The Agentic Toolkit in the Microsoft Ecosystem

All right. Hi, everyone. We are about to start in just one minute. Waiting for everyone to come in and join. We've got roughly half of the registrants already in the room, so we'll wait a bit more. Okey doke. Okay. Just one more minute to allow everyone to to flow in, and then we'll get going. Cool. Cool. Cool. Okay. All right. We're getting started. Today, we'll be talking about the different Microsoft agent technologies, and I'll start with a little story here. So this summer, we kicked off an AI project with a client who kept on saying, We want agents. We want agents. We want to deploy agents. And we worked up the statement of work aligned and everything started to build. Then roughly halfway through, we realized that what he meant by an agent wasn't what we meant by an agent. We're still working together, so it's all good. But that moment really stuck with me. I realized it wasn't his fault. It wasn't our fault. It's the fact that agent has become the most misused term in our industry. So today's objective is to bring clarity around that. I want to show you what the different types of Microsoft agents actually are meant to use each one of them and what they look like in the real world. So hi, everyone. I am Balaj Horvath. I'm the founder of Visual Labs, and I've spent my entire career working in the Microsoft ecosystem, ERP, CRM, Azure, and now in AI. And I worked with Cloud ERP since its inception in twenty fifteen, twenty sixteen. And in twenty twenty, we founded Visual Labs. The first cloud based ERP rollout in Hungary was done by us. And since then, we've grown into a team of twenty five people of consultants just fully focused on the Microsoft ecosystem. We're still covering ERP, CRM, BI. And what we are founding is that most companies want to be ready. They want to get AI value, but they their foundations are missing. Quite often, things are not in place. So, that's why we are, in the q one of twenty twenty six, we are running a webinar series that is focusing on the foundations of how to really become AI ready. And and let's let me just start sharing my screen, and then we'll really get into this. So what you will see here is agents among us. Okey doke. Alright. So why today's session is is important and and this whole series is because companies who start building agents early on will start to get value out of it a lot earlier, and they will start compounding. Microsoft actually has a study that shows that companies who are building agents earlier and they focus on the readiness piece will be able to build agents two point five times faster than the companies who don't focus on AI readiness. So this is why I think it's it's really important that once you get into that compounding effect, the early starters will have a lot bigger advantage. And today's session, you'll learn, why value beats, technology, what your options are within the Microsoft stack. I'll actually show you real live demos for every one of the tech options, that currently exists today in the Microsoft ecosystem, and you'll understand when to use what exactly what the the main differences are. And we'll talk a bit about the next steps for your business. I already introduced myself. Balaj Shorvath here. That's me. And this is our company, Visual Labs, with all of the certificates that our team members have, and we're quite proud to to to wear these certificates. And that allowed us to get all the partnership designations as well in business applications, digital app innovation, and, of course, data and AI. We've got clients across Europe as well as in the US. And let's get into the meat of it. Before we jump into the actual tech demo, I'd like to bring some clarity around technologies and terminologies. As already mentioned, that it it is quite confusing. So by now, we've got co pilots and agents literally everywhere in the Microsoft ecosystem. So I sort of think that Microsoft does make a mistake with with not really clarifying what Copilot is and using it both as a brand, as a as a terminology. But let's let's move past that, and hopefully, we can we can give you some clarity on on that front. The core piece is that AI is transferring what we build and how we build. So in the past two years, we've really seen that with the agent the advent of agentic AI and generative AI, we can now really build a lot faster. And how we are approaching problems, how we're addressing problems is also really becoming fundamentally different. So today, we'll focus on the how part, on the how we build without actually going into too much technical. So I'll we'll show you actual proper demos, but that's not the purpose. This is not a training session. It's more to bring to life the different technology stacks. So a quick definition of what agents are. So agents are AI assistants designed to automate and execute business processes working with or for a human. Right? And these are the components of an agent. We will actually see them. It's built on a foundation model, so something like OpenAI or Cloud or even Gemini. It's got an orchestrator which taps into different tools, knowledge, and the autonomy pieces it has available. That's think of it as the brain or the heart of the agent. Decides when to tap into what. It's called orchestrator for a reason. It is like an orchestrator of a of a big band. And it can be connected to other agents. And this is where it gets really interesting that an agent's user experience is actually optional. It might not have might not have that. So if we move forward with there we go. I'm back now. So if we move forward with with this, we have we can look at agents at a spectrum. So you've got the simple retriever agents, and you've got the fully autonomous agents on the on the end. And you've got different technologies that you can use to build these agents. And this is where it gets gets really interesting. So if you grab this simple advanced bar and move them across, We actually have m three sixty five Copilot, right, the one that we've all heard of and many of us are using on a daily basis. We can use that to build agents. We've got SharePoint, which we can use our SharePoint files to start building agents. We've got Fabric, which with which we can use data related agents. We've got Copilot Studio, which is a whole agent builder platform similar to Foundry, and I'll talk about the difference between Copilot Studio and Foundry. And we've got m three six five agent SDK, which is a software development kit. And these six pieces of of tech items are are what's available today for us to build agents. And what we'll look at is the the spectrum. Right? We looked at simple, complex, the audience, who's it actually for, who's it who's this meant to be used for, whether it's a declarative and an autonomous agent, and we'll actually define what declarative and autonomous actually means, whether it can be connected to other external items, what's the what's the connectivity capabilities, what large language model providers it can use and it can can sit on, and who what access where you can access these agent builder platforms and what licenses are required. K. So let's let's just first make sure that we are clear on declarative and autonomous agents. And for that, I will go back to the previous slide. So declarative agents are the most simplistic elements that are simple retrieval agents. You declare what they can do, and then they will just go in and and access the data, come back to you. It's practically a chatbot sort of experience. Whereas autonomous agents can actually tap into knowledge, skills, and would have autonomy, self learning. They know how to handle exceptions. They know how to do planning. So that would really allow them to to execute and to end processes in the background. Increasingly, when we talk about agents, we really do mean autonomous agents. Right? So this is a key piece that is that is different for every agent technology platform. So let's look at m three six five Copilot first. And with with this, we'll actually move into our demo bit. So m three six five Copilot is a no code agent builder platform. Its primary audience is business users. You can only use it to build declarative agents, so no autonomy, no connectivity there. That's why it's really easy and safe to give to business users. And what you'll have is a large language models you can use well, g p t four o is being deprecated. You can choose g p t five, and Claude has just come out last week as a sort of preview option for users to use here. And you can access m three sixty five Copilot agents, the builder platform through Office dot com, or if you are a Windows user, the Office app on your laptop. And you will need a m three six five Copilot license to be able to build agents. Okey doke. So let me pull up our Office three six five. Here we go. And move that over here. Do a reshare, and we'll be good to go in one minute. So while I'm just juggling the screens, We'll share the full window. Bear with. Okay. It's not playing ball, so I will do this. Alright. So what you can see up here is the office dot com site. So this is where you'll have all your agents coming in, and you've got your notebooks, chats, previous chats, all that good stuff. So this is how you can get get started here. And if I just go to all agents or here, actually create new agent, I can start creating an agent or start creating an agent here. What you'll find here is you've got all types of agents, whether they are built by Microsoft or by third parties like Canva, Atlassian, or Miro, and this is where your agents are. So you can see that I'm using this primarily as a playground. And I can start creating an agent very simply just by creating an agent. And I've got preconfigured templates, which if I just populate, it will pull in a prompt. And this is where I can give it instructions. You see this is a pre pre created prompt. But I'm I'm keen to show you a pretty good use case that actually provides a lot of value to our to our clients is our discovery simulator. So I wanted to train myself primarily on how to build, or how to run a good client discovery session so I can really understand what the business problems are that our clients are trying to solve. So I built role, practically a role play agent, and I used our discovery methodology, and I fed our discovery methodology to our to check GPT. But this is how I want to run discovery. Could you help me build a role play agent so that I can practice these these role plays? And it actually gave me the prompt. So this is actually something I would very much recommend that you will use ChatGPT or Copilot or whatever agent that you're using to start creating prompts. So rather than me having to come up with all of this, it typed it out based on my instructions in a way that is easily understandable for the large language model, and it gives the right context. So you can see you are a custom role play agent for Visual Labs, act as a client's stakeholder during discovery meetings, and it allows you to to to really, you know, define the core behavior, the dynamics, and create one, three scenarios here, either an operations director, field service, chaos, field service manager, or CFO. And we've got some guardrails. So, you know, don't don't reveal everything at once. Be realistic. If asked for examples, provide an example, so on and so forth. And you should add the conversation when when when you reach next steps, concrete next steps, and you've got conversation starters defined here. You could actually bring in other knowledge sources even from the public Internet. And and once you're once you're done, you can actually start testing out your agent very in a very simple manner. So here, I can just say, let's play the CFO scenario. It's processing, thinking. We've got the CFO scenario for an engineering consultancy. You can start discovery. Okay. What's the what's the business problem you are trying to solve? Right? Very, very basic thing. So I've played with it a couple of times, and then I'm you know, once you get into the habit, you you understand how it thinks, and then it becomes quite easy. And also, would recommend turning on voice mode so that you can actually converse with the agent, which is it gives you a lot more realistic scenario rather than having to type it all out. So you can see I could just keep on going. What's your what's your main KPI you want to improve? And I can just, you know, start going in and start discovering before I jump to conclusions. So I can update this. And what's important about this agent is that I can share this with others. So I built it on my own Copilot thing in my own tenant, and I can share it with other people. Currently private and available to you. But once I hit share, then it becomes a public public. So within my colleagues, within my tenant, I can share it with with others. So let's let's jump back to to the second agent scenario, which is can I share the entire screen now? There we go. Which will be SharePoint. So this is yet another no code agent, and this is also for our business people. What we'll do with this is is yet another declarative agent, very similar to to what you what we just built, what you just saw. It's not going to be an autonomous agent, and it's it doesn't have any connectivity other than the SharePoint site that you have. And it's also built on the same LLM model that that your Copilot agent, and it has access to your to your SharePoint site. So if I jump jump here, share the screen. Then here we go. So what you see here is we dumped most of our project files. Excuse the Hungarian. We jump jump dump most of our project files in here for each of our areas, and you can see different estimates, statement over kickoff, and then you've got actual files in here. And what I did within this SharePoint folder, I've you can see that I've already done this. This is project life cycle agent. I've created a through AI action, I can just create an agent, and I can give it a custom instruction if I wanted to. I can give it a logo. This can see that this is the the source that is built upon, and I can add additional sources. I can define the behavior, give it the custom prompt, and I can already start testing it out from here. So what are the main files here? Right? So this would be a very, very basic question, but I can I know that we have a statement of works here? What does a statement of work look like at Visual Labs? And just because this is a public webinar and this does have some client data, don't, and I did test this out. So don't expose any client name. And it will mention the sources where it came from. So you might see some client names there, but it's actually not referenced. It's not actually calling out different client names. Follows a similar structure, version control, executive summary, business requirements, technical there you go. So that's a SOW for us. And once I'm happy with this, I tested it out, I can just click create. And because I've created this so many times, I will want to close this. I can just very easily come back in here and show you that we have this previous agent, the project lifecycle agent that I've already created, and and it shows up among my agents. So this is actually really cool. And if I want to start a new chat, I can add an agent here as if I'm just adding one of my colleagues. So chatting with project lifecycle agent, give me summary of main artifacts of a project. And what will happen here is yeah. Again, obfuscate client data. Cool. So that way, I can just very easily build up a knowledge base on SharePoint, which is where we are already storing our files and click create agent, and I have an agent that I can use or share with other people. You can see that there's actually it's really interesting. It's actually created an agent type extension. And here you can see discovery deliverables, consider some of our main artifacts of a project, what we do referencing one of the clients. So that's agent number two there. And if you go back to, again, agent type number two. And then now let's let's bring up a fabric agent, which is a lot more interesting now. It is still a no code agent, and it's primarily to be used by data teams. But you will see that it doesn't actually require any specific capability to be able to build this out. It is still a decorative agent, and it is connected to, Power BI or fabric semantic models. So this is for you to essentially be able to chat with your data, build on top of your data domain. And you can't actually pick and choose the model, but it will it will give you a pretty good idea, that, you know, Microsoft uses, GPT four, and five. And you access this through powerbi dot com, and it does require a Fabric F1 or Power BI P1 non trial, so it comes for a paid license. Quite often, we see that customers want to chat with their data, access their data, give their data to agents or AI chatbots. But large language models are called large language models for a reason and not large mathematical models. So if you give too much too many numbers, too much data to an agent, they really start hallucinating. So what's key here is Fabric actually writes Python or DAX code that runs in the background and does the calculation. It runs the code, it builds its own little calculator, and therefore it eliminates the chance or opportunity for hallucination. So this is a best practice way of exposing your data for agentic AI. And what's really clever with the Fabric Agent is that you can link this with other agents. Okay. So that was the intro bit. And let me let me share my screen again. Sorry for this this back and forth. You've got Fabric agent coming up. So I'm at my fabric workspace. You can see issue tracking number four is the name of the agent. So this is a an issue tracker agent. We do have our issue tracking semantic model. It gives this is directly linked to our Azure DevOps instance, which is our ticketing system. So this this is an actual real use case, a real agent that we use. And here's here's a here's a prompt that I gave. So I just tried to chat with it without a prompt, and I realized that it doesn't actually know what's a current issue. So what is what is an open issue. So I had to define that issues in status of new and active count as open. And also because because of the data load, it's there's a flag that which which row is current. So I want it to pre filter or filter out any of the rows that are noncurrent. So it should only include is current. Yes. So this is something like someone would write in SQL or or DAX as part of the data engineering, but I can do it without having to write a single line of code. And when I ask about issues, I wanted it to give it in a table format. I wanted it to sort it by ID, give it a number, and these are the fields I want to see. So I can really define how I want it to to behave and discuss. And let me just clear this. And and let's start with a easy question of how many open issues do we have. And at this stage, it's actually analyzing the data source. You can see that it's thinking right? It's executing as per my actions, analyzing issue. And the clever thing is that you can actually see you will actually be able to see the code that it's writing in the background, and you can validate if you know how to read that code. What's what's what's what's in the back? So it actually wrote DAX code even with a comment. So if I wanted to, I could use this DAX code to build into my reports. But I'm not a techy person. I got what I wanted. So there are twenty two open issues. Where is current? Yes. And the state is either new or active. That is my definition of an open issue. And if I ask it, give me a list of open issues without client name and personally identifiable information. Please, of course, Skype. Right? So it's actually pretty clever in that even I'm sharing my screen, you won't be able to to get actual client and personal naming data, from it. And I also add it into the base prompt. So while that's thinking, the the really clever part is you wouldn't expect most people to come in here into sort of the back end, fabric workspace. There you go. So you can you can see the issues that came in. It came in as per the prompt. There's an item, a work item. There's a title. Again, that title is obfuscated. There's a state, and there's an age of these issues. So sensitive data has been removed or obfuscated. And I can just ask what is the total age of the issues. And where this get where this gets clever is you can use this as an automated prompt so you can have this sent to you on a daily basis, let's say, at eight thirty, nine o'clock, so on and so forth. So you are reliant less on actual dashboards and reports, and this is where people people are saying that the end of reporting, end end of dashboard design might be coming because you have an agent that you can just chat with. So there you go. You can just look at two hundred and forty three days is the total number of our aged issues, which is actually correct as I checked. Okay. So this is a fabric agent, and let's move on to our proper agentic platforms. But as I do, let's let's again jump back to our PowerPoint. There. Where's it going? Here we go. So now we are going to look at Copilot Studio and Foundry right after each other. So we'll go through these two in in one bunch. So Copilot Studio is still still considered a low code platform, but it is Microsoft's low code agent builder platform. It is for business people and makers, so it can get quite tacky, complicated. It's not as simple as it seems. You can build pretty simple declarative agents, but you can also start building autonomous agents that are triggered. Let's say when an email arrives or when a new joiner is joined or when a new customer is created in your CRM. And, you've got connectivity, so you can connect it to your CRMs, to your ERP, to your Outlook, to your Teams. You can even connect it to external sources like a REST API or or other, non Microsoft systems. And you can use OpenAI and Cloud, and you can actually link in other agents as well. So you can build a fabric agents that I just showed you, and you can bring that in as an agent that is sort of like a sub agent of the previous agents. You can really build an a team of agents with Copilot Studio. And this does require license as well, and the editor is accessed through copilot studio dot Microsoft dot com. And you can publish that agent, through to office dot com or into, the Teams Copilot experience, so you will really have one place to access your agents. And let me just cover Foundry as well, because Copilot Studio and Foundry are so similar. So with Foundry, you are now entering into the low code, but also the pro code space. It is primarily designed for makers and IT professionals. You will see that in the back end, actually the same technology applies as Copilot Studio, but it's a you have a lot more options, lot more guardrails, and also more complexity, to be honest, to to be able to build agents. You can still build simple declarative agents and also autonomous agents. And you can you would obviously have the connectivity as you would expect some such a platform, but you can bring in more than eleven hundred models. Plus, can bring your own model that you can bring in. And what's quite telling is that this foundry is accessed through a I dot azure dot com, and it is primarily using an Azure subscription. So you can see that this is for those people who are familiar using Azure Azure services, which is Microsoft's cloud platform. Alright. So let me first show you Copilot Studio, and then we will jump into Foundry. And we'll stick with a similar use case that we just had, which is our issues. Here we go. On top of let me just give you a quick tour of Copilot Studio. What would you like to build? So you can actually build two things, an agents or workflow with and and we'll be focused on agents. But with workflows, you can start build repeatable automations that you can give to agents. And I've already created my ICE dev agent. So we've got so many agents here. There you go. So I can just very quickly search for it. And and, again, you'll see that the rough idea is this is the same. You've got instructions, knowledge, and here you can really give it tools. You could give it sub agents, and you can define different topics. So what's what's interesting, what's important about Copilot about Copilot Studio is that Copilot Studio existed before generative AI. It actually existed before Copilot. It was part of the power platform, and it was called Power Virtual Agents. And you could use power virtual agents to actually build non generative chatbots. And you could really define the topics that came up. So you could have a goodbye greeting start over and link these topics up in a way that gives you that pretty poor pre generative AI chatbot experience. But Microsoft decided to leave this in, and I think for a very, very good reason. But you can now really just give it some tools and it will become a fully agentic thing that you can just access different tools. So you can see it is now linked to our Azure DevOps instance, but it's also linked to my planner or it's linked to our Dynamics three six five knowledge article. And I can give it custom prompts as well that it can tap into. So the use case is the same. I I gave it access to get a list of open issues, and I can actually use this agent to create a new issue within Azure DevOps. So it's not only retrieval. It's not only getting information, but it can also create and push information. That's why it's becoming autonomous. Right? So imagine an agent that we receive an email from a client, and we can start using the email income coming in and generate an issue on the back of that back of that email automatically and respond to the to the client. Right? So that would be an automation that that we can use with this. But yet again, please list open issues without showing clients and personal information. And what will happen here in this little chatbot piece is you can see that it actually pulled in list of open issues, and it started listing out, that connector, that tool. So it reached out to that particular tool, and it's pulled, our list of open issues, for different areas. And that's sounds as simple as that. It didn't take too much time to to write up. And what I can do once I have a good agent is I can actually publish it out to Teams or Copilot pretty much the same way as I have done with the SharePoint agent. I can publish it into non Microsoft technologies like Facebook, like WhatsApp, like Slack. I can really start using this to build agents that my customers can use or interact with or not only internally for my own people. Right? So that's that's one piece. That's Copilot Studio. This is going to be the the new cool kid on the block that most builders, makers are are playing with. And Foundry up next is going to be, again, if I go to the main page of Foundry, you can see a I dot azure dot com. Well, you can actually can see, but it's a I dot azure dot com. And we have our own playground for Foundry. And we have a foundry endpoint that we can start building out proper pro code items. But I can actually start building agents here as well. And we have a, let's say, let's look at the first foundry agent, which is very similar here. By way of any other agent, you've got tools, you can add tools, you've got knowledge, you can add knowledge, you can connect it to foundry IQ, which will come in at a later session when we talk about talk about governance and how we can govern what AI have access to. You can bring in memories, and this is really important. This is one of the things that we cannot do yet in Copilot studio is you can assign guardrails within which the agent can operate. So it really becomes a proper enterprise session. We can check and monitor these agents. And as we deploy them and as we make any modifications, we want to make sure that these agents are evaluated and continuously operating to the level of extent. And therefore, we would have automatic automatic and human evals evaluations, and we can run these, manually or automatically. And we can stop the release of these agents into production if they are not hitting a certain number of level of responses. And what's really interesting about AI Foundry is the the demos sorry, the the workflows. You can actually track the different sort of like a left to right flow to build agents. And this is more familiar if you've seen a agent builder platform like an a ten or even OpenAI's agent builder. This is this gives you that similar experience of what's the workflow that it's going through and and what are what what are the elements that it should be pulling. And this can escalate to a human. This can post a channel, a messaging to a channel and really, really drive that. So this I think this experience builder experience will really come forward. And there's also a lot of cool things here with the models. You can access different AI services like a speech to text model. You can just upload your own file and and get it transcribed. So there's a reason why this is called, like, an AI playground. You can really, really start, testing it out, what you can do and and discover, what your options are within the ecosystem. And this is where where I mentioned that whereas with Copilot Copilot and SharePoint, you can only use OpenAI models. Here, you can actually bring in several models, GPT to chat. You can use Cloud, DeepSeek, even Grok. I'm pretty sure, with the exception of Gemini, you can actually use most large language model model providers agents. So this is really becomes a a proper data science level, data engineer, level tool. And these are all the tools available with the connectors, again, Microsoft and non Microsoft. So it's it's a it's a really, really rich experience and gives you an idea of, how complex, this world can be. And that actually takes us to our last piece of technology for today, which is going to be Microsoft SDK. And I won't act like I'm a developer, but I would like to show you what the Microsoft SDK looks like. So if all else fails and you wanna go proper full code and you want to want to build your own agent, what you can do here with Visual Studio Code is build your there you go. Build your own agent in Versus Code with the m three sixty five add on. And if I click create a new agent or build a decorative agent, that's quite simple. I can just click build decorative agent. I could add an action if I wanted to, but it's going to be a genuine decorative agent, no action. And, yeah, I can use the default folder, and that will be the webinar agents. And what I what will happen is it created the file, the YAML files, README, the whatnot, and it will actually start deploying my agent using the the toolkit. And that agent will become available within within the Office three sixty five experience, so it's deploying it as we speak, and I can get a local preview of that. And if I wanted to, I can build agents using Python or dot net, but we'll leave that to the professional quarters. We'll stick with Copilot Studio or AI Foundry as a sort of an agent builder person. So that was we'll stop that. That was the main the six different technologies that I wanted to show you. And let's bring this back up. So as a summary, so we saw that. So the last one, Visual Studio add on, you can expose deploy agents into Copilot or you can bring your own models. And this also requires m three sixty five Copilot Copilot links, Copilot licenses. And if you go through the whole thing, you can see that it's quite a confusing, complicated landscape. You've got you can build agents on office dot com, on SharePoint, on power bi dot com using copilot studio dot Microsoft dot com. You've got a I dot azure dot com, and you can build Microsoft based agents with Visual Studios, add on. And the licensing can be quite different is quite different, and the different capabilities and the audience of these technologies is is different as well. So we recommend everyone to start as simple as possible before before you start going into into the depths. And this is where I'd like to bring it up a level. We went really deep into the different tech stacks, and let's look at how this would actually have an impact on an organization's life. So Microsoft saying that we increasingly an organization will be human led, but AI operated, and we are becoming frontier firms as they call these these these new type of companies, where you have a human interacting, with Copilot, Microsoft's user interface for AI, which has, several layers beneath it. And those agents are sitting on top of your current line of business systems, your ERP, your CRM, whether they are Microsoft or not Microsoft, and we will be interacting with those directly less and less. We can still do do that and go into our Oracle, our Salesforce, our dynamic system, and put in a record there. Or I can just ask my agent, hey. Create a new customer for me. Create a new opportunity for me and and go from there. So this looks really good from a conceptual diagram. But what I'd like to look at is what this actually looks like from a more practical point of view. So in my view, every organization has a communication layer, a documentation layer, and an application layer. Right? So how people communicate in the digital world in the Microsoft ecosystem is through Outlook and Teams. They store their documents mainly on SharePoint, and they're using apps like Dynamics, or they have built their own apps on Power Platform, or they're using other non Microsoft related applications. And if you bring all these together, that consists of the data layer of an organization. And we really need to make sure that the data layer is accessible for AI. Otherwise, you won't be able to get proper value from AI. AI will just remain a personal efficiency element to be able to chat with your own with with the employee can chat with with AI without actually accessing the right context. So in twenty twenty six, what we're going to really need to focus on is give AI the right context, the right data and information to be able to act autonomously and execute end to end business processes with human intervention only on an exceptional basis. So once we have those foundations, right, the communication layer, the document and the app layer with a with a solid data foundation, all these AI technologies will start making sense and start creating proper business value. And we can really easily hone in on what it is that we are build what we are building and how we can start building those out. So what I'd like to recommend is a couple of design principles to the end. So AppVisual Labs and Microsoft have have these two main principles of bring work where the user is. So increasingly, that's becoming bring the agent where the user is. So you can deploy agents into teams, into office dot com, into SharePoint, even into Dynamics. So there's an agent everywhere. That's why there's Copilot everywhere. And also Microsoft has such a wealth of systems and toolkits that we've had this principle of every system should be used for its intended purpose. But now every agent should be used for its intended purpose because it's quite easy to fall into the trap of building one super agent that will know it all, but that's actually a very common anti pattern that is doomed to fail. And that's where hallucinations happen. So you want to build small agents that do the job well for that small job that you give it. And you already saw how easy it is to build agents using these six technologies, and you can really connect them up. And when they are connected, our user will not actually see what agent technologies are being in the background. What they will see is that they are interacting with different agents through m three sixty five Copilot, and they can really just have a seamless experience doing their their day to day work. And the most important message that I'd like to leave you today is if an agent doesn't create value, it doesn't it's not worth building it. So just make sure that before you start your AI initiatives, understand what sort of business value it will provide, and only after that, should you get started. So this is my main message, and you can download this, the session, the part, the presentation, from this QR code. And I'd like to invite you to next week's webinar, where we will be starting our department departmental AI readiness sessions. We are starting with marketing and then going into sales and customer service. So in the next four weeks, we will be covering CRM related agentic readiness elements. And next week, we'll talk about what it means to be AI ready in the marketing space and what are the most common use cases the industry sees in marketing and what Microsoft provides within Dynamics and how we can use AI in a more general way for marketing using Copilot and other agents, and what are the big value levers that we can see. So hope to see you there. Hope this session was was useful, and and happy agent building. All the best. Thank you. Bye bye.