AI webinar series: Copilot in your business

Catch up on our Copilot webinar

In the third of our AI webinar series, we take a whirlwind deep dive into Copilot with the help of our friends at Pax8 - subject matter experts on all thing Copilot!

The topics we cover are:

• What Copilot agents are
• Writing better agent prompts
• Grounding agents in business data
• Automating workflows with tools and triggers
• Copilot Studio and Co-work demos
  
  

Transcript

Stacey Williams   
Thank you so much for coming, everyone. Welcome to our third AI Webinar in our Webinar Series. Today, we have a partner of ours, Pax 8, who are joining to share Copilot and do a bit of a deep dive into some of the amazing things that Copilot can do. So, we have Nigel and Sydney with us, and they're going to take you through various exciting features. Thank you so much for joining. It's great to have everyone here. As ever, we will circulate the recording after the Webinar. So over to Sydney and Nigel. 

Nigel Dangare   
Thanks, Stacey. Pleasure to be here. Really nice to meet you guys all virtually and welcome to our short but succinct Copilot in your business deep dive. So yeah, again, thank you for the introduction, Stacey. But I'd like to introduce you to me and my colleague, Sydney. So Sydney is our great solutions engineer here. She specialises in Copilot, just like myself. So she'll be here to sort of support from a questions perspective. If you guys do have any questions, we are happy to do sort of a Q&A. So any questions you do have, please drop them into the chat. If I have the capability to answer them live, I will do. If not, Sydney will look to get back to you as soon as she can. But in terms of me, my role is a solutions consultant. My background is modern workplace. And yeah, hopefully today we can kind of answer most questions you'll have and hopefully you will leave with more questions, which is always a good thing. But what is the agenda? What are we digging into? What does that look like? So we've gone through a little bit of an introduction, just sort of short and sweet, as we don't have much time today, unfortunately. But we're going to dig into agents. We'll then dig into a little bit of co-work and its capabilities, and then live demo. And then if time permits, we'll obviously open that up for Q&A. But as I've sort of said, please drop your questions in the chat so we can kind of go through Q&A as and when we're sort of discussing things throughout the chat. And if you do have any questions, then I'll be more than happy to answer them. So first things first, agents. What are agents? How do agents work? We're kind of going to dig into that in a little bit more detail, but essentially agents are helpful assistants inside Copilot, right? That's probably the biggest or the easiest way to explain it. They're not autonomous agents, so they don't have the capability to sort your washing out and do all the things that you might want someone to do. They are obviously set instructions that you will offer to this agent to do tasks for you in some ways automatically, in some ways not so automatically. So in essence, the agents can gather information, answer questions quickly and potentially look to complete workflows. That will be dependent if you have those workflows sitting in the background. But what we're essentially saying is it's a new way for you to look to complete work. I've actually created a little bit of an agent around optimizing some of my business processes and I'll kind of show you what that looks like, what the prompt for that would look like. And yeah, hopefully you guys will get an inclination of what or how to use your agents, right? But essentially, we're outsourcing our results. What we're trying to do is we're trying to take away the burden of the heavy things that we need to do, streamline that into an agent, and then essentially look to gain efficiencies from gaining faster results through using agents and things such as like. So, how do we use these agents? What are the agents supposed to sort of reference and how do we sort of create the right format for these agents? Well, that all comes down to what I would like to classify as the agent instructions, right? So these are really, really important and I'm sure Sydney will attest to this as well. Nothing will happen in a good way for you if you do not get a good grasp on prompting. So it sounds so basic. It sounds something like a child would say, but to be quite frank with you, a lot of the issues that people do tend to have with creating agents actually comes down to the instructions that they provide them. So the instructions usually are the prompt, and with the prompt, it's always good to be as specific as possible. Don't leave room for any error. Don't leave room for open-ended suggestions. Give it clear, decisive instructions as to what it can and can't do. Tell it where it shouldn't be and should be looking. And I can guarantee you, you will start to see the quality of your responses improve maybe 10 to 15%. If you start to add in certain elements like what's on the slide here, what you'll gradually see is that you will massively decrease the erroring that you will see within your environment. So once we've kind of got a grasp on instructions and prompting to be exact, then it's also about tying in the knowledge, right? So knowledge is a really important aspect of this. This is essentially the repository of data that your agent will look to fall back upon. Now, in most agents, that could be a SharePoint site, that could be a document library, or it could be a file, right? But in some cases, you can also look to send that out to the internet as well. So dependent on the knowledge is also going to be dependent on the types of responses that you're looking to get. A good case example is having worked with lawyer firms in the past, what they like to do is they will have some sort of off-site file server somewhere, and they will point their agent directly at that knowledge base. And the key around that is they want to ground their responses specifically in the data that they already have, as opposed to it going out to the internet to essentially say for example, tell me all of the information on this most recent court case. If it goes out to the internet, it may potentially pull back false information. So that's a key sort of thing to recognize, which is that sort of knowledge base, right? So once we've kind of got a grasp on that knowledge base, we then move into capabilities. So capabilities is essentially the ability for it to start to generate things. So maybe in some cases, the agent that you're creating could be a marketing agent, right? So if it is a marketing agent, what you would like it to do is potentially create images off the back of that, create potential source information for you if you're a coder. So essentially what you're giving it the capability to do is to go off and create things on your behalf, right? So the other side of this is when people are using the agent, how do we streamline and make their responses more efficient? Well, you could actually think of the end user as well in this capacity. And what that would look like is creating streamlined or maybe suggested prompts. And what this means is if we refer to the marketing agent, maybe when the person opens the agent on the banner, they might just see actually it gives them the option to say, would you like to create a document or maybe research a certain topic. And what they can do is they can click that suggested prompt, and they can fill in that information, which again helps to streamline the way that they work, which also decreases the time they have to spend going around searching, potentially questioning what is the right prompt. So, you also do have the capability to suggest prompts for the end users now. We get into three specific areas that once you have got the good grasp on those four things, this is where you start to extend the functionality of your agents, how they work, how they potentially correlate with other agents in your environment, and potentially how they start to increase your current efficiencies internally. So tools is a great one. This is essentially the ability for it to start to go off and work with other applications, right? So in some cases, you might give it the ability to interact with your office applications, or for example, you may have third party systems like a zero, right, or a Sage for a good example. So if you guys use Sage, you could actually give the agent functionality to go off and interact with that external system. What that is doing is that's decreasing the time that you would have to spend logging in, looking to find that information, and then pulling it back. passing it and then putting it directly into a prompt, all of a sudden now you can essentially have your agent speaking directly with that external system, collating and bringing back that information for you to then go off and start to do your other tasks within your agent. Now once we've been able to sort of expand the tool side of things, there's something known as triggers, right? So essentially what we're saying is that when something happens, it triggers an action. So if I refer back to that marketing agent, we could say every time we get a new request in or a service ticket to start a marketing campaign, the trigger could be that it goes off in the background and creates a new SharePoint site, right? So that site could be directly linked to this marketing campaign. That trigger will essentially create all of that information in the background, helping you guys centralize your information. But if you take a step back, what you're essentially doing is you're taking one aspect of your workflow out, automating that based upon a response or an action that you would insinuate within the agent, and then essentially you're now starting to decrease the amount of time it would take for you to complete your actions. And then last but not least, you then have the capability for agents to work with agents, right? And that is probably the sort of further end of this capability, where I would say once you've got a real good grasp on instructions, knowledge, and capabilities, then maybe start to look at suggested prompts. add your tools in always, because that's always going to be beneficial to you. But then with the triggers and the agents, that's really when you want to start to maybe orchestrate some of your workflows. Now we're talking about automation as opposed to AI, which is a different conversation that I want to make you guys aware of. But it's always good to understand that you do have the capability. ability for agents to work with other agents. So in terms of the marketing agent, you could create a child agent that basically goes off and creates documents for you. And you could have three or four of those agents. One creates a PowerPoint document, one creates A Word document, one creates a repository to store the information. And all of a sudden now it's essentially trickling down all the way to the bottom agent, where, for example, you have your top agent, you speak with the top agent, the top agent then delegates specific tasks that all responds back into that top agent, and then you just sense check, validate that, and if that is how you want it to be, you would then take that work and go off and do that. So, in terms of agents, that's just a little bit of background. I know I've gone through that really fast, but with time, I want to make sure that we're as efficient as possible and we can get everything in. So if you do have any questions, please do raise them. But if not, I'm going to move into the next section, which is just a little bit of a high level on Copilot, co-work and agents and then how you can start to take advantage of that. through our live demo. So in terms of all of that information, conscious that I am firing a lot at you, what am I actually saying to you, right? Well, what I'm trying to say is that you are now creating your own AI team and how you're doing that is through a combination of three things. It's through Copilot, it's through agents, and it's also through co-work, right? So within Copilot, you do already have some predefined agents, like the researcher, like the analyst, and some others that are already there that you can look to utilize. But it's now time for you to consider what is it that you need to streamline. Maybe you need a research specialist, maybe you need a strategist, or maybe you need an HR advisor, right? So if you had the capability to create these agents, what are you actually going to do with that, and how are you going to look to utilize those agents? internally. Now, in some cases, it's always good to make you aware that not everything requires to be an agent. So some of the things that you think agents require to be is actually just something you can do off your own back, and it's not going to save you time in the long run. So don't focus on the little things that are potentially problems, focus on the bigger workflows and pick out key strategic elements of that workflow to say, actually, if I was able to offload this, how much time would that actually save me? That's where people really start to take the onus and that's where they really start to get the most benefit out of Copilot and agents. Now, that's not to say that you can't go off and create your agents. Please go off and do that. Play around, test, see what works for you, as my recommendation might not be your truth. So it's always good to play around and see what works in your environment, as there may be things that are small to you that actually, if you were able to streamline, saves you 20, 30 minutes per day, which in the long run is how you start to get your efficiencies. As opposed to the bigger things, they take a little bit more time to scope out. So you might not be able to create the agent straight away. But as you tend to move into your agent building skills, you'll start to realize how to make those processes more efficient and where you can look to streamline. So that's just a consideration for you guys to take away and potentially look to question, where can I start to use agents? Now with Microsoft's global release of Cowork, this has actually drastically changed the way that you guys are going to start to work with agents. And what do I mean by this? Well, essentially, Co-Work is a team of specialists, right? So what it has the capability to go off and do is create, research, strategize, write, even provide you advice based upon the intelligence layer. So there will be a terminology that you're probably going to see quite often, which is known as work IQ. That is the intelligence layer that I'm referring to. Now for the technical people on the call, what we're essentially saying is that it's indexing within your environment using Graph API. which essentially for the non-techie people, what that means is we essentially have the capabilities for it to view your account, to see what kind of documents you access on a daily basis, see the way that you work, and essentially take that information back and apply it at the most relevant state. So for example, in this scenario, if I asked it to help me launch a new managed security service, what it would do is it would research my OneDrive, my SharePoint, look through my emails, and essentially look to collate all of the information around anything managed service-y, if that's even a word, that could help in creating this. It will then go off, do the research elements. So take a look at market trends, again, making sure to tie back into anything that I may have mentioned. Take a look at the GTM plan, potentially look to build a revenue model and then a marketing campaign. And then it will look to produce that content to me in the form of a document or however I would ask that to be. So if that's a PowerPoint or something else. So what we're saying is that through one specific prompt, you can actually centralize a lot of the things that you do just by using Cowork. Now, I did actually prepare a demo for you guys, and I'm just going to bring my screen up. So, I was able to go into Cowork earlier this afternoon, and as you can see, this is what Cowork looks like. So, essentially... When you have access to your M365 screen, you're essentially going to see two toggles on the left side. So you've got your chat, which is your internal aspect, and then you have your co-work aspect. So within the co-work, you do have my task scheduled and other elements, but you can essentially look to prompt directly into this section here. And as you can see here, I've created a pretty basic prompt, which is around cakes. The reason why is me and my colleague Sydney love to talk about cakes. So that's the sort of analogy that we have for you today. So what you can see straight off the rip is that it's actually going off to, prepare to work with Word. So it's recognized that I want a Word document. So in the background, it's starting to think. And in another word, we would call thinking reasoning. So with the reasoning aspect, that's where the tokenization starts to become quite a lot. So what I mean by that is if you give it multiple things to do, it's going to have multiple reasoning steps. Hence, it's going to consume more credits. So with this, this isn't a very heavy prompt, but as you can see, it's going to require it to go off and do research, to verify that that research is correct, to actually prepare that document, to then validate that document is correct, and then paste that document. Back to me, and we can see all the steps that it looks to go through right on the right hand side, and it will also look to show you any skills or plugins that it's looking to utilize. So, in this case, we can clearly see that it's looking to utilize Word, and as this is quite a small prompt, this actually finished. in about 3 minutes, which I would like to say is on the faster side. I have seen this take anywhere from 10 minutes to maybe 25 minutes, depending on the type of prompts that you run. So what I would say is take it with a pinch of salt. Not everything is going to happen as fast as you would like it. However, in most cases, it usually does look to get it right the first time, as long as you provide it with clear context, clear instructions, and essentially confidence gating. So confidence gating is what we like to call narrowing, and that is around how you would like it to complete the task, what it can and can't do, and essentially the areas that you would like it to look at. if that's knowledge, if that's specific repositories internally, or the kind of way you would like it to present the information. But as you can see, I've made mine very generic. I've not given it any confidence gating at all. So it will go through this prompt quite quickly. So as you can see, it took me probably about, I would like to say, a minute, 2 minutes at most to create that document. It's now just checking that the folder is created and it's saved in the right area. And one thing that I do want to make you guys aware of is that Cowork has this capability. For you to actually forward slash forward slash brings up this option menu where you can essentially pick skills, people, meetings, etc., etc. So you have the capability to go through multiple different areas, but in this aspect I would like it to give me a cost, so I also do run that prompt as well. And it does tell me I've spent 315 credits to actually create this document. And now we actually have the document created. So once it loads, we can see we've got a nice basic researching guide with some information, the researching goals, the ingredients, the essential equipment, the core methods to research and then how to's and other things such as, right? So in this case, not exactly a world beater, but as you can see for the purposes of what I'm trying to do, it was efficient and it was effective in how you look to utilize that. Now, I wanted to actually go into some of the agent side of things as well, as this is probably where you guys are going to look to spend a lot of your time in terms of how to actually build out these agents. What does that look like? Well, this is how you can build the agents. This is within Copilot Studio. You also do have the capability to build agents within your M365 as well. All you would have to do is you just go into your M365, under agents, you can see new agent directly there. You can essentially either pick a template or if you already have an agent, maybe you can pick a pre-defined agent or you can just look to directly prompt into there to go off and configure that. Now, if you wanted to skip that, you do have the capability to skip that and then you can configure it in this portal. Personally, I prefer this portal as it gives me a clear understanding of everything that I'm doing, but this is a personal preference. So that's only something that you should consider. So, if we actually come back to Copilot Studio, I've created an optimizing agent, and the reason why I've created this optimizing agent is I do a lot of things on a daily basis. Sometimes I don't know whether or not I can automate certain aspects of my job, so it's always good to understand if there is a capability to do this. If so, how would I do this? And what tools would I utilize, etc. So that is the purpose of this agent. And what you're probably going to see is a different type of prompt to the ones that you guys will have utilized in the past. So this is using the Risen framework. And my colleague Sydney can sort of drop the description in the chat on what the Risen framework looks like. but essentially it's role, instructions and a couple other things. But really what we're trying to define are key areas. So what it can and can't do, what its instructions are, how that agent works, how it correlates with other agents, what kind of information it pulls back. and the type of response we look to get. So here I've given it some clear instructions. I've given it a clear purpose. I've told it what its direct role is. So it's a best practice reviewer, not a salesperson. So that should remove its recommendations of third party options that aren't relevant to me. And then I've given it clear instructions that it must look to follow. I've also told it to classify things, which is really important. So this tells me exactly what kind of process it is. So maybe in this case, I just need assistance. Maybe it's an actual workflow. Maybe it's got nothing to do with a workflow and it's got something to do with permissions, or maybe it's a process improvement. So what that's clearly doing is it's telling me, actually, you don't need AI, you don't need automation, you just need to go off and improve your process. It goes off and it looks at Microsoft. and it will give me a recommendation based upon my requirements. And then it also looks to use decision logic. So dependent on the tool that it will recommend, it will also look to give me logic off the back of that to say whether or not it thinks this tool would be best. And as we can see, it's also grounded in multiple different areas. So source types, et cetera, et cetera. So if you build build out using that risen framework. Hopefully, your prompts look like this. To be quite frank with you, I build out my prompts starting off prompting in Notepad, and then I actually work with Copilot using the prompt buddy or the prompt agent to help me build these out. I don't write these out from scratch, but I do start with the baseline and then I get Copilot to help me. So that's a quick tip or trick for you guys to utilize. And then we've obviously got the confidence gating and a bunch more other stuff. So on the right hand side, we do have the capability for you to ground your data in the knowledge. So that's what I was referring to a little bit earlier. So if I click this, I have the option to either go off to a SharePoint or a OneDrive or a public website. If I click tools, this is obviously where I can look to pick some of the tools within my current ecosystem, or in some cases, there's a connector for a third party tool that I would like to utilize. So here we can see Adobe. If I wanted to work with Adobe, I would just click access to that. Skills can be predefined. So this is the capability for it to follow certain aspects. So for example, I could upload a skill on how to create PowerPoints. What that would mean is every time I ask it to go off and research something and to create it in a PowerPoint, it's going to look back at the skill and essentially reference the way and the format of how to create that PowerPoint referencing this skill MD file. And then last but not least, the memory IQ of the Microsoft IQ. This is the intelligence layer that I was talking about. So essentially looking at how you work, bringing your work context, your business data, and signals from other areas within your ecosystem to essentially improve your answers and improve the quality of actions that you do get. So now that I've created my agent, well, I could actually go off and start to preview what that looks like. So I actually created a little bit of a shot. prompt earlier and this is around organizing my inbox. Funnily enough, Cowork does have a capability to do this. So I kind of wanted to do some research to see would it recognize that Cowork is the best or would it actually tell me to go off and do something else? So what you can see is those areas that I informed you of it previously in the build section. So it gives me a TLDR. I absolutely love this. This helps streamline all aspects of the response. It then breaks it down into the individual areas. It then tells me the key question or the key answer that I need. Do I need AI? No, in this case. but it could be useful if I required it. And then it looks to give me the three options as to what I would look to do if I was looking to build out some form of a solution, alternatives, confidence rating, and then multiple different areas. Now, with this agent, I do have the capacity to go off and deploy this agent. So if I actually just click back, what you do or can see here, guys, is that my optimizing agent is actually available within Copilot. So this is the M365 version. So once you have created those Copilot agents within Studio, you can deploy those to individual users to then help. with that usage or maybe to extend out certain capabilities that you may have. So I do believe that brings us to time and I am sorry that it was a little bit rushed as there was quite a lot to try and fit in within that time frame. However, hopefully we've kind of gone through quite a few of the areas that you can look to utilise in terms of co-work agents, how that works, a bit of a baseline on frameworks around prompting. So hopefully you guys will be able to go off play and then if you do have any questions, you can forward them to Stacey and if it is required, we can obviously look to set up a conversations with ourselves.

Stacey Williams   
Thank you so much, Nigel. That was really interesting. And I think like the key for me is always at the end of that, you've mentioned like just go and play. And I think a lot of the stuff that I've done with AI as well has just been through trial and error. So I definitely would reiterate that sort of being able to play around with agents and Copilot and different tools is really helpful for just figuring out how it all comes together in the end.