Make your Own Agents in Copilot | Complete Tutorial

Agents expand the capabilities of AI tools,  allowing you, your partners, and clients to work   with customized AI assistants for specific jobs  or purposes. Agents can accomplish tasks within   preset specifications. They can be set up to  work independently or can create workflows across   multiple applications. Hi, I'm Nick. Welcome to  my full-length training course on using agents   in Microsoft 365 Copilot. This is a fairly long  video, but it is organized into 11 sections. You   can use the YouTube chapter markers to jump  directly to each one. In this first section,   we'll clarify what agents are and give you an idea  of what you can expect in the rest of this course.    Now, there are many different AI systems and chat  assistants, including Gemini, Claude, Copilot,   and many others. And most of those systems  have their own tools for working with agents.    This course will focus on tools from Microsoft's  Copilot brand. Naturally, this course will assume   that you have some familiarity with Copilot in  general. And of course, anybody can use Copilot   for free at copilot. microsoft. com. However,  Copilot agents are not available in the free   system and they're pretty limited in subscriptions  for individual users. The copilot agents and agent   building tools that we'll use in this course  do require a Microsoft 365 subscription from   a company or similar organization, also known as  Microsoft 365 business and enterprise accounts.    Now, there are three main goals in this course.   First, I want to help you understand what an   agent is and how agents can help you focus  or streamline your work in powerful ways.    Second, we're going to see how to work with  existing agents that have already been set up   and are available for you to use in Microsoft  Copilot and Microsoft Teams. Some agents are   built by Microsoft. Some are built by thirdparty  developers to support their apps and services,   but there are many agents that you can work  with today. And third, we're going to give   you the tools you need to build your own agents.   I believe that anybody who can use an agent in   Copilot can also build an agent in C-pilot. You  just need to be willing to do a little more work   and planning. However, it's very important  to recognize that there are different types   of agents. Building agents has a lot in common  with developing applications. No single training   course can teach you all the ways to make agents.   There are tools and systems offered by Microsoft   and others that require knowledge of application  coding and programming APIs. But in this course,   when we build agents, we will be using C-Pilot  Studio and the Copilot Agent Builder. Two options   offered by Microsoft which integrate into known  C-pilot systems, allowing you to build powerful   agents fairly quickly without the need for  coding or application development skills.    With that in mind, the agents we build here will  not be able to work independently and will not   integrate with thirdparty systems. Those are more  advanced features that are a little outside of   this course. So, with the stage set, the rest of  this course will be much more hands-on. If you're

taking your first steps into the world of C-pilot  agents, I want to start by helping you understand   why agents are so valuable, how they can save you  tons of time and get you much more focused results   from Copilot. As we do this, we're going to  actually go hands-on and use agents that Microsoft   offers inside of C-Pilot and in Microsoft Teams  at no additional cost. So, let's get to work in   Copilot. Since I'm using a Microsoft 365 business  account, I'll go to the Microsoft 365 website and   sign in. And I'll make sure to click new chat on  the left to work with Copilot. Now, let's say I'm   considering a career change and I want to work in  the forestry service. I could ask Copilot a simple   question about that. I'll tell C-Pilot to help me  find and apply for jobs in the forestry service.    After a moment, Copilot shows me websites and  gives me some examples of jobs that are open now,   but it's not really helping me navigate a career  change. C-Pilot can be much more helpful, but I   have to write a better prompt. You could apply  prompt engineering techniques, or you can just be   more detailed and clear about what you want. With  that in mind, I could tell co-pilot, "You are my   career adviser with over 20 years of experience.   Help me identify the challenges and opportunities   in making a career shift. Ask me questions that  I have not considered to help me identify what my   goals should be. After giving co-pilot a specific  role, instructions, and context, then I'll finish   the prompt with the same request. Help me find  and apply for jobs in the forestry service. And   now I'm not just getting a quick answer. Co-pilot  is engaging with me, helping me understand the   challenges of a career change and asking questions  to help me identify what I need. This may be the   start of a much more powerful conversation with  Copilot. But consider the work it took to get   here. This upfront work is one of the main things  that an agent can solve for. So in the Copilot   chat interface, there's a category for agents on  the left. So I want to click all agents. And these   are pre-made agents available for you to use in  Copilot. We'll start with the category of agents   built by Microsoft. And I'll click the button  that says see more. These are agents that are   already designed around specialized parameters or  specific jobs. For my career change, I might want   to try this agent called career coach. This agent  is designed to guide you through your professional   journey, explore career opportunities, and  give actionable advice. That sounds like   what we need. There's a button to add that agent,  but you should be aware not all of these agents   can be used by everybody. If you do not have  the full Microsoft 365 Copilot subscription,   then you may see this add button on some agents,  but others may have a button that prompts you to   upgrade your account. So, I'll add this agent.   And now the Career Coach agent is listed on the   left under the agent section. I can select new  chat up at the top when I want to work with the   standard co-pilot chat or I can select the career  coach when I want to work with that agent. And we   can assume that the agent has all the context  it needs to help with a career question. So I   can just ask the main question, help me find  and apply for jobs in the forestry service.    The agent gets to work as my career coach. And it  starts by asking me some leading questions to help   identify my goals. Once I answer those, it's going  to do some very specialized research for me. I can   expect this to provide great information, but  also engage me in a conversation to help me   identify what I need. Now, once you've added an  agent in Copilot, there are a few other ways to   work with them. If you click new chat on the left  to use the main Copilot chat, in the chat field,   you can type the add symbol followed by the name  of an agent. You can only choose an agent that   you've already added. And in fact, let's say you  don't choose an agent here. Maybe you just start   in the normal co-pilot chat and ask a question.   And if you don't get the information you want,   you can invoke an agent in a follow-up question  in the same chat. So, without starting a new chat,   I'll type the ad symbol, start typing in career  coach. It finds it there, and I can just click on   it. Then, I can ask my follow-up question. And  now the agent is working with me in this chat.    And if you ever want to remove an agent that  you've added in Copilot, you can just find it in   the sidebar on the left and point at it. A button  with three dots will appear. That will open a menu   and you've got the option there to uninstall it.   And if you work with Microsoft Teams, you can use   agents there as well. In Microsoft Teams, you can  click the apps button in the sidebar on the left   or click the button with three dots and then  choose get more apps. These apps are add-ons   that can give you more functionality inside of  Teams. And there's a category here for agents.    I'll choose one here. Let's try the idea coach.   I'll click add on this agent. And once it's added,   I'll click open with co-pilot. And that takes me  to the co-pilot chat here inside of teams. I can   see that co-pilot is selected on the left.   And the idea coach agent is selected. Now,   some agents will work differently in teams. Some  will create a dedicated chat in the chat section   and some will appear as an additional button  in the sidebar on the left. So, just keep an   eye out for different ways to interact with agents  in Teams. Now, one last thing. I'll go back to the   Copilot website and I'll select the agent that we  set up. And when you add a new agent that you've   never worked with before, I think it can be very  helpful to ask the agent what its parameters are.    Here's a prompt that I recommend. describe your  parameters including capabilities and functions,   your rules and constraints and your technical  configuration and the agent tells me about the   upfront parameters that it uses. We can also see  the functions it uses. One important thing to see   here is that this agent can use enterprise  search so it can access my data in my email,   messages, shared documents, and more from my  Microsoft 365 organization. We often refer   to this as the work mode in Copilot and it's a  feature that is only available for users with   the full Microsoft 365 copilot license. This is  something that we covered in my other course on   Microsoft 365 Copilot for organizations. Now, you  do not have to ask an agent about its parameters,   but this can really help you understand whether an  agent will do what you need. So, we've seen what   the career coach agent can do, but that's just one  agent that works to help with a specific type of   goal. And I also want to stress that this is just  the beginning of what agents can do. But you can   start to see the value of a co-pilot assistant  that is pre-programmed with specific roles,   instructions, and context. And if you stick with  this course, one of the things you'll learn is how   to build your own custom agent like this, but  with specific parameters that you define. And   once that agent is built, you can share with your  teammates, which can save a lot of time and effort   in your organization. I think it's a good idea to  get familiar with two very powerful agents that

Microsoft has built in and spotlighted in the main  co-pilot chat. We're talking about the researcher   agent and the analyst agent. These agents are  only available in the full Microsoft 365 co-pilot   license, at least at the time of this recording.   In the main Copilot app or website, we'll go to   the agents section on the left in the sidebar, and  I'll start by selecting the researcher agent. Now,   of course, most of the questions I ask Copilot can  be considered to be research. However, most of the   time, I'm also looking for a quick answer. The  researcher agent does not deal in quick answers.    It is designed to take extra time to compare  resources, clarify requirements in context,   and take precautions to avoid inaccuracies.   Now, when I worked with a new agent,   I like to ask what its parameters are. So, I'll  ask it to describe your parameters, including   your capabilities and functions, your rules and  constraints, and your technical configuration. And in the response, we see that this  agent does deep research comparing many   sources and it can also use information from  my organization. But there are some trade-offs   here. In rules and constraints, we can see  that the researcher cannot generate pictures   or images. All very important things to know  as I ask my starting question. And to start,   I want to refer to my employee manual. You could  drag a file to the prompt field or if you have   the full Microsoft 365 copilot subscription,  you can type the forward slash then find a   file from your one drive or shareepoint. I'll  find my employee manual. Then I'll tell it to   consider standard policies that should be  included in a corporate employee manual.    Tell me if there are any subjects that may be  outdated or be missing. The researcher will almost always ask one round  of follow-up questions to clarify your needs. So,   I'll answer these questions. Once I answer those  questions, the researcher will take a very long   time to research and come back with a response.   When you get the response, all of the normal rules   of Copilot apply. You can ask follow-up questions.   You can and should look at citations to check the   sources of where it found this information. But  when you want detailed information where accuracy   is vitally important, the researcher agent is  preconfigured for those needs. Next, let's look   at the analyst agent. I'll select that from the  sidebar on the left, and I'll ask my standard   question about the agents parameters. Basically,  the analyst is built around getting insights from   actual data. To give the analyst a try, I'll refer  to my company's product inventory spreadsheet.    Then I'll write my prompt. Several of our products  are backordered. Based on the prices and values of   the items in the inventory and information about  market trends, what insights can you give me to   determine whether we should restart production on  the backordered items? The analyst agent will also   take a long time to generate responses. And  let's be clear, you certainly do not need to   use the analyst agent for a question like this.   However, the analyst agent is already designed   with parameters that make it very effective  for questions like this, saving you the time   and effort of writing those parameters into a  normal co-pilot prompt. The results include some   actionable strategies for helping me decide  when we should restart production. It also   identifies some missing data. If I can provide  that, it may be able to provide more options.    And it finishes by offering to make a chart  or simulate scenarios. The researcher and the   analyst are two very powerful agents and they're  already set up for you as long as you have the   full Microsoft 365 C-pilot license. Now, I'm not  sure about this, but my guess is that Microsoft   identified some of the more powerful tasks that  people expect from Copilot and instead of making   the main Copilot chat tool more powerful and  potentially slower, they decided to offer   these preconfigured agents to help with those  types of tasks. and they're also very useful to   help us understand how valuable agents can be in  general. Within the Copilot app or website, there

are several pre-built agents that you can use  to expand Copilot's capabilities. Some of those   pre-built agents are made by Microsoft, but many  others are provided by third-party app developers.    These give you the ability to interact with other  applications within a co-pilot chat interface,   leveraging Copilot's AI features to supercharge  those apps and services. From the main C-Pilot   website, look for the agents section on the  left and click all agents. I'll scroll down   to the section labeled more agents and I'll click  see more to see the full collection of thirdparty   agents. There are many different apps and services  that can be integrated into Copilot using these   agents, but remember they won't all work exactly  the same, but we can choose one to get an idea   of how they work in general. I use Canva a lot,  so I'll choose that one. You may want to choose   one that you work with regularly, but keep an  eye out for details that may work differently.    And if you're not familiar with it, Canva is an  app for designing graphics, for social media,   marketing promotions, YouTube thumbnails, all  sorts of things. So, choose an agent, then click   the add button to install it. And now I see the  Canva agent is listed on the left and it's already   selected. And I always like to ask each new agent  what they can do. So, I'll ask it to describe   your parameters, including your capabilities  and functions, your rules and constraints,   and your technical configuration. I'll send that  prompt. And we can see here that this agent can   help me create new designs in Canva, retrieve  existing designs from my Canva account, or give   me suggestions about design styles. Great. So,  let's give this a try. Now, one of the best things   about using an agent in Microsoft 365 Copilot is  that I can reference other files. And if you have   the full Microsoft 365 Copilot subscription, you  can refer directly to files in your organization   or information from your meetings, messages,  and emails. So, I'm going to tell this agent   I want it to refer to information from my  email. So, I can type the forward slash key,   then start typing in a word that I think is in the  subject of that email. I'm looking for an email   thread about our volunteer day event. It found it  here, so I can select it. So, that email thread   is being referenced in this co-pilot prompt. I'll  tell it based on the information from this email,   design a promotional graphic for Instagram to  advertise the event. Now, the key to thirdparty   agents is that they will need to connect with  a separate app or service. So, I will need to   allow that connection to continue. But you should  also consider whether you trust that app enough to   connect it to your Microsoft 365 account. You may  want to get some guidance from your IT department   or security team about third party integrations.   In this case, I'll just click always allow. And   with most of these agents, you'll need to  sign into your account with that service at   some point. But after some time, it shows me some  design concepts from Canva. Remember, in Copilot,   you can always make follow-up requests. In  the Canva agent, I found that it doesn't   always like to make direct changes, but I could  ask it to generate new designs that include the   date. I'll send that follow-up request and give  it a moment. And here are those new designs.    Now, I say that the agent doesn't like to make  changes, but once you have a design that you like,   you can click the link to preview and edit it. And  it's at this point where I do need to sign into my   Canva account to continue working on this design.   And after I'm signed in, it opens that design and   I can work with it. You can make any changes  you want or download the final graphic. Now,   of course, Copilot itself can generate graphics,  but the Canva agent and Copilot can create complex   designs with multiple layers, including  stock imagery, adjustable color palettes,   and text that you can actually edit. Going back to  Copilot, let's consider privacy and security. At   some point, you may want to remove the permission  for Copilot to access this agent or app. Click the   button with three dots near the top right and  go to settings. then go to the subcategory 4   agents and I can see here that I have connected  co-pilot to the Canva app. I can click on that   and there's an option to reset which will remove  the permission for co-pilot to connect to Canva.    If I try to use the agent again, it will ask  me to grant that permission again which I can   always decline. And if you want to completely  remove the agent from the co-pilot interface,   reset to the main co-pilot chat, then find the  agent in the sidebar on the left, point at it,   and click the button with three dots to open the  menu. And there's the option to uninstall it. Now,   clearly I chose an easy app to integrate with. I  encourage you to look through the other thirdparty   agents available in Copilot. It's incredible how  we can basically turn C-Pilot into an AI powered   front end for each of these apps along with the  ability to reference the emails, files, messages,   and meetings in your Microsoft 365 account. The  agent mode in Excel is a specialized agent that

only works in Excel. It's designed to make  independent decisions based on your requests,   and it significantly expands what Copilot is able  to do in Excel. It's different from other agents   we've seen because you have to go to Excel to use  it and it mostly presents as an Excel feature.    You could open Copilot on the web by going to the  Office website, click the app button on the left   and choose Excel from there. Or you can open the  Excel desktop application on your computer. When   you open Excel on your computer, I recommend you  click the account button on the left and check   to make sure you are signed into Excel with the  right account. The Microsoft 365 account you use   for copilot is the account you should be signed  into here. Then go back to the home section. So   I'm going to open a file that I have stored on  one drive. In Excel you can go to the home ribbon,   click the co-pilot button and open the co-pilot  panel. And here you can make a general co-pilot   request or you can ask a question about this  spreadsheet. I'll ask copilot which are the   top five items with the highest profit margin. It  finds the answer to that question based on data   it found in this sheet. So, let's try something  else. I'll start a new chat and I'll tell it to   sort by retail price. But we'll find that the  co-pilot chat does not make changes directly to   a spreadsheet. It may answer the question in the  chat panel or it may offer a new Excel file with   the changes that you can download, but it will not  edit the original spreadsheet directly. Instead,   this is where we would use the agent mode.   The agent mode is able to make changes to   your spreadsheet, but that's just the beginning as  we'll see. So, I'll start a new chat, then click   the tools button in the chat field and choose the  agent mode. And now I'll tell it to sort by retail   price again. And after a moment, it does make that  change. This is similar to a feature that used to   be in Excel called app skills, but it's no longer  there. If you have a request like this, you should   be using the agent mode. Of course, you can always  undo the change or ask Copilot for something else.    I'll ask it to filter to only show accessories.   In the category column in my spreadsheet,   there are several different types of products.   Some of them are listed as computer accessories.    Copilot can identify those items and filter the  spreadsheet down to only show those. I'll reset   by telling Copilot to clear the filter. And next,  let's ask the agent mode to automatically label   certain data. I'll tell it for each item with a  profit value higher than $100, color the row in   green. The profit value is not in the sheet, but  Copilot can calculate it using the wholesale price   and the retail price, which are in the sheet. Then  it applied conditional formatting to label those   rows. Conditional formatting can apply a color  or other formatting element to specific cells   based on specific conditions. But I didn't ask for  conditional formatting. I asked Copilot to do a   task using natural language and C-Pilot decided  how to do it. It used conditional formatting to   highlight the items with the highest profit value  in green. Next, let's add that profit information   so we can actually see it. I'll tell Copilot to  calculate the profit margin for each item and add   that to a new column. And after a moment, it adds  a new column with that calculation. So, the agent   mode can handle requests that require changes  to your workbooks. But like I said earlier,   that's only the beginning. You can ask the agent  mode to analyze your workbooks or add specific   content. It can perform research to find what it  needs. It cannot make up information or draft new   original content like copilot Microsoft Word. But  it can project trends and it can do calculations.    and it can generate dashboards, tables, and  graphs based on data it calculates or finds   through research. So, for example, I'll ask it to  make a dashboard that will help me understand my   data and focus on giving me insights to help me  understand how much profit we lose by not selling   items in our inventory that are backordered. And  I'll even ask it to give me advice to help me   decide when we should restart production on those  backorded items. The agent mode will need to do   some research and do some analysis, which takes a  long time, so you should give it several minutes.    But when it finishes, I can find that dashboard in  a new tab in my workbook. There are several tables   that contextualize how much money we're missing  out on from our backordered items. And it also   includes recommendations and advice to help me  decide when to restart production as I requested.    And all the values in the dashboard are based on  calculations from the original spreadsheet. So if   I update that spreadsheet later, the connected  values in the dashboard will update as well.    Now this is just one example based on my simple  inventory spreadsheet. But the Excel agent mode's   ability to do independent research, evaluate  data, compare that data to information from the   internet, and its ability to modify my workbook,  and generate dashboards makes it an incredibly   sophisticated tool for data analysis. This is much  more than co-pilot responding to questions in a   chat field. This is independent action based on  my instructions. And that's what agent mode can   do in Excel. Once you see what agents in Copilot  can do, you might start thinking about making

your own agent. There are lots of different  types of agents, and some of them do require   specialized skills, but some types of agents  are surprisingly easy to build. The first tool   you should get familiar with is Microsoft's  C-Pilot agent builder, which lets you build   your own agents and share those agents with your  teammates. The agent builder is available right   inside of the C-Pilot interface for all users with  a Microsoft 365 business or enterprise account.    Just be aware some features are limited to people  with the full Microsoft 365 Copilot subscription.    Starting from the Microsoft 365 Copilot website,  we'll go to the agent section in the sidebar and   click new agent. There are two tabs at the  top, describe and configure. We'll start with   the configure tab, but after you know how this  works, you might prefer to start in the describe   tab. You'll understand as we go. Now, before we  do anything, if you have not already identified   what you want your agent to do, you should take  some time and really think about your goals. I   want to make an agent that can answer HR questions  in my company. It can save my HR department some   time by helping team members find quick answers  about company policies. Now, you could start from   a template here. These are mostly based on the  agents that Microsoft provides in Copilot. So,   if you want something similar to the Career Coach  agent, you might choose that as a starting point,   then set your own custom parameters. But I'm  going to make my own agent without a template. So,   I'll start by giving it a name. I'll call it  HR helper. And I'll put in a brief description.    Scrolling down, the instructions field is where  you do most of the work. Think about prompt   engineering techniques or just consider all the  detail you might put into a co-pilot prompt to   get a robust answer. You might give co-pilot a  role, describe a certain set of rules or context   to consider before answering questions. You can  put all of that into the instructions here. So,   I might add lines like, "You are an HR director  with 30 years experience managing HR departments   in large companies. " I'll start a new line, then  add, "Refer to well-known established HR practices   when answering questions. " or here's one of my  favorites that I like to include with most of my   agents. I'll tell it do not guess or estimate.   Check multiple sources to ensure accuracy. If   you do not know an answer for sure, tell the user  and encourage them to contact the HR department.    And if you're not sure of all the instructions  you should include, that's where the describe   tab can be very helpful. This is an AI assistant  within the agent builder to help you generate your   instructions. So, I could tell it that I want an  agent that can answer questions about HR policies   in my company. I'll type that in, then send the  prompt, and if it asks any follow-up questions,   make sure you answer those. But now, it says  my agent is set up. So, let's go back to the   configure tab and see what's changed. I can see  that it has added several more instructions, and   it even rewrote some of the instructions that I  provided. I strongly encourage you to read through   these and make sure that everything looks right.   And these instructions might give you ideas for   more that you want to add. But we're not finished  yet. Still in the configure tab, scrolling down,   we'll get to the section for knowledge sources.   And this may be incredibly valuable. I want this   agent to answer questions about HR policy at my  company. My policies are well documented in my   employee manual. So, I'm going to tell the agent  to use the employee manual as a knowledge source.    There's an option to add a link to a website as  a knowledge source. But if you also have the full   Microsoft 365 copilot subscription, then you  will have the ability to upload documents or   use documents that are stored or were shared with  you on one drive or sharepoint or teams. To use a   document from my one drive, I can click on this  field and I can start typing in the name of the   file. It searches through my one drive, finds a  file that I want so I can select it and after a   moment it's added at the bottom. But here's an  important variable. Will you be sharing this   agent with your teammates? If so, the files that  you attach must be accessible to those teammates.    So if I choose a file from my one drive, but  I have not shared that file from my one drive   with my team, then the agent will not be able to  access it when my teammates use the agent. So in   that case I'm going to click the X to remove that  file and confirm. And alternatively if you upload   a file from your computer it will be bundled  with the agent and any user will be able to   access and information from that file. So there's  an important trade-off between privacy permissions   and usability. I have a copy of my employee  manual and my company's social media manual on   my computer. So I'll upload those. And now those  files have been uploaded and will be included with   this agent. Before moving on, you should set  whether you want the agent to find information   on the public internet or if you only want it to  use the sources that you've provided. I'm going to   give it permission to search the web and move on.   Next, under capabilities, you can decide whether   the agent can generate documents, code snippets,  or images. And then there are suggested prompts.    These will appear below the chat field when  people use the agent. The suggested prompts   that are already filled in here are based on the  description I put in when I used the describe   mode. You can edit or delete the suggested prompts  here. So those are the core steps to building a   basic agent. There is a test version on the right.   So you can try some test questions and see how   it responds. And then you can make changes in the  configure tab. But in my experience, I think it's   safer to finish the creation process rather than  depending on this preview. So, when you're ready,   just click the create button at the top. And  don't worry, you can still make changes later.    When it's done, I'll click go to agent. And that  agent is listed in the sidebar on the left, just   like any pre-built agents that you might add. You  can go to the new chat option at the top when you   want to use the normal co-pilot chat, and you can  select your agent when you want to use that. And   I cannot stress this enough. You are not finished  yet. Testing and refining are essential to making   a good agent. You should try some questions in  your agent and see how it responds. So, in the   chat field, I'll type, I have a client who wants  to fly me to New York in exchange for publishing   some social media posts about our partnership. Is  that okay? I'll send that prompt and it gives me   a response based on the agents instructions as  well as information it found from the employee   manual and the social media manual. You should  test several prompts and if it does not respond   exactly as you want, you should edit the agent. To  do that, you can click all agents in the sidebar   on the left, find your agent and point at it.   The button with three dots will appear. You can   click that to open the menu and choose edit. Then  make changes to the instructions or the knowledge   sources. Update the agent and test it again.   Take the time needed to test and revise until   you're confident that the agent is ready. And if  you want to share the agent with your teammates,   there's a share button on the edit page or you  can go back to all agents. Find the agent and   open the menu and choose share there. You can only  share the agent with other users in your Microsoft   365 organization. If you only want to share the  agent with specific people, you can choose that   option. Then go to the field and start typing  in somebody's name. It searches your company's   directory. You can choose that person and you  can add multiple people or you could just make   it available to anyone in the organization. So  choose the option you want then click apply and   then you can copy this link and send it to your  teammates. When the teammate clicks the link they   will see an option to add the agent to co-pilot  in their account. However, you should be aware if   one person builds an agent and they use features  that are only available to people with the full   Microsoft 365 C-pilot license, then the agent  will only be usable to other people who have the   full license. People with the baseline Microsoft  365 account would see a note telling them that   they would need to upgrade to use the agent. So,  anybody in a Microsoft 365 organization can make   an agent, but there are some features and sharing  options that are limited to people with the full   Microsoft 365 C-pilot subscription. Making an  agent to share with your teammates allows one   person to do a little work that can save several  other people lots of time. But making an agent   just for yourself can also be incredibly valuable.   If you write long robust prompts to define   specific parameters in your co-pilot requests,  you might think about whether you can save those   parameters in an agent. So then whenever you have  a related request, you can go right to that agent   and ask a quick question without establishing the  context every time you write a prompt. If you get

to the point where the agent builder is not able  to do what you need, you may want to consider   Copilot Studio, which offers sophisticated tools  for building more complex agents. However, working   with Copilot Studio does require the participation  of your organization's Microsoft administrator,   who will need to set up licenses, pay for usage  fees, and help with the deployment of the final   agents. So there are some details that you will  need to work out with your administrator and we   cannot cover all of those details here in  this course. But in this video I will help   you understand how the licensing works in general.   This will help you decide if you want to jump into   Copilot Studio which we will do a little later  in the video. AI tasks can be expensive. So in   many cases your organization has to pay for your  agents in the form of co-pilot usage credits. And   you will have to consider two important things.   Who will build the agent and who will be using   the agent? So first, if you're the one building  the agent, the question is what type of Microsoft   account or subscription do you have? The baseline  Microsoft 365 business or enterprise account does   not let you build with Copilot Studio. There  is a free 30-day trial available which can be   extended by an additional 30 days, but after  that, these users cannot use Copilot Studio.    If you have the full Microsoft 365 C-pilot  subscription, then you can use Copilot Studio   to build agents. But we also have to think about  who will use the agents you build. If you have   the full Microsoft 365 Copilot subscription and  you are using the agents you build yourself,   then you do not have to pay for co-pilot credits.   If you share your agents with teammates in your   same organization who also have the full co-pilot  subscription, they also do not need co-pilot   credits. So when do you need credits? Well, if  you build agents and share them with teammates   who have the baseline Microsoft 365 account only,  then you will have to pay for co-pilot credits to   allow those people to use your agents. or if you  build an agent that takes independent action based   on program triggers, then those actions must  be paid for with co-pilot credits. And finally,   if you build an agent that can be used  by people outside of your organization,   those are also subject to the co-pilot credits.   So, how do you buy these co-pilot credits? Well,   your organization can purchase credits with a  prepaid plan or with a pay as you go plan. Of   course, you're wondering, "How many credits do I  need and how much do they cost? " Well, for that, I   will refer you to this pricing page with a link to  estimate your co-pilot studio usage. I know there   are a lot of variables to consider here, which  require a bit more research based on your needs,   but if you're just getting started with agents  in Copilot Studio, then these steps are something   you'll come to naturally further down the road.   At this point, if you are ready to start building

an agent with Copilot Studio, let's jump in. I'm  going to show you how to make an agent in the form   of an AI chat that can respond to user requests  within specific defined parameters. Just remember,   there are some features like independent action  and connections with other apps that are offered   by Copilot Studio, but they require additional  skills that do not fit into the scope of this   course. I'll show where those tools are, but  we will not be able to cover them completely.    And of course, if you plan to publish an agent  for other users, you will have to work with your   company's Microsoft administrator to do that.   But we'll see more about publishing as we go.    Before you do anything, I think that planning and  at least identifying your goals is very important.    I'm going to stick with the scenario we used  earlier in this course. I want to make an agent   that can answer questions and filter requests  on behalf of my HR department. This will help   my HR team save time because they won't have to  answer all of the first level questions that come   in. To get started, you can work with Copilot  Studio by going to copilotstudio. microsoft. com. I'm signed in with my Microsoft 365 account from  my organization. There are a few ways to start   building an agent. On the homepage, there is an  AI assistant. You could choose the agent tab,   then describe the agent you want. The assistant  will build out the features to start your agent.    Alternatively, you can switch to the agent section  on the sidebar on the left and you can click the   button to make a new blank agent and then fill  in every detail manually. Or you can choose a   pre-made template to start with. But we also have  the AI assistant here. I'm going to use that to   get started. So, I'll just describe the agent that  I want. I'll tell it to make an agent that can   answer HR questions in my company or elevate  questions to the correct HR representative.    Then I'll click next. This takes some  time, but it starts building the agent.    It gives the agent a name and a description. I  could click the edit button and change those,   but we'll stick with what it said. Scrolling  down, we have several configurations that you   should set. And many of these configurations will  look familiar if you've used the co-pilot agent   builder, which we saw earlier in the course.   But here's our first option that is not offered   by the co-pilot agent builder. You can choose  which AI model powers your agent. And there's   also some information about the strengths of each  model to help you choose. Next, we see triggers,   which can enable an agent to act independently  when certain events occur. As mentioned before,   we're only going to focus on the most essential  features. But if you are interested in setting   up autonomous actions, this is where you can find  those controls. Scrolling down, we see our agents   instructions. In plain language, you can tell the  agent exactly what you expected to do. You can   use well-known prompt engineering techniques to  give the agent a specific role, expectations, and   context. You can tell the agent what its limits  are and how you want it to respond. You can tell   it if there are any special rules. Now, in this  case, the AI setup assistant already filled in   several instructions for me. And you should read  these because they will help you learn the types   of instructions that you can give. but also if  there are any that you want to remove or change,   you can click the edit button to make that change.   Then we get to knowledge sources. Just like in the   co-pilot agent builder, you can tell your agent  where to look to find information it needs. Since   I want this agent to answer HR related questions  in my company, I want it to refer to my company's   employee manual where it will find the answers  to the most basic questions. Now, this is great.    The AI assistant has already searched through my  shared documents and identified suggested sources   that fit my goals, including my company's  employee manual. I could click add on one of   these suggestions and it tells me where this  file is located. But be careful here. If you   use a file stored on one drive or sharepoint  and you later share the agent with teammates,   but you have not shared the source file with those  teammates, then the agent will not be able to use   the knowledge from that source file. These privacy  protections might be exactly what you want,   but they can also affect the usability of the  agent depending on who you share it with. Now,   in my case, I'm going to cancel this and I'm going  to dismiss these suggestions. Then I'll click the   add knowledge button to add sources manually.   And from here, I think a specific website can   make a great knowledge source. You can also use  several other data storage systems listed here.    But for now, I'm going to upload the documents I  want from my computer, and that will bundle the   knowledge sources in with my agent, and they will  not be subject to privacy or sharing policies. I'm   doing this because it's easier, but you absolutely  must consider the sensitivity and privacy of the   files that you might include this way. Anybody  who uses this agent will essentially have access   to the potentially sensitive information in these  documents. So, I'll choose upload and select the   files from my document folder. So, I'm uploading  a copy of my employee manual and my social media   manual. Then, I'll click add to agent. Before  we finish with the knowledge sources, you can   set whether the agent can find information from  the general public internet. If you disable this,   it will limit the agent to only using information  from the knowledge sources you provided. Scrolling   down, we see the options for tools, agents, and  topics. We will cover topics in a separate video,   and we're not going to use the others. And  finally, we have suggested prompts. Most AI chat   assistants offer a few suggested prompts to the  users which helps the users understand the types   of questions the AI chat is designed to handle  and they can offer users a shortcut to asking   useful questions. I'll click the add suggested  prompts button and I'll just add one here. So in   the title field I'll label this one remote work  guidelines. Then I'll write the prompt describe   the company's policies for working remotely. You  can add up to six and then click save. We won't   really see those suggested prompts until the agent  is published, but you can always come back here   and edit them. Now, before we move on, I want to  scroll back to my knowledge sources. My documents   still show the status of in progress. My agent  will not be ready until these all say ready. So,   I'll skip forward in time. Now that my knowledge  sources all say ready, I'm ready to continue. Now,   testing your agent is vital, and that's why there  is a preview version of the agent on the right.    When you think you've done your setup that you  want, you should try some test questions. Ask the   agent questions or ask for help with tasks that it  should be able to handle. So, in the prompt field,   I'll say, "I have a client who wants me to  fly to New York in exchange for publishing   some social media posts about our partnership.   Is that okay? " While the agent is processing,   we see some of the background options on the left.   This might help you identify any problems that may   come up. And here's the answer to the question.   Including citations confirming that it is using   my knowledge sources to find this information.   You should try several test prompts. And if   there's anything that does not work as expected,  you should go back and edit the instructions or   the knowledge sources or whatever you need. You  should take the time for several rounds of testing   and revisions before the agent will be finished.   Now, clearly this agent has not been deployed for   my teammates or partners to use. We're going to  see how to publish an agent in a separate video,   but we have seen the core tasks required to build  an agent. You can go back to the overview tab at   the top to reset to your editing tools. And you  will also see that you can expand many of the   building tools using those tabs at the top. So I  could go to the knowledge tab to focus only on the   knowledge sources and go back to overview to reset  to the main set of tools. There's also a button at   the top for settings. You can see there is a lot  of depth here that you can explore as you continue   to learn. In the generative AI section, you can  set rules for how responses should be formatted.    You can set up content moderation or decide  whether your agent can process uploaded files. In   the voice section, you can set whether the agent  is optimized for voice inputs. You can go to the   languages section to establish which languages the  agent can work with. So, feel free to explore the   settings and click the X in the top right to close  settings when you're finished. And that will take   you back to the agent building console. Now, on  the left, I'll reset to the homepage. And from the   homepage, you can go to the agents section to see  the list of all the agents that you've built. And   you can click on one to continue working on it.   Generally speaking, an agent you build in Copilot

Studio uses AI to understand what a user is saying  or asking. Then it formulates a response. However,   if there are scenarios when you want your agent to  respond in a specific way to establish questions   or prompts, you can set that up using a system  called topics. We've already built an agent with   all the required features. Now, we're just going  to use topics to make sure it responds exactly   how we want in certain scenarios. From my list  of agents in Copilot Studio, I'll select the   agent I built earlier. In the overview tab, you  can see the main tools that you use to build or   modify the agent. There's a section here for  topics, but this isn't showing everything. You   can click see all or you can click the tab up at  the top for topics. And here you will find some   topics that are already listed and sorted into  categories for custom topics and system topics.    The biggest difference being that you cannot  delete system topics. You can modify them,   but you cannot delete them. I'll go back over to  custom topics and there's a menu next to each one   so you can delete them if you need to. But let's  take a closer look at one. I'll click on the topic   labeled thank you. Now every topic is invoked by a  trigger. The agent chooses trigger is very common.    This means when the agent recognizes certain  input from the user, it triggers the topic.    We see the instructions to the AI explaining what  it should look for. So if a user includes thanks,   thank you, or any of these other phrases in their  prompt, this topic will be triggered. Or if the AI   recognizes something that it understands  to be similar to one of those phrases,   that will trigger the topic as well. Then below  that, we see what happens next. In this case,   the agent will respond with a message replying  exactly with you're welcome. Each of these blocks   in this tree is a node containing a step in the  conversation process. This is a very simple topic,   but a topic can get much more complex. I'll click  the back button near the top left to go back to   the list of topics. You may want to go over to  the system category and select a more complex   topic. Let's select this fallback topic. Topics  can have different types of triggers. There can   be conditional nodes that split the tree into  different steps depending on certain conditions.    And looking at how existing topics are structured  like this may help you learn how to work with the   topics feature. But it will also help you if you  have some experience with programming language or   writing system configuration files with variables  and conditions. You can also click the button with   three dots at the top and switch to the code  editor. And if you have experience with YML   markup, that would definitely help you here as  well. But for now, I'll close the code editor   and click the button at the top left to go back  to my list of topics. So, if there is a specific   scenario and you want your agent to respond in  a specific way, you might want to make your own   topic, you could open the ad menu up at the top  and choose to make a new topic from blank, but   considering how complex the node system is, it's  unlikely that a beginner would be able to jump   in and figure it out. So instead, I suggest you  use the co-pilot assistant to help you build your   topics. So you can choose add from description  with co-pilot, then fill in the name you want for   the topic. In my case, my agent handles questions  related to HR issues. But if there's anything   related to reporting a conflict in the workplace,  that should go directly to our director of HR. So   I'm going to name this topic conflict report. Then  in the description field, I'll tell Copilot if the   user asks about or even suggests a scenario where  they may need to report a conflict, refer them   directly to the HR director. This will be our  starting point and we'll make changes as we go.    For now, I'll click create and it generated the  conversation tree for that topic. In the trigger,   the agent will choose to trigger the topic  when the conditions below are met. Basically,   when the user mentions something about reporting a  conflict, you can modify this description here to   dial it in. If that event is triggered, the agent  will reply with a specific pre-written response.    I would prefer to rewrite this response to assure  the user that we will protect their privacy than   giving them a more direct email address or  phone number of the HR director. But for now,   I'm just going to leave it as is. In most  situations, especially if you are new to this,   starting with a topic generated by the co-pilot  assistant is going to be far more practical than   making a new topic manually. You can and certainly  should go through every step of the topic and make   changes or rewrite the responses. Or you can click  the co-pilot button up at the top and you can ask   for changes there. But for now, I'll close this.   And when you're finished, make sure you click the   save button at the top. Then go back to the list  of topics. Now that I have that new topic set,   I'll go to the agent test panel on the right and  I'll tell it, "One of my co-workers is stealing my   supplies. Who can I talk to about this? " That  triggers the topic and it responds with that   pre-written response. Now, obviously, this was  a very basic topic tree. They can get much more   complex, but even the copilot assistant can help  you build the more complex topics. And these are   first steps. There's lots of room to grow and more  that you can learn about topics. For now, I'll go   back to the overview tab. Topics are just one tool  that can help your agent understand user input and   respond. Topics, knowledge sources, and general  instructions are the tools that you can combine to   make your agent respond accurately and efficiently  for your users. If you've built an agent in

Copilot Studio, chances are you're going to want  to publish it so your teammates or partners can   use it. In this video, we're going to see the  steps that you can take to publish your agent from   Copilot Studio. However, there will be additional  steps that your organization's Microsoft   administrator will need to handle. So, we will not  finish this video with a fully published agent.    But I believe you will have a clear picture of  the steps that you can take and what you'll need   to discuss with your administrator to finish  the process. From co-pilotstudio. microsoft. com,   I'll select the agent I've been working on in  this course. And first, you need to consider   authentication or who will be able to access your  agent. Will they need to sign in with an account?    Click the settings button at the top, then go to  the security section, then to authentication. The   easiest option is to set your agent to work with  Microsoft authentication, letting people use it if   they are signed into their Microsoft account.   If you plan to release your agent on the web   or in a third-party app and you want it to be  accessible by the general public, you can set   it so it does not require authentication. But it  is vitally important that you ensure that all of   the knowledge sources and information that may be  accessible through the agent is safe and clear to   be used by the general public. Do not disable  authentication if your agent deals in private   information from your company. There's also an  option to configure an authentication service   provider manually. If you have the setup for  that, the option is there, but I'll leave this   set to Microsoft authentication. And if you  make any changes, make sure you click save,   then click the X in the top right to close  settings. You can always make more changes later,   but if you've built and tested your agent, set  the authentication rules, and it's ready to go,   you can click the publish button at the top, then  click publish here to confirm. Now, publishing   an agent does not actually make it available to  others. It just makes it ready to push out in the   distribution channels that you choose. Take a look  at the tabs at the top. You may see a tab labeled   channels or you may need to open the menu to show  any additional hidden tabs and then open channels   from there. Channels are the different places  where you can make your agent available. In this   video, we're going to make the agent available in  Teams and Microsoft 365 Copilot. In another video,   we'll explore the options for making an agent  available in other services. So, I'll select that   channel and every channel will have completely  different configuration settings. This is just   the first step. So, first I need to confirm that I  want it to be available in Microsoft 365 C-pilot.    Then I'll click add channel. When that's done,  then this agent is ready for me to use myself.    I could click see agent in Microsoft 365 to use it  on the co-pilot web app or I can click see agent   in Teams to use it inside of Teams. That opens  Teams where I can add the agent. Once it's added,   I can open it and work with it here. And by the  way, I see a suggested prompt here. There's only   one suggested prompt because I only set one when  I created the agent. Now, this is all fine if you   just want to use the agent yourself, but you may  also want to fully deploy the agent so that your   teammates can use it in Copilot. So, let's go  back to Copilot Studio. And if you have closed   this channel information panel, you can always  get back to it by going to the channels tab and   click on the channel for Teams and Microsoft 365  Copilot. Since we've already added the channel,   the next step would be to click availability  options. And here we see the option to show   everyone in my org. This is where you go to  deploy the agent to the rest of your organization.    However, you can only start this process. You  can click the button and submit the agent to   your administrator for approval. I'll confirm  that that's what I want. And after a moment,   I see that it has been submitted and is  waiting for the administrator. After that,   your administrator has tools in the Microsoft 365  admin center to approve or reject your agent. If   they approve it, they can make it available  for other people in your organization. Your   teammates will be able to add it from the agent  store in Microsoft 365 or Teams. Just remember,   as we saw in another video, there are limitations  on who can use these agents in your organization   based on your C-pilot Studio license and any  usage credits your organization has purchased.

Aside from publishing your agent for teammates  in your organization, you may want to make an   agent available to clients or partners who are  not part of your Microsoft 365 organization.    As we saw before, you do have to consider the  question of authentication before publishing   an agent. And the authentication settings you  need will almost certainly be different if you   plan to publish the agent externally. Click the  settings button near the top right. Go to the   security section, then choose authentication.   Now, the reason it's not letting me change the   authentication settings here is because  I've already published this agent for my   teammates using this authentication strategy.   So making users sign in with their Microsoft   account was perfect in that scenario. But that  will not work if I'm sharing the agent outside   of my organization. If I want to change the  authentication, I'll need to reset my previous   publishing channels. So I'll click the X near  the top right to close settings. And at the top,   you may see the channels tab, or you may need to  open this menu to show additional hidden tabs,   then open channels from there. We'll be exploring  these channels more in a moment, but because I set   up the Teams and Microsoft 365 Copilot channel in  the past, I'm going to click on that channel. Then   I'm going to remove the channel to reset.   So I'll click remove channel and confirm.    It says that channel was deleted. So I can  close this panel. Now I'll go back to settings,   back to security, and to authentication.   And since I have reset my channel settings,   I am now able to change this. To deploy an  agent on the web or in a thirdparty service,   you'll need to set it to no authentication or set  up an authentication service manually. If you want   to set up an external authentication service,  you should communicate with your Microsoft   administrator or your IT department. This can  take significantly more work and research,   but depending on your security needs, it may be  necessary. Once you have those details in place,   you can connect that service to your agent here.   But for now, we're going to see the much easier   option. We're going to set this to require  no authentication. However, if you do this,   then all of the knowledge sources and information  that may be accessible through your agent will be   accessible to the general public. You should never  disable authentication if your agent deals in   private information from your company. But we are  going to do that for this agent. If you make any   changes, make sure you click the save button. This  gives a quick warning about publishing without   authentication, including a list of agent features  that you cannot use. Once again, I'll click save.    And with this authentication set, I'll click  the X to close settings. And back in my agent,   the next step is to click the publish button. This  gives me another warning about authentication,   but I'll click the publish button here to confirm.   Publishing an agent does not actually make it   available to others. It just makes it ready to  push out in the distribution channels you choose.    Once again, at the top, you may see a tab labeled  channels, or you may need to open the menu to show   additional tabs, or in my case, the channels tab  is already selected. We see some information about   my agent status, but I'm going to scroll down so  we can look through these channels. The channels   are the different places where you can make your  agent available. In the other channel section,   we see several thirdparty services. So, you  can make your agent available inside of Slack   or WhatsApp or one of the others. But there is a  lot of setup required for each of these thirdparty   services. Most of these will require you to have  an established account with that service. And each   one will have their own requirements for setup  and deployment. Also, almost all of these channels   require you to have a co-pilot studio license  where you have purchased a plan for co-pilot   credits. When external users work with your agent,  that incurs additional cost that is paid for by   the co-pilot access credits. Of course, different  channels require different configurations,   but this is where you can get started with each of  them. Since we can't cover all of them, let's see   how the system works in general by setting up a  demo website. This does not require me to have an   account with a thirdparty service. So, I'll choose  demo website. And here you have the option to set   a welcome message or some conversation starters.   I'll change the welcome message to welcome to the   HR policy agent. If you make any changes here,  make sure to click save. And in this panel,   there is a link to the website which you could  copy and share with somebody or you can just   click the button to open the website. And this is  a great way to get a finished agent available to   the public quickly. And it's an excellent system  to test your agent. I'll ask a question here. I'll   type, "I have a client who wants to fly me to New  York in exchange for publishing some social media   posts about our partnership. Is that okay? " I'll  send that question and see how my HR policy agent   responds. And of course, it gives me some great  information using the knowledge sources that I   provided when I built the agent. And for now, I'll  close this website and go back to Copilot Studio.    And I'll close this panel. And let's acknowledge  the demo website is by far the easiest way to   deploy an agent to the public, but it's also the  most limited and it's not really practical for   long-term use. I showed it so we can see how to  set authentication and how the core publishing   channels work. Of course, any of the thirdparty  publishing channels will have their own special   setup, but for now, you know where to get started  with those publishing channels. And that brings us   to the end of this full training course on working  with Copilot agents. Thank you for coming here and   learning with me. I hope you've got everything  you need to use or even build agents in Copilot.    You can always use the chapter markers to jump  back and review any of the sections from this   video. For more videos like this and to continue  learning, be sure to subscribe to this channel.