View Recording: Building Agents with Work IQ

Joe Steiner 0:06 All right. Well, hello, everyone. Happy Wednesday. I am Joe Steiner, and I today am going to be presenting with Corey Milliman. Our topic for today is building agents with work IQ. Microsoft has come out with the next layer of Intelligence on top of the graphs that exist already between Work IQ, Foundry IQ, and Agent IQ. Today, we thought we’d focus in on Work IQ, which is what most of Copilot Agents is working on top of, and the, you know, semantic layer that resides underneath. So I thought we’d talk a little bit about what Work IQ is, and then we’re going to be showing you some things about how do you develop using that and how can you leverage that to build agents from there. So thank you for joining us, and we’ll get started. So, Work IQ is that layer of intelligence that allows you to operate within Copilot and then leverage all the other data. skills, context, and memories about you that the system has and brings that all together so that it makes it easier for co-pilot to operate, but also as you’re developing agents for those to operate and makes development easier as well as provides for much more powerful capabilities within there. So some of the key layers of this are, again, the skills and tools. We’ll talk about what those mean, or, you know, kind of those special instructions and the different toolkits that are available for that context, which really is that memory, memory about you as an individual, preferences, certain people you’re working with all the time, patterns, and we’ll talk a little bit about that. And then obviously that brings in the graph of the underlying data there. And we’ll talk a little further what all is included there. You know, when we talk about Copilot, it’s not just Copilot in the web or Copilot, the, you know, app you may have. You know, Copilot’s embedded in many places, whether that’s in the Office applications with Word, Excel, and PowerPoint. And also then, obviously, within those, you have certain Agents that can be leveraged within the researcher and analyst or pre-built ones. that can be used there. You also have agents within SharePoint and within Teams that can be leveraged, as well as then using agents to build apps and workflows, all of which contains and can access that Work IQ layer that resides underneath there within Copilot. So, Work IQ kind of surrounds all the Agents, data, and apps that you that you have pertaining to you that you have permissions for within Microsoft 365. So, it adds this kind of overarching layer that allows you to again develop new capabilities within Copilot. But also use it using some of the core features of and capabilities of Copilot on its own. Yeah. How it works, again, you have multiple LLM models that are in here, and we’ve heard, there’s been a lot of debate lately about, okay, what’s, you know, we like, you know, Claude better than Copilot. Well, Copilot historically was primarily associated with OpenAI, but now Anthropic’s model is available within Copilot as well. There are also other models that are coming forward in there. And so this really isn’t, you don’t need to choose one model or the other. There’s auto selection that can be used and or you can select and say, I really prefer what Claude does with this. Let’s use that for this model. So It is a multi-model system, so you don’t have to, you know, pick Co-Pilot versus Claude. It’s embedded inside of there. The things we’re going to be drilling a little further today are, you know, the data, context, skills and tools that comprise Work IQ. Data comprises, you know, everything with Microsoft 365, but also Dynamics 365. There’s some Power BI and some external data that can be tied in there, too. So it can be pretty extensive in terms of the data that it can draw inferences from, ultimately. And we’ll talk a little bit about how that that happens. The context is one of the pieces of this that hadn’t existed prior to IQ in the same way, because there you have memory, and it’s both explicit and implicit memory, which we’ll talk about, and how it’s gathering kind of a basic understanding of you and the business. through the interactions that are happening naturally within the environment. So it can make better inferences and more informed inferences based on you and your organization. The skills and tools, we’ll talk about those, as well as then extending that as we kind of get into you know, how do you take what Work IQ provides and turn that into Agents and be able to build things off of that on top of that? One of the best parts about Work IQ and, you know, kind of the co-pilot construct is that all of this is secured based on existing permissions and identity. That’s why having all that Set up the right way is so important, because Copilot can very quickly expose some, you know, apps flaws or things you didn’t expect with permissions, but there is a it does leverage the underlying permissions that you have within SharePoint and OneDrive and all the data. that’s being pulled into this so that you’re only able to draw inferences on that data that you have access to anyways. And that’s all based on your Entra ID. And so it uses the existing security framework that exists to be able to limit what you can and your users can access as your developing agents. As far as data, obviously the M365 tenant data, SharePoint OneDrive get the most attention, because that’s where a lot of our files sit. But also, you know, your Outlook emails and calendars. There are skills tied to things that you can do with an email, things you can do with your calendar, being able to pull information from there, be able to summarize things, be able to get insights into that next meeting you have. things like that, that then can also be exposed to develop your own agents from there. The information within Teams, meetings, and messages is all readily accessible within Work IQ. The other thing here, it starts picking up the patterns of action, the people you’re collaborating with A lot. So. If I’m doing a lot with Corey, it will know that and be able to, at the right times, that information will be used to make decisions on what its recommendations are. So it does pick up on the patterns of collaboration and communication that occur there as it’s seeing the flow of information within M365. That’s how we get the M365 data in. To get outside data, there is a whole host of co-pilot connectors, both for third party and other external data sources that can be brought into this too, and it brings those into the underlying graph and into Work IQ so that That data, if you have those connectors set up, can also be used for any of the activities you have the agent performing for you. There’s also, there’s more and more connections into Dataverse, which underlies Dynamics 365. and the Power App space. So it can now pull from the information that’s stored inside a Dataverse alongside your files in M365 and anything else that you’re tying into, making it a very powerful platform in terms of being able to draw inferences off of this wide host. of data that you have out there. Again, to the connectors point, there are over 1,400 connectors that can be used to power agents. Now, these may be connectors, some may be MCP servers, which we’ll talk a little further about in a moment. But you can reach out to a whole host of cloud solutions and on-premise solutions there, too. Some of these, for any of you that have been around for a while, the same connector, it really began with the SharePoint search connectors way back when, kind of evolved into some of the connectors that you had with the Power Platform and being able to tie in there with some actions and but being also to draw data from there. And that’s really what’s kind of become these co-pilot connectors today is kind of a host of those things and connections to non-Microsoft sources that, or on-premise sources for that matter, that had been, you know, maybe cut off or felt like you couldn’t tie into today. So it allows you to develop pretty powerful agents just within the Copilot framework because of that, because now all that data can be surfaced through there. For context, this really focuses again on the views of memory. So explicit memory, when you hear that term within this context, pardon the pun, is about those things that you’re telling the agent and co-pilot to remember about you specifically. So I could say, hey, when I, when you respond to me, give me bulleted, I prefer a bulleted list, or I prefer more, you know, confined, shorter responses or versus longer for these kind of things, right? All those kind of things become memory there that you can tell it just by interrupting. interacting with Copilot directly and say, “Hey, remember this about me,” or you can store those in some of your preferences. Implicit memory are things that it’s remembering based on things that you have queried and your chat history with Copilot, so it will start to pick up those things that you may be asking for repeatedly over time, and that memory starts to develop around you within Work I. which again allows it to tune things to be like, hey, you didn’t tell me to remember this, but you seem to like this better than that. And some of that comes from the feedback mechanism that you can get within Copilot, the thumbs up, thumbs down portion that shows up on the responses. that gets pulled into that. And so it remembers things about you and hopefully gives you better responses the next time you work with it based on that. Again, it would be monitoring the workflows and activity that you have going across Teams, Outlook, Office, so that it’s remembering that those patterns of work that you have. And so then again, as it’s providing inferences, it’s taking that into account and maybe providing you a recommendation for, hey, maybe you should think about this versus that because of your history and your previous work patterns there without having to tell it. Finally, you know, in terms of business understanding and context, There is a semantic index that exists within there that allows for meaning-based retrieval of data. It’s not just, you know, kind of a keywords base or things like that. It actually has a little more meaning to it, which allows it to provide more intelligent responses. whether that’s data within M365 or via the connectors, and still maintains the security, privacy, and governance policies that are contained therein. You also have a semantic understanding layer, which it starts to develop ontologies and glossaries capturing knowledge from existing workflows. So it understands some of the terminology that exists within your business. And so then that allows it to, again, customize it further for you. In terms of skills and tools, the skills are those things that, you know, if, okay, I want to do something with my email, there’s a host of pre-built skills where it knows exactly how to do those things. Think of those like commands that you’re giving it and say, hey, I want to get this information from my email. That already exists and it knows exactly what to do. You just put in the parameters for that and it can go ahead and pull that in. So those are kind of pre-built commands that can be leveraged there. for greater speed and accuracy, because it knows exactly how to do those things. You’re not trying to tune that in the same way. It’s a very well-defined skill or command that underlies your agent. Tools, there’s a whole host of tools that can be combined into this. One of those are MCP server tools. different agent flows that can be leveraged, APIs, plugins. So host, and that’s an ever-expanding tool set that allows for the execution of those skills and allows you to build things leveraging those skills inside of there. So as we talk about building, what we’re going to kind of focus on today is Building, you can build within M 365 Copilot itself using the agent builder capability allows you to do things to be a natural language interface. So I can describe, hey, build me an agent that does this, and or you can then customize what it produces from there. If you need to do something more extensive. then you get into Copilot Studio, it can really build out some pretty sophisticated agents within there that then ties in with a host of other data sources and tools out there to be able to extend this even further. So with that, I’m going to turn it over to Corey, who’s going to walk you through where we take this from. Development perspective from there. Corey, are we good? Corey Milliman 15:02 We are good. Can you hear me and can you see my screen? Joe Steiner 15:06 Again? Corey Milliman 15:07 All right, very good. Let me go ahead and just maximize this window here. And I’m just going to go over a couple of slides before we actually jump into the demo today. So the tools that we have, right? So we have these tools and when it comes to our custom Agents, we really want to find the right models that are going through the line with our business problems. So say you’re building a customer service agent on your website. You might want a model that helps deliver maybe faster answers to your customers or one that gives you the most well-reasoned responses. So no matter which one of those scenarios you’re in, you can access the latest frontier models. with OpenAI, as Joe mentioned, Anthropic, we also have Grok, and we have many more. So we’re really not limited by one provider. And I’ve noticed in making these agents different use cases, different providers have provided much better results depending on what kind of agents we’re building. So when it comes to building these agents, customers do worry about how this will work with existing systems. So you can connect existing line of business apps and you get these faster integrations and less ongoing maintenance with these agent native servers. So again, if you’re building an agent to identify leads for your sales team, We can use an MCP server for our CRM to pull real-time opportunity data that helps qualify, I’m sorry, leads for our reps. So a lot of power with these MCP tools and the ability to intelligently wire these in without spending a lot of time on the development front. Another concern for customers is that agents might hallucinate. And so now we have automated evaluations that we can also run, oops, that help us actually evaluate and look at the responses that we’re getting and track these responses. So you can see here we have a screenshot of that, but we do have the capability built into Copilot Studio as well now. So we have this real strong ability to upload, say, custom questions or have AI generate questions for us. And then we can evaluate how the AI agent is doing and then take steps. Is that tweaking the model? Is that adding new knowledge? Is that refining topics or different things like that? So MCP versus traditional integrations. Today we’re going to be focusing on MCP. So of course we have our LLM, which is really limited to the knowledge it was trained on, right? So we have no out-of-the-box external system access. Think of it as maybe an isolated librarian that only has access to their books. So, in Copilot Studio, we also have connectors, and, as Joe mentioned, we have over 1400 connectors to all of these different line of business systems. Those connectors require custom integrations for each one of these tools, so each action I want to take, I have to define the input, I have to describe it in a way that AI is going to understand it. And I have to update it as new actions or definitions become available for that tool. So once we move into the MCP space, we are providing A standardized integration layer. So we are making, once we define our MCP tools, we’re going to be looking at out-of-the-box MCP servers today. That MCP, that once I define a function and expose it by the model context protocol, that function is instantly compatible with any MCP enabled LLM. So that is really, I’m integrating once in solving complexity and scaling and allowing for much less ongoing maintenance. So you can see here, this is an overview of different ways MCP works in Copilot Studio. We use actions, and today we’re going to be focusing on Work IQ MCP clients, and that covers Word, Excel, the Microsoft 365 stack. And it is really coming up to parity with the out-of-the-box connectors that we have as well that can be very complex to integrate. And then, of course, we have our custom MCP clients that are also available that we’ll be showing today that allow us to also expose these other tools like Azure App Service or GitHub or all these different repositories, so that we have one agent that has all of these different tools potentially to help us. go through not just one transaction, but an actual workflow and to actually bring that agent across my entire everything I’m trying to do instead of just one task. So I’m actually going to stop the slide show and I am going to switch over to Copilot Studio. And so here you can see that I stood up a very basic agent. And this only has two knowledge sources right now, IT documents and HR documents. So what you can do to add these Work IQ servers, we’re going to add one right now. I already have some that are configured. I’m going to go over here and I’m going to click Add Tool. And you go over to MCP, and you can already see if we do a compare and contrast. I have an MCP, I’m sorry, I have a connector for Microsoft Teams here. And so each one of these actions, if I want to add a member to a Team channel, I have to go ahead, I configure that connector, I configure the inputs, I configure the outputs. So you can see here, I would have to, if I want the same parity, say as the Microsoft Teams Work IQ MCP server, I would have to define all of these different skills connectors, and then map out what that looks like. It’s very complex. But now with MCP, excuse me, We’re going to click this button here, and I’m going to do a search for IQ. And we’re going to get a list of all the Work IQ ones that are out there. So we have Work IQ for SharePoint. We’re going to go through all of these and look at the different things that we can do with these. I’m going to go ahead and add Work IQ for Copilot. And so any of these that you select, Looks like it didn’t refresh. Let’s do that again. Go ahead and add a tool again. Looks like I still picked up my… Microsoft Teams that I clicked on at the beginning here, so we’re going to go ahead. Now, it’s going to ask for my connection. I’m going to go ahead and create that connection, use my enter ID, and go ahead and click create. That is all I need to do for the initial configuration of this, and it is going to ask me to sign in. And now we have that added and now we have the add and configure option. So while that’s adding, I’m going to show you a couple of the other ones that I’ve already added to this. So we have one for the calendar, which allows me directly through the agent to start interacting with the calendars. So think of scheduling, thinking of creating events, thinking of moving events, things like that. So all of these commands are available without manual configuration of a connector for Outlook. So it can find meeting times, and it can accept. All of these are available, and there is no other configuration I have to do here other than add that. For these additional details, if I were to publish this agent, I would use the end user credentials instead of the maker credentials. So that way, it’s respecting the permissions and our governance for our end user and security, and it is only acting on what the end user has access to. We also have Work IQ for mail, which is a little bit different here. And this one, you can see here that this is about writing emails, scheduling emails, sending attachments, creating, deleting, replying. So this is giving me, from an agent perspective, the ability to help me go beyond what I can do just with Copilot for Office 365. And we can think of this as also in ways as maybe You have agents where you want these agents to take actions. Do we want an agent to take action where it’s going to schedule an event? It’s going to send an email message. Those kind of activities previously, those took a lot of work. Now we have those available here. Same thing with SharePoint. So if we think about work on queue for SharePoint sites. Now we’re getting into the ability to save information from our Copilot Studio Agents directly over SharePoint lists and manage lists. So previously, we would have to create connectors. We would have to just go find our list, map our columns, and tell it how to do each action we want to take. adding information to a leaderboard or making updates or things like that. So those are things that now, just by adding this, I have the full ability through my agent to interact directly with SharePoint lists and libraries. Again, and we’re going to show how this actually works here live. And then Teams. Now, this is one that I’ve interacted with a lot from the connector perspective, and I really like this tool because now I can list out my chats, I can list Teams, I can think of this as a way to send notifications to people. I can actually post messages directly from this. If I want to give people updates or I want to go through my workflow through the day, I can actually use a Copilot Studio Agents and directly from this now, I’m going to be able to, we’re going to update a document, we’re going to save it over to OneDrive, we’re going to email it, we’re going to schedule an appointment with Joe, and then we’re going to post a message about it over to a Teams chat. So I’m going to minimize this. I see that we have… Yes. Joe Steiner 25:04 If I can, Corey, just real quick. Sorry, I just, I noticed somebody raised their hand. If you don’t mind, if you can put your question or comment in the chat or Q&A, we’ll gladly respond to it. We want to hear from you. We want your questions, but that way we’ll respond to it as we continue to flow here. Thank you. Corey Milliman 25:27 Awesome. Thanks, Joe. So here I’m going to do this over in the testing center instead of bringing this over, you know, through publishing and things like that. So as I mentioned, we have a test agent that’s out here. And I have HR documents and I have IT documents that are out here. So now I’m going to go. Back and find, gonna start a new test here. Yes, I believe I was in the wrong window. Here we go. We’ll start here. So first, to show different things. So I have a team site that’s out there, and I’m looking at an agent where maybe I want a quick onboarding. I want, I have a lot of it. We’re using HR information today, but I want to get a list of all of our policies. And procedures, as I think we need to update the email addresses we have for HR in our. Cash. Please check the demo. Copilot site, find all HR policy documents. And let’s prepare to change, change our contacts. So you can see here that first it kicked off Word. Now it’s looking at Teams. It’s looking at SharePoint. And if I now I have to click allow. It’s going to make a liar out of me here. Here we go. Going through, and it’s kicking off the model context protocol, searching my documents. Now I have a demo site set up, and I’m also a member of Concurrency’s HR site, so this is going to be interesting. It did understand that I was talking about a demo co-pilot site. Joe Steiner 27:18 Ann. Corey Milliman 27:37 And you can see here, it’s actually going through and looking for these different files. And you can see what’s actually going on. And it’s actually looking at our real HR content right now. And this is actually a site that I actually had put together for some other demos. So we have Getting Deck in. I found multiple policy documents. It found the correct site. We can see here that it filtered down, and now we’re going to wait for the model to respond. And it is going to now give me a list of these different documents that are out there that have an email address so we can start figuring out how we’re going to deal with those. And while that is going, I’m going to jump into another window, and we’re going to say, get me a list of all teams I am a member. And so this is showing how we can start interacting with different teams channels that are out there. It’s going to start looking across the organization here, concurrency, and it’s going to start showing me all the different teams that I’m part of, and it’s going to go ahead and do that search right now. So you can see here. So, we found and reviewed our HR policy documents from the Copilot demo site. Five key HR policy documents that contain HR contact support information. We have payroll at ParvezRajiwatecircanacom, HR, HR, HR, and HR. So here it’s showing all the different email addresses and where it appears, next steps. So it’s asking, do you want me to create a document? Do you want to update these documents with the new email address? Or do you want to search for additional documents? Actually, I’d like. To create a meeting with Joe Steiner to discuss these findings and create a plan. Create a task list. Emerging. Cash to meeting invite. And find time for me and Joe to meet before the end of next. Because I know Joe’s calendar is real packed. So we’re going to, while it’s going through there, it’s going to now go and look at what we’ve created. I asked it to create a meeting to go over our findings. I asked it to create a document for me that’s going to be a task list. I asked it to add it to the meeting invite. And to find a time for Joe and I to go ahead and meet. So now you can see it’s kicking off these different MCP tools that we have. And again, I didn’t have to go and create a connection to my tenant. I didn’t have to tell it in Outlook connectors or Teams connectors how I wanted to interact with OneDrive, how to interact with Teams, how to interact with our calendar. It is actually going to go out and do those things for me and create that. So you can see here, it has a list in my other window. I just wanted to get a list of all the Teams channels that were available to me, because I’m a member of a lot and maybe it’s hard for me to find these sometimes. So now let’s… Post an update to the Teams channel for pilot demo. Let’s post an update. of what capabilities are available in Work IQ for Teams. And please excuse my spelling as I hammer on the keyboard. So now while it’s going and doing that, I’m coming back over here and I’m kind of jumping back and forth here. So it’s gone ahead and it’s done 3 tasks for me. First, to create a task document. It says it created. Did it really do it? Let’s take a look. Give me a link to the document. It is a Word document. It looks like I created the document, of course, before it sent out the email. It gave me a list of things that need to happen. So I have that basis of what I discovered in this chat bot. So this can be something that’s much more complex, right? It could be, we found an error in a report. We have to go over something. Whatever information we have found, It went ahead and took care of that. So found that our next available time is Wednesday, April 23rd, already tomorrow. So I can actually take a look at my calendar and I can see here, it’s scheduled an hour with us for tomorrow. So this already went out to Joe. It wrote my meeting invite for me. It found a time that worked for both of us. And its writing, it came from me because I was the one using the agent. So I was talking about the key findings and it put an agenda together for me. Joe Steiner 32:32 Deb. Corey Milliman 32:49 It went ahead and put the task list and the link to the task list and created the meeting link for me. So I already went through Joe’s calendar. I didn’t have to go do that. And created the meeting, created the document, and did that by looking at all of the information that I had available just by scouring through those different Word documents. Things like that. So on this other window, you can see here, it’s… It’s gone ahead. I posted a comprehensive update about Work IQ for Teams to the general channel in the Copilot demo. So now if I go over to my Teams. Find my Teams window. I don’t usually go into Teams, of course, when I’m sharing my screen. So let’s find this. Joe Steiner 33:39 Mm. Corey Milliman 33:46 Ann. Go here. So here is an email update task. I need to find my Copilot demo site that is over here. So at 1133 this morning, which is just a couple of minutes ago, it posted a message for me here. And so there are some different ways I can do this, right? So if I have a team of people or if I’m working on a project and I want to send out a project update over to the team, I can tell my agent, look at my email, look at my SharePoint content, look at the communication around this customer, and now let me message my team with an update and next steps. So this is a really good way of automating and going through a lot of the things we do. We might look at a document. We might go to a SharePoint site, we might decide something needs to be updated. We identify a problem and then we communicate to everybody that this is the problem we need to solve. And so Copilot is helping us go all the way through that. And I’m just showing that we can interface Copilot Studio Agents so easily now to all these different tools. Whereas before, again, you would have to do these all manually. And I’m going to pause now because I know we had a question and I’m going to show some other things here, but I wanted to see what this one question was. We have some other questions as well, it looks like. I see some have been answered already. Joe Steiner 35:18 Yeah. Corey Milliman 35:28 Oh, yes, you can fire up a Teams message. You can send an email. We did cover that. So Jake, I hope that helped. And then, yeah. Joe Steiner 35:36 And if I can comment there real quick, Corey, I think one of the things that happened there that maybe wasn’t, may not have been picked up is particularly in the ones where we went through and you created the Word document and then found time in the calendar and sent the invite out. It’s actually operating against multiple platforms there as far as the system’s concerned there. and orchestrating all those together using just the combination of MCP servers that were in place there. So it made the decisions, okay, I need to work with Word, I need to work with Teams, I need to work with SharePoint to pull all of this information together and perform these set of tasks across multiple ports. platforms. This is all within the M365 space, but they are separate platforms within the M365 space. If you had outside systems in there too, and an MCP server for those, those also could have been orchestrated into there in the similar fashion. So Corey Milliman 36:40 Absolutely, absolutely right. So once we start MCP, once we start adding these, we add external MCPs as well. We are adding natural language over the model context protocol. So we can, just like in this, you can see I went through Word, I went through Teams, we can see exactly what it is. Joe Steiner 36:50 Oh. Corey Milliman 36:59 did here and what it actually did as… As I go through here, it looks like I lost some. Here we go. So you can see these are the different commands. And if I wanted to get a list of teams without using this MCP, I would have to go to Power Automate. I’d have to use the connector. If I want to list teams, if I want to list, if I want a conversation where the user can find a channel, I have to list the channels, I have to filter by the user. All of these things I have to build into these connectors. And then posting a channel message, which seems I have to go pull this channel ID, I have to specify a content type, I have to specify exactly how to format that, bring in the team IDs, all the internal grids and everything else. So now I don’t have to write these manual connectors. So if I want to interact with Teams, I can do that. And I can do that automatically through MCP. And this model context protocol shows, once we go through the effort of telling an MCP server, how do I handle a Team ID for a Microsoft Team? That’s reusable. So if you create your own MCP server for another line of business application and you bring that over as a tool, then your agents can use that just like they’re using these Work IQ. But we are spending the time today on the Work IQ because these are some really powerful ones that are available out of the box. And I see there was another question from Trent. that I do want to answer there. Joe Steiner 38:31 Deb. Corey Milliman 38:35 Yeah, I’m using the testing capability actually in Copilot Studio. So I’m cheating for a demo today so that I can have multiple instances and interactions with my agent without publishing it over to Teams. If you want to do something similar for that, you would probably have to use the web version. To get that open in multiple, it’s not really easy to start multiple chats. You would have to jump back and forth over here in Copilot once you launch your Copilot icon over from Teams. So not a really good way to do that for Copilot for Work, but you could get around it using the web interface. for Microsoft Teams and then have multiple web interfaces open with different chats and different windows. Joe Steiner 39:22 I think we have one more question from Eric asking if there’s a way to make this agent with MCP available to a subset of users via Teams. Corey Milliman 39:33 So, let’s… Think about that. So when I’m adding functionality to an agent, so there are a couple of different ways we can do this. If I only want some people to have access to that, I would only publish an agent that has this capability to that group of people. Because this is interacting with teams on their as them on their behalf using this just like we would with a connector. So that’s one way to do that. And that’s from an end user perspective. So the thing when you are adding these tools to your agents. We can also turn off different things if we don’t want our users to be able to have certain functionality. So by default, these are all log. So you can see here, maybe I don’t want all of my users to be able to do certain things. And we can go ahead, I uncheck this, this says allow all. Joe Steiner 40:54 But it, it. Corey Milliman 40:54 Add a tool. Go ahead, Joe. Joe Steiner 40:57 Yeah, I was just going to say, I think, you know, particularly I think the first thing you said there would probably help, Eric, if I’m not mistaken, where we’re only publishing, you can choose to publish the agent only to certain people. And that’s probably, you know, for example, with a proof of concept, that’d be a good way to do that and say, hey, this is the team that’s going to be testing this out. and just publish to them rather than to the whole work. Corey Milliman 41:21 Yeah. Absolutely. So there are some other Work IQ tools that are available. I’ve added quite a few to this basic agent. You can see there were no instructions. I have two knowledge sources that I attached, HR policy documents. And here I’m going to say, We need to update our PTO accrual. Policy. On the demo Copilot website, find our PTO policy. We want to increase referral by 25 percent, and we want to scroll over to 80 hours. Update in SharePoint. And we’re going to go through and let this expand again. And you can see here, it’s firing off one after the other. All the different MCP tools and deciding which one it needs to use. And it’s actually also looking at calendar here, but it is going through based on, now we can see here, that was actually pretty interesting. Find sites, it looked for my site again. Here’s the reasoning. The user wants me to find a document on Demo Copilot website, which I assume is a SharePoint site. I didn’t tell. Update the document to increase accrual, increase rollover. So now it’s breaking that down into steps. First, I need to find the SharePoint site. I need to search for the document. I need to read the current policy and update it with the new rules. And now we can go ahead and see what happened. And we can see here, it went ahead and it found the PTO document that’s up there. And I found and reviewed your PTL policy document. And here are new accrual rates. Here’s your rollover limit. And I actually cheated and I tested this earlier today. So if we go over to this site now, it should. Didn’t give me a link. See if it gives me a link, I can go out and find it while we’re waiting. So here it updated this document on the SharePoint site a minute ago. It did change the name to updated. And here’s a Virgin Dynamics PTO policy updated, shows it updated today. And the new bi-weekly accruals, annual accrual, unused PTO, contact information, and things like that. So it does a really good job of also, we can use it. thinking about if we’re hooking into, say, our ServiceNow system, and we are trying to update a knowledge base on SharePoint that our agent can use. I can use my external systems. I can understand trends and analysis and ways of solving problems. Say, go write a document, go write a KD article on this. And in turn, my agent is pointed to the SharePoint site. So this was, say I was looking at ServiceNow information, the MCP for ServiceNow, it brought up a bunch of information about recurrent password resets or some kind of issue that we have. Now I drew up a KB article, and it can use that KB article as knowledge. So we can do this a couple of different ways. We can use these out-of-the-box tools from Microsoft with Work IQ. When we pair them with external MCP servers, we query those external MCP servers for knowledge, extract the knowledge in a way that maybe Copilot Studio better handles sometimes, and start building up different assets across the organization. Word, Excel, PowerPoint, things like that, to also help us build those documents, schedule calls about it, and do different things there. So I think those cover… Oh, I see a question here. Saw something. Thick log. Let’s see, can you bring in an MCP server that’s not listed? Absolutely. So in order to do this, I’m not going to be able to bring an unlisted one into this demo because I don’t have one ready, but I’m going to show you how you would do this. So the question is, how do I bring in, say, my own MCP server, an MCP that isn’t listed? When I go to MCP or Model Context Protocol here, I have this list of all these different ones that are available, kind of off the shelf, out of the box. So if I want to bring in my own, I click what I misclicked on earlier today, model context protocol here. It’s going to ask me for the server name, the description, and the server URL. So it does require a streamable HTTP endpoint. It’s going to ask you for your authentication, API key, OAuth 2. And once you connect to this external MCP server, as long as it is compatible, it’s not the kind of MCP that you’d run, say, on your desktop with plot code. It has to be an HTTP endpoint. And once you’re able to hit that endpoint, you can connect Copilot Studio to an MCP server that way. And that’ll allow you to bring in whatever custom tools you want as well. See. Yeah, so if you’re using the MCPs and cloud that you install locally, again, you just want to make sure that it’s running as an HTTP endpoint. And that HTTP endpoint has to be available to Copilot Studio, which would mean it needs a public IP address. So it does need to be reachable from outside of your desktop if you’re running it locally. And there are a couple of different ways to do that. Let’s see, we have some new questions as well. So I personally, Joe, do you want to help with this one? Because I’ve not used workflows. This one question from Trent. Full disclosure, I have used connectors and when I saw Work IQ coming out, I said, I’m not using these other things from a Copilot Studio perspective because Joe Steiner 48:11 Derek. Yeah. Corey Milliman 48:25 It’s super powerful to me. I have not used the workflows that were part of Frontier though, Joe. Did you hit on that? Joe Steiner 48:29 So, yeah, workflows are, yeah, workflows are coming particularly with the addition of the further addition of Claude as a model within Copilot, and so there are certain things where it can, it more and more readily can take action on your behalf. and can be a little more sophisticated in what it can do there. And so, yeah, you know, essentially you can, and I think this will be one of the things, this is constantly evolving. As those workflows, you may find that workflows can handle certain tasks easier than pulling those in and having the way Work IQ does. That ultimately gets blended into this though as well, so that as those come out of preview, that’ll be just be, those could be called on just like we’re calling on within Work IQ. And those will ultimately likely be embedded into some of these work IQ, the work IQ functionality as that continues to evolve. But yeah, for right now, probably workflows can do certain things in terms of writing things into applications and things like that in a very sophisticated fashion. And so that is a great thing to be playing with. Work IQ has perhaps more capabilities at the moment in terms of breadth than the workflows do, but as those workflows evolve, they are a little more sophisticated in what they can do in the areas that they can operate in. I hope that makes sense. And then in the more narrow places we can do, they can do more, but Work IQ at the current state is a little more broad in terms of the things that I can touch. Corey Milliman 50:10 Thank you, Joe. Awesome. All right, so today we did show, let me just go through, we talked a little bit about what is available, kind of showed working through the Microsoft stack here. And so again, you have Word, SharePoint, Copilot, Calendar, Mail. Teams. Now, we didn’t go through user OneDrive. It can also create documents on OneDrive. The user I did not go through today because I did not have those rights in the concurrency tenant. So if we actually go through and the user agent is interesting, and this gets into something that might be more along. I want to build something for my internal IT team to start using. And so let’s go and bring that agent up. And I’m just going to show you the commands that are available and talk about that a little bit. If I have it available. See. I don’t have it available to me in this environment either. So I do have that in another Copilot environment, but that is available. It does do some things to help you with Copilot for Office 365 users, assigning licenses, looking at readiness, looking at update channels, a lot of those things that we do manually, getting a list of the users in Active Directory. There’s some different things there. I just don’t see that I have access to the user one. Try. Just a different search. Aha. Not sure how far that one’s going to go with my lack of permissions, but we will see. Some other things you can also do with all of these is start adding expected inputs. So if there are certain things you want your user to provide as ways of interacting with these different MCPs, we can actually start bringing over items from if you create a topic and you’re capturing A variable, maybe it’s a department name or it’s something that you want to capture and bring in programmatically, you can bring that structured type of data in as well, grabbing that from different contexts. So this is more about getting user ideals. user details, excuse me. So you can see kind of some different things that we have available. This tool has changed pretty dynamically, so I see it’s a little bit different even than what I looked at last week. Now it’s getting my profile details, UPN or Entra. is me for the signed in user, get multiple user details, get manager details, and get direct reports. So now this people version, now I can find out or maybe get information on an entire team, find out about the organizational structure. or make some changes there. But it’s giving me the ability to have this natural language search. I can go and get my user information. And I want to schedule a meeting with Joe’s boss. And then I would look up who does Joe report to and tell me and help me schedule an email or something like that. So there are also some admin tools that are available that I don’t have access to, but we’ve Joe Steiner 53:51 No. Kate. Corey Milliman 54:14 We’ll cover those in another session as well. Joe Steiner 54:17 No. Corey Milliman 54:18 All right, so… That is what I want to cover for the demo and to make sure that you just saw a repeatable framework or some kind of practical framework of how this can help throughout the day and how this can help from a Copilot Studio perspective. So Joe, I will turn it back over to you. Joe Steiner 54:37 Yeah, yeah, please. So really, you know, just to wrap this up, share mine. No, we appreciate the time. Thank you, Corey, for taking on that tour and always valuable to see it live, right? You can see exactly what this looks like. And he showed you a couple of great use cases there. Certainly, you know, take that where you will from there. If you’re interested in talking to us and say, hey, you know, I’ve got these ideas, but would like to, you know, maybe chat with you guys about that. explore maybe how we can leverage this further within the organization. Happy to talk to you. Please reach out to us. We’re gladly scheduled kind of a 20-minute follow-up session with us. Talk about what’s possible, what you’re trying to accomplish, and talk about how we can help you out. So thank you all for attending today. Please fill out the surveys. We are Your feedback is valuable. As we like to say around here, feedback is a gift. So we appreciate anything that you’d willing to share with us. And so I wanted to say thank you again for your time today. And yes, we’ll be sharing the slides out and look forward to hearing from all of you. Hope you all have a great rest of your day and a good week, and we look forward to seeing you again soon. Take care. Corey Milliman 55:52 Thanks, everybody. Thank you. Bye-bye.