Name: Workplace AI Agents in Action #2
Uploaded: 2026-10-08T16:46:12.327Z
Duration: 30 min 4 s
Description: Workplace AI Agents in Action #2

Transcript for "Workplace AI Agents in Action #2": My favorite time of day. Hang in for me. And we're live. And we are live. Let's see what we got in here. Forgot we had a fun icebreaker last time. Okay. So as you come in, share where are you from and then what's your favorite way of using AI. If it's something you have not done yet, we'll take that as well, something like a dream new state. Last time I said it's creating slides, I think now it's also creating like, financial models and projections because, it's just... It's crazy how much more detail and, like, the next step I can take things to instead of me having to do it on my own. So that's mine. And I'm from Chicago. For me, what's yours? It's Montreal, but mine is mostly... Give me what did I didn't know about the AI part. Nice. Okay. So asking AI what you didn't know about AI. Give it a minute or two, and then let's get started. Baltimore. Nice. Oh, Cape Cod. Fun. Yes. Design and spell. Gotta be one of my favorites. I don't know if you guys saw the, like, new updates on intelligent UI, but I just love the outputs I get. Like, now it'll, like, create something more designed for me so I can actually digest whatever I come up with. The clothes trial feature, I love that. It it kinda, like, sometimes it morphs my face or something, so it's, like, weird. But, yeah, those are actually really cool. I don't know if I've tried the Asus one. Alright. Let's do it. For me, it has some really exciting stuff to share today. So I think, I wanna jump in so we have some time and you guys have some time for questions. So hello, and welcome to the second session of workplace AI agents in action. So the last one was about... More so focused on how we can improve communications and standardize those. We showed you a page writer agent, content manager, broadcast designer. We had our friend Sarah O'Meara on, and, it was a great discussion. We had some good feedback as well. So we're continuing. And today, we're gonna show you how to create a policy agent. So policy agent is a specialized agent that basically, helps employees get accurate, trustworthy answers about company policies. And we wanted to share this one because many of our customers have started to build this out with Fermin, and it's just a very common need that, we feel like we're in a great place to address for you. So we'll spend a few minutes on why this is important. And then really, I'll, hand, hand the reins over to a friend, Fermi. He'll build this out. He'll configure the agents, connect into a few really cool approved policy sources, and then also show the employee experience. We'll have some time at the end for questions, and, yes, it will be recorded. Alright. So let's, let's jump into this. I'm sure you guys have had this familiar situation. You have employees that might be having questions about the remote work policy, their time off, travel reimbursement, maybe something related to information security. And when they go to go search for it, they might be looking through Lumapps, their Internet, Drive, SharePoint, anywhere else. And all of a sudden, you start to see these very similarly named documents. Some of them might be old. Some of them might be an old PDF. They go to their manager. They have to message someone and wait for an answer. And, of course, that is not great for many different reasons. And immediately, it sounds like a search problem, but it's actually a bigger operational challenge because not only are we losing time, but sometimes those are noncompliant policies. These IT and HR support teams, there's just more tickets created and more of a mess. And when someone's relying on that information, right, it can be a risk. So although traditional enterprise search can kinda find the document, it still leaves the employee to figure out what's going on and actually interpret it. Whereas with generic AI, it might generate something that's very confident and convincing, but it might not be approved in the latest, policies. So that's also dangerous. So probably probably good or probably right is not good enough when it comes to policies. And that's why we, see a policy agent and our customers see policy agent as so important. So what does a policy agent look like and do? It has a defined role, it has specific controlled sets of approved knowledge, and explicit instructions into what answers it should provide. That is what Fermin is gonna show today. Fermin, do you wanna introduce yourself? And then, yeah, let's get it started. Of course. And hello, everyone. Fermin Hernandez. I am one of the solution consultants here at LumApps. And, to Sid's point, one of the many challenges that you have today in any organization is that no matter how many, controls you have, you might have different scenarios. Like, I have all my content policy in a single place. But every time that I look for that information, I'm also getting mixed with a lot of results that are unrelated and whatnot. Another common scenario is that you may have, or you are an organization that is, that has multiple third party applications where your data might live either inside of LumApps or outside of LumApps. So the idea here is how do we make it easier and seamless for the users to find the data no matter where they are? As your intranet start taking space within your organization, usually this is going to be the very first place to go. But that present a couple of things. One, you already have an investment on your organization. You mean that you have a third party application that is outside. Two, you might just have your data already in LumApps, and you want to make sure that you target the content that you already have as soon as your users start typing. Three, it could be a mix of both, and you just want to make sure that the users find information. And, there is something that I always like to say, and that is, user will always be user. Meaning that I don't I don't necessarily need to know as a user where the information leaves. I just want to make sure that when I search for something, I get my results and I'm able to fulfill, the request that I'm doing to my Internet. So that's a little bit of, of the background on on what happens here. And what I'm going to show you now, let me just start sharing my screen, is how do we create a specialized... And please give me thumbs up if you can see my screen super quick. How do we create a specialized agent that is going to rely on, the data they already have. And, again, this might be different places. It could be inside of Flow Maps. It could be outside of Flow Maps. And the idea is let's just make sure that my users can find my information. Right? So So for that purpose, what we have prepared today for you is a typical Moab setup where you can go to your, back end and you may have something similar like a content type, in this case called policy. Just created this a couple of, minutes ago today. And I have a general travel policy and a general expense policy. So this could be extended to any kind of policy you may you may have. You may have multiple articles that depends on different policies. And and we're taking policies as an example. This could be as well your IT data or IT articles that you already have in here. It could be anything that has a specific audience target that you want to make sure the user has access to. So for this case, what we're going to do is we're going to use our two, articles or in this case, the content type for the policies. And we're going to create an agent for that policy. When that happens, the first thing that we need to do is, we're gonna go to our agent studio, and we need to create a knowledge source. In this case, the knowledge source serves as the main reference that our agent is going to use in order to make sure that it reads the right data. So when you click a knowledge source, and I already have it here. I'm just going to delete it to start from zero. So that way, you can see the process. I can click on new. In this specific use case, I have my data in LumApps, and I'm going to call it, like, guitar policies. And because I have a very specific content type that is called policies, then what I'm going to do is I'm going to tell my settings that this source is going to have a very specific entity. I am going to restrict my selection to my Internet side, and I'm going to pick my content type as policies. And that's it. Just by doing that, I am telling, my knowledge source, every time that you're going to be invoked or every time that you're going to be, asked for something, then this is the main specific information or source that you're going to use in order to answer my queries. Once that has been done, I can just click on create. And I have my HR policies with the provider being Lumapps right now. And then this is what I'm going to use to create my agent. Now for my agent, what I'm going to do is I'm going to go to the agent section. I'm going to create a new agent, when I call it an Ask AI assistant. And what I'm going to do is I'm going to call it, HR policies. I'm going to set my audience to everybody. And then I think that one of the big differentiators that makes a good agent be the standout out of a regular agent is the specific instructions that you give this agent. So it's pretty much saying, okay, this is exactly what you want to do or what I want you to do. This is how I want you to behave. And this is how, we need to make sure that your tone of voice is regulated and whatnot. You can have... And for the sake of time, I already have a set of instructions created. But as you can see here, in the system prompt, you have a couple of things, as an example. And one of those is you need to describe what your agent is going to be doing, how do you want the agent to answer those questions, what do you want that you do not want the agent to do in order to deliver those answers. So for this scenario, as you can see, what I'm telling here is, Sid, anything there? Sorry. I was just gonna say, so these are the individual agent instructions. You can also sell, you can also create global, set global instructions. So if you have brand guidelines or certain voice you need, that can also be respected at kind of the broader parent level and at the child level at these agents, you can have specific instructions. And that's correct. And that's a great, intervention there. The... Once we're doing right now because we're creating agents, we're giving the agent instructions. You may have or your IT team may have already a global policy agent on how you want the AI to behave. And then you can set it up right here in the settings, which I'll show you in a minute. But going back to the specific agent, what I'm telling my agent is, okay. This is who you are, number one. This is what your role is, and then this is what you're going to use as a reference. Right? So I am telling the agent who you are, what is your role, who are you, and how you're going to behave here. Then I am giving the agent the responsibilities. And here, for example, I'm putting, like, you're gonna help employee to find the HR policy content. You're going to provide, clear plain language summaries, the the details of the policies once you find it. And if needed be, you can ask follow-up questions. Then I've been giving a little bit more of crafting on how I want my agent to behave. Remember that, AI and depending on your organization, you can give very specific and detailed information on how you want that content to flow back to the users. In this specific scenario, what I'm telling my agent is if there are multiple items listed, then present them in bullet points so that way they're clear. And also, if there is too much information, break them down in different paragraphs. So in that way, the information is presented in a very clean format to the agent. Last and not least, what I'm giving my agencies... Okay. These are a couple of examples of questions that the user may create. So that way, you can use it as a reference. But also, you will be able to use them, as a start guide or get a little more context of what kind of information you might be asked by users. So once my prompt is is ready, the next thing that I'm going to do is add my source. So remember that was the first thing that I added in my knowledge source. And here, the... There is going to be a list of all the sources that are... That I created or that I have available. So I'm going to use the HR policies. And then once I have my source available, the next thing is going to be the specialization. And the specialization is really, when these agents should be treated. Like, what I'm going to use this agent for. And, you can, describe pretty much what you want to do. For example, here you see, like, all these questions about payrolls, benefits, vacation policies. I am going to say, you can use this agent when you ask about, policy rules, requirements, approvals, and employee benefits. You name it. Last but not least, we have our user experience. So let's say, the user experience becomes, some sort of suggested way and, questions that you can start. So if you already have, the type of questions that the users ask the most and you want to give them, like, a little style, a style ahead, you can add them here as a starter. So for example, we said travel policy. You can add it here. And you can add up to five. And what else that's going to do is that it's going to show up on the agents. So once the user opens the agent, it becomes like a guided questions that the agent can just go right away there. With all this, my agent is ready. I can do two things now. I can just save as a draft, and then all my information is going to be saved or I can publish it. Publish it go directly live in the platform and then you will be able, to get any specific information about the content and and start playing with your agent as you see fit. So we're going to publish our agent. And you can see our agent published right here. And that's it. Like, our agent is being published. And now, we want to make sure that we get the type of information or we can get to test it out. Ideally, when you deploy your agents, you might want to restrict your agent to a different audience, which is a test group audience just as a best practice. Right now I did it as an all so we can test it out. But if you have a a test group that you want a couple of people to go ahead and test it, you can define it here. And then, that means that your agent will be on the limited to, that specific, group of people there. Sid, before I test it, anything that you wanna add here? Yeah. This is great. As you can see, right, it doesn't require, like, a rockets, like a degree in rocket science to really be able to create these. Right? And that's part of the charm. Right? It's supposed to be easier for others to do, and then you have still central global guardrails defined by IT or whoever else needs to. So that's part of, what's great about this. And, you know, we have resources like for me, and we're continuing to templatize some of this So you don't have to be as responsible for the instructions and system prompts. So that will continue to improve. The only other thing I'll say is, yeah, let's, I don't wanna bury the lead for me. And I'm sure a lot of people are excited to see what you're gonna show in a second. Yeah. Of course. So my agent is published. I can go back to, my page. And then what I want to ask is, like, okay. Let's see, for example, what do I want to get on my agent? Then I'm going to start with that question, that I know that I know that exist on my on my specific information or my specific content. And I want to start asking, what is my, business expense policy as an example. And what the agent is going to do is, like, you see how right away is triggering my, HR policy, especially the information and making sure that I can get my information right away. So if you remember, and I'm just going to duplicate this tab, here. If you remember, that is the name that I added to, my agent when I named it. It's right here, the HR policies. And given the instructions and the specifics that I gave it right now, you can see that it's fetching the information that I have. And I can continue to ask and build on top of my agents asking more additional questions if I needed to be. The good thing about this is, it's going to give me not only the answer that I'm looking for, but also the, source of the page itself. So if I click it, I'm going to be direct to exactly where the page is. So you see that this page has a lot more information, but it's... Was very specific to the allowable, allowable, that all of these are suspense. And this is the information that I am getting here and not the rest of the information that exists in the page. So it's not going to extract anything or everything. It's going to stack only, the information, that I need from my page itself. So that is one example. Let's say for another example, and this one might be, might take a little time, which we don't have. But I just wanted to show you for the for the sake of, of knowledge. Let's say that your policy is not available in LumApps and that your policy is, available elsewhere. So if your policy is available elsewhere, then what you can do is you can connect, to that third party application. And we have multiple ways on how you can connect to it. You have the MCP, server. You have a two a. You have an ASCII micro app. And the idea is that here, you should be able to connect any third party application that you wish, and then you should be able to tell the agent. Okay? If I create an agent, then I go fetch the information, would also give me the information. So in this case, I'm gonna ask, in terms of questions, something different from HR to IT. And I'm going to ask what is my MFA policy? And let me see what is going to trigger the start of the agents. So, is... I can tell that there is a Confluence IT, agent connected here, and it's going to try to fetch that information. And right now, that information that is fetching the agent is not available in LumApps, but rather is available in Confluence, so an external source. You can see here that it's giving me the information. I can ask a question. And if I ask here, for example, okay, this is my source. If I click it, you see that it's going to take me right away to Confluence and then to the agent, or not the agent, the article that specifically, talks about my MFA policy. So as you can tell, agents are not linked only to internal but also to external informations. This specific configuration that I'm showing here is connected through, an MCP server, meaning that Lumen apps is using the server that is on the confluence side, is fetching that information, and then bring it back here. So that's another way in how we can do this. Another third way just to close it out, so we have, time for questions is what if I have the information living in multiple places? Well, when you create your agent, the sources are not limited only to one source or the knowledge is not limited to one source. You can connect multiple sources at once and then just define your agent, and the agent is going to have a mix of both of the informations that you have available. So for example, if I go to my, IT confluence, you'll see that here I'm using a tool as an MCP, but also I can use on top of it the HR policies as a source. And it's going to be able to read both of the information that exists whether it's internal and external and will behave, in the same way. That's all on my side. Sid, do we want to go ahead and and go to q and a's? Yeah. Let's do it. And, I know we had... We moved some... Through some of this fast, but, yeah, I mean, this is available. Right? And, you know, for me, and, of course, you you... You'll get help for some of this. So for me, it is one of the solution consultants. There are others. So you're not building all of this alone, and it's meant to be pretty plug and play at this point. So really exciting stuff being able to connect external sources. As we know, we're already seeing really good search results pop up. Right? For me, is that fair to say? Correct. Yeah. Yeah. The questions I've got brought up are very fair. So one of the questions is how is this different from Ask AI connected to multiple sources? So this specific example is if you are creating a box in most agents, right, for policies, you create one for IT support, others other, kinda use cases like that. And these are very specific prompt instructions. Right? So if you want a IT support agent that only references a specific service knowledge base and Jira knowledge base and provides an answer with troubleshooting tips and other other kind of instructions it follows, that That is an example of the specialized domain specific agents, whereas Ask AI is more of a general, general search option. Right? You can still create guidelines at the global level, but now you're making these separate domain specific agents that, respect different kind of sources and and handle queries and answers in a specific way. I don't know if there's something you wanna add to that for me. No. No. Very fair. Okay. What else do you guys have? You can use q and a or the chat. Oh, there are there are some q and a. Kevin asked, hi, Fermaine. Can you show the setup for the Confluence connection and how it connects to agent hub? Yes. If I can go back to my screen again. So for Confluence, what we are doing is connecting to an MCP server. Here... Right here, we have one of the MCPs and we have MCPs, for those who don't know, is a model context protocol. And that allows, your third party application to expose different parts of your third party application to external sources. So in this case, for example, in Confluence, what I'm doing here is I have a couple of tools that are activated already. And out of these tools, what I'm telling... Or when I connected to this tool, what I was trying to do is... Out of everything that you have as a Confluence, just give me a specific part of information that my users are going to use. But this is going to allow me to execute search in Confluence content and obtain that information back and forth. So when I have the information created as an MCP and I create my agent, the agent that I have right here, well, I'm telling the agent this is how you're supposed to behave, and then I'm adding the tools. So you saw that back then, I had eight tools that I... That were activated. I'm only using four of them. So I want to get the Confluence content. I want to get, the search content, the... Any resources and all the information that I would need in order to retrieve the information that exists in Confluence and back to my platform. So even though MCP has all the capabilities, there are only a couple of those that I'm going to be able to use, directly from, Confluence in this case. Send back to you. Was there question left? Okay. Yep. Yeah. Yeah. There's another one from Samantha. Do users only trigger the agent in the chat experience? Does the agent work if someone is searching in the global search bar at the top of the page? Very good question. Today... And by the way, our agents aren't in evolving. Like, we evolve our agents every quarter. There are two type of, search agents that you can use. You can use the Ask AI agents, which are the ones that you can search only for information, and there are also the, agents itself who are the ones that allow you to connect elsewhere but also execute actions. For example, if I want to create a ticket, that cannot be done through search, which is the Ask AI summarization of search that ask this to be, done through an agent. So if, for example, if I look for an information here, you're going to see that the amount of agents that I have available, done below here, is going to be only very minimum, only one. Whereas on my agents, I have a plethora of agents that I can use. So one is for search specific only. The agents themselves. are going to help you out with two things, search and taking actions. So, like, that's the difference in both of them. And the agents, if they are connected to external sources, they will not show up on your search, platform right here at the search now. level. Yeah. Yeah. Correct. So this is something we're trying to synchronize. Right? The search and chat experience at the end of the day should be kind of, work work in tandem. Probably shouldn't be a completely different answer or or a way of working. So that's something that's, already on the road map, something we're we're working to design out. But it's a very fair question. So definitely on our radar. I think those are some of some of the main questions that I saw. I do wanna... There is there something is wanna ask for. me? Yeah, no. There's one for Heather. How's this different from having us get connected to multiple sources? You can use both, Heather. The difference here is that agents are going to help you out with more direct actions. Right? So right now, you can connect multiple sources to an agent that is going to be able to trigger and and they just give you the information back. On the other side for, the agents themselves, they're going to extend the capability that you have, and it's going to limit it. It's not going to limit you only to search for information, but also to take action on those items that you have on the other side. Yep. That's perfect. So we are gonna... We are... We're gonna get cut off in a second, but hope you guys have great great engagement. I hope you guys enjoyed the session. As always, reach out to your account manager and see, customer success team to help out, and you guys can get access to some of this stuff. We are running POCs. And then, join in for our session three where we'll be having that. It's gonna be more focused on IT use cases. We'll have our head of AI there. So definitely looking forward to having you all there, and have a great day. Thank you all.