In this episode of the Copilot Connection, Kevin McDonnell is joined by Garry Trinder, a developer advocate at Microsoft focused on Microsoft 365 and Copilot. With Microsoft Build taking place this week, they recap the Copilot related news that came out with a special focus on extensions and custom copilots. Plenty of chat about why you should extend, the different ways to extend and when to extend or go custom. Plus find out how many times Copilot was mentioned in the first day keynotes thanks to Vlad Catrinescu.
Useful links:
M365 Conf sessions - some online now including the age of copilots -Introducing the Community News Desk with essential content from Microsoft 365 Community Conference
Vlad Catrinescu comedy recap Microsoft Build 2024 Keynote Recap in under 5 minutes!
Book of News - https://news.microsoft.com/build-2024-book-of-news/
GPT-4o demo by Seth Juarez - Conversational website: Satya Nadella at Microsoft Build 2024
Copilot+PC - Introducing Copilot+ PCs - The Official Microsoft Blog
VS Code for Education - https://vscodeedu.com
Note that you need a Copilot license to use this with Copilot Studio - Use graph connections with knowledge sources - Microsoft Copilot Studio | Microsoft Learn
https://aka.ms/extendcopilotm365/decide
Useful Build sessions:
BRK 150: Developers guide to customizing Copilot - slides available too
BRK 151: Developer deep dive on building plugins for Microsoft Copilot
BRK203: Building a copilot: Azure AI Studio or Copilot Studio
In this episode of the Copilot Connection, Kevin McDonnell is joined by Garry Trinder, a developer advocate at Microsoft focused on Microsoft 365 and Copilot. With Microsoft Build taking place this week, they recap the Copilot related news that came out with a special focus on extensions and custom copilots. Plenty of chat about why you should extend, the different ways to extend and when to extend or go custom. Plus find out how many times Copilot was mentioned in the first day keynotes thanks to Vlad Catrinescu.
Useful links:
M365 Conf sessions - some online now including the age of copilots -Introducing the Community News Desk with essential content from Microsoft 365 Community Conference
Vlad Catrinescu comedy recap Microsoft Build 2024 Keynote Recap in under 5 minutes!
Book of News - https://news.microsoft.com/build-2024-book-of-news/
GPT-4o demo by Seth Juarez - Conversational website: Satya Nadella at Microsoft Build 2024
Copilot+PC - Introducing Copilot+ PCs - The Official Microsoft Blog
VS Code for Education - https://vscodeedu.com
Note that you need a Copilot license to use this with Copilot Studio - Use graph connections with knowledge sources - Microsoft Copilot Studio | Microsoft Learn
https://aka.ms/extendcopilotm365/decide
Useful Build sessions:
BRK 150: Developers guide to customizing Copilot - slides available too
BRK 151: Developer deep dive on building plugins for Microsoft Copilot
BRK203: Building a copilot: Azure AI Studio or Copilot Studio
[Kevin 00:16]
Hello and welcome to the Copilot Connection. I'm here because it's just me today to share with you all the latest news, insights and capabilities that Microsoft Copilot ecosystem across the entire Microsoft stack. I'm Kevin McDonnell, MVP, Viva Explorer and copilot strategy and modern workplace AI leader Avanade. We'll be releasing episode as podcasts and on YouTube with insights from experts from the community and Microsoft on what the different areas of Copilot are, the impact they can make to you and your organization of what you need to do to prepare for them or start implementing now. And even as we'll talk a lot about today, how you can extend them. You may have noticed there's no Zoe here this week, but we thought with Microsoft Build 2024 out there, we bring in one of those experts that was talking about. So, Gary, would you like to say hello?
[Garry 01:17]
Hello, everyone. Thanks, Kev. Yeah, hi, everyone. My name is Gary Trinder. I'm a developer advocate at Microsoft, focused on Microsoft 365 and, as you guessed it, Copilot. So it's nice to be here. Thanks for inviting me.
[Kevin 01:31]
And so someone from Microsoft focused on Copilot, Blimey, who would have thought of that at the moment on there. So, and we're going to focus a lot on extensibility in the stories that have come from there from builds this week. I was about to say last week, it has been a long week, it's fair to say. And then there was quite a few other announcements and bits and pieces from builds. So Zoe, next week's going to do another recording and hopefully we'll have a well, we will have a guest, hopefully Chris Huntingford if we can pin him down at some point. But certainly Zoe will be on with someone to cover some of the other news on there. Just before we get into build a reminder about month of copilot will be in the newsletter on there. We've got a huge number lined up. I meant to see how many recordings we've already had sent through, but a good number already there and being lined up ready to share across all the different Co pilots and hopefully some special guests that we're trying to line up as well. Which reminds me, I must chase those from there. The other one on some of the other news. And and Gary, you weren't over in Orlando for the M365 conference, were you?
[Garry 02:43]
Unfortunately not, no.
[Kevin 02:47]
Yeah, it's sadly all that one, and neither was I, but they have made some of these sessions available, so we'll put links to these in the show notes. The reason is I can see there was one of the age of copilots reshaping Productivity with Copilot for Microsoft 365 by Bobby Kishore and Dan Parrish saw them speak of the MVP Summit. So if they cover any, any sort of thing like that, that would be a really good watch. Haven't had a chance to see these with all the things going on within there. But while this is across the M365 stack, some really great copilot sessions in there. So keep an eye out for those. But I think we're going to kick off with the build. And before I ask you on your thoughts, Gary, Vlad Katrinescu did a really good recap. And I know I asked you about this before the show. You said, yeah, I've seen a few people tweet that or watch it at some point and you hadn't watched it. So he he's done a recap and we'll play a little bit of this. Hopefully this comes through in the notes and just to kind of get some reaction. But I think this summarizes nicely from the keynote what was talked. Good morning. It's fantastic to be back here at Microsoft Build. The 1st is Microsoft Copilot Copilot plus PCs, Windows Copilot Runtime, Windows Copilot library. Hey Copilot, how's it going? copilot apps? Get up copilot, get up copilot, get up copilot, get up copilot for Azure copilot GitHub. I think she'd probably get the idea now. So Vlad had a very tongue in cheek fantastic example of that of, yeah, all the times in the keynotes that copilot was mentioned.
[Garry 04:34]
Yeah, it was quite a lot, right over 100. I think it was mentioned in about two hours.
[Kevin 04:40]
I I had a look on on Twitter and it was mentioned 139 mentions of copilot which is which is quite impressive.
[Garry 04:47]
I think, yeah, kind of joking aside, I think just just kind of shows the breadth of where generative AI is, is being placed across the Microsoft stack. You know, Azure, GitHub, M365, you know, wherever you're working, AI is, is is kind of going to be touching those areas. Try to try to help you give you different ways of of working. So yeah, it's, it's a bit copilot overload, but yeah, it's, it's all different.
[Kevin 05:21]
It's interesting because build traditionally has been the more developer focused conference. So for me, coming from that modern workplace for you looking at that copilot, we're kind of looking at this for this is things that are going to be coming into the the, the the core part, the platform soon rather than right now in many cases. Obviously you can use some of the tools they're talking about here to build your own copilot and whether we'll touch a bit on that later. But it's, it's, it's almost that forward view of what's going to happen, which is always interesting from, from this conference.
[Garry 05:56]
Yeah. So I think definitely this year there was there was still plenty of announcements around things that have landed and things that you can use certainly from a Microsoft 365 perspective. It was building on top of the announcements that you've probably already seen from Ignite. So new capabilities that are now going to be coming so that you can kind of say, OK, well, if you've been looking at like how to extend Copilot for Microsoft 365, then you might have gone, oh, I can't do this or I can't do that. You know, there's different options now available to you. So revisit them. Have a look at your fresh to say, OK, actually now that's interesting. Maybe we can do something like with these new features in the future.
[Kevin 06:41]
Yeah, absolutely. And if you want to catch up on all the things, we'll have this in the show notes, The Build 2024 Book of news. You know, I just thought I should have used Co pilots to check how many times copilot is mentioned in this, that that might be something I try while you're talking about later on there. What what I like going through this is again, lots of talk about copilot, lots of talk about AI, open AI and things like that. But there's lots of other things in there, you know, not something we'd necessarily focus on so much in the show. Let me see if I can get into some of it. AI was obviously up the top, but they were kind of new enhancements to Azure functions, things like the containers and Kubernetes video dumping. I loved this. That was kind of translating videos and things on there. So it's, it was good to see they haven't completely ignored all those other things that came through as well. And it definitely worth digging through that. And I'm hoping Zoe and Chris might talk about some of the fabric announcements, which did include some copilot stuff and other bits there. So there's plenty for non copilot focused people to to dig into and and have a look at as well.
[Garry 07:54]
Yeah, certainly. I think it's it's the, you know, it was a few days of of build with the keynotes. Obviously it's the highlights that kind of get talked about, but there is absolutely tons of work going on like across the Microsoft stack. So yeah, book of news is the is definitely the place to scour over the next few days and check out on the recordings as well, because you can't do them all right? You can't sit there and watch them all.
[Kevin 08:21]
And I can't remember if I've talked to you much about this as, as I did on one of the PMP sessions, you could obviously run a script, download all those videos, run a transcription service and point a custom copilot at it. Make sure you have plenty of credits and make sure you're ready to pay for that if you do, is all I would say to that on there. The I've distracted myself now by trying to run copilot against the book and you stop it, Kev, stop it. A few of the things I think I've managed to watch the keynote. I have to admit I haven't watched a huge number of the sessions. Too much going on, as it's always the case for me. There was one standout moment from the keynote, and we've seen a lot about GPT 4 O and the capabilities with that. I loved that Microsoft had GPT 4 O available in Assure Open AI the same day as it was available in Open AI. So I thought that was very clever. But what they showed was using that with a, a custom copilot built into website. They built up a mock company called Contoso Outdoor and they were showing instead of having a kind of chat bots you often have from websites, they had a video and it was a live conversation with a try not to say copilot or bots, a virtual assistant, let's call it to talk to and it was showing your video. And I'm trying to think who was Seth Juarez? Wasn't it walk in there and he was doing things like asking questions and the speed of GPT 4 O that it was a very natural conversation. It wasn't sat there waiting for things and he held a shoe and said I've got the shoe. Could you tell me a bit about what you have available? And it knew to go and get the information from the stock and be able to answer questions. He then flipped over to Spanish. I would say Angel deal, Carrero noted in a chat. Not very good spanner, very Latino Spanish. He commented on that, but it actually handled that and returns back in that. And I, I was just blown away that that just showed me that the power of some of these engines that are coming through there as well.
[Garry 10:36]
Yeah, I think the the big thing that that demonstration was something that was like very much like, oh, wow, you know, this is something different. But I think it kind of shows now like the multimodal approach that we're getting from generative AI where we've kind of moved then from, you know, putting things in a text box, putting things in like ChatGPT. But it's that's kind of limiting, right? You've got things around you in in the world. Isn't it great then that you like say just show us you and then the system being able to understand and take in that as context to then, you know, understand what you're doing. It's like it's better than just going, OK, now I need to type out my shoe that I've, you know, this is what I've got kind of thing. It's not always the best way of doing things. One of the things that really got me was the fact that yes, switching language was just immediate. The other thing was stopping the copilot as well from talking. It was like interrupting copilot and and that's interesting as well as that it can just stop and then carry on. And from a, from a developer perspective as well, I'm thinking, you know, think of the things that you would have had to have attempted to try and get that to work before. And now this is kind of like baked into the models that that you can use. It's it's really interesting to see where, you know, people are going to take this new technology.
[Kevin 11:55]
Yeah, what what I would say though, is I've seen Terminator, I've seen Skynet. Please don't interrupt Copilot too much. It it might remember it, it shouldn't. I know it doesn't remember things, but just in case, just just remember. That's got me nervous on there.
[Garry 12:14]
That's it's a copilot then actually it wasn't copilot. It was using just generative AI, but we're using copilot as the term.
[Kevin 12:21]
But it was a copilot, small C in that area. Good, good, good point, good point. Come back to the Copilot news. There was kind of a pre build announcement about this Copilot plus PC, this amazing new hardware. And our former colleague who's who's still a colleague of yours alertly was very excited about getting a new laptop and then found out he didn't have the very latest one with that hardware. But it's all about having those NPU, having those neural processing unit chips on that device that will help offload locally a lot of that capability to run. So I know there's been talk of things like background blur within teams actually goes up to the cloud. I think more the Teams premium where you're kind of replacing with there that uses AI that goes up to cloud, uses a lot of battery life. By bringing that onto that focus chip, it's not only increasing performance, it's also increasing the battery life. And also what was announced was the kind of Windows dev site to make use of that. So bringing to that local development model within there. I know you're going to talk a little bit about some of the excitement with that, but I was really intrigued by that. And it was, it was good at the keynote, you could hear the old skull Windows devs who are getting very excited. It's, it's good to know they're still about from there and still making use that it's not just web these days. So I really interested to see how that world evolves. That to me was a classic build. We haven't got anything using this yet. Here's the capability going wow us with what can be done. So really, really looking forward to to hear him on that.
[Garry 14:00]
Yeah, the hardware side of it is like maybe we don't talk about it because we just think of kind of like, Oh yeah, AI is just cloud services, right. But gain the benefits of running things locally, not having to send things over, over the wire, wait for the network capacity as well as we know, you know, it takes a lot of resources to power these these systems as well. So offloading it to the devices makes total sense because you get better experience as well.
[Kevin 14:29]
Yeah. And it does make me feel old when you have this cycle of it's all going to the cloud and being sensual, it's coming to the local machine, it's all going to the cloud and being central. I won't say how many cycles of that I feel I'm on at the moment, so.
[Garry 14:41]
Eternal battle, right? Yeah, it's making.
[Kevin 14:43]
Me feel old definitely on there. The flip side to being old is being young and learning and education. And there was a small bit of news, but as a Cub Scout leader really excited me. And that's the VS code for education. So really kind of having that learning capability, especially in this world, you know, pythons really becoming the language that a lot of people are starting with things like scratch connecting over with Python And having a, a specific VS code for that. So very small comment on that. I'll put links in the show note, but really excited about what that means to kind of educating the young with that. And then the other one, I was intrigued, not specific copilot, but they they talked about GPT 4 being able to be trained with your organization's content. And it to me, it also talks about having a almost a voice of your organization loaded in with that model. So you can have that central model. I've loved this idea that where we have brand assets, you know, we have images, we have fonts, we have logos that organisations are very focused on. To me, the same should be there for prompts. This is how we should talk as an organization. This is how we should expect, whether it's internal or external things to come through. And I love the idea that you can now kind of load that into the, excuse me, that GPT 4 model itself still a large language model, still looking that kind of cloud hosted, but be able to train that with things specific to your organization. I thought was was very interesting because, because I like those large bits on there as well. Gary, any any particular things that caught your eye on the broader aspects?
[Garry 16:24]
Yeah, so the, I've been tracking these things for for a little while, but it was nice to see the 553 models being talked about, which is the opposite to what you're talking about Kevin, about large language models. This is the small language models and the benefits that you can get from from using these the smaller models which can, you know, perform as well as some as these larger models, but because they're.
[Kevin 16:51]
I'm going to put you on the spot slightly for those who who haven't heard about LLMS and SLMS. What What's what's the difference apart from one's large and one small?
[Garry 17:03]
I guess in the sense of to kind of really dumb it down is, is that you need a hell of a lot more resources to run something like, you know, a ChatGPT model, like you mentioned running in the cloud, right? There's a lot of power that it that it needs. It's very general. So, you know, it's trained on public data as well, masses of of data. So it needs those, those resources to give you responses in like an adequate time. Really with the small language models, it's it doesn't have and not as big, but what we're finding is that the responses can be as good as right as good as you would get for the GPT 4. But maybe if falls down in in other areas before. If you're wanting to go down the route of maybe making a specialized copilot, right, you could start with an A an SLM, which you can run just on a normal desktop, right. I have 5 three running on my desktop and it is very quick. Everything is local.
[Kevin 18:03]
Phones at times as well, aren't they?
[Garry 18:06]
Exactly, I missed The thing is that it's it's moving LL Ms. sorry. It's moving the language models down from like big cloud services down to almost the level of you can run it on a desktop, you can run it on a laptop, you can run it on a phone and have those generative AI capabilities locally to your device, which solves quite a bit of problems. Because you know, there's always talk about, well, where's my data going? What is being trained or you know, all these conversations that that that people.
[Kevin 18:36]
Low bandwidth situations for frontline workers.
[Garry 18:40]
Exactly, disconnected devices as well where it can run locally and you can do things like rag on that as well. You can fine tune it. You can even train it yourself on your custom data, your private data, which I look at that as as a as a vision towards the future.
[Kevin 18:58]
Yeah, that, that's fascinating. That ability to kind of train it with your data and have that model. I thought that was really nice.
[Garry 19:05]
And one of the things that we announced was the AI toolkit for VS Code as well, which helps you do that. It helps you like be able to fine tune and train these models. The training obviously needs a lot of resources and you can use cloud services to do that. But it's I see the the real benefits of using these, these small models in, in, in these other use cases are really interesting. It's like it's, it's, it's something to keep an eye on basically because, you know, I think on device AI is, is, is probably the where you know, in my mind, definitely it's like that, that to me, the future, but it remains to be seen, right? Interesting. So if you've not heard about them, not heard about 5-3, definitely go check it out and and just try it is all I'd say.
[Kevin 19:59]
Yeah, and and I think also what what I liked and I think I might be wrong, but this was the first announcements of some of those multi modal SLMS as well. So there's five three vision. So that ability to kind of view, you know, put images on there and recognise images locally as well, which I thought was really interesting. So I think I know many of do we have some people who listen who are on the development side would be happy to go and play with these others. There's certainly bits of five, three you can go and in in the same way you use ChatGPT. There's easy ways to wrap around that and. And use it which are worth looking at and understanding again how close they are. What what I'm finding fascinating one things I love about working at Avanade is we have such a broad range of people. We have the modern work the the kind of users and adopters and extensions of Copilot and things and we have our data and AI team. We're really getting to the nitty gritty and and trying to get that understanding of yeah, I should use an existing copilot. Yeah, I should extend yeah. So customize it well, yeah, I should go full in training this 5 three model and using it there specific for there and working out those use cases. That's that's a really interesting point and I think a really evolving point for for everyone because these things are moving so fast that everyone's kind of learning in their little worlds on there. And there's a few people I think so and I hopefully fit in that bucket of trying to connect to those things together and sort of say, but when would you do this? When would you do this? When would you do this? And trying to bring that understanding a little bit. And it's very hard when these things are moving so quickly on there. That fascinating at the same time, maybe talking about specialising. We we did say that we'd focus more on extensibility. So I think it's probably now a good time and maybe to kick that off, Gary, you know, the build is about what you can do. It's about all those magical things that gets announced and excited on there. Before we get into what, maybe we should start on why. Why we're talking about extensibility so much.
[Garry 22:05]
Yeah, sure. So I mean, extensibility is something that, you know, it's obviously talked about at build, but I think there's a few people questioning, well, why would I need to do this? Because surely I, you know, it's Copilot's AI, surely AI does everything, right? And whilst it does a lot and certainly Copilot for Microsoft 365, right, it's it's already connected to your organizational data in, in Microsoft 365 is that the reality is your organization doesn't have all of its data in Microsoft 365. You have other systems, you have live business.
[Kevin 22:43]
Systems. Some people don't have their stuff in Microsoft.
[Garry 22:48]
Exactly, you see, but it could be, you know, a line of business systems that you have built yourselves running on, you know, maybe your own infrastructure or actually, you know, running in the cloud. It could be third party services as well. And that is your company data. So what you would want is, well, when I go to Copilot, I want it to understand where all my data is so that it can connect to it. So that when I ask it a question, it's reasoning over all of the data in our organization and then being able to give me a more accurate answer so that we don't have these these gaps. So we talk about extensibility. We're really thinking about how do we connect copilot to these different systems that we have in some cases that might be taking some data and bringing in bringing in that data 365. It could be, you know, just going and getting the data as and when we need it. There's lots of different options. There's also different use cases as well now of like, OK, what what do we want from a end user experience as well? So moving away from OK, I just have this one copilot of of I have a text box. It's like we've got copilot in teams, we've got Co pilots in the different office products as well. And as the products evolves, you know those different kind of entry points to Copilot. If you like, you'll want to have the knowledge of your organization available so that you know it doesn't matter where you are. You ask a question, Copilot will be able to go and access your data and then provide you with a good.
[Kevin 24:29]
Answer yeah, absolutely. And, and I think the, the nice thing I liked about builds was they're starting to evolve that story about taking action with those plugins as well. So getting that information, getting that knowledge, but also taking action. So things like ServiceNow raising a task within their logging a case within things like Salesforce and, and as you say, your own line of business applications within there. That was interesting. I think also when it comes to the why I'd like to think about this is scaling out what you've talked about there is bringing opening up to more information, not necessarily bring it in. I caught myself there on there, but being able to connect to that wider world that's outside your scope of the Copilot, which could be Microsoft 365, it could be your Dynamics and your, your for service, for sales, etcetera, for finance almost lets things slip there shush. But the the other side for then is scaling in. So because Copilot has all this information, there are times it's not always perfect at getting to the right thing. If we, if we look at things like search, if we go back to the old SharePoint search view, it would search everything, but it wouldn't necessarily find the right thing if you were looking safe for a particular policy. And we had bookmarks and the various different things. It's been called over the years that you say if someone searches for this phrase, I want to pin this towards the top and we're scaling in. It's a little bit like that with the extensibility. If someone's asking about policies, whether it's HR policies, IT policies, I want to be able to scope it to that kind of bit of knowledge of those actions that are taken within there. And I think the extensibility also allow those conversations come in. I've been trying to tell this story. I change it slightly depending on what I'm talking to, but I started off thinking about Hamleys that I want to go there. I want to find out about some Lego and what's available. So I go to Hamleys giant toy shop in London, if you haven't heard of it, and I go to walking through the door and I ask someone there. Now they, they know a little bit about everything, but they won't know that specific information. So they'll say you want to go and speak to the Lego person on the I think it's the 5th floor. You can tell I've been there too much and you'll go up there, speak to that person. They may have the knowledge or they may have to go and kind of look it up from there and return. You know, maybe it's is this in stock? Is it available? When's it going to be coming there? They may know it or they may extend there. And the kind of copilot story does that because with copilot, I don't need to actively go and walk in there. I can ask that question and behind the scenes it with the extensibility, you can connect up those different things to give me the answer straight away. But actually there's lots of different things happening behind that to make it happen. So from the end user perspective, there's a single pane of glass to go to, but we can add that specific expertise and route people towards the intent of what they want to get with that as well. And, and that's why I love the extensibility because we're not necessarily in, in many cases, not teaching people new ways of doing things. They just have to ask the questions that would be natural to them, and we're helping Copilot get to the answer in a more effective way.
[Garry 27:46]
Yeah, that's definitely, I absolutely agree with what you're saying. It's like at the moment having a search capability that goes everywhere. But if you're looking for a specific area, you can't just go just just look in that bit, right? With search, you can, you know, you can filter down and say just search within this particular site, for example. But Copilot is very much like it's going across everything. It's always going to your emails, it's always going to SharePoint sites, documents that you might have, you know, played around with or commented on. It's going to do all of that, but that's not necessarily an efficient way of doing things. Like I said, sometimes you already know where it is, you know the expertise. So I want that expert right within my organization to go ask about and and that's some things that.
[Kevin 28:30]
Pay vacation copilot, How much leave have I got?
[Garry 28:33]
Left exactly. And it's it's that where we're starting now to see that that change. So to kind of step back a bit in terms of the whole kind of extensibility story that was taught about at at build and we talk about Copilot extensibility as copilot extensions now. So when you hear copilot extensions, that's the term of here are my options that I've got right. And these are options that you can extend copilot as a whole. So like the global search, if you like, but then how you can extend Copilot to have these really focused, specialized, tailored experiences, which has a difference on the user experience as well. So if we we step back and and look at things that you might already know this, but if you don't, just to recap, you've got you've got the ability with graph connectors to bring data into M365 S It becomes part of Co pilot's knowledge, right? It's it's inside your data boundary, inside M365 from your external system. You need to keep that up to date, maybe refresh that, but there's no user interaction required. They search and they will get answers and copilot will go search that external data. The other option was plugins. Plugins were is is a way of connecting in real time. So making a real time call to a external system that you've got get data from a database really good for data that is, you know, frequently being updated and you just want the latest information. And that is brought into the copilot conversation that you've that that you've got and as context. But that requires you as a user to say, Hey, I need to enable this plug in. So I've got options now. And in my prompt, I need to think about how that plugin kind of gets invoked as well. So those are the the two options. But the third one now is kind of like grouping those together. So we've got a declarative copilot as it's called now that is a tailored version of the larger copilot. But the benefits here is you can give that copilot instructions. So this is a copilot, this this copilot, I want you to behave in this particular way. So you know, it might be customer service, product support queries or something like that. So you can really kind of make it into, you know, helpful assistant, but specialized, you can help with these queries, blah, blah, blah. So that defines how it should should interact with the person who's who's using it. The next part is what knowledge it has. So rather than it having all of the knowledge of the organization, you can say you just need to know this information about products and we have our product.
[Kevin 31:26]
Information.
[Garry 31:28]
Exactly, we have that product information maybe brought in by a graph connector or it's stored in a SharePoint site or, you know, other data sources which which you can use. But it's that idea of of your constraining the copilot so that it it doesn't go, you know, I'm gonna check all your emails and everything and find, you know, absolutely everything and return it back. It's like, no, here's here's our like maybe you've got a curated set of documents in your organization. You only want those documents to be part of the response. Then a declarative copilot will will, will give you that. And then to lay that on, you've got plugins in in declarative Co pilots. But because the copilot knows which plug insurance it's got available, it's not an explicit action on the end user if they're just there, right? So again, from an end user perspective, it makes it a little bit easier because you're working in a, a particular domain, right? That that copilot is a, is a specialist in. So when we talk about declarative copilots, like what you said, Kev, it's like, you know, HR copilot, finance copilot, whatever you need, right? It's a it's a tailored experience that that you can create either through, you know, Visual Studio Code or even like Copilot Studio.
[Kevin 32:51]
Yeah. And one, one thing, I would the other thing I'd say this change that if people might have looked at plugins in Copilot for Microsoft 365 in the past, there's a little teeny box and I'm lucky enough to be in the MVP sandbox, but last time I counted, I think there was about 60 plug insurance. And you scroll up and down and it may be a little nervous because as organisations get more and more plug insurance connecting to more things, it's going to get harder and harder to find and turn those on. But I think there's been a shift to that model that's going to be a bit more like Copilot Pro, which is the the Microsoft Copilot, the one you're seeing being on there, that kind of consumer facing one where you can see those plugins down the right. They were called GP TS within there. I think where I've got to make sure I get the naming convention, but I think they are now being called, is it extensions with those on the right? I know Jeff Teper talked about those as as part of the keynote. So it's going to become easier to discover those and, and again, I think you touched on this that you can @those as well. So where you know the name of them, you can bring that up. It isn't going to be the only way to invoke those. If you, you know, if you say, what is my vacation on their copilot will still have that intelligence to root towards that plug in. So you don't have to always @it, but if you want to be very specific and make sure you do, you've got that capability as well. So I really like that change to the model to make it easier to, to bring those plugins front and centre for, for everyone. I felt they were, they felt quite technical and you had to really make an effort to get towards those. I think with this evolution there, there's going to be less of that and and it's easier to see what's available and, and what you can bring into there as well.
[Garry 34:39]
Yeah. And you know, mentioned Jeff and and in his keynote then it was the Co pilots in SharePoint, which was, you know, incredibly easy intuitive of that. You know, you've got a document library, you're already working in SharePoint, you want to be able to reason over those documents. You know, you can now click of a button, create a copilot that's grounded in in that data and then share that with others as well. And then once that's created, there's then abilities to extend that with Copilot studio. But it's, it's making it very easy for people to be able to create these kind of experiences, like a specialized tailored experiences that that, yeah, helps you get more relevant, accurate answers for, for, for, you know, whatever questions you want to ask rather than, you know, oh, I need to search everything.
[Kevin 35:39]
Absolutely. Now one of the things I love about not working for Microsoft is that I love spreading the joy of these things, but I can also highlight some of the concerns. So feel free Gary to, to not comment on some of these bits because you, you mentioned that I can go in and have this, these Co pilots in SharePoint and I can create that and share with everyone. So we, we're kind of saying that anyone in a large organization could go in and create the SharePoint and share that. And I'm sure many of you are going governance, governance, governance, governance. How are we going to maintain those? I think that's going to be an interesting thing to watch. And I, I hope that we will see some more tools. I, I, I'd like that Microsoft's going to kind of make this available and get people encouraging them to use this and get moving, not blocking it from the start. But I hope we'll see that fairly rapid follow up in ways to kind of look at is it being used? If it's not, can we remove it and hide it? Can we maintain, I'm sure third parties, you know, we have Michael Bisarik from orchestra. We've got people like Rencor have points, many others that are, that are out there will, will hopefully follow up and look at these as the governance. But I, I do have a slight concern with that. I I think helping organisations to manage these effectively will become something very important. Following up soon on that as well.
[Garry 37:05]
Yeah, of course. And you know the things that have have been shown at build, a lot of these things that we've talked about are private preview. So things that we have announced they you know are not complete, we go through private preview with specific customers. We then do like public preview before then obviously getting to GA where things then are more stable. So and that is the time for feedback. So we need to hear this feedback to make the changes that that are going to be needed for our organizations to use this effectively. The one thing I will say is it does come back down to the basics of what is your security like as well. What is your governance around the documents because you know, you as a user still need to access those documents. It's not that it's, oh, now we've got a, a copilot. It's now open. You know, it's still grounded in the security models in, in M365 as well. That's the benefits of, of, of using AI with Microsoft 365 because you, you get all those security controls out the box, which we've been using and accustomed to for, for years. And that still applies to, to copilot. So yeah, I think that would be the one thing is make sure that, you know, house is in order as well.
[Kevin 38:34]
I completely agree. And if if people want a deeper dive in that, go back and check out our episodes with Michael on Copilot readiness looking at looking at the importance of that and different ways to do it within there as well. I think we we covered a lot in there. I think still there's the again, look at the MVP sandbox and the number of Copilot plugins in there called demo or test things.
[Garry 38:56]
Like.
[Kevin 38:56]
That and things like policy copilot, you know, we've seen the same thing happen with teams where he's my finance teams like, oh, is that the IT finance, the sales finance, the finance, finance and things like that. So I, we, we see these things happen. I to me, I think it's important we get this capability out and we have people to challenge and say these are things that need to follow up pretty soon afterwards. But let's, let's not build all the things in there straight away and block, block the innovation ideas from happening until we really need to. So hopefully, hopefully we'll see that follow up come through. And and yeah, completely agree with with sorting out the security that picks up as well. So in, in terms of creating these, and I think you touched on this a bit, Gary, you've you what I really loved from build is it felt like there was a, a more solid pro code story. So you've got those ability with things like teams toolkits to create these graph connectors. You've got their teams message extensions available within there. You've also got Co, you've got power platform side and you've got the copilot studio way of building on these as well. So there is that low code and pro code and it it it was great hearing Jeff in the keynote pushing both sides. Because I I feel copilot studio at times has had a a little bit more focus in in the other areas. So it was good to see both both of those mentioned within that. One thing I got very excited and I'm going to bring up my my screen is you touched on that ability to add knowledge and I always love it at build when they announce something and make that preview available straight away. Copilot Studio updated So I think it was on Tuesday overnight. I, I will give a shout out to Mark Rackley for his bravery on Wednesday. He was doing a copilot studio training session to several 100 people at Avanade and he had to rewrite pretty much all of it because suddenly it all changed. So a huge kudos to him to to actually get that running and show some of the new stuff that had come in there from that. But you can see it's bringing more of those knowledge, basic knowledge sources. Where before there was kind of SharePoint of one drive and files and public websites, there's now that preview of data verse coming in there, but also those enterprise data elements, which is a combination of of those power platform connectors that are available and those graph connectors. And, and I loved seeing this for those looking screen. I can see I've had my Viva learning app that's in there by Pokémon Connector that I love demoing on there, the Salesforce connector, the ServiceNow and it it's so good seeing the kind of speeds with which these things come available. I have distracted myself slightly by seeing ADO gets which seems to be git and as you are DevOps I mean treat I guess that's the the Git side of Azure DevOps that you can connect to versus the wiki side, but I noticed it called that before.
[Garry 42:01]
That has been Azure Repos inside Azure DevOps, right?
[Kevin 42:07]
Yeah, that seems to be a rename anyway. Absolutely. What I love about this is before, if you were talking about graph connectors, there was a lot of kind of code involved, You know, it became a load to pro code. Now we're starting to see more of the wizard connectability connections with this, it's become a little bit easier. Having played around with some, there's a little way to go. I wouldn't say it's easy easy, but it's it's certainly a big step to towards that and and worth looking at. I would give a caveat when looking at these. If you are looking those graph connections, they will only work in Copilot Studio if you have a Microsoft Copilot license. We assume that's a Copilot for Microsoft 365, but have asked that question to Microsoft on there. But if you want to use those graph connectors in Copilot Studio, you need that license on there. I would also see notes and I'm still testing to get this working, but you need to set up authentication. So there's a few steps with running that building app registrations that are a little bit more complex than click, click, click. So I'm hoping to get some videos over the next few weeks. Well, 8 get it working and B then get some videos and share some things on how to do that. But I thought it was really good to see that there there is really that support for that low code and pro codes capabilities coming through still.
[Garry 43:33]
Yeah. And and on that point, I think that you know, it's, it's the different organizations have different needs. There was a great slide actually in the developers guide to customizing Copilot session run by Jeremy Fake and Barnum Bora that goes into, you know, the code approach and the low code approach and maybe the considerations that you would want to have in mind when using either approach. So, you know, it's great that, you know, things are made more accessible through the user interface, like what you just showed there, Kev, and that is, you know, clicking around, getting things set up. But it's like there's some downsides to working a platform. Sometimes you don't have the full visibility. What what about source control, other things that you might need if you're an independent software vendor, you know, do you want to use power platform or do you want the control of using code? You know, there's all sorts of different considerations and there's different ways of of, of going about this. You know, if you got multi tenants, have you got single tenants as well? So keeping all that in mind as to which approach is best for you is something that, you know, encourage everyone to to try out. Definitely use copilot studio. It's the fastest way to to to extend Copilot thinking of you know, there are certain limitations that you may hit by building in a platform, but there is the pro code side to fall back on as well. So definitely keep both options in mind.
[Kevin 45:08]
I'm going to hide the slides a SEC. You've reminded me. We're going to talk about this and I completely forgot because Jeremy's made those slides available and we will put the put those into the show notes. What I would say is go and watch the session because the the conversations that happen around it. And again, we'll put the slides in there around that as well. There we go. That's a slide I was looking for because because what Jeremy adds towards this is, is well worth watching the conversation. And just because, Gary, you were talking about that, that kind of what to use when there's some great examples. And, and this is some of my favorite. I've seen various people talking about. Well, if you've got loco people go loco. If you've got pro go that this was the the great one that picked up a few other things. So you know, if you've got existing power platform solutions, you want to move fast as you were touching on there, you've got that managed infrastructure, great copilot studios there. But if you've got things like a multi tenant deployment and you want to deploy across tenants, then exactly you were talking about that source control, that ability to maintain that branching those releases, being able to manage that infrastructure and scale that to the way that you want the control, then you're edging towards that pro code teams toolkit. So well, well worth digging into that session and and having a look at that. I'm going to jump back to to what I shows and we'll talk through this for everyone. I know many people listen rather than watch, so we'll make sure we cover the different things on there because I love to also that he was talking about the kind of different layers. And I know Satya talked about the layers of copilot of having those models and things, but this was more about I've got too far now. Where's it gone the the fundamentals of those extensions? So I don't know, Gary, do you want to talk through these at all?
[Garry 47:08]
Yeah. So it's.
[Kevin 47:09]
It's good to know.
[Garry 47:10]
It's good to know the layers right of of kind of like it's almost a way of thinking well, what, what am I getting through copilot for M365 as well? And what am I building on top of? Because they're on one side, there's extending, you know, copilot, there's the other side of it where it's going and building your own, right. And going and building your own is like, OK, yeah, I can just go and do that. But then there's other things that you need to think of. So kind of starting at the bottom with the foundational models. You know, this is where we're talking about the large language models and you know, ChatGPT, you know.
[Kevin 47:43]
All the small language models potentially.
[Garry 47:45]
All the small language models, yeah, that foundational model, that generative AI model that that you're going to use. So obviously in Copilot, that is something that is abstracted away from you, you know, that's part of the service. But if you're building your own, hey, you get to choose whatever you like for whatever your needs that you want. Then you've got the orchestrator on top. So obviously Microsoft 365 has an orchestrator that will go and take your prompts and say, oh, you know, I need to go and find some data from different sources. Is the sauces that I've got available, Which one should I choose? You know that decision maker if you like.
[Kevin 48:21]
And I'd like to talk to it as that user intent mapping. What is the user trying to get to? It's it's the bit that works out that that connects to whether it's that knowledge blah blah blah.
[Garry 48:32]
Yeah, exactly. And and you know that reasoning part of understanding and then acting upon what has been been asked, obviously the next side is then the knowledge. So what? What data is available to to to that system as well? Obviously Copilot for Microsoft 365, by default its knowledge is Microsoft 365, but that could be also external systems as well. And and being able to access that triggers being able to kind of run automation. So when something happens, use generative AI to understand, OK, something happened, what do I now need to do? And that could be to, you know, start a workflow and, and, and automate that kind of side of things, but using natural, natural language actions as well. So you know based on that, that prompt that the users put in maybe like invoke a plug in for example, like oh I need to get this data from this external system. I have a plug in that can do that. I need to pass it some information to get some information back. I it makes that action on your behalf and then the instructions as well, which is, you know, how should you behave? And these are different things that are relevant to the different options. And I think the next slide that that you'll show Kev is, you know, here are all the different options that we've that we talked about using these extensibility options in isolation with just graph connector or a plug in. And then how they actually combine when being used in a declarative copilot and then versus the, you know, if you decide to go on your own, like you want to build your own copilot that is not using copilot for Microsoft 365, you know, where are the differences? Where do you have to then think about, you know, which orchestrator do I use? Which LLM do I use? So this, this is the layers that we're talking about. And you know, if I feel it, we're kind of back to a bit of the remember pizza slide, the IRS and Payas. Sorry, you know, it's like it's a piece of the, the, the pizza, right? That some bits you have to deal with of the bits you don't the benefits of using a manager system versus maybe the disadvantages. All these things need to be taken into consideration when you're kind of thinking down this route. And I've seen a lot of people go in Russian building their own custom Co pilots, like maybe because, you know, they don't want to pay a license or it wasn't, maybe we didn't understand, you know, what the capabilities of Copilot for Microsoft 365. But you know, on the flip side, you've got a lot of work that you need to manage and maintain off the back of that. So, you know, we're back into scenario of the, you know, do I use managed? Do I use, you know, do I build my own, which is again, the like centralized and decentralized that we talked about before, right? It's the eternal bow. But yeah, hopefully you know the information from Bill is helping people make better decisions as well as to which option is right for them or for for their customers as well.
[Kevin 52:01]
Absolutely. And I think that's that's important thing is to take in those considerations for these who can't see the screen and we'll put in the show notes. There is a link to AKA dot Ms. slash extend copilot M365 slash decide. And the the team that that put this session together have got a great page of all those different things you should should think about. And you know, there's a value to this. There's what is going to be the cost of going copilot for M365, the cost of extending and building. There's a cost of building your own custom copilot versus the benefits you're going to get, which I I think all need to be to be thought about because they're there's certainly some interesting costs you need to think about and modelling when you're bringing these custom Co pilots and doing things like as you're open AI, how many calls are you going to make? How many hits is it going to get? Do you need to be consumption based? You blah, blah, blah, blah, blah, blah, blah, Which are, you know, worth thinking about and digging into? And I'm not going to cover that because lots of consultancy will go, Hey, that's our job to have that. But there there is lots of materials for people to go out and have those thoughts and and conversations. So I think it's great. I feel that Mike Sauce making that that consideration a lot easier for people. They're starting to help guide people towards that more effective because they're learning and everyone is still very much learning at this stage. And it, it, it was really good to see that that presentation very, very much enjoyed that.
[Garry 53:31]
Yeah. In terms of the going down the custom copilot route, if that's something that you're interested in, there is tooling available to like build custom Co pilots inside teams or custom engine Co pilots now as as we're referring to them. So if you hear that custom engine copilot, then it it's, you know, build your own essentially we've got teams AI libraries that helps you build AI capable bots in teams. So it's it's very specific to teams, which makes.
[Kevin 54:02]
Things quick quick question, is there a rule that every time someone says bots they have to down a shot now because it's it's banned or?
[Garry 54:09]
Is that still allowed? But yeah, there's, there's the tooling that's there. And then you have the, you know, the services from Azure, so Azure Open AI service, I said the open AI models that are available, you know, like I said, the the new model is available straight away as one of the the announcements that it's there, it's available for you to to use. And then it comes with, you know, all of the safety checks compliance that that you get with, you know, using a service in, in, in Azure. So you know the all of the tools are there for you to to go build these custom Co pilots as well.
[Kevin 54:56]
Yeah, I'm probably worth mentioning AI studio as as well. One actually one decisions are put in there that I haven't had a chance to watch. But Hank Bowlman and Dan Leskovitz did a session of copilot studio versus AI studio. I mean they seem to have missed the whole team's AI and the other side, but well, I've had that argument with Dan already, but do want to watch that. So I think that's an interesting one. And and for me, I see a lot of that of if you look at employees, you've got to focus on teams, you've got to focus on that internal view of things. If you're looking external, then the Azure AI studio comes in there and copilot studio stretches a bit across both. So think about those things is key. We're kind of coming close to the hour. Was was everything else that you particularly wanted to wrap up on Gary?
[Garry 55:50]
Not particularly. All I'd say is, you know, if you weren't able to see some of the sessions at build, there'll be sessions that were put in in the notes, I guess the, the key ones from the copilot extensibility perspective and custom building custom Co pilots as well as yeah, just go back through the the schedule. The recordings should now all be there by the time this is published. So you should be able to go and view them, download them as well. And, and you know, go to speed on on this. But you know, things are in preview. Things will be, you know, coming over the the coming weeks and and months. So yeah, it's a here are some announcements, but it's, it's, it's not the not the end the, you know, there are more things coming and just things change, right? If you tried Copilot 3 months ago, try it again, chances are you'll get, you'll get different. OK, OK. Any copilot, we're in the world of experimentation. But you know, it's major improvements in, in AI as we've seen, you know, just in general. So, you know, make sure you you're going back and just trying things again because, you know, things are only getting better as as well. So yeah, yeah, just keep trying it and, and, and see where you are. And if you have any questions, then bombard us with your questions. Yeah, absolutely.
[Kevin 57:25]
Great. And and you know, wrapping up the show, let us let us know what more you want to hear or if you want to hear more on custom extensibility, particular areas and voices. I'll try and see if I can get down maybe even Dan and Hank to come on the show. That could be quite fun to grab them together and share that one as well. But thank you, Gary. Where one thing's trying to kind of highlight of events coming up. We've got comms verse in I think almost exactly a month's time over in the UK and I know there's going to be a lot of good copilot sessions and some dev sessions on there. Are you coming to Scottish Summits, Gary in October?
[Garry 58:06]
Rare situation. I don't have any events planned in so I'll yeah nothing.
[Kevin 58:13]
I I'm definitely going to be hustling people and getting you to come to ESPC in December as well, which are there.
[Garry 58:19]
I think.
[Kevin 58:21]
On that I would say one other session, June the 20th, I'm part of the Experts Live UK organising committee. It's a community powered set of user groups. We're looking to move towards having a conference at some point. Nothing confirmed yet on there, so we're going to run some user groups and June the 20th we're in London at the Advania HQ. We got Amy Boyd coming talk about AI and Steve Goodman's going to be talking about agents and spring some with him to talk about auto Gen. and the the elements there. And I think agents, one of the things we haven't touched on too much in this episode, I'm hoping Zoe and Chris will cover a bit more, but it's a really interesting space that's evolving there. So if you have a look at expertsliveuk.co.sorryexpertslive.co.uk is the website with information on how to register on that. So keep an eye out for that. And a final plug for Month of Co Pilots coming up in June, go to copilotconnection.com/month of Co Pilots or just go on there, the main page and click from there and you can register and be be alerted when all those sessions come up. And also, if you want to hear more than please, please, please, please subscribe to us. We're on LinkedIn, we're on X, we're on YouTube. Go and follow us there and be the first to hear us. Our videos for Month of Copilot and all these podcast episodes come through from there. So tell your clients, tell your friends, even tell your partners, drop some reviews in there. I need to have a look before to see if there's any reviews, but please put all those things in. I want to say huge, huge thank you for to Gary for stepping in and coming join this chat today. Really, really enjoyed that. Remind me of a good old days.
[Garry 01:00:03]
Yeah, thanks. It's been good to chat.
[Kevin 01:00:09]
Absolutely. Otherwise, thank you very much and see you all next time. Bye bye.
[Garry 01:00:15]
Bye bye.