Build AI Agents with Microsoft Copilot Studio – No Code Guide

Hi everyone, Kevin here. Today, we're going  to learn how to use Microsoft Copilot Studio.   With Copilot Studio, you can build your very  own AI agents that answer questions using your   own documents, automate repetitive tasks, and  even respond to emails, all on their own. And  

the best part, you don't have to write a single  line of code. You just use natural language,   like you're chatting with a friend or helpful  coworker. We'll walk through everything step by   step, so by the end of this video, you'll have  your very own AI agents working on your behalf. 

Let's dive in. To get started, head to the  following website. You'll find a link right   here at the bottom of the screen. Then sign  in with your Microsoft work or school account. Unfortunately, this does not work  with personal Microsoft accounts.

Once you sign in, on the left-hand side of the  screen, click on chat and then all agents. Here,   you'll find a variety of ready-to-use agents, each  designed to handle specific tasks. As an example,   right here, we have a brand-new agent called  the researcher. You can feed this content like  

documents, reports, or articles, and it'll  analyze everything and give you a detailed   comprehensive breakdown. This one is perfect  if you need to dig into complex topics quickly.   Right next to that, we have another new one called  the analyst, and this one does data analysis on  

your files. You can upload, let's say, an Excel  sheet, and it'll summarize trends, flag outliers,   and even help create visualizations. This one is  well worth trying. Now, you might be wondering,   how are these agents any different  from something like ChatGPT? Well,  

ChatGPT is what's called a general-purpose  AI. It's great for answering all sorts of   questions. These Copilot agents, on  the other hand, are task specific. They've been trained and optimized to  handle a particular kind of work. Let's  

say I want an agent that can answer customer  questions about the Kevin Cookie Company. ChatGPT won't know our secret cookie policy,  but I can train my own agent using our company   documents, and it'll give accurate answers every  single time. Let's now go ahead and create our own  

agent, and we have a few different ways to do  that. In the top right, we can click on Create   Agent. You can also scroll down the page. Here  you see the option, and you can even do it over on   the left-hand side. Lots of different ways. I'll  click on this. This drops us on a simple version  

of Copilot Studio that we can use to build our  AI agent. Later on, we'll look at a more advanced   version of Copilot Studio that gives us even  more control. Right down here, it says, "Hi,   I'm here to help you build an agent." Well, that's  very nice of it. You can start by describing what  

you want the agent to do. I'd like to build a  simple customer service agent for my favorite   company, the Kevin Cookie Company. You're welcome  to follow along, or you could even use your very   own example. Right down below, we could also  start from one of these templates, but I know  

exactly what I would like to build. Down below,  we have this text field, or the prompt field,   where you could describe your agent. Let me type  in my description: create an agent that answers   customer questions about the Kevin Cookie Company  using our frequently asked questions, or FAQ,  

and company documents, and let's call it Chipbot.  I love that name. Once you finish typing in your   description, let's send that. And check that out.  Chipbot is now alive, and we could even start   asking Chipbot questions, but it probably wouldn't  provide very helpful information. And the reason  

why is we haven't yet told Chipbot how to answer  questions. It doesn't even know what our FAQ is.   Over on the left-hand side, here Copilot Studio  asks us for additional information to continue   refining how Chipbot should work. As an example,  here it says, "Could you provide more details on  

how Chipbot should handle customer questions."  Or, "Could you even give me a link to the FAQ?"   Those are all great questions, so let's give  it some more information. Down below, I’ll say,   please provide concise and friendly responses. Add  some humor in as well. We are, after all, a cookie  

company. If you don't know an answer, don't make  up information. Instead, direct customers to our   support email address. And then, most importantly,  please use the kevincookicompany.com as a source   of information. Let's now send that. And right  up above, we can now see that Chipbot knows how  

to respond and also what information to use in  its responses. We are making fantastic progress.   Now that we've gone through the initial setup,  right up on top, let's click into Configure.   This is where we're going to fine-tune Chipbot  and make sure that everything is set up just  

right. Let's click here. You'll notice that most  of this information is already filled in based on   what we entered earlier. Still, it's a good idea  to review each section and make any adjustments to   match exactly how you want your agent to behave.  Now, first off, we have to change this image.  

Let's just say it's very neutral. It doesn't  exactly scream, "Cookie expert," does it? To   change the image, let's click on this pencil icon.  Right over here, let's click on Change Icon. Now,   just navigate to where you have your image. Here,  I see Chipbot. That's kind of what you'd expect  

a Chipbot to look like. Right? As a quick tip,  you can use AI to help generate an image. Simply   describe your agent and you'll get a nice-looking  image. Once you select it, let's click on Open.   And here, I see Chipbot. Now, let's click on Save.  Now that Chipbot has a proper identity, let's take  

a quick look at the name and the description. And  I think both of those look good. Next up, we have   the instructions. This section guides how Chipbot  responds to questions. The best way to think of   it is like you're writing onboarding notes for a  new team member. You want to be clear, specific,  

and also kind. This is where you shape the tone.  Maybe you want to limit the responses to only   sources that you provide. Like here, I call out  the KevinCookieCompany.com. Or maybe you want to   include fallback guidance in case an answer isn't  available. Here, we tell customers to reach out to  

support at KevinCookieCompany.com. This is where  you provide all of that information. Scrolling   down, we arrive at the Knowledge section. This is  where Chipbot learns what it knows. I've already   added our company website, but we could also  include additional information. Right up here,  

I'll click on the search field, and you could pull  in things like an Excel sheet, a Word document,   PowerPoints, PDFs. You can pull in websites,  SharePoint, Teams, and also emails. But I want   to pull in my FAQ. So over here, let's close out  of this view. And on my computer, I'll select my  

FAQ file, and here I could simply drag and drop it  in. Right down below, you'll see all the different   file types that you could pull in. And there,  I'll release. And just like that, it becomes   part of the agent's knowledge base. As a quick  note, because I uploaded this file, if any of  

the information within the file changes, I'll have  to upload a new version. If instead, I use say a   website or a SharePoint file, it'll automatically  capture the most up-to-date information. So just   consider that as you're adding information  to your agent. Next up, we have capabilities.

These let you enhance your agent with features  like the code interpreter or the image generator.   These are great if your agent needs to calculate  things or create visuals. But in my case,   since Chipbot is just answering customer  questions, we'll keep it simple and leave  

these turned off. Down at the very bottom, we've  got suggested prompts. These are sample questions   that help users know what they can ask. You can  edit, delete, or even add your own here. When   someone starts a chat with Chipbot, over here on  the right-hand side, you can see an example. These  

prompts show up right there to give people a head  start. Let's now test it out. Right down below,   let's type in a question. And what are the store  hours for the Kevin Cookie Company? And let's see   if it knows the answer. And here we go, right up  above, it gets our hours, Monday through Saturday,  

nine to seven. And here it calls out that Sunday  is closed. And look at that, it even included some   humor because even cookie monsters need a day  off. Nicely done. One of the things I really   like is that it even includes a citation to where  it got this information from. I can click on this  

icon and that opens up the original source Word  document where it calls out those hours. That   way I have confidence in the result that I'm  getting back. I really like how this works and   I'm now ready to publish this so other people in  my organization have access to this agent. In the  

top right-hand corner, let's click on Create.  The agent is now live, but currently it's only   available to me. Luckily, we could easily change  that. Let's click Change Sharing Settings. Here,   I have the option to share with specific users  in my organization, or I could simply share with  

everyone in the organization. I'll click on  Anyone. And right down below, let's click on   Save. Now I can share this agent with anyone in my  organization. I could copy this link. I'll close   out of this. Also, over on the left-hand side,  here I now see Chipbot showing up as one of the  

agents in this organization. Here, if I click  on All Agents, I see it as one of the options.   That's beautiful. This agent is already working  well. It understands questions and retrieves   accurate answers using knowledge that we gave it.  Now, that'll probably take you most of the way,  

but if we want to take it a step further and  we want this agent to, let's say, take action,   like sending an email or retrieving live data,  let's head over to Copilot Studio. You'll find   a link right down below, and that'll allow us  to unlock even more advanced capabilities. This  

drops us into the more advanced version of  Copilot Studio. Over on the left-hand side,   if you'd like to recreate your agent, you could  do so right here. I've already done that, so I'll   click on Agents, and right here, I see Chipbot.  To create Chipbot here, I basically went through  

the same flow. I provided the same instructions  and uploaded the same FAQ file. Right over here,   I'll click into Chipbot. At first glance, it may  appear that we have similar functionality here.   I see the image of Chipbot. We have a name,  description, and also general instructions.  

If I scroll down a little bit more, here too, I  have those same files as part of the knowledge,   the FAQ Word document and the website. But  let's click on Add Knowledge. Right away,   you'll notice that I have many more sources here.  I'm not just limited to uploading files or linking  

to a SharePoint site. I could also connect to  things like Dataverse, Salesforce, ServiceNow,   and if I click on Advanced, you'll see even more  data sources. I can retrieve information from all   these systems and use it in responses. So, let's  say that a customer wants to check the status  

of their order. I could connect to the right  database. Let's say it's an Oracle database,   and then I could have Chipbot pull the information  from there directly. That is so powerful. Now,   I don't need to bring in any new sources, so in  the top right-hand corner, let's close out of this  

view. Within the knowledge section, you'll notice  that we also have something called Web Search,   and currently it's toggled off. This means  that Chipbot won't pull in anything from   the internet. It'll just use the  information that we provided. Now,  

that's exactly what I want for controlled on-brand  experience. Next, we have a section called Tools. Now, tools let your AI agent go beyond just  answering questions. They allow it to actually   do things, like let's say sending emails or maybe  updating records. Let's try adding one. Right over  

here, let's click on Add Tool. You can connect  your agents to over 1,500 different tools. In the   top left-hand corner, let's click on All, and here  you'll see how comprehensive this list is. Chances   are if you'd like to use a service or a tool,  you'll find it here on this list. I'll go back  

up to the top and then let's switch into the Flow  view. You can also call a Power Automate Flow. If   you're curious about what that even is, you can  check out a video in the description down below   that gives an introduction to Power Automate.  Let's click back on Featured. Along with these  

different tools and Power Automate Flows, you can  even connect to your own backends and even APIs. In this case, I want Chipbot to  send an email. So right here,   I see the Office 365 Outlook  connector. Let's click on that.

There are tons of different actions available  within Outlook, but I simply want to send an   email. So right over here, I see the option  for Send an Email V2. Let's click on that and   then let's configure it. When a customer has a  conversation with Chipbot, I want Chipbot to ask  

the customer if they want an emailed copy of the  conversation. If the customer answers yes, then   Chipbot will use this tool that we're creating to  email a summary to the customer. Right down below,   let's click on Add and Configure. At the very  top, we can give this tool a name. I'll set it  

to send an email, and also a description, Email  Response to Customer. This will help the agent   know when to use this tool. Let's scroll down  a little bit more to the section titled Inputs,   and this is where the magic happens. We're  telling Copilot Studio what should go into  

the email. Here, we see the input name, to, and  currently it's set to dynamically fill with AI.   Chipbot will figure out who to send the message  to based on the conversation that it's having   with the customer. We also have the subject, and  currently that's set to dynamically fill with AI.  

The AI will generate a subject line that fits  the topic of the conversation with the customer,   and lastly, we also have the body, and again,  this is dynamically generated. This will include   the original question and Chipbot's response  in the email body. Now, you could also set it  

to a custom value if you want to set a fixed  subject line or maybe a body in the message,   but I'm good leaving it set to dynamically fill  with AI, so over here, I'll click out of that.   Once everything looks good, in the top right-hand  corner, let's click on Save. And just like that,  

Chipbot can now send emails on command. Now  that we've added our Send an Email tool,   we need to tell Chipbot when to use it. To  do that, we need to update the instructions,   so up on top, let's click back into the Overview.  Back on the Overview page, let's click on the Edit  

icon next to Details, and if we scroll down just a  little bit, we could see the general instructions.   So right here, let's add an additional step. I'll  say after answering the user's question, ask if   they would like the response emailed to them. If  they say yes, use the Send an Email tool to send  

a summary of the question and the response. I  think that sounds good. Let's now scroll back   up to the top and let's click on Save. Let's  now test it out. Over on the right-hand side,   let's ask a question. What are your hours?  And let's see what it says. Right up on top,  

it returns the open hours, and it also asks  me if I would like this information emailed   to me. Perfect. So right down below, let's type in  yes. And just like that, I received the answer in   an email message, just like I was expecting. Back  within Copilot Studio, over on the left-hand side,  

we have what's called the Activity Map. This  shows us step-by-step how Chipbot or your   agent is processing the conversation. Right up on  top, I can see the knowledge that it referenced   in its response. Then I said that I wanted the  information emailed to me. So here it called the  

Send an Email tool, and here we see what values  it filled in into the email. Then at the very end,   it completed its response, just like I expected it  to. Now that we've given our agent its first tool,   let's see if we could have it work autonomously  or without us needing to be involved. To do that,  

we're going to add what's called a trigger. Up  on top, let's click back into the overview. Back   within the overview, let's scroll down the page  until we see a section called Triggers. Triggers   let your agent get to work or basically activate  when something specific happens, like let's say a  

new email arrives or maybe someone submits a form.  Let's click on Add a Trigger. Here, we see some of   the featured examples. Right down here, when a  row is added, modified, or deleted, maybe a new   customer order is added and we want an agent to  take some action. We could do that. Over on the  

right-hand side, when a new response is submitted  via Microsoft Forms, you can then have your agent   take some action. Now I'd like to respond to  an email message, and right down here, I see an   option that says when a new email arrives. Let's  click on this one. Then down below, let's click on  

Next. On this screen, you can give your trigger  a name. I'll stick with the default. Then down   below, you'll also have to make sure that you're  signed in to all the different services that are   being leveraged. I've already done that, so at the  very bottom, let's click on Next. On this screen,  

we can set up filters for this trigger. So, does  the email have to be in a certain folder? Should   it come from a certain person? Right down below,  should the subject include certain words? I want   this to trigger any time I receive an email in  my inbox. My inbox is purely meant for customer  

questions, so I would like the agent to respond  to all incoming email. So, for now, I'm going to   leave all of these set as is, but remember, you  can go through and you could customize all these   different values. Now, down at the very bottom,  let's click on Create Trigger. It has successfully  

added the trigger, so let's now test it out. Over  here, let's click on Close. Before we can test a   new trigger, we need to remove the instruction  where we ask the customer whether we should send   the email. I simply want it to respond to all  emails. To edit the instructions, let's go back  

to the Details section and then click on Edit.  Let's scroll down just a little bit, and here   we see those general instructions. Let's remove  this step, and then once we do that, go back to   the top and click on Save. Now, let's scroll down  the page, and right within the Trigger section,  

we have the option to test the trigger. Let's  click on this. Now, I need to trigger the event,   so I'll send an email to the inbox. Here, I've  composed a message. Let's send that to the new   ChipBot. Back within Copilot Studio, it looks like  it detected the email message with the question.  

Let's now click on Start Testing. Back within  my email, here, I have a response from ChipBot.   That's pretty cool. Back within Copilot Studio,  you can also set it up so your agent can hand   off to or collaborate with other AI agents, but  overall, I'm satisfied with the way ChipBot turned  

out. It's trained, it's connected, I think it's  ready to go, and now it's time for the final step,   publishing. In the top right-hand corner, let's  click on Publish. Here, I get some additional   information. This all looks good. I'll click on  Publish, and just like that, ChipBot is now live,  

and it'll respond to any email that I receive  in my inbox. Of course, you can also make it so   people can access your agent elsewhere. Right  at the very top, let's click into Channels.   With channels, you can define where people can  interact with your agent. Here's an example.  

You have Teams, SharePoint, a web app, native  app. You even have third-party chat services   like Slack and Facebook Messenger. To activate  any one of these channels, simply click on it,   and then run through the different connection  steps. Before we wrap up, let's take a quick  

look at how to track your agent's performance,  because even AI needs a little bit of supervision. In Copilot Studio, at the very top, let's click  on the Activity tab. Here, you'll see a history of   recent interactions. You could even click into any  one of them to see exactly what the user asked,  

how your agent responded, which tools or knowledge  sources it used, and if anything went wrong. It's   kind of like a flight recorder for your bot.  It's great for debugging, refining answers,   or maybe just being nosy. Next, let's head over  to the Analytics tab. Right up on top, let's click  

here. This gives you a bird's eye view of how  your agent is doing over time. You'll see things   like the total sessions, usage, trigger counts,  and how often it couldn't find an answer. Now,   if you notice common questions going unanswered,  that's a perfect cue to add new content or tweak  

your knowledge base. So yeah, ChipBots working  24-7, but you still get to be the boss. And   that's a wrap. ChipBot is up and running. And  honestly, it might be the most helpful member   of the Kevin Cookie Company team, mainly because  it doesn't eat the inventory. If you enjoyed this,  

please consider subscribing. If you have questions  or ideas for future videos, drop them in the   comments. Now, if you'll excuse me, I need to  debug why ChipBot is recommending oatmeal raisin   as the best cookie that we offer here. I mean,  we all know that it's double chocolate chunk.