# Build your own AI Agent in Copilot Agent Builder (October 2026 UPDATE)

Channel: Lisa Crosbie
Video: https://www.youtube.com/watch?v=21SVXuwK50g
Duration: 28 min
Language: English
Words: 5267
Transcript page: https://viewrankai.com/tools/youtube-transcript/21SVXuwK50g

---

[0:00] In this video, I'm going to help you understand what Agent Builder is inside Microsoft Copilot, and we're going to go through a tutorial with the use case of building an agent that can act as your thought partner. Now, we're working here in a business context. Some of you will be working with the basic or free license that's included with your work subscription to Microsoft 365, and some of you will have the paid Microsoft Copilot license that sits on top of that work account. Either way, I've got you covered here because the fundamental capability is actually the same thing, but with the paid license, you get some

[0:36] extra stuff. So, I'm going to take you through the tutorial that will work for everyone, and then we're going to call out the point where you need the paid license to do more and make that comparison at the end to help you understand. Let's build a thought partner agent together. Now, before you get concerned about the word agent and that this is going to get technical, it's really not. This is designed for everyday business users. We're going to create something that's quite capable, but it's all just by describing what you

[1:04] want. So, this is for you. Now, first up, the concept of Agent Builder inside Microsoft Copilot because we use the word agent to mean so many different things. A few years ago, everything was an app. Now, everything's an agent. Now, the word agent here is being used for something really quite straightforward. Agents are also things that can run entire processes and do complex things, but like I said, this is our no-code everyday knowledge worker

[1:31] zone. So, the Agent Builder inside Microsoft Copilot is essentially a way to narrow down your experience. We're going to be talking about building an agent, but realistically, this is narrowing down Copilot. So, if you are on the free version, you're going in and chatting and getting responses all the time. The way that we're narrowing this down is by

[1:53] providing it with a set of instructions. So, we're building here a thought-partner agent that is given a set of instructions about here are different sort of thought processes like six hats thinking or five whys thinking, all of those different types of things. And when I call upon you, thought-partner agent, then you're going to understand the task that you exist for and you're going to do it in the way that I've asked without having to write

[2:20] a prompt to do that every time. Now, if you're on the paid version, the premium version, same same. The difference here though is that your version of Copilot when you're on the paid license has access to all of your data at work. That's essentially the main thing, the value of what you're paying for, this work IQ layer. It has access to all of your Teams chats, your emails, your uh what else we got? All the documents, everything in SharePoint, OneDrive, all of these things, like this very, very rich contextual data. So, when you're building an agent, you are narrowing down the instructions and the task at hand the same way we just

[2:59] described for the people on the basic version, but also narrowing down the data that the agent has access to. So, if you're writing a prompt in Copilot and it's got access to all of this work IQ, everything you've ever done, it's a lot. You have to go, "No, I just mean that document or those documents or only check my emails or don't check my emails." So, this agent also allows you to narrow down the instructions and the task of how you want it to behave and what you want it to do, as well as what data it has access to when you're on

[3:31] that paid premium version. Enough [snorts] of the concepts, let's get into it. I'm starting here in Copilot, and what we're going to do is go across to the agents menu. If the navigation has changed since I recorded this, have a look in the description and I'll tell you where that's gone now. And we're going to go across here and select new agent. Now, first thing I'm going to do is show you the uh template agents both in the free version and in the paid version because that really helps you with the concept. So, starting here in the free version, now remember if you're on the paid version, you can do

[4:05] everything and more. So, this is this part is for everyone. So, what we've got here is you start by describing it in natural language, but I can come in here and have a look at the templates. So, we've got templates here that can help you with various different types of things. One of the good ones to have a look look at is this prompt coach because writing better prompts is a useful skill that we've all got. If you go into your main chat in Copilot and say help me write a better prompt, it can certainly just critique the prompt for you, but you'll see here that this

[4:38] prompt coach, like we said, it's a narrow down task. So, its task is to help you write good prompts using the GCSE goal context source expectations framework. And you'll see that it's got all of these different things and a description of what it is that the agent exists for. Now, this is sort of at its most basic. An agent can just be a set of instructions that says, "This is your special task. This is what you're here

[5:06] to do." We're just going to click create because these ones that are the template versions, you can take them or leave them. And so, now if I go into start chat, you'll find that comes up over the side here. If you don't see it, go back into your agents menu and pin it and now you'll find it on the side there. So, if I go into the main chat and just say help me with a prompt, it will certainly do that, but if I go in here, I give it my prompt, and it will actually come back with this framework. Remember, because this agent exists to be able to

[5:42] give you that specific framework. So, it's looking for goal, audience, contact uh goal, audience, context, sources, and here's the information prompt. So, it's working in a very structured way. So, with the free version, this is one of the easy things you can do. This one's good regardless of which version you're on. Let's take a look at the paid version by contrast. If I go in and do the same thing here, you'll notice that the templates are a little bit different. The difference with the paid license with the premium license is that it is connected to all of your work. So, the examples here are things like plan

[6:17] my day, get a status update, executive briefing. Let's go in and take a look at one of these. Because with this license, it can read your emails, read your documents, read your Teams chat. So, this is an executive briefing agent. Again, it's got these different sort of purpose and steps and what we want the output to look like. So, we've got those pieces in there. But, this is the difference here with the paid license, the premium license, is that it is working across all of these different things. Now, let's say I wanted to create an agent that only reads my Teams chats and I don't want it to look at the

[6:55] web, then you can narrow that down. You can also come in here and say I want this just to be certain files. So, this is the difference. This part not available if you're not on the paid license. So, in both cases, let's not create that one. In both cases, the agent builder is this narrow down set of instructions with option for knowledge if you're on the premium license. So, what we're going to do now is show you how to create this thinking partner agent. And I'm going to start from the beginning over here in the basic one and then we're going to switch across. So, what you need to do is come across here

[7:33] and say agent, new agent, and then describe in natural language what you want the agent to be. I'll put these prompts in the description so that you can copy and paste them. Create a thinking partner agent that helps me work through problems, decisions, and ideas. Let's let that run. Now, if you are familiar with building agents and you've done this a lot of times before, you can actually skip this step and get straight to the next part and fill the form from scratch yourself. However, this part is very good, regardless of which license you're on, at kind of taking your concept and turning it into a bunch of things. So, it is a much

[8:10] easier starting point even if you know what you're doing. So, here we go. It's drafted the thinking partner, so it's called it that. We have a purpose, general guidelines, so it's asking focused questions, step-by-step instructions, frame the issue, interaction patterns. Okay. So, so far so good. Let's take a look at a couple of other things we can do here. We're going to have a look at skills. If you're not familiar with the idea of

[8:35] skills, I've got a whole video on there. Have a look at that and come back. Skills are like a set of instructions. It's like a standard operating procedure or a recipe for the agent. So, you'll see here in the description that we have got, for instance here, step-by-step instructions, but what we're going to do next is give it a whole lot more specific things. So, what I'm going to do now is describe something else. So, you can edit this, but actually, the easiest way to edit is to come over here and give it another message on the side. I want to the agent to give the user options to

[9:13] use different thought models. Create a skill for each of these. So, six thinking hats, SWOT and uh I'm going to press enter and try and find my words. Six thinking hats, SWOT analysis, pre-mortem, and five whys. These are all very standard things that the AI models will know the methodologies behind these. Now, I've explicitly said create

[9:36] a skill because I want it to work here. Even if I hadn't said that, it would probably have decided to take that route anyway. The skill is for when you want the agent to follow a methodology, a procedure, your way of working. So, with all of these particular things, they have specific methodologies that I want the

[9:57] agent to work through. I'm also implying without saying it that I want the agent to update the instructions so that it knows when to call each skill. So, instead of having a great big long list of instructions here where you say you are a thought partner, if the user asks for this kind of thing, you should use six thinking hats, and

[10:17] then you should do this, this, and this. And if the user asks for SWOT analysis, because then it has to sort of process and think about those instructions every single time. What this has done is put in a skill for each of those, and hopefully here, choosing a thought model. So, it's updated those instructions. If I explicitly ask for one of those thought models, then use it. If I haven't chosen one, recommend it based on which model

[10:47] is the right job for the right task. Now, this example works really well in either version of the agent builder because this is using something that's quite well known. It's on the internet. It's known to the language model. But you can do these things with something that's very specific to your business as well. Let's take a look at what the skill actually is and then jump into the next one. So, let's take a look at the six thinking hats skill. So, this just

[11:14] looks like another set of instructions. It's like a subset of instructions, if you like. It has a purpose, and that purpose helps the top level of the agent understand when to call this skill. And now we've got the six different colored hats, if you're familiar with this, and the output and what it should be. Again, the more information you give this, the better it's going to be. I've just kind of thrown bullet points at the language model and it's figured it out. Let's

[11:42] navigate back here. Now, if I have something that is a specific thing that we do in our organization. So, I'm going to come in here and say create another skill called assumption stress test. So, this is another thing here where we say, I'm giving it my own instructions. Again, this is not terribly original, but you get the idea. So, something like when I

[12:06] say six hats thinking, that's out there. Now, I'm actually giving it my own instructions and saying create a skill for that. Skills are a really useful thing to get your head around. As I said, pop over to my other video if you want to kind of really go on a deeper dive of this. But, you'll see here that we're starting now to create an agent that doesn't just do one thing. It's one job is to be my thinking partner, but I'm asking it to help me think in now five different ways and depending on what problem I give it or what model I ask it to use as a

[12:38] thought partner, it then goes off and kind of follows that little set of instructions. Think of it as like a person who's been given five different instruction manuals. And so, now we have got the assumption stress test in there and then it's actually kind of updated those instructions as well. So, this is you can see why I'm

[12:56] recommending editing over the side here. You can certainly edit directly in the text here if something isn't quite right, but when you're doing the natural language chat, it's actually quite sophisticated about putting the right thing in there. Now, there are different language models that we can use. So, while I'm over here in my chat, I'm using the auto model. I'm doing something that doesn't really require heavy thinking in the building of [clears throat] it, but you could

[13:23] actually also could different models. So, I could choose a think deeper model because this is what's going on over here. I have got also within the thinking partner itself, when I build this agent, I actually want it to think deeper in response to my user question. So, this model is choosing the experience I'm running through to build the agent, but this one is choosing what the agent will do. So, I'm going to choose a think deeper model for the agent. So, every time somebody uses this one, it will default into kind of using that more complex reasoning way of thinking and to, as the says on the label, think

[14:05] deeper. Now, let's just scroll down a little bit further here. You will see I'm still on the basic version. Everything that we've done so far equal on on whether you've got a a paid license or not, all the same. This is where it gets different. While I'm working here in the basic license, in the knowledge, remember before I saw all those options for emails and documents and things, you have to have the paid premium license to have access to that part. If you are in the basic license though, you can choose to say add knowledge. So, the first option here you've got is the web search. So, do you want this to use the

[14:38] public web or not for this one? Yeah, maybe thinking through brainstorming things, that's a useful thing. Or I can come in here and add knowledge, which is a link to something. There's very little going on here. You'll see I can't kind of upload documents or do anything else. The knowledge settings, it will be able to create documents and charts. So, if I wanted at the end of a conversation to say turn that into a word document, we can do that. I can choose to create images, and this one is less relevant here. Discourage model knowledge, only use the selected sources. We don't really have much in

[15:15] the way of selected sources here. So, you can toggle those options on or off. And then we have our suggested prompts, and these have been determined based on the conversation that we're having with the model. What are the different things here? Here are the different messages. These become sort of quick buttons that you can press. You can choose to delete, reorder, add more. You can have between zero and six. Typically, if you've done the work over here, you'll end up with a bunch of things that look sensible. So, what we're going to do now is create

[15:45] that one. So, we'll switch across now into what this looks like with the premium license so that you can understand the difference. And if you like, you can play spot the difference with me on this screen here. So, first of all, you'll notice we've got a configure, preview, and monitor. On the last screen, we only had configure and preview. So, monitor actually allows you to see what's going on. I built a version of this as a testing thing beforehand, so you can see that we've got the sort of duration and the sessions and what's going on and which knowledge sources we used because remember, that's core to what we've got

[16:20] going on here. But, you'll see that the ability to say "Think deeper" is exactly the same. We've got the instructions and the purpose. We've got all of the same skills. I went through exactly the same process and got to exactly the same state. But, here's where we've got this big difference in the knowledge. I can say I want this to have access to all of my cloud files, everything in Outlook, everything in Teams, everything on the web. If I've got connectors to third-party business systems that have

[16:47] been set up by my admin, I can do that. You connect to Dataverse if you're using Dynamics and Power Platform in there as well. So, there's a lot It's not just Microsoft things. If your admin has set those things up, you can connect to those third-party sources. You can also choose to upload files. If you've got something that's sitting on your desktop and you want to add it to the agent, you can do that. You also have more options here in the settings. So, we've got the same things we had before, but there is also this option for people data, which understands the org chart of the people

[17:20] that you work with because it has access to all of that as well. So, if it is a use case in the premium version where it is something to do with your collaboration or the hierarchy in which you work, then you can also switch that one on. So, what this means, and I'll show you the results, is that the fundamental principle of I've built an agent that's a thinking partner. It understands different models of thinking and it has these skills where it's going to be able to choose the right skill and help me think in the right way. You can do that with both the free license and

[17:52] the paid license. However, thinking in the context of everything you do at work is what you're getting with the Work IQ layer that only comes with the premium license, so adding knowledge to it. So, as you're thinking about use cases here, if you're on the basic version, these kinds of things are awesome. The thinking partner agent can actually work really well as an agent that doesn't need to be connected to anything, but if it is connected to everything, it knows my entire context, and if I'm asking it to solve work-related problems, it's got access to all of those things. So, I'll show you the difference there. Suggested

[18:27] prompts exactly the same, and then this ability to go back and monitor and see how it's going is also part of the premium license. And what I'm going to do now is choose to um choose to run something here. So, the first thing I'm going to do is pretend that I'm going to start a blog. Should I start a blog? I am actually genuinely thinking about starting a blog as if I don't have enough to do. So, it

[18:52] your agent is updated successfully. Okay, great. Start chat. So, we're going to switch back across to our friend over here who's on the basic license. And the first thing I'm going to do is run, so you'll see these are the suggested prompts that come up. We're going to run a pre-mortem. And the first thing I'm going to do is to say, I started a blog on making sense of AI work and let's run this. Now, this version with the basic license doesn't have access to my emails, doesn't have access to my documents. So, it's just running based on the instructions and it's going to, you know, go through all these

[19:26] different things. So, it's saying you've started a blog, I'll assume a 6-month goal, it's 2027. Here's my pre-mortem. So, you'll see it's going through the entire thought process, which is which is really cool. And let's see what Let's see what recommendation it it comes back with. Let me know in the chat if you think I should start a blog cuz honestly, the human the humans will be the one that I that I hope read it, not the AI. And maybe depending on when you're watching this um this pre-mortem will be irrelevant. All right. So, we have now got the final result here. So, it is giving sort of actions to take now,

[20:02] create these things. So, it's you know, like it's actually incredibly useful. Let's take a look at the difference of what happens when I do the same thing in the premium license. Now, over here, I have got this one connected to a whole bunch of stuff. I've got some things in my OneNote where I'm brainstorming this idea. I have got some kind of memories and context that I fed to Copilot along

[20:24] the way about what I'm thinking about. It knows me, knows my content creation world. So, there's depth here. Now, this is my tenant that I use for demos and for content creation. If I did this on my real work tenant, I've got years and years worth of collaborative data and proposals and all sorts of things. So, I would get something even more, but not for this particular ask cuz it's not related to my work. So, the more of that work IQ layer you've got when it's your real work scenario, you can go quite deep here. But, you'll see straight away it's finding 99 relevant internal files,

[20:56] several emails, seven meetings. So, it's pulling context from all of those different things. You will see that I've actually got something here where I was thinking about the positioning of stuff that's coming from my OneNote. This is primarily in OneNote where I've been working with this, but you'll find there are different sources along the way depending on what's happened. If you go all the way down to the bottom on this one here, it's actually coming up

[21:22] with a series of things. It's giving me kind of very specific feedback based on the brainstorming that I've done, not just general things from the AI model, and I do still have a further thinking to figure this out. But you'll see it's got a bunch of different sort of thoughts here, and personalization will show you the kind of specific goals and ambitions and things that I've asked Copilot to remember, but I might just keep that part between me and Copilot if we don't

[21:50] mind. Let's take a look at another example here in the free version, not connected to anything. Help me explore a difficult decision using six thinking hats. Keep each perspective distinct. Should I be creating content on Claude? And so, now it's going through the thought process. It's called that skill. I'll assume your goal is to decide on this. Here's the white and what we know about it. So just we'll have a look at the white hat in both. What we know, this is what's going on. If I go across into the version that's connected to all of my data at work, you are deciding whether it's part

[22:22] of your creator strategy. It's got all of that in there. This is the tension about what I'm doing, meetings and files. This is linking to those emails and PowerPoint decks that I've got and all of these different things that are in there. So I hope that helps you understand the difference between those things. In the end, you've got this capability to create agents that narrow down your experience and focus on a specific task, but if you are in the premium license, you can think about those tasks as things that go much deeper into your actual work-related world. Now, if you want to edit and share these things,

[23:00] let's come across over here where you've got it on the side or you can find it again in that main agents menu and you're going to click on the three dots for more and you'll see we've got options here. You can click edit. So this is usually an iterative process. It's rare that you just follow the instructions once and do it although by all means grab these instructions and have fun with it cuz you probably will go quite far, but if you're doing something where you're being a bit creative or you're experimenting or trying your own ideas and I really encourage you to do that cuz you can

[23:32] just uninstall these things if they're no good anyway. Go ahead and come back in and you can choose to edit. So you can chat with it on the side or you can edit things in here. So as you try it out if the responses are coming back in a way that you don't like the tone or the style or the way it's handling it, come back in and do that iteration, click the update button and have another go at it. Now the other thing here is about sharing and expanding on these things. So let's have a look at sharing first. First of all, let's switch back into our friend

[24:04] over here who does not have a premium license and we're going to go back into the edit same experience. Now this person can come in here and they do have the option to share. The sharing from here, if you've got a free license, you're just basically sharing the set of instructions and you can copy the link and the link goes out to

[24:26] people. If you've created something that is using your premium license, you won't be able to share it with people who don't have a premium license unless your organization has set up pay as you go. So you can't be paying for something and sharing it with people who aren't paying for it is a general principle of the licensing here, but if you create say an agent that asks questions about the HR leave policy or something and the leave policy is in SharePoint, you want to share that with everyone, then your organization can set up a pay-as-you-go license so that each time the person who doesn't have the paid license uses it,

[25:04] they just pay for that usage. Just be careful with that because, to be honest, the paid seat license, which is the US $30 per user per month, has so much value in it now, like so much value, that if you are only building one or two agents and the person just asks a couple of questions every now and again, pay-as-you-go is probably fine. But, if you're distributing an agent that goes everywhere and people are using it constantly, the balance of when it's worth just buying them a license to be able to do unlimited and all sorts of other things, just keep an eye on that balance. It's really only worth paying

[25:42] for the pay-as-you-go if it's an infrequent, like if it's for a lot of people who don't have any other reason to have a license and they're using it infrequently. In almost any other circumstance, that value [clears throat] balance is going to tip pretty heavily towards considering the paid license for those use cases. Now, the other thing you can do here, obviously sharing with the organization depends on the permissions that have been set up. Are you allowed to sort of share across the whole organization or you're adding a group or an email in there? The sharing here is basically just copying a link and sending the link

[26:16] to somebody else or adding it into the agent store, which is sort of in this agent catalog. The other option you've got here is copy to Copilot Studio. Now, Copilot Studio is a whole other box and dice. I've got videos on that here as well. Copilot Studio is a low-code maker platform that is for people who have a higher level of skill to create things that are more

[26:42] than everyday personal productivity. We're starting to talk external-facing agents, agents that automate and replace entire business process flows. That's not necessarily for everyone, but the handoff point here is if you need your agent to do more than anything on the screen. So, what we've got here is the ability to create something that has instructions, that has skills, that has knowledge if you're in the paid licenses and basic monitoring, but it lives, breathes, and exists inside this Microsoft Copilot world. If you want to be creating agents that can live on other surfaces and be published elsewhere including externally, that can be run on a schedule, that can have

[27:27] actions to actually do things like send emails or post data to other business systems or retrieve data from other business systems and do more complex things, that's Copilot Studio, and that's where you need to be. But, if you start here and you like the idea and you want to grow it up into there, then

[27:45] you've got that. So, that's Agent Builder, free version and paid version all in one go. Thank you for watching, and I'll see you in the next video.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt