wrap up, the four types of agents: operators, researchers, builders, and automators. Operators are using the mouse for you, researchers are doing deep analysis, builders are creating digital products, and automators are connecting different systems together.

All of these have really powerful use cases for DMOs, and the future is moving toward a world where we have swarms of these agents working together, coordinated by humans who are acting as strategic directors.

And then the final piece is making sure that these agents are connected to sources of truth through technologies like MCP so that the information they provide is accurate and reliable.

And so my encouragement for all of you is to start experimenting with these tools. Start small, think big. Try out one agent in one area of your organization. See how it works. Learn from it. Iterate. And then start to expand and think about how you can use agents across your entire organization to transform how you work.

At Brand USA, our mantras when it comes to AI are: start small but think big, embrace both top-down strategy and bottom-up experimentation, and remember—using AI at work is not cheating, it's evolving how we work.

So thank you so much for joining me for this session on Introduction to Agents. I'm really excited to see how all of you are going to be using agents at your destinations, and I can't wait to hear about the successes that you have. Thank you.