Ep 26 - A prompt dive into prompting

In this episode of Copilot Connection, Kevin goes solo and dives deep into the world of Microsoft Copilot, exploring the intricacies of prompt engineering and the practical applications of prompts in enhancing productivity. He discusses various techniques for effective prompting, the significance of community contributions, and how to utilize the Copilot prompt gallery to streamline workflows. The conversation emphasizes the evolving nature of AI and the need for users to adapt and refine their approaches to maximize the benefits of generative AI tools.

Takeaways

• Prompt engineering is best for effective AI interaction but not essential.
• Understanding agentic threats is essential for security and OWASP can help.
• Community resources can enhance learning and application.
• Utilizing the Copilot prompt gallery can streamline workflows.
• Advanced prompting techniques can lead to better outcomes.
• Regularly revisiting and refining prompts is beneficial.
• AI can assist in complex problem-solving scenarios.
• Diversity of thought is important in AI applications.
• Practical applications of prompts can save time and effort.
• Engagement with the community fosters knowledge sharing.

Useful links

News:

OWASP agentic threats https://www.linkedin.com/posts/jsotiropoulos_aisecurity-agenticai-owasp-activity-7297385436925095937-vQmg

Beginner guide for agents https://www.linkedin.com/posts/donasarkar_happy-valentines-day-friends-as-a-activity-7296237085194084352-db7o

New AI models include a gaming model https://www.geekwire.com/2025/microsoft-reveals-new-gaming-focused-generative-ai-model-muse-that-could-revive-classic-games/

Fabian Williams bring us some agentic magic with CAPs https://www.linkedin.com/posts/fabiangwilliams_autonomous-multi-agent-group-chat-working-activity-7297051837969358848-R4zG

Prompts:

Crafting effective prompts https://techcommunity.microsoft.com/blog/microsoft365copilotblog/getting-the-most-from-the-copilot-prompt-gallery/4383106

Prompt Engineering Guide | Prompt Engineering Guide

Best practices for prompt engineering with the OpenAI API | OpenAI Help Center 

Stefan Bisser’s tool to write prompts https://www.linkedin.com/posts/stephan-bisser_agent-instructor-visual-studio-marketplace-activity-7298292501763780610-OAYW

Prompts of the week pnp/copilot-prompts: Examples of p

Transcript

Kevin McDonnell (00:11)
Welcome to the Copilot Connection. We're here to share with you all the news, insights and capabilities of the Microsoft Copilot ecosystem across the entire Microsoft stack. I'm Kevin McDonnell and I'm the Copilot Strategy and Modern Workplace AI leader at Avanade. We'll be releasing episodes as podcasts and on YouTube with insights from experts from across the community and Microsoft on what the different areas of Copilot are, the impact

They can make you and your organization what you need to do to prepare for them. We'll start implementing them now and even how you can extend them with some beautiful agents. For those eagle-eyed among you, you may have noticed that it's just me today. Unfortunately, Zoe has been caught out with some work. One of joys of Copilot is that you can get more productive. But if you work in the Copilot industry, it's also making a load more work.

which is adding to the fun. So apologies. It's just me here for a solo show. Why am I apologizing? mean, you've got one whole half of an amazing duo coming up. So I'm not going to apologize too much. Now, I've actually already recorded a little bit that I'm going to share in a minute where I decided to do a bit of a dive into prompting.

And you may go, everyone knows prompt engineering now. Why is Kevin's droning on about that? Well, the answer is because there's been kind of updates to the prompt gallery that I don't think people are aware of. this is kind of a time to go. Yeah, when when generative AI first came out, you may have looked into all the different ways. But have you looked recently? Have you looked at how effective your prompting's being? So it decided to dig into some of those things.

how you can use that and how you can get some more effective prompts, primarily for Microsoft 365 co-pilot, but a lot of the other ones such as Power Platform and in fact any generative AI system and especially the agents within there can make use of these techniques as well. hopefully you'll enjoy some of that. I thought I'd share a little bit of the news. I'm looking over at my notes so I keep looking off to the side, but

To be honest, there's not huge amounts out there. think we're seeing, you see this kind of peaks and troughs with news. We saw that kind of beginning of year, January had a lot of announcements. It's that kind of post-Christmas, let's get everything out that's kind of been building. We hadn't quite signed off on there. Let's have some exciting marketing moments. You can get to February and it's, I don't know about anyone else, but here in the UK and I think a lot of Northern Hemisphere, it's a bit meh.

a bit grey out there, it's a bit wet, it's kind of cold and starts to warm up and then gets colder again and I think we're seeing that in the news. But more importantly we've also got kind of events starting in March, the conference season in March and April, we've got the MVP Summit coming up so I know there's teams working to kind of get content ready for that, which will be a preamble to build happening in May and various other events that happening all around the world on there, so well worth looking those, I know we've got the European

collab summit, we've got the M365 conference power platform conference coming up, agent cons happening all over the place. That reminds me, I must put some notes on the agent con dates into the show notes. So I'll make sure those go in there as well. So there's kind of different things happening as well. On that.

I've got to pick up a couple of bit of news. Strangely, one of them is about prompting, so probably should have included in the prompt section. because I'm recording this in a weird order, it doesn't matter too much, which is great. So a little let me share my screen for those watching on camera. A lovely post from Stefan Bisse. Now, he he mentioned this in some other chats recently I saw.

that he was working on this, but I hadn't seen what it was, but he's created for those building declarative agents, he's built a way to kind of get the instructions to get the kind of common prompts for that in place. And he's created a VS code extension for devs building those agents. So if you're using VS code, there's something called agent instructor, instructor, just the two installs. Tell you what, I'm going to try this right now. So thanks, Stephen.

If you're listening, I'm going to hit on install and make that third one. You're not going to see that because I'm only sharing my screen, but that's just clicking on those buttons. It is opening up Visual Studio code for me and has, let's see, once it's opened on my machine, which is getting far too slow, has installed that already. So I've clicked on there. I've clicked trust on it and it's bringing that through. And let me just see what if I got open.

I've got my Salesforce agent open, which I've been working on and I can, yeah, I can share that screen. Let me just bring that window up so you can see it's opened up the window. Let's make it a little bit. No, it won't let me zoom there. That's a bit annoying. there we go. So it's put the agent instructor within there. I've got it set to auto.

auto-update, it's got some descriptions on there, but it evaluates existing agent instructions for clarity and potential improvements, which sounds very nice in there. So you can see it's asking me to look at instructions.text. So let's look at instructions.text. there's copilot there. No, that's GitHub copilot. This is...

going to be awful and poor Stefan's going to go no it's meant to be easy why can't Kevin do it I think it says more about me than you Stefan don't worry because I'm not very good at reading instructions there we go so what I need to do is actually describe the agent purpose you can see here it gives me the purpose hopefully you can see that nicely along the top and it's actually building out my instructions on there

So there's some of the configuration. So I actually need to set some things up within that. So I might go back and have a look at that later because I need to point it at OpenAI for that. So you can see here, well, we'll give it a quick try, but I don't think it will work. But if I have a look at my instructions from within here, then in the palette above, let's

See, it's That's why shown run confirmed. Let's have a look for the agent structure, analyze instructions. Now, as you can see, 401, I haven't set up the agent yet, so I'll give that a try a bit later. But big shout out for Stefan. I think it is one of these things that we need to get people moving. This is to make it even easier for people to create prompts. So I really like that that's come through there.

Now, if that was a little bit early for you in your journey with declarative agents and you're going, Kevin, what's an instruction? I have no idea what you're talking about. Then what you need is some guide on creating agents. And our good friend of the show, Donna Sarka, has exactly what you need here for this, because the team has released a free AA

AI agents for beginners course. And within that, it's got details on intro to agents, exploring the frameworks, understanding the patterns, looking at the tool use, agentic rag. I'm so happy to see this section building trustworthy AI agents, are you taking on all that responsible elements? Think about your design patterns, bringing multi agent design patterns. It's got to hit metacognition design patterns.

I am very much going to watch that course because I do not know what metacognition is. So there is something in here for everyone on there that's worth looking at. So check out that lovely course that's available. While we're on there, I'm also going to share this window here. In fact, let me share the whole window so I can jump between a couple of different posts.

This was a fantastic one around the guide of staying ahead of agentic threats within there. So OWASP has been putting together a lot of security oriented material, especially around generative AI on that. So let's download and choose the top 10 that I wanted to link. They've got a guide on the top things for agentic to take look at for that one.

So when you're building an app, using agents, thinking about those different considerations, I'm sure some of this will come into that training course. But specifically with things like threat modeling, mitigations and recommendations, the agentic security landscape, putting together all these best practices will make sure that you have an agent that is usable and relevant to other people. So go and check that article out. It's absolutely fantastic from there.

got one more agent and then one more models as I say there was no copilot news that we saw at all out there which is a little bit sad but this is kind of copilot news from good friend Fabian Williams who's got caps so caps are and I can never believe so for copilot agent some things I think I hope it's going to say there

Copilot agent plugins, that's the word I'm looking for. So if you've heard of semantic kernels, a way of building custom engine copilots, what Fabian and the team at Microsoft is build a way to kind of connect to your Microsoft 365. So contacts, calendars, emails to integrate that. And he's been putting together some fantastic examples of that. So do go and check that out.

One last announcement, I'm going to put the main article that Sacha Nadella shared about a lot of new AI models, but there was one that particularly caught my eye called Muse, and they had something in Geekwire about it. Because I think it's interesting, I haven't seen as much around generative AI in gaming and how it can be used there, whether it's the cost of kind of building something with generative AI that runs there. I don't know whether it's something that's

is still evolving, people are working out how to use it effectively. I have seen someone recreated Minecraft in the browser where effectively it would kind of build images every 10th of a second. So as you browsed around, it was using generative AI to build an image of the world you're looking in. And it did mean that occasionally it would hallucinate and completely add something else in there. But generally, I had a

It had a bit of memory of what was on there. So it was kind of an interesting idea and proof of context, but I haven't seen it in reality in many games on there. This model that was created in conjunction with the Microsoft Research Lab in Cambridge and one of the Xbox subsidiaries called Ninja Theory has kind of taken trying to create a more human like CPU opponents within their using generative AI. So they'd be

working on training this model that can use it, taking on some of those abilities to learn from what others are doing on that. So am I going to try this out? Probably not. Am I very excited to see what's going to come out from this? Yes, yes, I very much am within there. I think we'll see more intelligent games, more interesting articles from that. So really.

powerful, not really particularly co-pilot-y, but definitely in that agent space. I'm going to put a few other news articles into the LinkedIn newsletter that I put out there as well as the show notes. So do take a look at those. But for now, I'm to do a very strange handover to me a few hours ago. And me a few hours ago didn't know how I was going to introduce this. So you're going to get another introduction and things on there.

Please do keep an eye on us. We'll be at the Microsoft AI tour by Zoe and I. We'll be the MVP Summit to come and say hello. I know Zoe is going to be at Collab Summit and M365 Conf. So get in contact, come say hello. Otherwise, handing over to other Kevin. Thank you.

Kevin McDonnell (12:46)
So I'm not entirely sure how this section has been introduced because I thought I'd record it and then see what happens. So sorry for any overlap here, but what I'm going to talk to you a little bit about is prompting. And we're going to look about this as a general perspective and then dig into a bit around M365 Copilot as that kind of front door to a lot of the other capabilities, especially with agents. I think it becomes the most

easy way to look at the project. There's other ways. Obviously, you can look at things like the dynamics. There's a whole load of prompts, a power platform that help there. But I really wanted to dig into some of the ways of getting started, because I think a lot of people are just doing prompting and haven't stopped and thought about it a bit. And I'm very lucky where I work. We have a fantastic prompt course that's got me thinking in it. It kind of has naturally helped me evolve how I do.

prompting from just a kind of single line to, many cases, I like to tell it bit of a tale. I often describe to people and clients who ask how do they do prompting? I say, well, just imagine you've got a summer intern with you and you want to get them to get you some information or you want them to do something. How would you describe it? And then I pause and then I say, OK, OK, step, step back a sec.

Just imagine that you were a summer intern and you wanted someone to give you instructions. What would you like? Because yeah, the reality of what we're doing right now is the instructions that people are giving. Sorry, is those bad kind of hitting instructions, those prompts that aren't always the best. But think about how you'd like that. And that would probably work quite well. So that's, if you're not going anywhere else, that would be my fundamental take on things.

There are actually some techniques and there's different ways that work quite well. Whether you're there as the average user of copilot, of chat, GPT and things like that, or whether you're looking at kind of building into your developer flow of building a little bit more information and content from there. There are ways to dig a bit deeper. And so I thought I'd look for a guide for those watching video. You'll see this for those not well, I'll just try and describe as much as I can as I go along.

as I always do. So let me just get the window up there. So I decided the simpler thing to do would be to ask Copilot what is the best guide to prompt engineering? And I've had a look through these and two have really jumped out as being useful. I will be honest, because I still haven't built the habit. My first thing was to go and search Google. And there are so many recommendations for prompt engineering out there that

I didn't really like any of them. Some of them had nice takes, some didn't. Went to Copilot and genuinely I have much preferred the results from here. So this has really helped me on that. Let's let's have a look. First one is OpenAI's guide. OpenAI's best practices for prompt engineering. And this one is is really nice. This is very it feels very designed for developers on there. So it gives a bit of a

an overall rule of thumb on their interesting show. I don't have so young for this bit because it says rule of thumb number one use the latest model. Who would have thought that open AI who makes more money from the latest model would recommend actually using the latest model. So it's something we've talked about in the show before. When Scott Hanselman was at the Scottish Summit, we noticed that he was using GPT three point five. So the question there was

What's what's why why are you not using the latest model? And his answer was I don't have to I don't need to why would I use the latest model unless I had to? So I kind of agree that you will get more capability absolutely from these latest models, but only use it where you have to so don't jump into spending more money and putting more load on the data centers.

impact in the environment. So think about that on there. It talks in here and I'm not going to run through all the guides, but it talks about using hashes. So three hashes and three quote marks to separate instructions and defining the text. And what's useful to that is, for example, where you're putting some text in there, being very explicit to whatever you're using to say, this is the text that I'm using as an example, or this is my kind of phrase will help it in

interpret that as well. It talks about being more descriptive on there to articulate what format you want as the output on that. And I tend to use examples. So I was building an agent the other day about taking my waffle into a status report. I wanted it to be in the kind of format of a daily stand up. What did I do yesterday? What am I planning to do today? Any blockers? And I gave it some examples of that. I showed it the format I wanted and also included some examples.

within that as well. Now, there are some details here. It starts talking about phrases like zero shot, few shot example and things like that, that are quite helpful. So zero shot is give an example, sorry, don't give an example on that. A few shot gives you a few examples, fine tune on there. Now, I'm going to pause because this is where I kind of went with the OpenAO1. I love that it's talking about those, but it didn't.

really described it very well. Instead, the other recommendation from Copilot was this promptingguides.ai. And this one is fantastic because it's got all the different sections in there. It talks about the settings you can configure in your LLM, whichever one it is you're looking at using. So temperature, the top P on there, the nuclear sampling as it is, the maximum length, et cetera, what all those things mean, which

The majority of you just using Copilot or using ChatGPT won't look at, but you can go into things like the GitHub playgrounds, the Azure AI, the Foundry playgrounds, as it now is the labs that are in there, that you can go and fiddle with these things if you want to. And it's got details of that. But this guide also starts from the very basics. So the prompt is, for example, the sky is and the output is blue.

Not here today, it's great, but I get where the thinking is going on there on that. And it kind of talks you through the different ways of doing that and then trying to improve it. So the next example is complete the sentence, colon, the sky is and the output is blue during the day and dark at night. Good point. Never think about that. So it starts to take you through the different formats of how you put across questions, how you suggest answers and the different details within that.

But it also includes within there the different types and describes those zero zero shot prompting. So for those who don't know, fact, I'm just going to read this to make sure I get it right. Zero shot prompting means that the prompt used to interact with the model won't contain examples or demonstrations. So you're just giving it that detail on there. The other ones you're looking at.

are you can also look at few shot prompting so you give it a few examples. Another really interesting one for when you're feeling particularly adventurous is chain of thought. So chain of thought is particularly useful when you're thinking through those reasoning capabilities. So maybe you've got some examples in the case it gives here the odd numbers in this group add up to an even number.

And you give it the answer when you add them up, it gives 25 and the answer is false. And then you choose another one and you say it's true. So you give it examples and you give it the answer and then ask it to reason on within that. I think and he'll probably if he's listening, correct me. But Abram Jackson talked about this of the reasoning models and the capabilities there where he gave examples of the connections for the New York Times quizzes. We have four four blocks with words in

And you've got to kind of build the connection between four sets of those. And traditionally, the models couldn't handle that. It couldn't build those connections, couldn't see those details. But the latest reasoning models, when they came in, could do that. And this is a kind of example. You'd explain what you wanted it. You get some examples where things do match. You give some examples where things don't match and then try out with a model. So is this something you're going to use?

day-to-day with copilot on a regular basis? No, probably not. Is this reasoning model, is this ability to take examples and build it out something you could use to solve a complex problem? Hell yes. This is the magic that you can go for those a bit deeper on that as well. Some of the other examples, I'm trying to see which one, which I've suddenly forgotten, but things like prompt chaining.

So you can loop together those different examples. You can build in that first prompt and then take the output of that prompt and load it into the next prompt. So you can start to build those things together. I'm trying to remember, I think it is the prompt training, but a lot of those I've seen a fantastic prompt that takes you through a guide of things.

on there, it was kind of to help you, I think, for your goals. So think about your OKRs, your challenges for the year. And within that, sorry, Zoe is distracting me on this. it, Zoe, stop distracting me. It can help you think about things and with the right prompts as a kind of mega prompt.

Thank you, Paul, if you're listening for that phrase with a mega prompt, it can kind of automatically take you through the next phases. So it could say, you know, what are your aims in one sentence? But because of the prompts you've to find at the beginning, it will start you talking through thinking through those and what they mean, how how do they work towards your longer term goals and things like that. So that's a really good, powerful example that you can use. Would you necessarily write one of those now? No.

But if you go out on the Internet, you will find an awful lot of ones available. So I really need to stop saying so. Those are some some nice examples of how you can kind of find out more about prompting and how you guide on that. What kind of triggered this thought today was actually this post, which is in the show notes about getting the most from the Co-Pilot prompt gallery. And this is where I

I do have a slight confession. I don't use this as much as I should. I still haven't built that habit of going, this is good prompts that worked. I must save this. But you can do that now. The way you have that copilot prompt gallery available within there, looking at your kind of regular example prompts, you can start to curate your useful prompts that happen within there. So there's

There's a whole load of predefined prompts in there, which I always recommend are a great starting point because Copilot has been geared towards those actual prompts that are available, excuse me, that are available and work within there. So to start with those, and it's got things like stay informed, what's the latest from person? I think you're meant to replace that with the name of a person on there. What's the latest from person? Organized by emails, chats and files, a lovely, very simple prompt on there.

Create a presentation from Blar. And by Blar it means another file from there. Get a to-do list from my Fabrikam client meetings, for example. Draft an agenda for, and you can actually reference a specific meeting within there. So you've got some nice quick examples. It's kind of a pain because these examples, in some ways, if you were making your own, making your own prompt, these are awful.

because they're short, they're not descriptive, they're very concise. But what's happened is within this prompt gallery, Microsoft's kind of geared copilot to work well with these so that they do work well. It's not really example if you were thinking up your own is a great idea to do. You need to get a little take those a little bit further. So you can you can start with those and customize them and add a little bit more information so that

that one I talked about just now of what's the latest from Person, organized by emails, chats and files. You could say what's the latest from? the example it gives here, human resources regarding changes to the annual review process organized by emails, chats and files. even though the prompt just said person, that's not to say you can't start to add your own information. And I've done things like, for example, I had this email from Blah that mentioned this.

Could you find more examples that added to this as well? you're again, as you would with a summer intern, giving a little bit more information. If you said to them, what's the latest from so and so? They will probably stand there. Look around a bit. Looked panic. Go back to their phone and check things on their stand around. If they were feeling particularly confident, they might ask someone else, but probably not.

So think about the same things. How do you give more information? Because you're not going to see that glazed look of confusion and bafflement in Copilot. Because Copilot is far too confident. Copilot is a summer intern that's been to Oxford and thinks they're the greatest thing ever on there. Nothing wrong with people who've been to Oxford. That was me once.

But they have that confidence. They're going to tell you the right answer, whether it's right or not. So you might need to give a little bit more guidance to make sure it is on there. Now, once you have that lovely prompt and it may be something you want to use on a regular basis, you can now save that prompt. If you hover over the prompt you've made. So once you've submitted it and run that article, go back to the the block that's at the top.

generally in blue on Copilot. And if you hover over that, you will see that the little bookmark icon appears along the top. You can also share that with people. There's a little link icon so you can share that link to other people. And if it's enabled for you, which I know is not that many people yet, you can also schedule that prompt. And I love this one, especially where you want to get the latest, you know, what's the latest conversations on Copilot.

Can you summarize all the things that I've missed and haven't done yet within there? You can now schedule that prompt so you get it nudged to you still comes through as an alert within copilot. So you're not not sending this to email, increasing your email traffic still there where you where you mostly would expect it. I'm hesitating there and getting myself in trouble because I'm hoping we'll see a slight improvement to the alerts and nudges on that. That's a conversation for another time.

ask me about that in person, I'll be a lot more effusive on that. But anyway, you can get those alerts and you can see those results and you can take those off into Copilot pages from there. But the save prompt then brings you back when you go into the Copilot gallery, you can actually filter on your prompts and you can see those. And in fact, I mentioned those ones you can share, those shared ones will appear in your team prompts. So other people you're working with

can see those as well. So if you work in a project team, fantastic. There could be find the latest on this project. Help me generate a status report based on the content here, here and here. That sort of thing. And you can actually share those to a team, a Microsoft team team from within that as well. So I think the potential that we'll see and we're still seeing improvements happening with the prompts gallery are absolutely fantastic. So well worth looking at.

and sharing, building your own community internally where you share those prompts. You could have that ability to share a team that could be a prompt team. So within your organization, you could have a center of excellence, a center for enablements, whatever you want to call it, that is there sharing those prompts that everyone can see those benefits from there. Or you could take that wider community that exists around the fabulous world of

Microsoft and actually someone has already created these. So I'm going to jump back to the main one as part of the one day I can remember what this called. The Microsoft is not the Microsoft tech community, the Microsoft 365 and Power Platform community. I'll be honest, it's the artist formerly known as PMP. Let's let's just call it what I know it as within there and turn off my alerts coming through.

From this, we've got some co-pilot prompts within there. So Rabia Williams has been fantastic at curating this. I'm just going to see if there's a list of any other names before I get in trouble and forget to credit anyone else. No, I think it is mostly Rabia who's been spearheading this. From within here, you can see there's some samples on here that give you instructions on that. And anyone can...

contribute to this. Just need a little bit of GitHub knowledge. Look out for the caring and sharing sessions. If you aren't aware of GitHub, feel free to reach out to me for any guides on how to contribute using GitHub. There are guides within this as well. But if you're confused, because it's not the easiest thing for the average non-developer on there. Now, if you go into this, if we browse into the samples folder, you can see that there's a whole load of examples in there. There's some Git

GitHub copilot ones. There's a whole load of M365 one. There's some PowerPoint whiteboard and word ones within there as well. And the key to these, let's have a look at something. There was a couple that caught my eye on this one. This one I really liked. And I'll be honest, I hadn't looked through here for a little while and hadn't noticed this one before.

But the prompt for those listening is show me the list of attendees for the meeting Project Nexus in initial resource planning. Show me the list in a table with name, organization name from the email domain. If they're external, show it as a globe and it's got a little globe emoji. If the user is internal, show us home. Following should be the name of the columns, name, organization type and user type within this. Now, I really like this idea to get that list.

I kind of like the idea and reading this, I'm kind of thinking you could extend that and maybe ask it to say how you know that person, because it could look at your emails and your connections to see where you know them. You could look at their profile and see which area of the organisation is, what it is they do. So to have a prompt that helps you understand who it is in the meetings, what engagements you've had with them before, things you might have forgotten so that you know when you get to that meeting, they hijack you again.

Kevin, have you managed to get this PowerPoint presentation together yet? And you go, I'm so sorry, I'm a God. Hopefully don't do that. You know, obviously, I know people are in that phase, but don't try not to do that if you can. Now, with this, you can see for each of these, it's got the prompt in text, so you can very easily copy them out. It's got a description of what it does within there as well and a summary for that.

Let's have a look at some of these. Yeah, you can see another one that lists the RSVP status for there. I'm not going to worry too much around that one because I have heard that you can do that in Outlook, but there we go. Let's not go through that one. Now, there was a lovely couple of ones. This one, think, yeah, was Nico who put this together fairly recently and made me think of this.

For those of you out there who have to do timesheets with your large organization, with your consultancy, timesheets are one of those pains of your life. Now, this prompt says, can you compile my timesheet for and you give it a date or a week, focus on emails, meetings and chat messages. Do not divide between my emails, meetings or chat messages, but give one general overview and a table with current start time of the meeting, email or chat, end time, total time spent expressed in hours.

and the subject, sort the table according to the current time that chat email took place or when the meeting started. It's very possible that several things overlap and have taken place at the same time. Give the total time spent from below. hallelujah. Thank you, Nico de Claire for this one. This is going to look in your calendar and try and break down some of the time and details from that for your timesheet. Is it going to be perfect? No.

Is it going to help accelerate some of your queries when you're in a rush and need to get your timesheet in before the deadline? Yes, hopefully it will. Obviously, if you are a consultant working across lots of different tenants, it might not help you quite so much, which is a little bit of a shame, but that is this multi-tenant world we live in. So take a look through some of these different prompts, these available items that are there.

There's different ones. I love that there's some ones here for whiteboard of how to generate a comprehensive list for a company intranet. So using the ideas to trigger your workshop on there. was another one I saw earlier that really caught my eye and I'm completely missing now from there. I like the copilot as an Excel tutor. So I think we've mentioned so Paul Bullock has found that the copilot chat.

interface is better at kind of guiding you on Excel than often the Excel one is there. So looking at some of these prompts in there, I'm not very good at using Microsoft Excel, helping by generating a sample table and show some go to formulas supplied in table. And that's a good point. I didn't touch on that in the prompting earlier, giving you bit of a persona for things. So creating, I've been generating some sample content at the moment.

I've been trying to get it to create some HR knowledge articles and policies and I started off by saying you are a HR specialist you've been working this large organization for many years within there you have a deep expertise in this you hate it when people go get things wrong so giving it a little bit of a story and a something to work on a background separating that out nicely with here's the background here's what we want you to achieve

can help get the most from it, but a really nice way of doing that. So explaining I'm not very good or why you're doing things as well. Think about user stories. Think about that persona, what you want to do, why you want to achieve it that will help guide it towards the right thing. But again, that wasn't the one that I was looking for. No, I'm going to have to give up and remember it later.

Again, there's one here around extracting unresolved issues that wasn't very exciting on there. I quite liked this one here, which is about comparing technical proposals to find the suitable proposal based on the defined criteria. So where you've got two different proposal, maybe you're running an RFP and you want to kind of deep dive on particular subjects and you can get very opinionated from reading that and

Organizations often submitting in different ways, depending how prescriptive you've been. even when you're doing an RFP, organizations haven't got the same history and maybe approach it from different ways, which is a fantastic way to get some diversity of thought. But when you're trying to compare two different ones and they're in different formats in different ways, that can be hard. So looking at this and asking it to focus the example here.

highlight the risks and challenges in both approaches, which is the better governance and secured approach. So if there's something you think is more important, you can ask it to compare on those particular outcomes from that as well. And also, I'd be interested, the one problem with these prompts is they don't necessarily give examples of how this works. in this one, it says,

Also, let me know which proposal, according to you, is most suitable based on the following criteria. The issue I have with this is Copilot and the chat GPT's and other ones often won't try and give recommendations, so I try and avoid those where I can of asking it to pick which is best. But I think this is probably my my closing thoughts of this. Give it a try. There is no.

real harm in saying, let's give these things a go. Let's give it a try for different things. Let's see how it works in different scenarios. I love that in Copilot, you've got those conversations. So even if you haven't saved that prompt into your gallery, you can go back and have a look at those and try them again and sort of see what works well. make sure if you're doing something new that you've given a bit of time for it, that you've given a bit of thought within there and pull that together. So

these things out, check out the links in the show notes, just try and remember if there was one other thing, no that wasn't the thing I was going to show here today, check out these samples, check out the guides on the different types of prompts and think about where those different ones could use it, and most importantly make the most of the prompt gallery when you're using this in Copilot because you can save and share those and even either now or soon.

have that capability to schedule that and see that coming through. So I hope that has helped to make you think a bit. I find for me personally with prompts, I blurred your load of things out. I put some things together I don't necessarily think about it every single time. But every now and then I read an article that makes you go, I need to be a little bit more intentional with how I do prompts. So we decide to go in and do that and pull it together.

So I hope this guide has been helpful. I'm hoping that we're back doing this together with Zoe. We have a couple of guests lined up coming up soon. We will be at the AI tour again to come and see us there and let us know what you think. But for now, thanks very much and bye bye.

Kevin McDonnell (40:27)
Thank very much, other Kevin, for that fabulous take on prompt engineering. I'm just going to close here and say thank you very much for joining us. Please do like, subscribe, give us comments, ratings. If you don't want to do that, you can just ping us and let us know what you think. Good or bad, what should we do more of? Should we do more of these solo shows? Or please do not want Kevin droning on you, just him. You need that kind of...

lovely leads accents to kind of interrupt me time by time. Let us know and otherwise we will see you again in a couple of weeks time. Thanks very much.