chapters transcript notes
click any line to jump to that moment in the video
0:13 Hello everyone and welcome back to a genetic thinking with me. I have a wonderful PM, a program manager from Microsoft. Just send her welcome to the show. I'm very excited. You were talking to me a little bit over teams about this 0:24 new feature that you're introducing here. Tell us more about this really exciting feature and let's unpack this. Over to you. Sounds good. Thanks Mike. we've 0:34 recently announced co-pilot in web modeling in preview for Power BI service. And essentially it's an AI powered assistant within Power BI 0:44 service web modeling view that lets you analyze and improve your semantic model. it's built on top of our modeling MCP server which you may have already tried. We've gotten lots of 0:54 great feedback on that it. It's good. we wanted to bring a very similar experience to in product and just help users to get started really easy. and 1:06 some of the key capabilities that it has is model analysis and guided recommendations. if you have some inconsistent naming or just structure that you want to improve in your model, 1:17 co-pilot can help you do that, suggest improvements and implement them as . You can do model editing there as you can create, update, manage tables and relationships. And you can 1:27 also improve your model discoverability as such as implementing clearer descriptions, proposing display folder structures and just recommending better technical and more user-friendly 1:38 names as . . I really that feature that you mentioned there because I'm having a lot of clients that I'm working with where I'm explaining it this way and you can 1:49 you can resonate with this if you or not and you can tell me what you feel your feedback on this one but if I give a model to another user in my organization And they just simply don't understand what tables are there and 2:00 what relationships are across the different tables. And if I pick a specific measure, can I use that measure with a different dimensions? If I don't communicate through the semantic 2:11 model to my other business users, how on earth is an agent or someone else going to pick up on these things? , really enhancing the model with better descriptions, explaining why the DAX is 2:22 there, what's the business logic behind it, we can capture a lot of really useful business information inside the semantic model. And we really And I think this is another great 2:33 opportunity. I also don't love typing in little things each time. , you know, having an an agent in front of me that I'm a much better reacting to a 2:46 statement than I am making a brand new statement from scratch. , even just suggesting what is why this table exists and how it got there, letting the agent look through the code, provide a really 2:58 good description or at least an initial summary, I can react to it and say, "Oh, that was really good." and just approve it. Or I can say, "No, no, that's not what I wanted." and then I have more of an opinion about how we we shape 3:09 and name things. , that's I think that's really amazing. I want to pick up on something here you mentioned just briefly. You said really quickly in passing the MCP server. 3:19 is that Again, I'm just trying to mechanically put these together. I'm talking to a co-pilot, but you're bolting on that MC MCP server capabilities. It has the same 3:30 tool sets, tool calls that the MCP server has. Yes? Correct. Yeah, it's built on top of our modeling MCP server, which we've heard lots of great customer feedback 3:41 on. And we just wanted to provide those same experiences within Power BI service. Just making sure that the experience is very easy to use. You don't have to download the MCP server, 3:51 install anything. It's just there within service. , we're providing same capabilities in service. I love this one. And , Matthias Tierbach, another gentleman who on the 4:01 show who helps us do the Indent Thinking podcast, we have done a very clear demonstration. We had a full pre-con day all day on the Fabric 4:11 conference in Atlanta. And we did an 8-hour session on nothing but the Fabric MCP server both remote and local on inside the session. , we are super 4:23 positive on the MCP server. Very capable. It makes modeling, I think, a lot more fun, in my opinion. you can go around. You can ask it to document things. I was 4:34 having some demos around, , understand where this table came from. Build me a mermaid diagram or document that supports that. And , the MCP 4:44 server is a huge advantage here to be able to directly link that to the Copilot agent, which I absolutely love. Awesome. Yes. All . Let's I think you've got a 4:54 demo for us. We have a little video that's pre-recorded, we're going to put that on. We're going to go through the demo here. And then let's talk about or wrap up any other final features and where the community can get involved with this as . , let's go 5:04 over here and let's run the demo. Here we go. In this demo, I'll show a few of Copilot and web modeling's capabilities. I'll start from the ribbon and Copilot will 5:14 provide prompts for me to get started to improve my semantic model. I'll select improve table relationships and then click on allow to provide permission for changes to be made. 5:24 What's great is that as Copilot is processing, I can also see a set of reasoning steps that it's going through. And it looks it's done analyzing what the current state is and what table 5:34 relationships I'm missing. And it looks I'm missing a total of five relationships between the GL fact table and the surrounding dimension tables. And it specifically calls out which 5:44 columns they should be drawn from. I'll go ahead and say create exactly those five relationships. And in just a little bit, I'll see that those five relationships 5:55 have been created and updated in the model view. It also suggests what I should do next. And I want to go ahead and hide some of the key columns in the dimension tables. 6:05 I notice it's not hidden for department and division. And I've told it to hide it and I can see the I icon show up indicating that they're hidden. 6:15 I also have a an account _name column that's a duplicate and I want to delete it. I'm going to provide confirmation to go ahead and do and just that 6:25 you'll notice that in real time it's updated and it's deleted from my model. Lastly, I want to show you a capability that is coming soon to Copilot in web modeling, which is the 6:35 ability to suggest and set AI instructions and schema to help prepare your model's AI readiness through prep data for AI. I'll ask Copilot to suggest me what AI 6:45 instructions I should add and it provides me a variety grouped into different categories. I'll start by saying that I just want to add a few of these. I'm going to say 6:56 add two through six. And in just a little bit, you'll see that it has successfully added those instructions and I can go to prep data for AI. 7:07 And I can go ahead and click on add AI instructions to see that those have been successfully added. And I can go ahead and make modifications later as I . 7:17 To recap, Copilot in web modeling improves my semantic model according to best practices and helps to make it ready for AI. 7:28 wonderful demo. Really what you were doing there. This is great. let me ask a question here cuz Tommy and I run the Explicit Measures podcast and we talk about this a lot. One of the things that we were trying to unpack was 7:39 where does instructions go? Where does the business logic of the model live? And you said prep data for AI. Correct me if I'm wrong, that's a 7:49 markdown file in the folder structure of the semantic model, correct? Yes. Yeah, and that prep data for AI is coming very soon. That support to be 7:59 able to add AI instructions and modify your schema selection. , stay tuned for that support very soon. And just to be again, help me understand about that one as . , is that in 8:10 today in desktop, can you do that? Is that in a desktop feature ? To be able to use Copilot to set AI instructions, not yet. But, you can use just manually prep data for AI in 8:21 desktop. I don't want to do that without having I want the Copilot to write the instructions for me. , this is while while I understand I understand the feature and and I'm going 8:31 to 100% be in the service using Copilot with AI to write these initial Again, it's the it's the first pass, ? I A lot of people would talk about 8:41 this whole concept of AI slop and it doesn't make good things. pause there for a second, ? I do expect you to read the instructions. I do expect you to understand what 8:51 business things it's doing, but the fact that I can then give a first pass of the Copilot and say, you go through the model, you understand what's going on in the model, give me that list of 9:01 instructions. And I do I really do think Again, we're really beating up this idea on Tuesday. This is just Tuesday, yesterday. We're beating this idea we really need some place to 9:11 put these instructions and have them in there. , giving us ability to do this in the web, having the ability to see those instructions inside desktop is incredibly important. And 9:21 I think that's where business context should live. we really do need that really rich enhancement, ? From from users to be able to 9:32 apply all the extra instructions that AI or agents need to understand about a particular model. , for those of you who are listening to us on Tuesday, we were just complaining about it and tada, 9:43 on Wednesday you get the feature. , apparently Tommy and I need to do a little bit more homework and make sure we understand all the features that are coming out . But, this is amazing. Thank you much for this demo. 9:54 Anything else you'd to cover off with? I think you have a couple more links for us. Yes. what else can we share with the audience here? Yeah, I wanted to share with you all our official documentation on Microsoft 10:04 Learn. please check it out, learn more and we are very excited for you to try it out. We also have a blog to complement that as to just dive into more of what our 10:15 feature is about. And if you have any feedback that you to share, we're really eager to hear about it. , there is a community post page yeah, I have a page that you can post 10:25 on the Microsoft Fabric community page just to share any feedback you have. If there's any improvements you want to see, anything that was working really for you, we would love to hear it 10:36 that we can iterate on this feature and help improve the experience. Awesome. For those of you who are listening and watching the show, all the links are in the descriptions. , in 10:46 the description of this video, there's all all the links that you are going to find are there. Also, for those who are listening and watching live, all of the chat windows for most of the platforms, I think X is the only one 10:56 that we don't have on here , but , Twitch, Facebook, YouTube, LinkedIn, all of the links that you need will be in the chat window as . , feel free to hit those links, check them out. Make sure you 11:07 give feedback. I again, Microsoft listens to these things. The reason the feedback forms exist, Justina, you're going to read those things. You're going to go back through and comb the details and figure out what people are asking for, what do they want to go do. , 11:18 make sure you're vocal about what you and what you wish it would do because that might be another round of a feature revisions on how we can leverage this inside the service. 11:29 Exactly. Thank you much for this demo. I think this is going to be a super fun feature. This is what we're talking about, agents and things are really weaving their way into our workloads directly inside 11:40 fabric, which I'm finding extremely valuable. I love the fact that you're using the MCP server again and reusing the best practices there and leveraging it in other tools. I think that's a a 11:50 huge unlock for many organizations and no matter what tooling it is. I've been working with a lot of organizations that don't have a co-pilot or a GitHub co-pilot or an an 12:02 ability to turn on an agent. And , being able to provide direct localization to co-pilot for this feature directly, I think it'll be immensely useful and people can leverage 12:12 that really rich MCP server that you would really to talk to and make a lot of simple changes across your model. Just send a thank you much. Appreciate your time. We'll have to do 12:22 another one. Let us know when the next feature comes out. Thank you for joining on the show. We appreciate you all listening. Thank you much and we'll Thank you.