Claude Design & Power BI Embedded – Ep.533

Claude Design opens up a much bigger conversation than one new AI feature. In this episode, Mike and Tommy unpack how design-first workflows, embedded thinking, and lightweight HTML prototypes may change the way Power BI teams plan, prototype, and eventually build reports.

News & Announcements

This episode starts with a broader reality check on how quickly AI tooling is moving. Mike and Tommy talk through the current wave of design-oriented tools, the growing role of agents in day-to-day work, and Mike’s ongoing push toward more agentic reporting workflows tied to Fabric, embedded scenarios, and rapid prototyping.

They also frame Claude Design as part of a larger shift: AI is no longer just helping write code in the background. It is increasingly being used to shape interfaces, generate wireframes, explore style systems, and accelerate the earliest stages of report and application design.

Main Discussion

The main discussion centers on a practical question: what can Claude Design and adjacent AI workflows actually do for Power BI teams right now? Mike and Tommy agree the opportunity is real, but they spend most of the episode debating where the value is immediate, where the tooling is still immature, and how different the experience is for embedded developers versus traditional Power BI Desktop authors.

• Claude Design is most useful as a design and wireframing accelerator. The hosts describe it as a way to work with fonts, colors, components, layouts, and visual direction before worrying about the final implementation details.
• Power BI Embedded is closer to the current AI sweet spot than Desktop-only workflows. Mike sees embedded and web-based reporting experiences as the place where design automation, HTML prototypes, and agentic creation can move fastest.
• Context still matters more than hype. Tommy repeatedly emphasizes that good report design depends on understanding business questions, stakeholder needs, and the meaning of the data—not just generating something that looks polished.
• Prompt engineering is only part of the story. The episode draws a useful distinction between prompt engineering and context engineering, arguing that the best results come from pairing instructions with the right transcripts, prior examples, and business context.
• Wireframes are the practical bridge for most teams today. Rather than expecting AI to fully build a production Power BI report, the hosts recommend using AI to create design inspiration, rough layouts, and single-page mockups that help teams get unstuck.
• HTML prototypes are becoming a useful intermediate format. Mike argues that asking an agent to generate a single-file HTML page is an approachable way to test layouts, styles, and report ideas before final implementation.
• The debate gets spicy around what a report actually is. Mike pushes the idea that reports, apps, and websites are converging, especially as embedded experiences become more interactive and action-oriented; Tommy pushes back from the perspective of classic dashboard design and decision support.
• Mainstream adoption still depends on better tooling. Both hosts acknowledge that most Power BI users are not yet working with the same agent stacks, context libraries, or technical workflows that advanced builders are testing today.

A key takeaway from the episode is that AI does not need to build the final dashboard to be valuable. If it can help a team move from vague stakeholder notes to a useful wireframe, style direction, or prototype faster, that alone can improve the reporting workflow.

Looking Forward

If you want to experiment with this workflow yourself, start small: take a meeting transcript, a list of priority metrics, or a reference image and ask an agent to build a simple wireframe or single-file HTML mockup. That exercise alone can help you test design ideas faster, clarify stakeholder expectations earlier, and prepare for a future where report design and embedded experiences become much more agent-driven.

Episode Transcript

0:07 measures. Pump it up high and lighting up the sky. Dance to the day to laugh in the mix. Fabric and A. I get your feels. Explicit measures. Drop the beat now. Feel the crowd. Explicit measures. Tommy, good morning and welcome back to another episode of the Explicit Measures podcast. How you doing, Tommy? It feels good to be here. It feels good to be with you as always, my friend.

0:38 to be with you as always, my friend. We live in exciting times these days. I think everything has been moving very fast, super quick here. It’s hard to keep up with everything. I feel like I’m trying to spend late nights catching up on things and making sure I’m familiar with all the stuff that’s coming out. It’s really hard to keep up with all the different features and designs that are coming out these days. I think this is partly accelerated Tommy, my opinion here around the reason this feels so fever like fever paced right now is because AI agents are now building a lot more code. It we physically are getting

1:09 more code. It we physically are getting more code generation, more projects, more GitHub. It’s just the agents are able to do more and people are just leveraging them three, four, five, sixes of them at a time to build stuff. I think that’s that’s why we feel overwhelmed because there’s physically more things to look at and review. And really too there’s the other side of the coin too and what we’ll talk about today is it’s not just doing code if you’re not an application developer. Well, it can also help design and that’s a huge part too. So I think we’re seeing this whole forefront of one aiding my

1:43 this whole forefront of one aiding my ability to get started but then two to your point anything I can think of anything I can think what would be cool can more or less be developed which is insane which is so cool. It’s really neat and I think I think we’re entering a new So when code is no longer scary to people to write or build and when you start programming in your native tongue whatever language you speak in versus actually having to write

2:13 speak in versus actually having to write software language. Yeah. I think we’ve gotten to a point, Tommy, where some of these more advanced models that are coming out now, if you have access to them, there there’s not this scary cliff of I can’t understand what the code’s writing. And there’s there is I think there is this moment of hesitation for people in this space. There’s a little bit of hesitation like, well, I write code. Code was my job. Writing code and building software was what I thought I was going to do. And I

2:44 what I thought I was going to do. And I think that just changed for a lot of people which is making them having to rethink their careers to some degree. I I would actually ask you a better question or ask you a question here. Do you think there’s more hesitation around developers or around people who never coded before? I I think you’re seeing what I think we’re seeing here, Tom, is we’re seeing a very technical space called development and software and codew writing being open up to even though I’m

3:14 writing being open up to even though I’m not writing the code, I’m able to generate the code with the tools that I have, right? Give Tommy a $300 a month subscription to some agent and now Tommy can now build websites. Right? Previously, you would not do them. It was too hard. It was too difficult. Unless you had a project to start from, you wouldn’t build all these different things. Now you’re able to,, not write the code, but you can then describe to an agent like what you want it to build and how you want to build things. things. So, So, totally. totally. I think I think what we’ve done is we’ve taken a very technical space and opened

3:44 taken a very technical space and opened up its availability to more people to to more. It’s going to be there’s there’s more more I’m going to say the word Tommy. It’s I think you’re going to it’s going to rub you the wrong way. We’ve opened up to more developers, but we’re not saying developers in the same way. The word developer means something different now. It it’s it’s less about can you write a function in a certain language or code. It’s more can you articulate a prompt to an agent to produce the right output. And that could be graphical, it could be code, it

4:15 could be graphical, it could be code, it could be could be something else. Like so it’s it’s it’s this new medium that we’re trying to explore. I’ll do you one better because I actually was talking to someone on Friday and you’re gonna like this where they said, “Hey, we’re basically looking now to hire professional Q&A.” Like that’s really the role right now is rather than developers like we’re just prof we’re hiring professional Q&A Q&A people ra and you talked about middle management before where I think a lot of it is I’m not necessarily building the

4:46 it is I’m not necessarily building the code to your point or and again I I when I say code I’m selling it short because there’s sure graphic design sable diffusion actually did better in in AdSense for like display images on advertising And people did they did the study and it’s cheaper. cheaper. Sure. Sure. You’re finding it with the content and then you’re finding with obviously application development but the person I was talking to said someone needs to review it still and I need professionals who can do that.

5:19 This is interesting. Let’s get into the main topic today. So let’s Yeah. Yeah. I like what your statement is. I need to react to this. I’m still trying to un I’m trying to unpack what you’re saying here a little bit because I think in parts of it I agree. Some of I maybe have a slightly different perspective on some of this. That’s all for an episode. Yeah, this might be another whole episode. You just PowerBI Q&A for now on. So that’s good. That’s I’m putting that in the backlog. So okay. So okay. Okay. so the main topic for today is claw design and PowerBI embedded or just in PowerBI in general. What does cloud design do to this new space? Claw

5:50 design do to this new space? Claw design. Let’s let’s maybe we should unpack that a little bit. Tommy, do you want to give us a quick one, two of like what is Claude design? Have you played with it yet? Have you used it yet? Yeah. So, I’ll let you talk about a piece. Yo, yeah. So, Claude design is actually something that just came out. Still in beta. I think it’s still available for just a few select people. But what it allows you to do is for those those I think it’s in beta and it’s not I don’t think it’s for select people anymore. I think it’s more widely they might have just relaxed that. So,

6:20 they might have just relaxed that. So, okay. okay. I think so. I thought you needed some subscription. It wasn’t just for free users, users, but regardless, it’s coming out for everybody. Yes., it’s it’s a design framework that sits on top of Claude. It was a custom UI or custom,, part of their tooling that was there, but I think it was like on the higher end licensing subscriptions like it was given to those users first. Yeah. And what it actually allows you to do is if you have a design kit, you kind

6:50 do is if you have a design kit, you kind do is if you have a design kit, you have your colors, your fonts, if you of have your colors, your fonts, if you want to start building an application and really focusing on what the functionality would be, not the code, not the the back end, but really how do you want it to function? If you want to do wireframes, if you want to do proof of concepts, it’ll actually help you build that. And Mike, it is pretty incredible. And I’ve seen this for applications. I’ve seen this for reporting and let’s take this idea and well before we go into that anything you want to add about cloud design

7:21 you want to add about cloud design just make sure that we cover all the basis there on what it is. Yeah, I I think all the points you’re making are very valid here Tommy. I think I think the idea is this is you think I think the idea is this is there are style guides there’s know there are style guides there’s things you’re going to want to build in your company. What’s happening now? I I think there’s a general trend just not just cloud design but this is just a general trend across all agentic coding experiences where previously we were writing a bunch of code to have AI agents build just code and the back end of things and the code could produce websites and we

7:51 could produce websites and we had some way of like screenshotting or give it an MCP server like the agents sometimes can see the front end of what they build. Yeah. for websites. That that feels like what we were starting with that that felt like from,, late last year till like March. Something happened in the middle of April where people building video is now much more of a main topic. Video content, websites, it’s more graphical in nature. And so something shifted and then very quickly

8:22 something shifted and then very quickly the wind became I’m not just adding an MCP server to like Figma or Adobe’s products or in design or whatever the whatever the graphical program is. You’re not just adding an MCP server to it. You’re actually focusing directly on building graphical and video objects now. now. Yeah. And I’m seeing now this month, May, we’re seeing just I’m seeing a proliferation of like project after project of this clip editor, this short thing, this thing, edit my videos, do

8:52 thing, this thing, edit my videos, do these things, give command to these other tool. Like there’s a lot of tooling coming out that helps people build videos. And so I think design is one of these steps of people are now starting to figure out okay we want to build goodlooking things but everything in AI world when you say build me a website looks so the same thing over and over again. So it’s this like purple bluish colored thing I guess always hap for whatever reason it’s trained on purple and blue. and blue. It’s very functional. Yes. Yes. But it’s not necessarily

9:22 But it’s not necessarily AI generated. It has like a like AI generated things have a style to it. I think what what Claw Design is trying to attempt to do here is look let’s let’s step back and reimagine how we build things designwise. What does what we’re reimagining that for like an agent. How would you do that? And to your point, Tommy, it’s it’s a lot of this bu design a style guide, build something that makes sense to users, users, but then translate that into now, okay, now I’ve got this really amazing

9:54 now I’ve got this really amazing framework to your point, font, colors, styles,, components, build those things, and I’ll stitch them together. And I think that’s where claw design comes into it. Yeah. And I would I can’t wait to talk about this in the realm of PowerBI because Mike, there’s a few things I want to make sure that we cover today is obviously the Reddit page and just the ability to actually design something using cloud design in the embedded world. But Mike, I also for those who maybe still not AI adverse, can you use

10:25 maybe still not AI adverse, can you use cloud itself to help you build wireframes for PowerBI even just in general? So those are the topics that I want to make sure we cover because I think there’s a lot here where we’re finally getting to the point, Mike, where AI may not build your dashboard for you per se in PowerBI. How and we’ve always seen we’ve seen C-pilot try to do this. We’ve seen Copilot try to do it again. We’ve seen all these other designs but nothing’s really saying hey for PowerBI in a

10:55 really saying hey for PowerBI in a PowerBI report can I build using AI design design and I think we’re finally getting to a point where it can aid or help build we’re scratching the surface now like we we we are literally getting to the place where and now I’m seeing things where people are giving an agent a full website and saying hey here’s the website that I like the style of just rebuild just make a new make a new style and a new website on top of this and it’s and it’s doing a really good job of mimicking and and mirroring and giving

11:26 mimicking and and mirroring and giving the same look, style and feel that are together and I’ve been playing with it. Armando, which is another gentleman I do some podcasts with, he and I when Claude design came out, when they was announced the day of Armando had it on his Claude Code subscription, I think it was his Promax subscription. Yeah. Yeah. We stepped in and started building websites immediately. We gave it some prompts. I had some images, some reference items that I was giving it and I had it start building websites for me. And now I’ve turned those images, those initials mockups of like what the website was going to look like into real

11:56 website was going to look like into real websites. Now it’s actually becoming,, I can take those things and from,, an hourong video, not even in 30, 15, 20 minutes, we go from, hey, I want this website to looks like the style. And let’s let me get you some weird things here, Tommy. The style I was looking for was all generated by an agent to begin with. I I went to So the initial images were I went to VS Code. I started an MCP server. I went over to Figma and said,

12:26 server. I went over to Figma and said, “Okay, build me a design for the website.” And I described what I want. Here’s the style. Here’s the vibe. Here’s the feeling. Build me a mockup page. Just build some stuff. And then I say, “Okay, now great. Take that design and build components from it. Things like this, this, and this.” And I built all the components. Okay, great. now make thumbnails from this, make the thumbnails in the same style. And then I added my own little flavor. I tweaked a couple things. I maybe added a couple flares here or there more of the branding of style that I wanted, but it gave me like a really good 90% out of the box of just getting started. Then I took that entire

12:57 getting started. Then I took that entire design and gave it to Claude Code Design and said, “Build this.” So I I think

13:01 and said, “Build this.” So I I think we’re really close to the space where we’re going to say, “Here’s an image. I like this image. build me a a report, a background, an image, a theme, a dashboard specifically for that. One thing I think that’s a miss here, Tommy, and I just want to be very clear. There’s this idea of a template layout or design. I’m very motivated to focus on the design side and not the data side because I think there’s a neat interplay here of you need to build the Lego bricks of your report first before you

13:32 bricks of your report first before you put it all together., and what by that is you have a design guide, style, feeling, whatever that looks like. And that’s what we do today in Power Designer today currently in fabric. You can go get this now. Did you just call Oh, oh, I was like, you mean PowerBI desktop? PowerBI designer. designer. Power designer, the tool that we built. Yeah. No, no, I got you. I got you. The workload that we have has a bunch of design files, right? So, it’s all it’s all design based. Colors, graphics, images. is those are those are starting points of like what

14:02 those are starting points of like what looks good to the human eye. But you take that information and you marry that with report modeling, data modeling, and you marry that with measures and fields on visuals and the combination of those three bricks, those three Lego bricks, you have to puzzle them together in a way that makes a full report. And so what I think we’re finding right now, what I what I see happening is this is just honing in on the design part of that story. We already have a different story we’ve talked about in the past, which was the MCP server for data modeling, which is honing in on that

14:35 modeling, which is honing in on that portion of things. So, let’s unpack things. All right, Tommy, what do you think? Yeah, because I think I’m going to caution you here because you got the podcast, the Aentic Thinking Podcast, and again, that’s focused purely on AI developer, right? It’s it’s well, not let’s let’s I’m going to rephrase it. It’s anyone who’s interested in using AI. Okay. Anyone who’s interested in AI I would want to take Yeah. I want to take the word developer out because I think that’s I don’t want to lock you in to be like I

15:06 I don’t want to lock you in to be like I need to write code and I’m a developer and I’m going to use AI. I think I think the the Microsoft calls us developers too guy. So I know we’re definitely not that. That’s for sure. Well, I remember we were at the fabric conference or the Microsoft Ignite in Chicago. They’re like and all those developers. I’m like I guess that’s what we are. We finally have a name. but but regardless it’s a general again and I’m being very wide here. It’s a general anyone who’s interested in AI to whatever they want to build. I want to caution you on this podcast on our conversation today on is the same

15:38 conversation today on is the same workflow the same for a PowerBI developer builder right and I think there’s two avenues here Mike that are really important because I think the workflows are different. There is the PowerBI embedded which I think that has a much more of an overlap to a lot of things you’re talking about but talk to someone in the world of in FTE talk to someone who’s living in PowerBI desktop or PBIR and living in the service I think there’s still it’s a very hard to

16:09 think there’s still it’s a very hard to do that same workflow or more importantly to equate what what you’re saying to their workflow to that workflow to the embedded world there’s a lot of easy easy ways to go. I can see how we can play this in. But I think there’s we got to again shout out to all those people who build PowerBI desktop still. There’s also still that element there. So I think there’s really two avenues here. And also too, regardless if you’re doing embedded or design or or a

16:39 doing embedded or design or or a desktop, you’re still dealing with data and even though it’s application like like the context of the data matters so much more than just in an in a website, right? It’s not just I need a navigation and I needed to do this. Well, is what are you conveying communicating on that screen? That to me is where we’ve seen the biggest gap with AI building designs and reports. It’s not just build me a line chart. It’s this client has this need. This is

17:10 It’s this client has this need. This is what keeps them up at night. How can I give my corresponding visuals across the board to help build that right report? And I think that is the holy grail that we have yet to reach. I I think you’re going to have to divide a lot of these topics into what is good that an agent can do versus what is good that a human can do. Right. I think your your note here Tommy around if I’m talking with another

17:40 around if I’m talking with another person about getting requirements like what keeps you up at night that feels like a workload that is much more equipped for people talking with people and relating to people using your emotional intelligences to talk to I don’t to be honest like if I’m trying to build a report I don’t want to sit down and go to a command line and tell it what I want. It just seems so impersonal. Right. Yeah. Right. So So there’s a part of there’s a part of that which is love that. Love that. Communicating with you and be able to like really pull some like human like it’s the it’s the the body language.

18:11 it’s the it’s the the body language. It’s the facial expressions. It’s the hand waving. It’s the can you explain more? I didn’t understand this. The way you explain things I I can reason my way through those things. And there’s there’s a certain level of reasoning there that I think makes sense for people to connect with people in that in that aspect. But once some of that’s known, then what? And I think Tommy, I think what you’re pointing out right now is what I’m describing in and what you’re describing here as the power the workflow that’s embedded much more codecentric, much more tooling heavy in

18:42 codecentric, much more tooling heavy in that workflow of building reports. The tooling isn’t really equipped for general PowerBI users, right? right? Like you’ve already worked your you’ve already worked hard on building things in PowerBI. I do think though that the hyperpersonalization of what agents can provide for you is going to change things. And let me just hang with me just for a moment. Let me go on a tangent here slightly and come back to this one. Nothing new.

19:12 Nothing new. Previously, before I started really diving into agents heavily, all my work was done on large screens with big computers. That was where most of my work was coming from. Okay. What I have realized recently though is I’m shifting how I like to work. I’m working more and more on mobile. I’m working more and more on texting an agent. Like just just text back and forth, just talking to it because my main my main focus of medium between me

19:42 main my main focus of medium between me and talking to an agent. Let’s, Tommy, let’s imagine you had just as an experiment. Let’s say you had you worked at a company and video format communication teams and zoom was just not invented yet. All you and and everyone was around the whole world and you couldn’t go there like you you literally can’t go to places right so if you imagine a company where all you can do is communicate via communication

20:12 do is communicate via communication texting basically you don’t really need the big machine anymore. I could communicate to my team via text. I would write emails. I would a lot of the correspondence would happen just in words on a page. This is eerily similar to episode one. And this is what it’s feeling like to so my workload is shifting from I I sit down at a desktop and I do work here on lots of screens. I’m shifting more and more of my work towards I have one or two agents GitHub co-pilot. I might have

20:42 two agents GitHub co-pilot. I might have multiple sessions running. I might have different things going. I’m you literally using my phone more and more to leverage me getting stuff and work done. done. I have an email come in. It’s a it’s a lead. I’ve made a repo that has a skill in it that I can go use. So, I copy the information from the email. Literally highlight the text. Here’s the guy’s signature at the end of the email. I was doing this literally last night. I ran my skill and this is some of the skills I’ve learned from you, Tommy, because you do a lot of statements of work with with skills as well. And I

21:13 work with with skills as well. And I literally wrote my slashskll, right? Company research slkill setup new client. And I just feed it the pile of text information and let it sort it out. And while I’m,, getting ready to go to bed, you laying down and getting tired, it’s now writing my statement of work. It’s doing a first pass on an MSA. It’s doing a first pass on a draft project. It’s doing pulling their website and researching their website. What do they do? How do they work on things? provide me a business summary. That way when I show

21:43 business summary. That way when I show up tomorrow for the call, you’re ready to go. It’s all ready to go. The the text part is ready to is ready to happen. So then what I’m doing now is taking the video call, the human part, getting the transcript and summary of that, and then feeding that back to the same message, to the same AI. Hey, here’s the conversation. Here’s the summary of what we did. Now, write a statement of work. Focus on this area, this area, this area. This is how I want it structured. take a first pass at it. So, more and more of my workload is moving away from computers and I have

22:14 moving away from computers and I have this vision, Tommy. I I yesterday I was thinking about it. It’s we’re getting closer to it. More and more I’m talking to my phone. I I write text messages to my family. I’m hitting the little the the voice button and I’m just talking to it. Hey, do this, this, this. Hey, can you pick up this at the store? Hey,, here’s your chores for the day. What all the things are. That’s happening more and more and it’s converted to text and the text is being sent to the family. Okay. Now, listen., there’s nothing I disagree with because very briefly to your point, like I’m every

22:46 briefly to your point, like I’m every meeting I have now that gets subscribed and going to go through my own funnel. I am talking to my agents like they’re my assistants. Like, hey, we just had the meeting with tech team. Looks like we need to change a few things. Go ahead and take a look at the milestone. What do we need to do? And and it’s really I’m talking more than I have in the last three years. And that includes more meetings I’m having now with people like and it’s great. Now I’m intrigued. Take that for me and tie that back into the PowerBI workflow design of someone

23:17 the PowerBI workflow design of someone building a dashboard not just an AI but yeah yeah great point and this is I wanted to volley it back to you ideawise before I love you. I love you but how much you love AI too. Yeah. So I here’s where I think the the weakness is right. So, how do we follow this back over to the PowerBI story of things? things? I believe the PowerBI story right now is we actually have inadequate tooling at this point in time today, today,, May 20th of the time of

23:48 , May 20th of the time of recording of this video. We don’t have adequate tooling for us to do what I’m describing to to verbatim automate the design. Is that right? Yeah. to to go from I want to send a not a not parag I don’t want to send paragraphs and paragraphs paragraphs information right I want to send a minimal amount of text to something or even just in general Tommy I don’t I don’t even talk to my reports the tooling today does not exist now the

24:18 the tooling today does not exist now the reason I say this is because my team is heavily working on our current workload which is called power designer you can go build designs with the UI it it simplifies a lot of the theming file files, a lot of complex things that we try and simplify what they simply use UI. We are aggressively attacking the design aspect with agents. Now, we’re building brand new workloads with agents and using agents to help you assist with getting this stuff done quicker. So Tommy, the tooling does not exist today,

24:50 Tommy, the tooling does not exist today, but I can demo for you things where we’re leveraging AI and talking to things that it understands the underlying code, the underlying framework. And if you’ve seen it on YouTube already, people are already building reports, PowerBI reports with technical tools like they’re still going VS Code. They’re not building simple things. We just did one on the Agentic podcast called AI buddy, which is about a lot more around using the MCP server, but talking to it and using like a chat window as opposed to building a full-blown,, extension for VS

25:21 full-blown,, extension for VS Code and pulling all the code open, right? So, the tooling is getting there. People are building things and I think with agents available to us, you’re going to see a lot more tooling continually simplifying these experiences and you’re going to be able to describe what you want in simple terms. the AI will interpret it and then build out a lot more of what you need and then give you previews back. So the handshake the tooling is does not exist today to make this a reality. But I in the next month or two we’re going to see a substantial change in how

25:51 going to see a substantial change in how this development cycle works. think I think we’re getting there and I think if you are very in-depth in the AI world you can absolutely get a lot of

26:00 you can absolutely get a lot of automation done because I think so there’s a few things people have done online where they’re using D3 and embedded with cloud design but again that’s so that’s so that does not meet the majority of the people I think in PowerBI it doesn’t meet the majority of people in general like like the only people who know about that who do a podcast about it right? Or or or you’re a nerd like I am on on these things. I don’t think that’s right now today, you have to know a lot about

26:31 now today, you have to know a lot about a lot to get that done to have that automation. That is not a setup. It’s a lot of prompting. It’s a lot of things that need to be set up already. But I want to introduce to you before we dive into what’s poss Okay, I just want to react to that comment real quick. So hang on to your next thought. Write it right there. Oh, forget it. I want to go to your next thought. something goes mainstream when your parents start picking it up. Okay. Okay. Okay. So, I’m I’m saying this because they’re not come my my parents and my

27:02 they’re not come my my parents and my wife’s parents are not building reports. Let’s just be clear. Okay. Okay. What they are doing is they’re now sending me chat GPT messages around how things should be done. Sure. Okay. That’s chatbot. That’s our level one we’ve talked about. Yes. But the fact that that exists, the fact that now the generation behind me, in front of me, what I don’t know which way direction you’re I don’t know what direction you’re looking, but when when when the generation the older generations are embracing the technology

27:33 generations are embracing the technology and they’re using it as part of like, wow, this finds value for them because there’s my understand my feeling here is the older generations typically are more resistant to change in general. It just feels like that’s how it is. Even I’m as I’m getting older is becoming more resistant to change. I’m I’m I’m finding myself being like, “Oh man, why do we have to change all this stuff? It’s just can’t we just leave it the way it was? It was great when I was a kid.” You It was great when I was a kid.”, I’m I’m seeing myself say these know, I’m I’m seeing myself say these things. But when that generation starts adopting things, to me, that’s a signal that says this is going a bit more mainstream. So fine,

28:04 mainstream. So fine, when I look at young generations now, like my intern at my company, my son who’s 16 getting to work on these things with agents, I’m introducing him to agents and how things are working. Their brains are already thinking this way. They’re already being reprogrammed. our kid like the young kids now that are starting to show up and say like right now my children are talking to the the hey Google thing and saying turn on the TV, turn off the TVs, turn on play this, play that. Like there’s now

28:35 play that. Like there’s now automation and and inf like not infinite knowledge but like wealths of knowledge behind this little speaker box on my t on my table. So this is where I think we’re going Tommy. It’s We’re not there yet, but I think we’re so close to the tipping point of what this is going to look like. And when this tipping point occurs, occurs, we’re going to run into Jav’s paradox again. We’re we’re just going to make report creation as easy as possible and it’s going to go like wildfire.

29:05 it’s going to go like wildfire. I got a few problems with what you said. First, and and I think it’s pretty clear here. First off, anytime my parents adopt something, it usually means I need to stop doing it because it’s old. So for some once they start doing it’s like all right well this is no longer relevant first and second off and you relevant first and second off and this better than anyone know this better than anyone chatgpt chatbot is not an agent. Okay so I’ve set up pro I have chat gtp projects for my mom she does retreats and all these things and that’s not an agent though and you’re talking about

29:37 agent though and you’re talking about cloud design with cloud code. If my parents start doing that yeah I think I think the line is blurring. Holy crap. If my parents ever I’m gonna disagree with you there, Tommy. The blind is blurring faster than you think nowadays. I just got a hands on I just got a hands on Grock. so Grock has their own coding system. Yeah. And I believe Grock just announced they’re going to go buy Codeex. I think Codex is being bought out by Grock. What? Codex? From OpenAI. No, that’s OpenAI’s cloud code. code. They just made a $50 billion bid on some

30:09 They just made a $50 billion bid on some program company. So, okay. Anyways, XAI is taking like a really big stance on like absorbing a lot of these AI coding companies. Fine. Fine. Grock has also released this. Grock, their whole AI agent now has these things called connectors. Those are now added. So you say agent and traditionally it was and even goo even even Google has announced a brand new but we are used like the chatbot experience that we are used to is so

30:40 experience that we are used to is so rapidly changing that I think the chat window the window that you would just type a question in and get a response from like like we normally used to with chat GPT when it first came out that is right now that’s no longer the case there’s already connections and extensions. I can go into that chat window and say read my email and summarize it for me and it will read it. That’s an agent. Yeah. Yeah. Okay, I hear you. So def So then my then what I guess I’m

31:10 def So then my then what I guess I’m saying is I disagree with your definition of an agent now because in my terms of agent agents can go do things and actions on behalf of you. It’s not just ask a question, get a response back. It’s look at my email, make a calendar invite, add this to my grocery list. agents are doing those kinds of things. That’s what I would call an agent. Now, those tasks may be lightweight, but an agent to your I think if what you’re describing here, Tommy, an agent could also be like, “Build me a design file, build me a website.” These agents are doing that

31:40 website.” These agents are doing that now. And like that, that’s the chat window is still there, but very quickly that chat window is creating to connections and doing things and making things happen. The very first day I saw chat GPT, I was like, “This is awesome. This is so amazing. It’s it’s useless to me because I can’t make it do anything. I the only part and and I want to get back to when we’re talking about for people designing is just for the sake of argument I guess but when we those are agentic tasks but an agent to me actually is has its own context with its

32:10 actually is has its own context with its own set of instructions in a certain harness. So it’s not just doing a task, right? It’s not just chatting to I listen but yes, but when it does not have its own set of instructions oneoff they do the one I’m looking at right now. I have it in my hands. The Gro one right now has this thing called projects. You go into the project the project has but that’s that is the instructions. Those are the custom instructions that are there there and it’s just skill and MCP servers. All we’re doing now is

32:40 MCP servers. All we’re doing now is you’re adding connections. You’re adding MCP server things and you’re adding skills and instructions on top of it. All of that now exists. So the common tech is there the technology exists. The other hurdle Mike and you tell me when your parents start doing this or my parents too. I don’t mean to say your mom thing but but your mom your mother your mother use agents. Yeah. thing. So but but let me know when the main Your mom used to use agents. Yeah. Your mom used to chat. I feel there’s like a your mama joke in here

33:11 there’s like a your mama joke in here somewhere. That’s what I was about to say. I’m like didn’t mean to get so personal there. Whoa. Tommy, easy. Let me know when your parents are picking. picking. You’re throwing some hot comments there now. now. Yeah, right. You’re getting fired up. But let me know when the mainstream is comfortable and aware of all these things. Just because the technology exists does not mean it’s going to be is widely adopted. Maybe. And again, I don’t disagree. we’re going to get there. But most older people will say

33:42 there. But most older people will say the the the people on the outskirts of AI are using AI like Google, right? An advanced Google. I disagree with you. Anyways, so I I’m going to disagree. I understand your point. I hear your point. I’m gonna acknowledge the fact that you said it. Fine. I’m going to sorely disagree with that point because I think things are changing so fast and the big three, the Grock, the Google, the Facebook, and the

34:12 Grock, the Google, the Facebook, and the u anthropic and and CL not big three, but the big guys, they’re already adjusting that chat with me experience so fast that people are just using it in a very different way very quickly. So, so then tell me Lyn then let’s walk through then if we have an agentic design to tie this all in to building a report and building the design I want to introduce to you potentially the future and I think what things that are being built but for those who are still stuck in the rut for those even if you have

34:43 in the rut for those even if you have agents Mike even where we are at today to your point May 20th what can you do to help design your dashboard this is something that you and I have poo pooed. We’ve we’ve talked about the inefficiencies with co-pilot to do this. Yes. Yes. And honestly, when we had this topic set up when we set this up like two weeks ago, I’m like, I don’t know how I feel about this. However, However, something’s changed in the last two weeks. Truly, man. I’m there. I’m building stuff with it. Like, it’s it’s happening for me. Like, I’m I’m actively

35:13 happening for me. Like, I’m I’m actively building reports and components and styled things and I’ve stopped using desktop all together. Okay, it’s it’s D3. It’s React viz. It’s Vegaite. I know what language I’m saying to get these things started. Great. Great. At some point, it’s going to be so easy where people don’t have to know that. And it’s going to be the skills and the language of things will just be able to produce it on their own. And I won’t need to have that extra little bit of technical knowledge. 100% 100% that’s the future. I think that’s still the future for the majority of the masses of people in in a company

35:45 of the masses of people in in a company using these tooling. We’re not there yet for the masses. for the Mike Carlos and the Tommy Pulas who can install and get started and buy whatever subscription they want. Truly, and if they want to, they can. I But for the masses here, Mike, there’s still a here and now that I want to introduce to you that’s still possible. If I am someone who has limited capabilities or resources, right? What can I do to the design?

36:15 right? What can I do to the design? Because again, I still think you are at the forefront. You are in the frontier. You are the person,, crossing the Mississippi River for the first time and looking for gold. Basically, one of the first 49ers, so to speak. But that’s again when everyone’s going to be living in LA, so to speak, and everyone’s in California, we’re we’re not there yet. And we’re whether we’re close or not, we’re not there yet. So, there’s still that gap, right? And would you at least

36:45 right? And would you at least acknowledge there’s a gap right now? If you’re talking to all the people who listen to the podcast and all the people who you meet at user groups in the conferences, would you say that they’re all doing this? No. No. I’m not sure where you’re going with this, Tommy. Like you’re you’re So, let me introduce So, let me introduce something for you because Yeah. Yeah. The reason I’m confused about this is because the way you’re the way you’re describing it is like it’s this all or nothing stance on it. It it feels

37:15 all or nothing stance on it. It it feels like how you’re describing it to me is like it’s this everyone needs to use it or no one uses use it. And I don’t think it’s that it’s that clearcut right now. now. Yeah. Yeah. And I’m I’m looking at clarify. Yeah. There there are some headwinds I would argue that are against us right now and being able to do this. One of them is just does your organization even have the ability to turn the stuff on. So that that’s one of the complaints that I have right now., one of the things that Microsoft has definitely most common pushed has been this whole era of the

37:47 pushed has been this whole era of the co-pilot. Co-pilot isn’t everything. We didn’t ask for it. It showed up in every single program. Paint has it. We didn’t ask for it. Everything it just showed up. Agents were just everywhere. They’re just there’s an integration with everything. Now, my opinion as an outsider looking at this one going, it felt like they were just throwing co-pilot with like it was like an Oprah show. You get a co-pilot and you get a co-pilot and you get a co- and there wasn’t this coherent story around what the co-pilot was doing. So, now that I’ve been into the space a lot

38:17 that I’ve been into the space a lot more, I look at it going, all these co-pilots, all they are is a harness for some large language model. Mhm. And for me, I look at it going, Microsoft can’t economically throw chatgpt chatgpt 54 at every single user using all of the agents at like they Microsoft just cannot physically dollars. It was too expensive to throw the highest model on every single user every time Copilot showed up. So what happened was lots of instructions, lots of handholding, lots

38:48 instructions, lots of handholding, lots of guard rails were built around what the agent could do to not give it bad answers. So it’s running a lot of these lowerend note effective models and so we Tommy us in the tools that

39:00 and so we Tommy us in the tools that we’re using today co-pilot’s there it’s not able to handle what we want it to do the capability is for us not effective and so we just walked away from it what’s happening now I believe is the harnesses and the models have gotten so much better continued infrastructure is getting built out now the price of tokens has not been going down yet I think it’s still I think it’s actually still ramping up. Oh, they’re going to treat that like text messages when text messages came out. out. Yes, it it’s going to take some time for enough inference to get built so that it

39:32 enough inference to get built so that it actually can start driving down the cost of the tokens. But we’re already facing the attention war between the big companies., GitHub is changing its pricing because they were too cheap., Anthropic is trying to incentivize customers to stay with them by giving two months free tokens., Open AI is saying you get a a month free of something else or we’re going to lower your rates for some,, half off on things. So, these large companies are fighting for your attention right now to get you in their world by hoping that you would like their models. You stick

40:03 you would like their models. You stick with them. You build structure and tools around them. But in reality, Tommy, the MCPs, the skills, these are all transferable plugins to any of these harnesses. It doesn’t matter what you use. So, use. So, I don’t know where I’m going with this one. I just got myself blocked. I just ran into the woods and I got really lost, Tommy. I don’t know what my point was. was. You you made the argument. So, you were confused about what I said or disagreed with what I said about it’s either all or nothing with AI. Yes, I think it’s a very it’s more of a progression now. Like it’s it’s more of we are Yes.

40:33 we are Yes. The headwinds are companies aren’t turning it on and what we are being given by Microsoft isn’t capable enough for what we want. Period. Exact. and especially in the in the realm of fabric and power powerbi design really is where we see the biggest gap. My argument and and I’m I’m going to clarify here is being on the forefront like we are to see the design you can automate the design with cloud design cloud code sure sure is not speaking to the masses right now again because they don’t have the ability they’re not comfortable with it

41:05 ability they’re not comfortable with it or they don’t have the resources their company’s not providing so I want to introduce it may be coming in the future but what can I do now right if I’m someone if I have Again, are you building all your dashboards in and using that automation in agents right now? Probably not. Maybe you’re testing it out. But what could I do now? And I think there’s a lot to be said about even what’s available if I even have a chatbot or an agent because I really think this is important. I’m rapidly changing what I do though,

41:35 I’m rapidly changing what I do though, Tommy. Like my workflow is rapidly changing in this in this arena and no longer am I spending a lot more time in desktop., I’m spending a lot more time focusing on it’s let me just say it this way. We had five minutes to wow before. before. Mhm. Mhm. That was on a framework that had been built by hundreds of people to build hundreds of things that I could click and drag and drop and build stuff. Yeah, it was it was effective. I think we’re going to cut that time in half. It’s going to be

42:05 cut that time in half. It’s going to be like two and a half minutes. It’s going to be two minutes to wow with some of these things. Now, all this being said, I understand what you’re going to like if you don’t have this is and this is the part that I think is the bugaboo that what you’re putting on is if you don’t have access to the model to a harness that is given to you as a user, this basically removes your ability to do any of this. So, and the skill that you have, Mike, too. The experience and the skill that you have is also important here. It Yes. But less so, I think. I think I

42:37 It Yes. But less so, I think. I think I think it’s you said earlier there was this we’re we’re Q&A you made the note you comment around we’re the Q&A the Q&A professional Q&As professional Q&As that has changed into prompt engineering they call it that’s like what they would call a prompt engineer right I don’t really know if I like the engineering part but there you’re right there is a science to changing how you think about what you ask or tell or put into the prompt so that you get a better output.

43:07 that you get a better output. And I’ve seen this with people who try to write songs like I use a lot of like suno and I write songs on grock and then I take the words and I put it in sunno and I build the actual song. There is a bit of an art to it of what you say and how you say it and it does take multiple tries on the same thing over and over again to get the output that you’re desiring or the output you’re looking for. I have reprogrammed my mind to get closer to I know what to say in a window that gets

43:37 know what to say in a window that gets me a better output. I know to your point I know what instructions to fabricate a little bit better to communicate with an agent to get better outputs out of it. And there’s an additional layer here though when you’re dealing with design. And the thing is you’re talking about a song which is we had John Cle PowerBI jokes right on one of our podcasts. I know what you’re saying, but I need think I think you need to introduce another skill human skill layer here where it’s context engineering. What con are you providing the right context along with the

44:07 the right context along with the instructions because that’s so essential if you’re going to build a design. I don’t care if you’re using agents or a chatbot if you wanted to build your D3 PowerBI embedded or help you with PowerBI desktop. It needs the right context about your data, about the use case, etc. And I think that’s also again something that most people would say that they’re a beginner at. So what’s what’s your definition? Actually, real quick here. This is this is an appropriate time to put in a yo mama joke. joke. All right. So from from the AI agents

44:39 All right. So from from the AI agents that are in the internet, yo mama so slow the AI agent finished her entire task lick, generated a report, took a coffee break even before she finished her first prompt. Where’s the applause? There is. Yeah. Hold on. Do I EVEN HAVE IT? YES. YES. THERE WE GO. All right. All right. So, that really like that joke that that was a fell flat joke. All right. So, anyways, back back to So, I wanted to ask you there was two items you you mentioned there. You said cont

45:09 you you mentioned there. You said cont prompt engineering and then context engineering. Two terms you’re saying. Yeah. How are those different in your mind? What is the difference between Yeah. Yeah. prompt engineering and context engineering? Prompt engineering is the practice of how you actually write to and provide instructions to an agent or a chatbot. Context engineering is how you actually structure and manage the information that the AI,, harness is actually seeing to ensure that it’s providing the right output. So it’s not just what my instructions are

45:39 it’s not just what my instructions are and how I provide those instructions, but it’s also I provide it the proper context whether it’s coming from connectors, whether it’s coming from the information that I send it. So prompted veneering is about just really the building of the instructions for usually it’s a single task tax but the that harness or environment whatever you that harness or environment whatever if it’s a notion or whatever you know if it’s a notion or whatever you want to call it is all the things the external documents you may have in projects right the system prompts you may have the your user particular

46:10 may have the your user particular data those are all the things that would make up context engineering which to me is essential if you want to build a good design. So again prompt engineering is my particular instructions for a set in time. Context engineering is over a set of time with external documentation or external information on what it sees. So when I can say hey help me build my statement of work right prompt engineering would be the whole instructions at once and context engineering is the skill. I have a skill

46:40 engineering is the skill. I have a skill for my things is looking at my previous statements of works. It knows the sections that I like. So there’s a huge difference there and you have to incorporate that when we’re talking about design and I really think this is a huge point here. The point I’m trying to make here is I think that’s essential right now today if you are going to automate your design using cloud design using cloud code using MCPS if you want to try to design it right you’re going

47:10 to try to design it right you’re going to need to use some CLI they whether it’s curers you can’t not your or code again I’m making the argument Mike that you are the first across the Mississippi you and a few others right? To know all these things to be skilled in context engineering. I cannot assume today that all people can just spin up an agent provided the right context and it’s going to spin up the right dashboard. And I’m not just trying to do the

47:40 And I’m not just trying to do the negative here because I want to introduce what you could do if you are a beginner or again you don’t have the capability to do context engineering because you don’t have the tools or the systems to do that. Mike, there’s this element here. There’s this bridge that I think we need to introduce. And the bridge I want to introduce to you is the ability for wireframes. We’ve not talked about this at all. We’ve I feel like you and I have done the all or nothing approach with design

48:10 the all or nothing approach with design with PowerBI. It’s either can co-pilot or something build my design. If not, then forget it. Right? We haven’t talked about the ability to for for the inspiration side of things and I think we I this is where my aha or my epiphany has been in the last two weeks because I’ve been very much like in this space of man copilot stinks at building dashboards or it builds something but I don’t want it. It’s like well why does

48:40 don’t want it. It’s like well why does it have to build it at all? If I have all the context on what my semantic model is and what the client or my stakeholder is looking for, couldn’t I just spin up a wireframe, even if it’s not in PowerBI, an image on just an idea of how I can start building. Most people in our space struggle with the design side,, because you’re the data people and design’s hard. That’s why you built the tools that you did, right? Because people had a hard time with

49:10 Because people had a hard time with colors. colors. So I want to introduce to you leveraging AI for even just the design, the wireframe, the inspiration around, hey, these are the three metrics that matter. Here’s the last three meetings on when we talked about what they want to see. Can you help me spin up what a dashboard will look like? And you may not use half of it. Maybe you may not,, you may not use all of it. Maybe use half of it. But just to go,

49:40 Maybe use half of it. But just to go, oh, I can put the KPI cards there. Hey, well, that’s neat. If we had an alert that did this. I didn’t think of that. So, I’m introducing that aspect here because that to me is something you could do right now if you have Claude or Chat GBT, right? And you have a meeting transcript. That’s something you can do today without having the skill and I think what you and I possess and the amount of work that you possess. So, that is what I’m trying to that’s

50:10 that is what I’m trying to that’s literally what I’ve been trying to push for I think the masses and what I’m trying to say to you. So, I want to know what do you think about that? Would you do that? do that? Where you at? What’s your take? Yeah., lots of thoughts. I’m just taking notes here while you were talking here, so I’m just going to distill them down. On one hand, Tommy, I’m a little offended. I’m a little offended. And and it’s because Tommy, what I’ve been building for all these years. It’s incredible. Well, I’ve been building a lot of stuff.

50:41 Well, I’ve been building a lot of stuff. And it’s not just stuff. I’m building wireframes. I the the whole the whole Power Designer product that I have built is a wireframing tool. Like so to your point, Tommy, like you’re right, there’s nothing agentic about what I’ve built in my tool. Stay tuned. Okay. Okay. This is going to change rapidly. Well, let me say I’m GSA. I use PowerBI themes all the damn dang time. Sorry. Wow. I am fired up. So, So, well, okay.

51:11 well, okay. Thank you. I appreciate that. But it’s going to get better. And yes, you’re right. I’m on the front line of heavy code CLIs. So, you you made a couple Let me just I’m going to address the comments that I just wrote down in order. One of the first ones you mentioned was you’re going to need a C a CLI or VS Code to do these design or wireframes. I already disagree with that today. Currently, I don’t think you’re going to need that. And to your point,, you do need a framework. There needs

51:43 , you do need a framework. There needs to be some skills or something from the community you can go lean on to bring it forward. Already there’s been a lot of YouTube videos about people vibe coding and building dashboards and wireframes and samples of things already. So, I already think there’s enough

52:00 already think there’s enough interest in this topic to get people to build skills that you’re going to be able to go use. Now, does anyone know how to use them? Is there a tool that you can plug them in easily? I don’t think so yet. I’m also trying to address that problem because I know this is an issue. What you described is exactly the problem I’m trying to solve in some of my new workloads. So, that’s coming., you you may another one here that said you can’t assume anyone can just spin up an agent and then do context engineering. And there’s a little bit more there. I think this to your point Tommy

52:31 Tommy a month ago that was impossible. Now with AI Foundry and the recent announcement of hosted agents, this is now really possible. I think AI Foundry can spin up hosted agents for you whenever you want. I’m going to propose an idea here, Tommy, that’s going to be wild. Okay, what if your hosted agent that you’re producing comes from AI Foundry and you go into Fabric and there’s just a CLI tool. It’s just an item in the workspace

53:02 tool. It’s just an item in the workspace called CLI. You click on it, it spins up and what it does is it spins up an Azure Foundry hosted agent for you where you can step into that hosted agent and it’s just a UI. It’s just a chat window between you, the chat window, and then that hosted agent. That hosted agent would then have be able to be granted permissions to do different things. It could build what you want to build. It could have skills and libraries and things already on it that are ready to go and hosted for you. So when I look at this, I go that’s

53:33 So when I look at this, I go that’s something that’s reality. Like so the comment of anyone can’t spin up a we are like the the infrastructure in the UI elements in fabric or PowerBI does not exist to do this yet. This is one that I think needs to happen and I firmly believe that this is going to be on my radar to build something simple that at least proves that this is possible and then we’ll see what the community does with it and take it from there. So I I also I think I think we’ve already tipped over the threshold or

54:03 already tipped over the threshold or we’re so close to the edge of anyone can spin up an agent anywhere and get it. Agree. Agree. With this being said, I really like my open claw. I really I’ve just started testing Hermes which is another agent the Microsoft team and one of the the VPs at Microsoft is trying out this new internal pilot called claw pilot which is like an open claw type hosted agent for them inside Microsoft so you can just go ask that agent to go do things. things. I’m now integrating work IQ directly

54:35 I’m now integrating work IQ directly with my agents trying to figure out how to leverage that. Work IQ lets me say it’s almost like an MCP server for my email, for Excel documents, for SharePoint, for like I can pull business context into what my agent is doing, which is awesome. So all this down and then my final comment here was the the wireframes piece that you you speak of, Tommy. I to your point Tommy I’m going to double down with you and I’d agree with you that doing more wireframes having more examples generating more of these things

55:06 examples generating more of these things programmatically with is is going to be incredibly important. And then we’re not there yet. We we don’t quite have a system that lets users generate wire wireframes that you can then just easily use and and manipulate. However, what I will say is I have a workload for that coming very soon. So, so stay tuned and this my workloads may be out by the time this podcast actually lands. So, if if I can actually get something that we’ll start regardless, Tommy, I probably done

55:37 start regardless, Tommy, I probably done you a disservice. I probably need to get you testing some of these things in your tenant as well to see what it’s some things are still see if your opinion will change. But I sure sure because I’m so early to this game and I’m seeing the vision for where this agent stuff builds, everything I’m creating now has to have everything I create makes it better to work with when I have an agent. And and anytime I look at anyone who’s building a product or something, if there’s not some sort like there’s I don’t know how to describe

56:07 there’s I don’t know how to describe this, Tommy. I I I I think you’re just trying to be a little a little safe here on what you want to say because I think I know what you’re trying to say. Anytime there’s a UI with lots of clicks and things, I feel like there’s a better way to do that by giving that what you’re doing under the hood behind that a an agent on next to it. Right? I I don’t want to go I’m physically saying I don’t want to go through the multiple menus of clicking like set this up,

56:38 menus of clicking like set this up, click these buttons, set these settings, do this thing, do this thing. I want those series of UI menus to have an under the hood definition of what you’re setting up, what parameter, like I was just watching a YouTube video recently from someone who’s who’s built like a rules engine. Like it’s basically run rules against your DAX statements and then those rules get built into something. Okay, interesting. But I don’t want to do that. I want to just describe the rule to the AI agent and say make 10 rules that verify this

57:09 and say make 10 rules that verify this on these columns and recommend from the through the model. Build all five tools at once. I don’t want to type them in. I don’t want to do it all manually. I want everything to be automated in a nice easy concise way. So when I look at this, I’m like, I don’t really love what’s happening, how it’s appearing. So for me, I’m looking at this going, there is additional context that I want to just give to an agent and have it just figure out how to build the underneath code that I can run over and over again. So I want to double down on this concept of the creator agent

57:40 of the creator agent agents that are used to create. That’s that’s to me that’s the money shot. That’s the money ball right there. when you can incorporate agents to help you build the deterministic piece of code, the deterministic thing over and over again. Now, like we talked on the podcast maybe a long time ago where I said, “Look, you can you’re going to get custom websites where the website doesn’t exist until the agent builds it for you in front of you as you click on things.” Yeah.

58:10 you as you click on things.” Yeah. Right., you’ll be able to ask an agent, hey, what’s the,, here’s a a a website that’s selling products. You click on the product, it just generates a page for you automatically on the fly. Now, the agent is generating the page, but you’re not talking to the agent to run the page, and then over time, the agent is now generating hundreds of pages around a product detail and doing automatic analytics, like which one performs the best. If someone asks a question about what’s missing on this page, like, “Hey, I really wish I knew how much the product

58:40 really wish I knew how much the product would weigh, it can add it immediately right there to the page. Oh, let me add that for you.” Boop. It’s just there. And now you have this hyper dynamic. The the agent is going to build the website for you, but it’s going to do I’m not talking websites. It’s the same thing for a report, Tommy. A report is a website. It is. It is 100%. It’s D3 under the hood. It’s JavaScript. under doesn’t matter. It whether you click the buttons or not, whether you do it in desktop or not,

59:11 whether you do it in desktop or not, it’s all a website. It all is. It runs on power. com, meaning it’s a website. It’s just a very narrowed down specific version of a website. Everything you’re building is websites. Period. To build the back end of that. Okay. I Yes, it it does have a URL. We’re talking reports. We’re talking No, hold on. Hold on. We’re talking reports. We’re talking wireframes. We’re talking all the front end side of things. things. Yeah. Yeah. All of that’s website. That’s very

59:42 All of that’s website. That’s very wellnown, very well documented. It’s just a website. I get there’s code on the back end. I understand that. Who cares? Yeah. Which by definition, I don’t care if it’s built in Rust, HTML, JavaScript, TypeScript. I don’t care what it is. It just needs to have the visual on the page. And when I click on something, it better do something. Like that’s what we’re defining. We’re defining the user experience and that’s it. Everything else is like you’re like there’s a whole bunch of other techn I don’t care about

60:12 of other techn I don’t care about writing DAX. I don’t care about the semantic models. I don’t even care about the DATA STRUCTURES LIKE WHOA WHOA WHOA WHOA THOSE ARE those are all there to serve the end of the HTML page that is the report. report. That’s that’s the land like that’s the landing goal of what we’re doing for a lot of the things on the reporting side. I’m I’m absolutely shocked by some of these statements here because when AI agents can build all this for you, I just did a I just did a video demo on Tuesday with Alex P’s tool building building a

60:43 building building a so building a wireframe of a fabric environment that said, “Hey, I have SharePoint files that I’m pulling in. I want to build this, this, and this.” And it built all the infrastructure. It built the entire folder. It it built like six or seven items in fabric for me. and deployed them all. My eye is switching. My I think I think my eye is switching right now because you just said some really more than just hot takes, burning takes here about the semantic model not being as important or the DAX. I don’t want to build that. But

61:14 the DAX. I don’t want to build that. But unlike unlike the agents are the agents are obiscating that technical away from it. I understand what you’re saying from the building side, but whether or not it actually does something help I I sound like I’m sports radio right now. Goodness. No. Whether it actually helped Derek Ger Yankee. No. Yeah. Good. Why can’t you win a World Series? This is Mike and the Mad Dog. No, my my my argument here, I don’t disagree. He can build all the things. Okay. It can

61:45 can build all the things. Okay. It can do it. Great. Cool. I’ll go get coffee. I’ll go ice skating with my daughter. The point I’m trying to make here is unlike a website, the application of a website in terms of what is the purpose of a website, right? It’s to click on things, to do things, interact with it. The purpose of a report, of a dashboard, don’t care what you call it right now. Where the purpose of that is to look at information, consume it in a way that helps my team. Five minutes to make a

62:15 helps my team. Five minutes to make a decision. What’s the rule? 3 seconds, 30 seconds, three minutes. that that information on there is relevant to me for three seconds when I have to make a You just described to me a website. You just described me a website. You did a website does the exact same thing. Take that 30 seconds, 60 seconds, 90 seconds. Take that same language and use it for any website you go to. I need everyone who’s listening to this, go to YouTube, listen to this podcast, please put your comment. Are we talking

62:45 please put your comment. Are we talking the same thing here? Is it a website or is it a report? Are they different? I need to know. I need to know from other people. If I’m wrong, if I’m wrong, I will say I will wear a t-shirt saying Mike’s right, I’m wrong. Okay, but they’re they’re the same thing. It’s the same thing, Tommy. Okay, so let me ask you this. We I feel we could do this all day. If I’m a listener right now, wait. And wait. And before you go there, tell me tell me the business things you just said. I want to write them down. I want to literally I want You said reports are this, this, and this. You

63:16 reports are this, this, and this. You said that you said some very clear things you needed for reports. A report is for a user to make a decision in 3 seconds, 30 seconds, and 3 minutes. It’s not just interacting, but it’s for the context and relevant to me. Information. It’s relevant information for me. Okay. Okay. I know what you’re going to do with this and I am not going to like it. So, So, I already I already know why. I know I’m getting set up. Continue on with your points, Tommy. continue on with your points cuz you’re

63:46 continue on with your points cuz you’re just proving my case right now to I know you’re feeding this to AI., I’m not feeding it to AI. I’m doing this is all human effort here, Tommy. This is all human related. You haven’t done human effort since 2024. Okay. Yeah, you might be right. But but keep going, Tommy. Make finish your point and then I’m going to rip it all apart in a second. Keep going. Do it now because I have a question and we’ll we’ll do a closing. Okay. Just just rip the band-aid. Okay. Let’s hear it. it. Okay. When I go to amazon. com, am I not

64:16 Okay. When I go to amazon. com, am I not looking at a lot of reports, visuals, product details, information in order to make a decision about whatever I’m purchasing regard and that’s just a website. And I would argue every website I’m going to, it’s some form of making decisions on information I’m provided. Period. And that’s and that’s and so the the line between whether that’s a bar chart honing in on my business process and showing me data

64:46 business process and showing me data specific to my business or whether that’s a product image with details and a short form video. It’s just more refined in how they’re advertising to their thing. If I have a website for my

64:58 their thing. If I have a website for my company, I’m pitching you here’s what we do. Here’s why I would meet your needs. here’s what you you want to work with my business for. Every single thing we’re doing, they’re all reports. They’re just like we’ve just blurred the line. And the and the other part is websites are birectional, meaning I can get data out of them and I can insert data into them. And now it becomes a two-way street between information out to me to make the decisions, which we don’t. We get really well out of PowerBI, but you

65:29 really well out of PowerBI, but you don’t get really well with PowerBI by entering in information. I don’t have random fields. I don’t have random calculations analytical. Yeah. Yeah. I can’t in my report today with PowerBI. I can’t real time adjust DAX in order to change the output of the visual or measures or calculations. It just can’t happen. It does not exist in the current form of the way the reports being built today. So the reason I state this is because with this new agentic space wireframing and agent space here, you’re going to have the ability to make

66:01 you’re going to have the ability to make the website for information for reports. So the the report moves from a report to now it’s a product, it’s an app, it’s a website. It’s truly going to become that direction moving forward. Let me rip your band-aid off with another band-aid. Okay. Okay. Every website requires me to act on it. Meaning I have to actually interact or click something on the website or it serves no purpose. Amazon I have to I have to click on add to

66:31 I have to I have to click on add to cart. I have to buy something. You can’t be any different for reports because you want a dashboard to provide the information when I go to it. What’s our number? Everything we’ve talked about since day one. one. I don’t have to click on anything on the website website like like No, no. Everything that we’ve talked about for every report, we boil it all down to does this report make me money or save me money. Save me money by just looking at it. There is no website in the world that you just go to and just look at the first URL and don’t click on anything.

67:02 first URL and don’t click on anything. But the the actual dashboard is something that you go to go there’s my numbers. Maybe I can explore more or discover, but I shouldn’t have to do all these different clicks. Tommy, you were playing right. You were saying the exact same thing that Microsoft is saying. Microsoft is saying reports are no longer read only. That that’s a what you’re you’re painting a picture of a report that is so narrow and so close-minded now because reports are are dashboard starts with that. What I’m saying you can’t do more insights to action. That’s what Microsoft’s tag is insight to action

67:33 Microsoft’s tag is insight to action somewhere else. Sure. But yeah, but that’s just that is a the action may be somewhere else. Okay, fine. That just means your report doesn’t do everything you want. That means your report isn’t fully capable enough. Your report should do more. Why doesn’t this is exactly why that’s the only thing it does. I’m saying by default by default the purest definition okay of a dashboard or a report. Right? So obviously you can add all the bells and whistles that you want

68:04 all the bells and whistles that you want to it. I’m not saying you can’t do that. But what I’m saying your definition of what is the difference or what is in its most pure form a dashboard. It is something someone looks at, consumes information to act on somewhere else. What is my number? Are we meeting our requirements? Great. Okay. I just disagree with it. Like so like it you’re describing step one of multi-steps that should dashboards and reports should be doing. They should really be operating like web pages. That’s what they should be acting like. You should be able to then see the

68:34 You should be able to then see the information you need and take action. This the reason I and the reason I say this Tommy is because Microsoft has proven my statement because insight to action is what they language the whole reason translitical task flows appeared was because reports doesn’t let you do the actions you need inside the context of the report we talk I get it but that’s that’s the direction Microsoft is moving towards which is stay in the context of the data of the report make it more like a web page and now it becomes actionable right next to

69:05 now it becomes actionable right next to what you’re And then when we’re talking about the Amazon analogy, that’s just I’m still using it as a dashboard. I’m still using it get information, but the tooling, the website, the app provides me the ability to take action directly inside the product. That’s what we’re getting to do. We’re going to be doing more of moving forward with reports. It’s they’re going to change from reports. They’re going to move into apps. apps. Clearly on this, I don’t think we’re going to see eye to eye. Again, please, if you’re on YouTube, I will admit if I’m wrong, I will wear the t-shirt saying I’m a dummy. Okay. So, I’m going

69:36 saying I’m a dummy. Okay. So, I’m going to do that. No, no t-shirts or however you want to do it. I will introduce myself on the podcast. Hi, I’m Tommy and I got dashboards wrong. Okay. I’m I’m going to This is so important right now. This This is a poll or this is Yeah, Yeah, Tommy. This moment, this episode, this is going to be an episode. We are going to retalk this episode. 533 Claude Dashboard Designs. We’re going to retalk this. Okay. Okay. Sooner than you think. Okay. No, I can’t wait. I can’t wait. I

70:06 Okay. No, I can’t wait. I can’t wait. I can’t wait. You’re going to talk about this again sooner than you think. Okay. We’ve gone substantially over on time. Make your final Let’s do final thoughts here, Tommy, and we’ll wrap. Final thoughts. I want I wanted to say this and and I’ll ask you this if you want to put add this to your final thought, but if I am a listener and you’re asking why are mom and dad fighting and you wanted to say how can we sum this up here is I I we haven’t had a good fight in a while. This is great. This is a spicy episode. You You called this one, right? You did. You did see that at the

70:36 You did. You did see that at the beginning of the episode. You said this is going to be a spicy episode. It was spicy. spicy. I’m just letting I’ve missed this. I’ve missed this type of argument. So,, but if a listener was to try oneclad workflow this week, like what would you actually do? And and I think that’s a a takeaway here. And I and I’m thinking about this and again I’m going to say if you actually have a transcript if you just did a discovery call you have the data provide to an agent just the transcript on what would

71:06 agent just the transcript on what would a wireframe look like. Hey just build what a dashboard design would look like. I don’t think you even have to go as so far to provide the semantic model information. Just the d just hey I know they really care about these things. here’s how ask me some questions use the grow me skill to ask me questions about the what what I’m telling you and then help me build some designs from there I think that’s a great place to get started using AI for a design point of view

71:39 I’m trying to be thoughtful about my answer back to this one so what’s the one what’s what’s your one let me go back to you Tommy you ask me a lot of questions and I usually have to give my answer first what is your one thing you tell users this We walk away from this episode. What I just said, okay, it’s it’s use a grill me skill. It’s it’s build What are you doing? Build an image. Build a screenshot or build a a wireframe for Yeah, I just read an article that said the shoot, what was it called? It was something something to the effect of

72:11 something something to the effect of HTML is undervalued. The the incomprehensible value of HTML is what it said. Basically, agents are so good at building website pages. I just read that too. that too. Yeah. Yeah. Okay. So, our algorithms are synced because we’re reading the same stuff. So, I the article title was so catchy. It was like the the I get the confounding value of HTML. It was something silly like a really interesting title. We’ll see if we can find it here or not. Yeah. Yeah. Let me give you the gist. The gist of the article was HTML is way better to

72:42 the article was HTML is way better to look at than markdown. You can write a bunch of markdown files and it can be very helpful for you to write those. However, HTML is much more human readable. It’s much more you can make design choices. You can make organiz organized information. You can put little infographics on it. HTML is really useful. Now, I’m saying the technical term HTML. Oh, Tom, you did find the article here. Let me see if I can pull up the name of it specifically. The same one, I think. Let Yeah. the unreasonable effectiveness of HTML. Right? My task

73:14 effectiveness of HTML. Right? My task for for users this week would be if you have access to an agent, which again that’s dependent on whether or not your company’s giving you access to or you on the side are buying one and using it on your own. your own. We know your mother does. So we know you would know your mom has HTML and a cloud code agent. so so that’s great. design work with an agent and say and be the the prompt the prompt engineering I’m giving you right now is I need to build a design style diagram for my report I have and if you have a reference image

73:45 I have and if you have a reference image you like if you have a color palette you like just give the agent the image screenshot it take a take a snippet of something say I want to make something like this and say in the prompt I need you to design a sample report that has these styles and elements ments on it and you just describe what you want and you say make it a single p this is the critical part make it a single page HTML file single file HTML page I said that wrong make it a single file HTML page you say

74:15 make it a single file HTML page you say that it will make a single file for you and it will generate a page you can look at now the trick of this is if you’re using like a chatbot like in like chat GPT you’re going to a window if you tell it this there really is no file system there so it’s going to what it’s going to is going to spit back to you just the code of like hey here’s the code that I made for you and Tommy to your point earlier in the conversation which was well agents are really around like a file system that can make a file and say here you go I’ve made it now download it so so in in using this it’s going to be

74:46 in in using this it’s going to be advantage to you to use like VS code a command line interface something that that talks to it where you can actually get it to make a single file for you that you can download to your computer and run it. You may find there’s a little problem where it has some like corores information whatever blah blah blah just give the error back to the agent and say hey there’s an error fix it I can’t render it whatever it’ll it’ll fix it for you it’ll fix yeah but just doing this just spending the time to to spend a little bit there

75:17 time to to spend a little bit there you now have bridged the gap between things you may want to wireframe and what the agent can start doing. This is a fundamental skill we’re going to have to pick up and the tooling will get better. I promise you this because I’m working on stuff like this that will make this tooling much easier for you to use. And so you could just say, “Make me this thing. Make me this thing.” It’ll just point you’re learning about how to actually utilize that too. I like that a

75:47 actually utilize that too. I like that a lot. lot. Yeah. So I think I think there’s something there that is really useful in this space that’s going to be be leveraged. So to me there’s a we could take this a hund different ways. There’s a lot of other things I could tell you to like do this this and this. There’s all these design patterns and like day one that’s not helpful. Day one is get started making designs on your own and building your own little like learn how to build a style library, learn how to build a sample report, learn how to build a wireframe of some information.

76:17 build a wireframe of some information. do that with an agent. Once you master that skill, then you can incorporate all these other things and then you’ll start getting the aha moment of like, oh, this really is important. This is going to be super impactful moving forward. So, I think that’s the the learning the lessons to be learned here to really push this forward. Love it. I I actually really really like that because that helps people also get comfortable with it, too. Yes. And that being said, we’ve gone way over on time. These episodes keep getting longer and longer. Tommy and this is we’re like we we’re not going to do news. This is going to be a short

76:48 do news. This is going to be a short episode. Nope. It was longer than all the other episodes we’ve ever done. So, that being said, thank you very much for listening to the podcast. We do appreciate you. If you like these heated discussions, please make sure you give us a thumbs up for this one. like it, let us know, share it, please, if you don’t mind. We really appreciate that. These are fun conversations that we’re having. We hope that you’re also enjoying the fun conversation as well. Tommy, that being said, where else can you find the podcast? You can find us on Apple, Spotify, wherever get your podcast. Make sure to subscribe and leave a rating. It really does help us out a ton. You have a question, idea, or topic that you want us to talk about, maybe talk about this

77:19 us to talk about, maybe talk about this more, head over to powerba. tips/mpodcast. Leave your name and a great question. And finally, join us live every Tuesday and Thursday, a. m. Central on all of PowerBI. tips social media channels. Thank you all so much and we’ll see you next time. Tommy and Mikey, dance in the mix. Fabric and A. I get your fix. Explicit measures. Drop the beat now. Pockets

77:50 measures. Drop the beat now. Pockets can’t steal the crowd. Explicit measures. Drop it.

Thank You

Want to catch us live? Join every Tuesday and Thursday at 7:30 AM Central on YouTube and LinkedIn.

Got a question? Head to powerbi.tips/empodcast and submit your topic ideas.

Listen on Spotify, Apple Podcasts, or wherever you get your podcasts.