Building Intelligent Agents with Microsoft Copilot Studio | #CopilotChronicles

Key Takeaways

[Music] [Music] Heat. Heat. [Music] Hey everyone, hope all is well with you all. Uh welcome to yet at yet another episode of Copilot Chronicles. I am Paru events and program manager for uh Microsoft Reactor in India. Uh but before we start our today's event, uh let's go through our code of conduct. We are all here to learn together, participate together. So please be respectful of other people views understanding the differences. Be kind and considerate in a way we all engage. We do encourage you all to participate. Please drop all your questions in the comment section and we will pick it up from there. Now I not like to delay further and would like to bring in Kamalashri uh the host for Copal Chronicles. Kamalashri is also a developer advocate from the team M365. Hey Kamill, how are you? I'm good part. How are you? I'm all well. Uh all great. Uh it looks like pre- monsoons already started in Bangalore and uh we have a very exciting session today. Uh like it's uh like evening snack with agents. Yeah, definitely definitely I really liked it. Oh. Oh yeah. But uh the session is another part of it. And in the past 3 days we were like totally occupied with the build. So I'm sure the audience must also know what really happened in the Microsoft what's the new release. So did you get time to watch it or do you have any thoughts on it glimpse? Oh with build we have planned couple of uh events that will be on reactor meetup. Audience please do be very keen on reactor meetup. We have sessions uh planned with the post build activities like what all things in the build announcements and all with the different teams. So all these sessions are lined up at reactor meetup. Please go visit reactor meetup Bengaluru and you'll find sessions lined there. Awesome. Awesome. Well, definitely I was totally amazed with couple of um uh topics that was covered and definitely events play a lot of important role because keeping the sessions and not talking about it it's not going to reach the audience. So just few things about it. Uh so welcome to co-pilot chronicles audience and uh looking forward for another episode but before that if you haven't watched build I think path will post a link where we had a keynote from Satya and uh he did cover a few um topics and what is upcoming at the Microsoft and uh just to give you a a nitty-gritty details about it I would say the GitHub copilot just got really smarter and uh there is the gro model from XAI. Have a look into it. And also if you're more into developer side, you would find more new tools on the edit on Windows and the NL web project for the natural language UX and definitely M365 copilot tuning. So I will share the links if you have missed or you want to rewatch the keynote from uh Satya and the new updates from the Microsoft build 2025 which happened at Seattle 3 days back and um yeah without further ado let me move on to the speaker of the day where we have uh Sarup who's going to be talking about building intelligent agents with Microsoft copilot studio. So I have received a lot of requests. So before that welcome sab um I have received a lot of requests saying kamal I don't know coding but I want to build an agent how do I do it can you come up with sessions can you tell us how do we start from scratch and um like a foundation principle what are the specific business scenarios without requiring any coding expertise or how do I automate flows I'm not into coding I'm into a sales but I have an idea I want to build an agent I have a requirement. So I think today all these questions will be answered from um creating intelligent agent using natural language or trigger phrases. What is from design to deployment? How do you do it using studio? Uh the path that is used for custom agents or how do you expand agent functionality that is also another important aspects. So all this will be covered by our very own speaker SU Agarwal who is a passionate technologist with over 17 years of experience almost touching two decades. So definitely Sab this has got to be one of the amazing interesting session who has an experience driving digital transform. He has a deep expertise in Gen AI co-pilot M365 and power platform. So Sab without further ado um the floor is all yours looking for your session and thank you so much for being a part of Copilot Chronicles. Thank you Kamill and thank you Pat for having me here and uh good evening everyone and welcome to Microsoft Reactor's Copilot Chronicle series. Uh so I'm uh very thrilled to be here today to talk about a truly transformative technology which is Microsoft Copilot Studio. So the world of AI is evolving at an incredible pace and Microsoft is at the forefront of this revolution. So today we are going to dive deep into how you can build and extend intelligent agents that can automate task, streamline processes and truly empower the workforce. So basically today's session I have divided into two parts. So in first part I will be discussing more around diff different aspects of copilot and copilot studio and in second part I will be showing you how you can build an autonomous agent using copilot studio. So therein I will be taking an example of a automated onboarding experience for the new employee who is going to join the organization. Okay. So let's start with the fundamental question. What exactly is copilot? In its essence, a copilot is an AI powered assistant designed to significantly enhance our productivity and creativity across various task. At its core, copilot leverages power uh powerful large language models like those developed by OpenAI and which are incredibly adept at understanding and generating humanlike text. So the functionality of copilot is quite broad. It can generate content whether it's drafting an email or writing code snippets or brainstorming ideas. So it can summarize vast amount of information helping us quickly grasp key points and it can answer complex questions in a conversational manner and critically it can automate repetitive task. So that's freeing up our time and maintenance energy and uh so now now that we have a foundational understanding of what this copilot is so let's look at the broader co-pilot ecosystem within Microsoft so in Microsoft has strategically integrated copilots across various layers to assist everyone in different ways. So uh basically to cater different needs of the different users within an organization. So first we have a firstparty copilots. So these are AI capabilities directly embedded into your familiar Microsoft applications be it like word, excel, powerpoint and think of them as intelligent assistance enhancing our productivity right where uh where we are working. So for instance, copilot can help uh help you in summarizing a long document in a word or generate creative slide ideas in PowerPoint all within the uh application or office application itself. Next we have Microsoft 365 c-ilot based co-pilots. So these goes a step further. So they leverage the vast ocean of your organization's data within M365. And this means that copilot is in teams can help you catch up on a miss meeting by summarizing key decisions from the transcript or copilot and outlook can help you draft a email referencing an in internal document. So basically they are grounded in the company's specific context and the knowledge sources knowledge base and making them which makes them incredibly powerful for enterprise productivity. And finally what we are focusing on today the custom co-pilots which are built with the Microsoft copilot studio. So this is where uh uh you even like even you don't know coding you uh you just know uh the business processes you can take control of this through the copilot studio. So basically copilot studio empowers you to create highly tailored AI agents which are designed for your unique business scenarios. So whether it's automating a complex HR process or providing specialized IT support or creating a unique customer service experience. So Copilot studio allows everyone to build agents that understand and act on your specific requirement. So it which basically integrate with your existing system. So these are uh truly your specialized AIES I would say. So design which are designed to solve the unique problems or the specific business scenarios for a particular organization. So let's start by understanding the big picture. We have seen AI evolve rapidly from simple tools that perform specific functions to something far more profound which are intelligent agents. Think of them not just as a tool but as a digital teammates capable of understanding complex request making decisions and even initiating actions autonomously. So recently at Microsoft build Satya spoke extensively about Microsoft's vision for an open agentic be a future where AI agents interact seamlessly perform task and drive significant value for individuals and organization. So this isn't just about large language models. It's about making those models actionable. The promise of agent is image. They can take on repetitive time- consuming task allowing our human team to focus on higher value creative and strategic work. It's about augmenting human capability. It's not meant to replace the human. And at the heart of this building these intelligent agent is Microsoft copilot studio. Basically this is a powerful low code platform that democratize AI development. What does it mean? It means that you don't need to be a data scientist or deep coder or procoder to build sophisticated AI agents. Copilot Studio allows both citizen developers, those with the business expertise and professional developers to create e and extend and manage AI agents. It integrates seamlessly with the entire Microsoft 360 Microsoft ecosystem which include all the M365 applications like team, shareepoint, one drive and power platform and it also enable integration with other thirdparty line of business applications. thus enabling the true end to end automation. So crucially, Copallet Studio isn't just for building simple chatbots anymore with recent advancements especially those announced at the build this build this year which have been just uh which is still going on it will conclude today. So it's becoming a central low code hub for building agents capable of executing complex workflows and intelligent decision making. So thus bridging the gap between low code and proc code development. Now to build effective agents in copilot studio we need to understand its core building blocks. So at the fundamental level everything revolves around topics. So think of topics as the agents which has distinct compet competencies like different departments in a company each handling specific type of request or information. Each topic is activated via trigger phrases and these are the natural language inputs that a user might type or say to the agent prompting it to initiate a specific conversation. So for an example, for an IT support topic, trigger phrases might include my laptop is broken or I need a mouse or how do I reset my password. So once a topic is triggered, the agent follows a conversation path which is a series of nodes. So these nodes represent the flow of the conversation allowing the agent to ask questions, provide information, collect data or importantly perform actions. So actions are where this agent truly becomes powerful. These are the steps an agent takes to do something whether it's calling up power automate flow or to update a database or making an API call to an external system or it can be a sending a notification somewhere and this is how our agent moves beyond just talking and it starts automating task. So copilot studio offers both system topics which are pre-built for conversational events like greetings or escalations and custom topics which you can create from scratch to address your unique business needs. And a significant enhancement especially highlighted at build 2025 is generative orchestration. This means that isn't isn't just looking for exact keyword matches. It uses advanced AI to understand the intent behind a user's query and intelligently determine which topic is most relevant even if the exact trigger phrases isn't used uh to define this. Now it's crucial to understand a key distinction which is the difference between a conversational agent often referred to as a chatbot and an autonomous agent which is what we are truly enabling with the copilot studio. So put it uh if I put it simply conversational agents are like using a GPS. You tell the GPS where you want to go and it provides directions step by step. But you as a driver are still in control, constantly interacting with it, interpreting the instructions and making the turns. If you miss a turn, you need to tell it or it will ask you to correct. It's a reactive in kind of nature. So basically relying on your continuous input to guide the generary to guide the generary. And if I uh compare it with the autonomous agents which are like using a self-driving car once you set the destination that car takes over. It makes independent decisions on how to navigate, accelerate, brakes or change lanes all without your direct input unless there an unforeseen anomaly in the system. It proactively works to achieve the goal you have given it. And the core differentiator is that an autonomous agent has the capability to take action and make decisions indep independently to achieve a defined goal much like a self-driving car. While a conversational agent might tell you how to request a new laptop or like an autonomous agent can actually initiate the process of ordering and assigning that laptop for you which is what we will be demonstrating today in our employee onboarding scenario. So building on our distinction between conversational and autonomous agent, let's now dive deeper into what truly defines an autonomous agent and how they handle variability and complexity at scale. So as you can see, autonomous agents are designed to be far more sophisticated than traditional BS. First, they independently begin work based on autonomous triggers. This means they don't always need a human to initiate every action. They can be triggered by events, schedules, data changes or even by a specific business condition being met. So this proactive capability is the foundational. Secondly, they are built to automate long running processes. We are not uh talking about single question and answers here. We talking about multi-step workflows that might span hours or even days such as our employee onboarding scenarios. Crucially, autonomous agent dynamically reason over their capabilities. This means they can choose the best course of action from their available tools and functions adapting their plan based on the situation. So they aren't just following the rigid script as we do in the uh uh like uh power automate based workflows. So they are also designed to learn and improve over time often through human feedback loops and by processing more data which is becoming more efficient and effective with each interaction. An extremely important aspect for the trust and safety is that they follow human guard rails and know when to ask for help. Autonomous doesn't mean uncontrol. So they operate within the predefined boundaries and are programmed to escalate to a human when faced with ambiguity, complex decisions or situation which are outside of their scope. And finally, a truly powerful capability especially highlighted during this build 2025 is that their ability to orchestrate other agents. Imagine a team of AI agents each specialized in different area delegating tasks to each other to complete a larger objective. So for instance an HR onboarding agent asking an IT agent to provision a laptop and a training agent to enroll a new hire in the courses. So this is what the future of collaborative AI looks like. So all these capabilities are becoming increasingly accessible through Microsoft copilot studio which is leveraging leveraging generative AI and sophisticated orchestration features. So we have talked about what an autonomous agent is and you might be thinking like how is this different from power automate flow which also automate things right. So a cloud flow is static in nature. This means every single step in that flow is scripted in advance. So it's a predefined path and if your process changes or if you encounter a new scenario, you typically have to go back and manually update or even rewrite that flow. In contrast, an autonomous agent is self-directed. While it utilizes underlying actions, the agent itself can adapt and orchestrate those actions without requiring you to rewrite the entire workflow. It interpret intents and dynamically choose the best path making it much more flexible as your business needs evolve and it figures thing out on its own. Next, uh uh if I take a look at the deterministic versus adaptive, a cloud flow is a deterministic. It's built for a well- definfined process with predictable steps. So if there is a trigger A, if if the trigger A happens, action B will always follow then action action C. It's excellent for consistent repetitive task where the outcome is always certain given the in given the predefined input. An autonomous agent however is adaptive. So this makes it ideal when the workflow itself can change dynamically. So this could be due to unclear user inputs, unexpected external interactions or situations which require real-time problem solving. The agent can reason throughout these ambiguitities and adjust it approach to still achieve the goal. So to summarize, cloud flows are fantastic for automating predictable, static and deterministic task. They are the brawn and executing precisely. Autonomous agent powered by copilot studio bring the intelligence to handle dynamic self-directed and adaptive scenarios. They are kind of a brain that int intelligently orchestrate the drawn of the cloud flows deciding what action to take and when. So in in Microsoft build uh which which has just happened and it's brought some truly groundbreaking advancement for the autonomous agent in copilot studio. So this isn't just iterative improvement I would say it's a significant leap forward. So massive announcement is multi- aent orchestration. Imagine a team of AI agent each agent specialized in different area working together to solve a complex problem. This capability currently in private preview will be transformative for scenarios like complex onboardings where multiple systems are involved. Another incredible capability is computer use in agent. So this allows your copilot studio agents to interact with desktop applications and websites just like a person would clicking buttons, navigating menus and typing into fields. So this opens up an whole new world of automation possibilities for tasks that previously required human intervention. And for those who need even more control, Copilot Studio now supports bring your own model from Azure AI Foundry, giving you access to over 11,000 models and the ability to fine-tune them with your specific enterprise data for high domain specific responses. And the model context pro protocol is Microsoft effort to create a universal standard for how agent interact with external data and tools making integrations even more seamless and powerful. And to tie it all together, the new agent feed in Power Apps provide a central hub to monitor your agent teams, see task in progress, and quickly identify any exceptions where an agent might need human assistance. It's about providing oversight and control over your digital workforce. So uh so far I have covered the foundational principle of Microsoft copilot studio uh and uh how it enables powerful agent creation and explored the exciting advanced feature especially those which are hot of the press from Microsoft build 2025 which enables truly autonomous agents. So the key takeaway is that copilot studio empowers you to build intelligent agents that can understand act and automate complex processes across your organizations. It's about transforming how work get done by providing that brain to orchestrate actions in real world system. So, so now theory is good but seeing is believing I think and in the second part of my session I'm going to walk you through a practical demonstration showing how you can build an autonomous agent using copilot studio. So for this particular demo I would be taking an uh uh employee onboarding uh experience uh or building an agent on employee onboarding and uh so for this onboarding uh basically I have uh what we have done is uh we have uh created a list of trainings uh which you can see on the screen. So basically whenever any new employee would join in the organization based on its department it will go through different trainings which are uh and there are like specific duration or the uh the frequency of the training is deterine uh defined and basically this is kind of a one knowledge source which we will be providing to our uh agent. Another uh another knowledge source that we will be providing to our agent is the uh devices that will be allocated to the uh user based on the department. So again we have taken uh these uh these five or six different uh devices which will be allocated to the user based on its department. Okay. And uh I have also set up a list in the shareepoint which basically uh which basically uh act as a uh backend database to store the device allocation request and uh this autonomous agent will be designed to trigger whenever we add a new item in the list called uh new starter training. Okay. So now uh when I go to this copilotstudio.microsoft.com it will open this homepage and here I can see uh uh like there are different uh predefined agent templates that are there which you can choose uh to build on top of these and then there are different learning resources. So if you need any help you can go through these learning resources. Okay. Now let me click on the agent create button to create an agent. So again as I mentioned earlier like you can you can search or you can build on top of existing agent templates that are defined here or you can go to create a new agent. Now I clicked on new agent and uh here again you will see you see this there is a copilot inside this copilot studio. So here you can define what you want to achieve through your agent and based on this description it will create a template of the agent which you can further modify according to your requirement. However for this particular demo I am going to skip this. I will directly go to the configuration part where I will going to manually configure this agent. Now when you manually create this agent you see these fields like you need to give the proper name the description and the instructions. So I will come to each of these points one by one. So for example if I take uh uh if I take my agent and uh uh so my agent is about creating a new uh on employee onboarding u agent. So let me just give it that name. So one important thing important thing is uh uh that you need to make sure that uh you are giving the proper uh proper descriptions or proper names to your agent because it uses those details uh while performing different actions. Now again I will be giving some meaningful descriptions. So as a new employee starts at this organization they need to complete onboarding process and uh this involves uh so this involves securing a device and shoulduling the required training which is delivered over Microsoft teams. Now coming to the instruction part. Uh so instruction is very important here and basically while giving the instruction in so basically instruction determines the overall uh overall flow of the agent that you are going to design or going to develop. So basically if uh from the instruction part it will be basically divided into five different uh areas. So first one is the rule. So basically you need to define what is the purpose of agent. Then second part is the scope where you define like what what this particular agent would do and what it should not do. Third you will define the context like you define what does the agent need to know like from where it needs to fetch the information and all those things. Fourth point is around you need to define the tone of the agent and finally uh you need to define the error handling wherein you describe like uh what this agent should do in case of any error or any issues. So let me uh quickly pick one predefined instructions that I have created for this particular training. So if you see like I have given this instruction like you are you are to organize new employee training. This varies by department. Again as I mentioned like uh for different department there is a different set of uh trainings. So based on the users department you will select training from a list of available training sessions and then create teams meeting for the new employee. So you see I am giving the instructions in the natural language and uh through this instruction it is going to pick what step it needs to perform during the course of action. So you must also uh here again you must also find a suitable device for the new employee again based on the department. So for example IT team generally require high-end devices whereas uh marketing uh team or sales team require uh not so high-end devices. So again this I have given the instructions when a new employee is added use the training knowledge to build the list of required training for that person. Uh okay so when a new okay when a new employee added use a training knowledge create teams meeting for each training session using the schedule meeting action. Again I will be defining the schedule meeting action in this uh in this agent and ensure that it the team's meeting is created on the corresponding week day. So as I've shown you earlier that I have predefined knowledge uh where we have uh mentioned like on what day of the week the training is scheduled for a specific department and then using the approved device knowledge select a device appropriate for employee based on their department. So for example, IT employee require higher specification devices. Sale employee require lower specification divid devices and you can look for additional specs and information about the device and confirm it is in stock from the business device data source. Again for this business device data source I'm going to point to the Microsoft store site which basically so this agent will basically look into this Microsoft store and check for the availability of that approved device and if it's not in stock choose another device. Next is create a new device order request for that selected device in step three using the create hardware request item action. Again this action I'm going to create and answer grouping data which has a tag authoritative with a header line containing this following followed by the based on official sources. So data with no tag must be grouped after adding a new line at the. So these are the different instructions. Again there are like for conversational agents we can provide the starter prompts which basically give clue to the user who is interacting with those conversational prompts like what they want what they can start from. So uh for this particular we are going to create autonomous agent. So I will leave it blank and knowledge section I will again come back and uh add the different knowledge sources in this uh particular agent. Okay. So let's click on create. Just give it a minute and uh it will set up and let's Okay. Okay, it's loading. Okay. So it has created that um the template for this autonomous agent and then you now see there are like additional options available where knowledge was there previously but there are additional options like I can define actions. I can define trigger points. The starter points was already there. So, and for orchestration, we need to make sure that this generative AI based orchestration is enabled so that it automatically uh pick the best possible action uh based on the instructions that are given into this section. Okay. And in the right side of this uh screen you will see that you can you have this interface where you can test your agent at a at any given point of time. Again now I will go back to the knowledge section and uh let's define let's add some knowledge here. Okay. So first I'm going to add this business devices. Okay. So click on this. Now you see uh when we when I clicked on this add knowledge. So there are different uh knowledge sources which are available. So you can ref you can point it to public website. You can point to any files or libraries or the shareepoint sites or you can also refer to the Azure AI search. You can refer to data stored in the data or dynamics 365. And if I click on advanc there are some other options or connectors which are available. And in addition to that you can define the knowledge source in form of a file. So you can upload the file here and uh which basically act as a knowledge source for your agent. Okay. So let me go back to this. Let me select the public website. Okay. So again for public website uh the name of the knowledge source I have given in the struction was business devices. So, okay. So, okay. I need to give the URL first. No worries. Okay, I can give click on add. So here again you need to give the proper name and description because based on that it will pick uh pick that detail. So I've given the name as business devices. Okay. Second, you need to make sure the description is correct. So basically this knowledge, this knowledge searches to see if Microsoft devices are in stock. Okay. and uh can click on add. Now I add my second knowledge source which is the file that I just show you. Okay. So I click on select to browse. Okay. So, where are my files? Let me see. Okay. Where is this? Okay. So, let's give it a name. Again, I need to give a proper description. So I give it something like this. This knowledge source contain information about required ratings for new hires by department. Okay, I clicked on add. So you will see like uh it is uh taking some time to load and uh now after uploading it is showing in progress. So we are basically in the back end it is creating the semantic index for this knowledge source. Uh in the meanwhile we can upload our another knowledge source which was the list of devices. Okay. So let me bring it here. Okay. Again, we need to give the proper name and description. So, let's copy this quickly. Okay. Click on add. All right. Give it a minute. Okay. Okay, again it is in process. Okay, while uh so we can move on to creating an action. So while it creates the index in the back end. So again uh I will create an action. So here you will see uh again as as I mentioned earlier like there are different actions. So basically these are all actions which are available in power platform and power platform currently supports around 1,300 plus actions. So you can search for a specific action which you want to uh uh execute through this copilot studio. Again there are like different other options like you can connect uh you can use the connector you can use custom connector or you can call a cloudflow or power automate based flow or you can you can use the uh AI bot skill set or so for this particular uh I'm going to call an action for creating a list item in shareepoint. point create item in SharePoint. Okay. So again here you can define the permission you can give you can add the user through which you want to execute this action. So again I need to give the name of this and the description. So basically the purpose of this is to create a hardware request. Again I'm going to get the right action uh right description. So basically this is going to create a new hardware request in the SharePoint hardware request list. Okay. So in the authentication part you will see there are like two options whether it can run through the user authentication. However this is an autonomous agent so I will use uh the authentication of the copilot author which is this is basically me. So I click on add action and there are like options of input and output that will go after this is created. Okay. So now if you you can edit this action by create clicking on this action. So again I need to give the name of the action which is which I can copy from my display name. So here it is. Okay. Description I have already given copilot authentication. Okay. Connection. Now I come to the So before that I can click on save. So again there is an option you can do the go to the code editor but that is not required at this stage. So if I go to inputs uh so I need to provide the site address which is basically set as value and I can select the onboarding user one. So I created this site for this. So in this I can select I need to select the list name. So again I select status value. So you see that when you're changing input it asks for the confirmation. Okay. And uh I can select the hardware request. Okay. Click on save. Now this hardware request if I go to the hardware request list you see there are like different uh columns uh which are there. So I need to define the value uh define what should be the values of those columns. So I need to uh define that in this particular action. So I just click on add and I first I will give the title which was the first column in this list. If you see this is the manufacturer which I renamed uh later to manufacturer earlier it was title. So I need to set like what will be the title. Again I need to give the description on what what value should this particular column have. So okay here it is. Then you fill in the manufacturer name of the device from the approved device knowledge source that is one. So there is more. Okay. Model again I need to define the model uh what should be the values in the model column. So here you see here I am selecting this dynamically fill with the best option. So basically we are giving the this in the description we are giving the instruction like what could be the value and the auto this autonomous agent automatically determine what will be the right value. Okay. So next is uh recipient. So I need to again give this. So here you see I have uh in this I am giving this field with the name of the user the devices from when an item is created ticket. So there is again a trigger that I will be creating. So basically in that trigger whatever the name of the user it is pick it will pick from there. Similarly I'm going to create two more uh which is cost. Okay. Again there is a I need to provide the description. So again fill the cost with the device from the approved device knowledge source in the USD format. Okay. Now the next one is the description. [Music] Okay. So for the description I'm going to take it just defy details of the device that so basically hard what is the size of RAM hard drive and all those thing and finally the last one is the asset tag. So in the asset tag I am going to ask this to generate this asset tag randomly uh with some uh with the asset as a prefix and some six-digit uh numbers at the end of this. So these are the different columns that uh that basically this action will autoop populate and uh it will create this new item in the hardware list in the shareepoint now. Okay. So now we come back to this. Okay. Go back. Now I have to uh uh like set up a meeting and uh uh uh like we will send the notification to in the teams to the user. So for that I'm going to create a uh power automate flow. So again if I click on add an action and here you will see this flow option. Okay. So basically in interest of time I am not going to create this. I will just show you how this looks like. So basically I will uh I will use an existing flow which which I created earlier. So here you will see this schedule meetings flow. So basically okay I click on add action. So basically you once you click on that uh add flow it will open this make.p power automate.com and it uh let user create a flow. So let me quickly show you how that flow looked like. Or I can open it directly from here as well. So if I go into this, so you see this link. So once you click on it, it will open that flow. Okay, let's let me leave it. Let's quickly Okay. So again for this uh schedule meeting I'm just going to describe the different parameters. Just click what are the different inputs. So again there are like uh it takes like input uh as a email address, start time, end time and the training title. So let me quickly add those details here. So if you see the start time I have given this uh like some formatting instructions like how the start time will be formatted. Similarly uh for the end time in the uh in the training list I have given the duration. So how it calculates the end time based on that duration. So this is given here and finally the what will be the training title. Okay. Click on save. Okay. So, and now finally we need to uh add the trigger. So, uh just in interest of time I will uh just open one of the flow that I created. Uh okay. So, I created this agent and you see uh let me quickly open it. So you see like the different knowledge sources that we defined the actions that we have created and the trigger like when an item is created in this list it will automatically trigger this workflow. So uh to test this I just uh I just need to uh go to this uh you can see this there's this test trigger. So I click on this test trigger. So basically what it does is like it similar to power automate it opens the uh like last executed run. So you can pick up from there there. So you can just click on start testing. So basically it picks that particular data and uh initiate the trigger and it opens this activity map. So here you see in the right side of this it shows the step-by-step execution. In the left side you will see like how what it is doing. So basically it start it uh with the knowledge source. It explored the knowledge source and then it figured out like what it should do like it should meetings uh different meetings that are defined for that particular uh employee who has joined uh the organization. And uh then you will see like uh once the schedule meeting is complete it again check the knowledge source and uh it is again checking that knowledge source what it should perform next. So again it has determined that it has to create a hardware request item. Okay. Okay. And in the right side you will see like uh what are the different uh knowledge source it is processing and uh so basically for this particular employed market as a marketing as a department. So it uh determined the different trainings and the device and the set tag and all those thing which is generated. You see the session is marked as complete. And if I go to the team's chat, I can see this uh the it has pinged us sent that message communication around the trainings which are scheduled for this particular user. Okay. So yeah any questions guys? Yeah sab. So uh we do have questions. So are we done with the demo or do you have few? Okay great. You're done with the grammo. Okay great. Thanks Sab. Um that was definitely an interesting session to understand or the flow part even I had few questions on it. Now just a very general question. Now if I have to edit a particular part of a flow say for example I need to add something in the between where I got a new requirement from a client is that possible? Right. So as I mentioned like uh first part is like autonomous agent are very flexible in nature. you just uh if there is a specific requirement uh first you can simply go and update the instructions. However, there is any specific things or specific activities need to be established. So for example in this particular demo I scheduled a meeting in addition to that if you want to send the email also uh to to his N plus one. So you can add an action. So you not to modify the entire flow or anything. you just go and click this add action and just add an action and it will accommodate that change. Okay. So in the flow uh I mean the previous um screen that you had shown where you had the complete flow. Yeah. The activity that was the activity map. So basically it shows uh the like while executing this particular autonomous agent that particular stance of flow what all different activities has been performed. So through this uh it you can determine like if in case of any issues or errors you can see like what is the overall execution steps? Mhm. Is there a possibility to edit it here directly is that UI provision available basically action item or something like that? Yeah. So yeah so basically this is particularly the blueprint of what has already been executed for that particular instance. So if you want to modify the flow, you can go and edit the flow for the future runs from here. From the first screen, you can add actions, you can add triggers, you can add more knowledge sources or you can simply update the instructions. Perfect. Perfect. Uh we do have couple of more questions. So we have a question from the audience Deepak who says I need to make a bot that helps in filling the data in a predefined Excel file. Is this possible? Yeah. So again uh you can you can achieve that. So for that what you need to do is like as I use power automate based flow uh for some activities or some part of it you can use power automate base first uh for first part of it where you define or you determine the content that needs to be populated you can use uh uh the uh like a standard instructions or other actions and for populating that data into the Microsoft Excel you can use power automate flow and you and link it to this actions part here basically which will create the Excel sheet. Okay. And uh definitely you have shown us how to build an agent. So we have a question from the audience Rahul who says how do we publish an agent um in Copallet Studio right? So again once you publish it there are different channels that are available for your uh for your agent. So basically you can publish it to say telefony you can publish it to teams or m365 copilot you can publish it to be app or these are the different channels that are available and autonomous agents are kind of a backend agent which basically get hosted in the copilot uh Microsoft copilot uh stance which does not require any front- end publishing platform. So for conversational agent you can publish it in any of these channels that are available. Okay I think Rahul uh Sab has answered your question and you can see the screen for publishing it. So we will take another last question which is from Brutin J. He asked how can we ask the end user to provide the authentication and process the flow in teams. Yeah. So uh if you remember like during this uh when I was uh creating an action there was an option where I can choose from the two different option one is the uh copilot author uh authentication or the end user authentication. So once you publish it to any of the channels say for example Microsoft teams it will automatically uh and if the user is accessing that agent from the teams uh channel then it will automatically takes the authentication details uh of the logged in user in the teams. Okay. Got it. All right. I think we do uh have few more questions but I will leave the social handle of SAU where if you would love to know more about it or you want to know another session you want to come up with another session if you are interested let us know what topic you want Sora to cover we will definitely vouch in for another session from Sorup um and thank you Sorup for such an insightful session on copilot I mean building agents on copilot studio and emphasizing more on the autonomous part. Uh thank you so much for being a part of it and uh stay tuned for another episode of Copilot Chronicles next Thursday. We will meet you and uh please do watch the build keynote and other various uh sessions if you have missed out and part thank you so much for being a great host. Thanks Auro. Thanks all. Thank you. Thank you everyone. everyone. Thanks