Fabric Apps Fit into the PBI World – Ep.563

Explicit Measures spends this episode on a question with overlap on both sides: where a Fabric App fits now that Power BI Desktop can take an agent through the Desktop Bridge. Tommy Puglia and Mike Carlo put it to Kurt Buhler of Data Goblins, and the test they keep returning to is whether the Rayfin app does a job the report still cannot do well.

News & Announcements

• Get ready for table discovery in OneLake catalog search (preview) — Tables become standalone searchable objects in late September, and the metadata you can discover expands on the permissions you already have. Tables from semantic models, lakehouses, and mirrored databases show up as their own results. You can search by table name, description, or an exact column-name match, while the columns themselves stay out of the standalone results. Search is available in the Fabric global search UI, the OneLake catalog search API, Fabric MCP servers, and the Fabric skills library, with exact phrases plus multi-character and single-character wildcards. Mike Carlo treats exposing semantic-model tables as a win and expects the catalog to show how thin the descriptions are. Tommy Puglia’s picture is a user hunting every table about the shirts the company sells, then a Teams message asking why there are so many of them. Both hosts want the OneLake catalog to be the place you look, and both say the BI team, or whoever owns governance, has to be ready for that exposure. Releasing the search and walking away is the move they refuse.

• Lineage-aware AI with the Fabric item relations API (preview) — A new REST API returns the upstream and downstream relations of any Fabric item. Each response includes the items in the lineage graph, typed edges for association, shortcut, push data, and orchestration, and the workspaces those items belong to. Tommy reads it as the portal’s visual lineage turned into a programmable surface, and he files it with the other Fabric work Microsoft is building for agents: an agent can walk the dependency graph and ground an answer in the real relationships. Mike has not used it. He likes the API and is wary of the AI label, the cost, and the chance that this is a fancy lookup. The REST surface itself is what he calls important, because it is how you start a lineage view and ask what moves if a table changes, the same impact question he associates with Measure Killer. Tommy’s comparison is the Power BI Desktop Bridge. People are not living in that terminal. The CLI is there so an agent can call it. Mike’s guess is that the relations API lands in the Fabric MCP server the same way, so an agent can fetch an item and report the impact.

Main Discussion

Topic: Where a Fabric App belongs once an agent can build the Power BI report too

Mike has a series on building Fabric Apps with AI, and the Desktop Bridge has started pulling that same loop into Power BI Desktop. He calls the balance a teeter-totter. Tommy opens with the Steve Jobs iPad test he uses whenever Microsoft ships something that overlaps a tool people already have: it has to do a job better than the phone and the computer, or it has no right to exist. Kurt Buhler takes the question from the visualization side. A couple of months after Fabric Apps arrived, the three of them try to say what the job actually is.

• The pitch is a contained web app, and a KPI page is the wrong first one. Kurt ranks Fabric Apps, after the Fabric CLI, as the feature in Microsoft Fabric he has been most excited about. The draw is control of every pixel, so the interactions can match the question and respect the audience’s time. An agent can produce a D3 visual to a specific instruction, which is still a fight inside a Power BI report. Past the chart, he wants custom integrations: planning, writeback, and management of something that already lives in Fabric, delivered as a web app that keeps Fabric’s distribution and permissions. An executive dashboard with a few KPIs and breakdowns is the case where he leaves the Fabric App, and Rayfin, on the shelf. Interactivity and writeback is where he would look at a Fabric App ahead of a translytical app. He is plain that teams should not plan to replace reports with this tomorrow.

• Power BI’s constraints are doing work, and PBIR is what lets an agent join that work. Kurt’s line on report building is direct. A decent visual, short of an exceptional one, still means working the report and the semantic model over: SVG measures, HTML, error bars used as a target line. Those limits buy consistency, a result you can reproduce, and a cap on how wrong a low-maturity build can go. Mike saw the same pressure in the world championships at FabCon Atlanta. The winning report was beautiful, and a large part of the polish was HTML an agent wrote into a measure and rendered in a visual. He still wants Power BI as the wide tool, the one a new user and an advanced user can both get value from. What the Desktop Bridge changed for him is the click tax. The agent takes settings off his hands, and he can see the edit sooner. Kurt’s correction is that PBIR is the unlock. An agent could already publish into the workspace. The bridge closes the loop: the PBIR CLI he built with Maxim can refresh Power BI Desktop and grab screenshots, and the agent is spared the legacy JSON. The tradeoff stays personal. Every author has a gut feel for how many elastic bands are acceptable, and when that author leaves, someone else has to reconcile them.

• A Fabric App is a different craft, and the skill is systems thinking plus visual literacy. Tommy rejects the ladder that puts Fabric Apps at level eight after seven years of Power BI. Kurt’s reason is that the app is code. A Fabric App can take a supply chain attack. A Power BI report cannot. A LinkedIn screenshot can be a YOLO. Something you put in front of people cannot. Mike’s version of maturity is the story you pull out of the user before anyone writes code, and the example he keeps is Injae Park’s app: a 3D body you click through, with parts highlighting as it moves. A 3D building for a construction company that never touches the day-to-day work is tokens spent on something shiny. Tommy’s practical version is the analyst who does not want a second career in web design, choosing among D3, Plotly, Deneb, Vega, and Vega-Lite. Kurt names that visual literacy, and whether the organization has it. A team that already struggles to land a scatter plot will struggle more with custom interactions. Demos are easy to make interesting. Useful is the bar he wants, and Tommy’s word on top of that is consumable. The cognitive-load series the two hosts already recorded applies the moment every app looks like a different product. Tommy’s working picture is ad hoc use, chosen with the restraint of someone at a buffet.

• Few BI teams will build one without an agent, and the agent has a bill. Kurt’s prerequisite is that very few data teams have the skills to build a Fabric App without AI tools. That does let developers who were outside Fabric participate, which he welcomes. A BI team promising custom experiences will be leaning on those tools, so the tools have to be allowed, governed, and watched for cost. He is frustrated that so much AI guidance is produced on subscriptions and then offered to enterprises that pay per token. A prototype can stay ephemeral. A dashboard for a team-building day can look good, start a conversation, and be thrown away. A month-to-date sales view against budget and forecast needs ingredients Power BI did not ask for: a design system, rules for how the apps look and behave, and expertise the model cannot supply. Mike asks what they actually run. Kurt mixes Claude Code, Codex, and OpenCode, with more local models lately, and he cares more about the context than about the harness. His habit is to plan with a stronger model, execute with something in the Opus class, and do the small iterations on a lighter one. Tommy keeps that context in Notion. Skills sync from there, Rayfin skills and Fabric skills included, and a master sequence of pages walks Claude from data discovery through visualization into the design and system phases, updating Notion when a phase is done. Claude and Opus are what he reaches for, and Cowork’s sync with Notion is why he has been living in the CLI less.

• The terminal, and a Linux laptop at home, are Kurt’s workshop. He likes what Cursor did for managing agents, and he has mostly moved on. Claude Code in the terminal is the daily tool, including at Tabular Editor, because he wants several agents running and able to talk to each other. Tommy points him at Clink, the Windows terminal autocomplete he uses so a CLI such as the PBIR CLI can suggest subcommands the way pip does. Kurt’s machine since the start of the year has been a Mac, and more of the work now happens on Linux, on Omarchy. Power BI Desktop runs there in a Windows VM he describes as basically the real experience, containerized, and he has had the agent on Linux control Tabular Editor in that VM through an encrypted tunnel. The disclaimer is the sentence Tommy was waiting on. This is a personal experiment for an old laptop at home. It is too early to take to a decision maker as the company standard. The feeling he warns workshops about is the one he calls token cocaine: the stretch where every idea feels possible.

• Have the agent build a deterministic tool, and notice when you are machining a screwdriver. Mike’s placement of AI, in Power BI Desktop, in Fabric Apps, and in the data story, is that he does not ask an agent for the number. He asks it to build the repeatable thing. Same inputs, same outputs. His picture is a calculator that needs extra buttons for one field of math: describe them, let the agent build them, then use the tool. Kurt calls that the Pareto take, and a Fabric App can be that tool. The failure he has watched, and owns, is using a power drill to invent a better screwdriver. The PBIR CLI experiment with Maxim last December could drive Power BI reports programmatically and still could not get pixel-perfect control without custom visuals. The chart he actually wanted had already been a D3 problem for a decade. The exception he keeps is a problem that is not deterministic. Unstructured data, an unstable ETL notebook failing almost daily early in a deployment, and an agent you deploy to watch the pipeline, attempt a fix, and bring it back for approval. A comment in the chat lands on the same nerve: the explorers who only ask whether they can. Mike’s account of the psychology is the shortened win. Hours of being stuck, then the euphoria of a thing that works, compressed into minutes, which invites a bigger build. Kurt still wants the ambition. He is still critical of AI. The three words he wants people to value are critical, curious, and creative. Shorter build cycles do not lower the quality bar. Put the friction on planning and design, prototype inside that process, and throw the prototype away before it has to carry a product. Tommy’s worry is building for the author. Kurt ties that to why Power BI spread from the bottom. Jack or Janet in finance gets five minutes to wow, and their colleagues are right to be proud. Fabric Apps can do that on another level, and the result still has to be purpose-driven and gated. The visual you slaved over, and the stakeholder who asks for a table, is the same bruise in a Fabric App, in a report, or in a notebook.

• Nobody has the environment yet. A research bench is the honest version. Asked what a successful setup looks like for the developer and the consumer, Kurt says he does not know. It is too early for evidence. Mike turns that into homework for Injae Park, for Kurt, for Marco Russo, and for the rest of the community: write what worked, what failed, and where an app was hard to maintain. His own bright spot is Fabric as the backend. Once the data is in a Fabric SQL database or a semantic model, he can push it around. An app for an ops team in a warehouse that ships product is a maybe, after more attempts. Tommy wants a research garage. Build quickly, stay critical, and keep the experiments from simply getting loose. Then run a real request through an intentional process: how this Fabric App would actually be tackled, and what it needs. Kurt will not call himself an expert, which is why he holds back a recommended personal workflow. Tabular Editor already splits the work into an innovation team for rapid prototyping, with room for outlandish experiments and a duty to kill them. His takeaway for a BI team, a center of excellence, or a business team living in Power BI is one or two people responsible for the trying, bringing it back every week in the open, so everyone else can say whether it is valuable.

Looking Forward

Leave the month-to-date sales report in Power BI until the design system, the permission model, and the token bill are real enough to carry it. Give one or two people a research bench, have them bring a Fabric App back every week, and kill the experiment that only impressed its author. Mike and Tommy already have Kurt Buhler back for more of this, including how artistic the building gets once the agents are in the loop.

Episode Transcript

0:01 I get your fix. Explicit measures. Drop the beat now. Pumpkins feel the crowd. Explicit measures. measures. Drop it loud. Hello and good morning. Welcome back to the explicit measures podcast with Tommy and Mike. Hello Tommy and good morning to you. to you. Good morning Mike. How are you doing? These days are going well. I think things are very interesting. I’ve been

0:31 things are very interesting. I’ve been having a lot of fun playing around with different AI things. it’s this is going to be a good conversation today. Today our main topic is going to be discussing a tangential area of AI. maybe we’ll we’ll get into a little bit of this one, but there’s this new thing called fabric apps that have been out for a while. I was extremely excited about this. I’ve done a whole series on building with them and how to use AI to help you create things. We’re going to unpack where does a fabric app fit in lie of what we can do

1:01 fabric app fit in lie of what we can do now with PowerBI., and I think some of the recent developments of PowerBI, especially with the desktop bridge, bring some of the AI development building experiences directly into PowerBI desktop. So, I think this is for me a teeter totter a little bit. I’m trying to balance where things need to fit these days. Anyways, that’s our main topic. Tommy, I know you’ve got a little bit of news for us today., give us an article. Let’s go through some of it. So, we got two quick things that we’re going to talk about. First is the

1:31 we’re going to talk about. First is the fabric has a new update for the one lake catalog which is for table discovery in one lake catalog search. And basically what this h what’s changing is tables become standalone searchable objects in late September. So outside of semantic model which is really interesting here it expands discoverable metadata based on the ex existing permissions and how it works is simply tables from semantic models lakeouses in mirror databases will appear in independent search results you

2:01 appear in independent search results you can search by the table name description an exact column name match which this is great Mike even columns themselves won’t appear in the standalone and search is available via fabric global search UI onelay catalog search API fabric MCP servers and even the fabric skills library. Mike, so Mike, this is important to me. This is really as we get to the Garmin side. There are a few things too with search operators, exact phrases, multicar wild card, single

2:31 phrases, multicar wild card, single character wild card. That feature I thought you’d pick up on because that feels a little bit more like what a Tommy would would look for in a search feature, which is okay, can I search it like Google? Can I can I do I have wild cards? Can I do spaces and gaps and complex reax searching for things? I felt like you would pick up on that feature and I’m glad you mentioned that one too, Tommy. Yeah, I think Tommy all this is telling me this is going to further expose how lame your semantic model definitions and descriptions have been of our of our

3:02 descriptions have been of our of our tables. I think it’s going to further expose our issues with that part of the world. world. Well, and this is part of the governance, right? So, I can’t Am I excited? I don’t I’m excited for it for myself. And I think about this from like a user saying, “Let’s find all the tables that are for our products, right? Or the shirts. What columns do we or what tables do we have that talk about the shirts that we sell?” I think that could be really important. But I think the onus here is on the BI team or the

3:32 the onus here is on the BI team or the if you have data governance in place or this is one of those you’re going to get a lot of teams messages going hey why do we have so many shirt tables in here and you’re like come back up somewhere. Yeah. So I think there’s going to be a few things here. I think this can be really impactful for organizations but it’s not just releasing it and letting people do it. Yes. I I think one lake catalog I think is really the right place to start really putting a

4:02 right place to start really putting a lot of data or a lot of things about your system in the same place. I I I really like what one link catalog is doing. So I’m very pleased with how that’s working and very excited to see more development on the one lake catalog cuz I you build a lot of semantic models, you have a lot of lakehouse tables. Where do you go to find all this stuff? You should you should go to fabric. You should be in the one lake catalog getting things out. Anyways, I like this solution. I like this idea. I’m happy to see them continue investing in this one. Exposing tables that are in

4:33 in this one. Exposing tables that are in semantic models. That’s a win in my book. This is a great feature. I love this one. This will be fun. All right. The next one we got is more AI and fabric. So, this is lineageware AI with fabric item relations API. And with there’s a new API that enables the ability to do upstream and downstream relations of any fabric item directly via REST calls. And each response includes items involved in the lineage graph, typed relation edges, association, shortcut, push, data,

5:04 association, shortcut, push, data, orchestration, workspaces those belong to. And this actually turns the portal visual lineage into a programmable surface for automation AI agents. So again, lineage answers to critical questions. We know that. But do this for context for AI. And again, this goes back to something you and I have been having almost every episode of conversation about that teams and Microsoft especially is beginning to build a lot of things in fabric for the sake of agents, not for the user. Again,

5:35 sake of agents, not for the user. Again, this is really going not that it wouldn’t be helpful for a user to use the programmable REST API, but the real value here, the value ad is for AI agents to go through the dependency graphs and be able to answer questions and grounds AI reasoning and actual dependencies. So, what’s your take here? I I Tommy, I’m gonna have to play with this feature. Any I like the idea of it. The again, this is feels like the

6:05 again, this is feels like the difference. Is this marketing or is this a real feature I’m going to want to use? Anytime I hear the word AI and something Microsoft has put their hands on, I’m a little bit hesitant about it because it feels like I like the feature. Is it going to cost me an arm and a leg and use? Is it is it really going to resolve to me the information that I need or is this just a really fancy lookup and we’re using some level of AI in the middle here to get the feature out the door? So, Tommy, this is one I I to be honest, I haven’t directly played with this one. So, I don’t have an opinion yet. I’m not No. So, yeah,

6:36 No. So, yeah, I’m not going to push you one way or the other. However, the fact that there is an API, a REST API to give you upstream and downstream connected items, I think is incredibly important. I think this is the beginnings of someone being able to build out a lineage view. Speaking of lineage view, there’s a great ex there’s a great item that you can go get from power designer which is doing lineage view and we are using the API to go get upstream

7:07 we are using the API to go get upstream and downstream items and we’re going deep on this lineage. But regardless lineage stuff I think is incredibly important. where the AI fits into this one I’ll need to play with it. Well, I think it’s like PowerBI desktop bridge, right? The CLI tool. I don’t think anyone is going in and using the terminal to use PowerBI desktop bridge to me. to me. Mhm. Mhm. So like I think again it’s meant for Asians. So I see the same here. So

7:37 Asians. So I see the same here. So definitely check take a look at both of those out. Awesome. Yeah, I think these are these are definitely going to be very worth it as well. and this will give you a lot of good metadata. So again, teacher AI. Okay, so this will be I think I’m guessing because this is an API call, this will come with the fabric MCP server because the fabric MCP server I believe lets you call fabric APIs. So they’ll be added either added to that or eventually will make it in there. Your agent will understand how to make an API call. It’ll be able to get the item you’re looking for and find the upstream

8:08 you’re looking for and find the upstream or downstream items from that level and say what’s the impact analysis. This feels a little bit like it’s stepping on the toes here of measure killer slightly., if I change this table, what happens upstream? What happens downstream? What’s attached to this? this? So, those are things that I think that are going to be really interesting there. there. Okay, Okay, that being said, we get to our main topic today, Tommy. Let’s do it, Mike. Okay, drum roll, please., we have probably the premier blog personality that has been

8:40 premier blog personality that has been on our podcast more than any other blog poster in the world., we’ve talked so many about these articles. We we’ve give a mad amount of praise to this individual. So, I’m really I’m going to hype them up because I am very excited to have with us Kurt from Data Goblins with us today. And let me get him on the screen here. So, Kurt, welcome. Hello., this is Data Goblins. He is with us on the show today. I’m very excited about this. Kurt, I know you’ve had a a very busy day. Kids are keeping you up

9:10 very busy day. Kids are keeping you up early and you’re working hard in long days. So, I do appreciate this is your evening. Thank you so much for spending a little extra time with us today, Kurt. Welcome to the show. All good. It’s all good. Thanks for having me. Appreciate it. Appreciate it. And of course, also appreciate like the fact that you have discussed and read a lot of my blogs. Like it’s been there have been a lot of times where I’ve been writing and I’m like,, how will Mike talk about this? Oh, no. Oh, no. I should probably add some clarifications here. Yeah, I should ask a bunch of dumb questions around something. I hope you don’t change don’t change it for me. Yeah, it’s it’s very instructive

9:42 for me. Yeah, it’s it’s very instructive though to like hear people talking. Like it’s also of course obviously like extremely flattering and I’m very I feel very privileged that to to to hear someone talking about your own work. Like there’s nothing more satisfying about it. But it also teaches you very well how does it look in the eyes of someone else and it helps you see things from a different point of view. So that’s extremely valuable. Like there have been times like I remember for example like when I worked on the BI strategy guidance for Microsoft and you guys had some lengthy discussions about it. Yeah. Some some of

10:13 discussions about it. Yeah. Some some of it got a bit picant as we say in Dutch a bit spicy and in in direct response to like some of the discussions you had. I did end up like rewriting one of those topics. topics. I remember that. Yeah. So I Yeah. I’ll put it to this context. anytime there’s a new article that you come out with, Mike and I are not saying whether it’s going to be on. We just say put it into the episode schedule because we know that great. It’s so yeah, we have loved your content for a long time

10:44 have loved your content for a long time and it’s not just because it’s the how-to, but it just fits perfectly on a conversation that honestly needs to be had. had. Yeah. Yeah. Yeah. Appreciate that. You’re so one, thank you for all the community input that you’ve been giving. We absolutely love it. The content is is very wellreceived. I love continually unpacking it, thinking about it. the lot of the watercolor cooler conversations we have here on the show are very much related to what your articles are. And I think Kurt, you resonate with with the audience. So

11:14 resonate with with the audience. So we’re extremely happy to have you on here. Tommy, I’m going to kick it over to you a bit here. Look, we’ve got a topic here that’s a little bit maybe edgy a little bit. This is a new world in how we build reports and I know Kurt, you’ve been blogging about this a little bit and unpacking this. We’ve talked a little bit on on the side here about this as well. We have had PowerBI for 10 years. It’s been out for quite a while. Just this last year, we’ve been able to receive this new world of where we can build fabric apps and we’re using our Aentic experiences to help build those things.

11:44 experiences to help build those things. So, where does this fit? We’re a couple months out since this has been announced. We’ve been able to play with it a little bit here. I think we’ve got our hands on things here. How does this fit? What does this look like? Tommy, anything else you want to add here for the context of our conversation today? Absolutely. So, Kurt, if you have listened before, one of my favorite things when Microsoft comes out with different products, I always go back to Steve Jobs keynote on the iPad. It’s one of my favorite things in the world, how he introduced it to the world. And his idea was if we’re going to have

12:14 And his idea was if we’re going to have a tablet, it has to do things better than our phone and our computer. It should be better than browsing the web, than a phone, but it should be better than watching a movie on a computer. If it doesn’t do anything better, it has no right to exist. And I apply that philosophy, so to speak, to a lot of the products Microsoft comes out with, especially when there is overlap. Holy crap. When you look at fabric apps and you see what it can do, I think this is a perfect conversation here in terms

12:45 is a perfect conversation here in terms of how do fabric apps actually fit into this PowerBI world that we have both from the reporting solutions and the sources of truth both from me as a developer but also from an organization as a whole. So that’s the we’ll say the groundwork for today. So I will open the floor to you

13:06 open the floor to you to you Kurt. So fabric apps first of all I think after the fabric CLI are definitely the feature I’m most excited about in fabric or have ever been excited about in fabric or or PowerBI I would say. So, and the main reason for that is the the potential and possibilities that it unlocks from a visualization and a UI UX point of view. like the fact that you can literally control every pixel of the experience

13:36 control every pixel of the experience in order to create exactly the bespoke thing that you need to create in order to in my case,, related to my particular interest like to be able to visually represent the data and create interactions that facilitate the exploration and understanding of that data in a way that’s respectful of your audience’s time., so I like know your audience’s time., so I like for me that was what was most exciting especially if you if you if you look at like you talked about the PowerBI desktop bridge and stuff like this like and of course I’ve been deeply

14:07 and of course I’ve been deeply experimenting mainly last year with agentic development of reports. if you look at trying to use AI to create reports it’s extremely difficult and you could we could talk about that for hours. there’s a lot of there’s a lot of reasons for that. there’s a lot of constraints, there’s limitations, there’s there’s complexity. an AI can just,, if you’re using an agent, it can create something in, for example, like D3. js or whatever, very quickly and easily and it can be

14:40 very quickly and easily and it can be able to tailor that specifically to your instructions. So suddenly you you can control all of it. But all of a sudden when you remove the fence and you have the open wide field in front of you, when those constraints are gone, it creates a different problem. So yes, it is a lot to unpack, but but it is something that I’m very excited about. I think it has very high potential. I think it has a lot of potential value. I I don’t think it’s something everybody should like jump into, and I

15:11 everybody should like jump into, and I don’t think it’s something you should say we’re all replacing reports with this tomorrow. There’s so much there’s so much to unpack here. So I I feel like I’m the I feel like I’ve just read two paragraphs into a really good Kurt article and data goblins and I’m like okay okay I have so many mixed emotions already you feel the emotions. I’m not expressing it well in my voice but you can feel can feel I know but it’s okay as you’re saying things I’m like

15:41 okay as you’re saying things I’m like yes 100%. totally agree with you and I know this is going to come up at some point and I think what you’re what you’re alluding to here is with much power comes much responsibility is what I’m hearing a little bit of your language here right right yeah so yeah so yes like a lot of what I’ve been thinking about lately and like we’ve talked about this in the past is constraints what is the what is the purpose of a constraint like in PowerBI and and I know this is going to sound a

16:11 and and I know this is going to sound a bit direct But we are all aware of the fact that PowerBI is an extremely constraining tool when it comes to creating good visualizations. It is very hard and the things I have to do to both my report and my semantic model to make what qualifies as a good report, not even something that’s like exceptional, but just decent. should be hung up in some,, hall of fame for butcher shops or something like it’s it’s pretty scary. So and and you don’t want to do

16:42 scary. So and and you don’t want to do that but you have to do that. so so but but right those constraints serve a lot of purpose. It ensures consistency. it ensures a certain reproducible experience and visual. it’s it’s very functional and it also helps to facilitate creativity debatably and it can prevent things from going you and it can prevent things from going too too wrong especially when the know too too wrong especially when the maturity is on the low side. when those constraints are gone all of a

17:14 those constraints are gone all of a sudden all kinds of crazy things can happen. if you have a hornets’s nest in your backyard it’s one thing to be able to like take specific poison for those hornets versus to take the shotgun and shoot some slugs at it. So, so and I think the the big thing is okay so what gap then is it filling here is it from that from the statement you said it’s the ability to create in a sense the perfect and perfect report subjective but the in a sense the most

17:44 subjective but the in a sense the most expressive way that you want to build a report with PowerBI you are limited from the design you’re limited from the visuals that you can choose yes we know there’s third party visuals but I would argue or not argue with you, but I would make the argument that for the most part with a PowerBI report, you from a utility point of view can provide the right information to the user. Okay. Okay. I don’t think fabric apps though they’re the biggest selling point would be the fact that I can build whatever I want from a design

18:15 build whatever I want from a design point of view. I would lean. Okay. Yeah. So, I think we’re going to be on the same page here. I think it’s that other element of the application. So if you were to basically explain the proper benefit of fabric apps to a business or to our listeners, what would be your in so many words elevator pitch? So like I said like my focus is primarily on delivering a specific, you primarily on delivering a specific,, data experience like you you you know, data experience like you you you really want to be able to portray to people in a visual and interactive way

18:46 people in a visual and interactive way an answer to their business question as as effectively and efficiently as possible. And like so for me that is the value that I can do that. But then from from a more generic point of view, the fact that you can create these tailored interactive experiences that are not just going to be about visually representing data, but are going to be able to have custom integrations and custom,, interactivity to be able to do all kinds of different things like to facilitate planning to be able

19:18 like to facilitate planning to be able to help management of,, even something inside a fabric. the fact that you can basically like it’s a web app, right? So you can build a contained web app experience that benefits from all the distribution and permissions management of the of the fabric ecosystem. so so so the potential is is very enormous. It’s just about finding the right application for applying

19:49 right application for applying that potential like you don’t you don’t want to say okay I need to create an executive dashboard I want to show a few KPIs and breakdowns like okay let’s let’s whip out the fabric app not necessarily but but if you have a case where you’re like okay we need to have this interactivity maybe we need some ride back like certainly I would be looking at fabric apps like over something like translitic apps for instance instance I would agree This is a good point., I want to maybe touch on some points you made here, Kurt, a little bit earlier around the PowerBI as

20:20 little bit earlier around the PowerBI as a framework. And I I think Tommy, early on, we really talked in the podcast, and this is a point I think you’re bringing up here, Kurt, is the idea that PowerBI is a framework. It it has boundaries on which it’s good at and what it’s not good at. And and I want to bring up some just observations that I’ve seen around just PowerBI desktop or building PowerBI reports. One of them is, have you ever wa have you participated in watching the world champs before? Did you go to there’s Fabcon at Atlanta? At Fabcon Atlanta, there was like a a world champs. Watch it live. Did you see that by chance? Okay.

20:50 that by chance? Okay. No. No. Okay. I I was able to attend one. It was fun, high energy, loved loved the competition. The winner of the world champs, beautiful report. I absolutely loved it. But to your point earlier, Kurt, around hacking the PowerBI system to get it to do what you want, there is an immense amount of HTML being written with an AI agent and then being shoved into a measure and the measure was rendering inside a visual. So there was a whole bunch of like not I wouldn’t say it was cobbled together,

21:21 I wouldn’t say it was cobbled together, but it was it was the tool had limitations and the only way to work around them was like a 15 little step thing that was like very specific to making this report look one beautiful polished edges again it was it was bringing a lot of the application like experience directly into the report. Now there was there was great storytelling. There’s a lot of things there. But I really like what you said here and I really resonated with there are different levels of data maturity and building maturity of reports and

21:52 and building maturity of reports and visuals as you look across your organization. Yeah. And I think the when you’re looking at the widest product you can throw at a team and say use this everyone new users all the way up to advanced users can at least get some value out of this looks like PowerBI right any user at any scale can can build things in PowerBI and polish it make it look amazing can get basic visuals on a page and start

22:23 visuals on a page and start communicating some stories So I think you sacrifice some of that really fine-tuning capability by constraining the PowerBI around a specific way of doing it. Right. Right. I will just add one little note here and I’ll kick it back to you Kurt here. I feel like we’ve gotten a little bit of fresh of a breath of fresh air by having desktop bridge being added to my PowerBI desktop experience. And I one of the things I’ve always complained about PowerBI was the heavy click tax.

22:54 PowerBI was the heavy click tax. Everything felt like a thousand clicks to do anything any setting. And so now that I have Bridge, I feel like I’m getting a little bit less of that and I can focus a little bit more and provide a little bit more of that refined experience, but the agent can take some of that weight away from me and I don’t have to do all of it myself. Let me pause there. Kurt, what do you think? think? So, first the bridge. I have a really hard time seeing why the bridge is so valuable there if I’m honest because it’s you could already do that before just and then have the agent publish it into the workspace. It’s now

23:25 publish it into the workspace. It’s now the the the loop can be closed more tightly and more quickly. That’s what I would maybe I point on that because you’re right the PBI is the unlock for that, right? Really? Exactly. Exactly. Okay. Okay. So So So let me let me let me say that way. Let me rephrase my phrase because because I think you’re right. The PBIR is the big unlock for me that agents can understand how to build reports programmatically with code. Yeah, Yeah, I like the bridge because now I can see the changes that it’s doing in a much faster cycle and that’s why I think to me that’s a closed loop now like oh this

23:56 me that’s a closed loop now like oh this feels much more natural to build with agents. Sorry, go back to you Kurt. Yeah. No, exactly. So, and like with the PBR CLI that I built together with Maxim, like we we also incorporated the the PowerBI desktop bridge. So, you can refresh PowerBI desktop and you can get the screenshots all from that CLI. And yeah,, it’s it’s the primary way in which I can make changes to,, programmatically to those files while also ensuring that the agent isn’t going to have to be reading a whole bunch of JSON files and potentially making a lot of mistakes

24:27 potentially making a lot of mistakes and stuff like this. but definitely like PBR is is is huge and if PowerBI didn’t have PBIR I think it would be in a in a very different position today where we’re trying to work with the the legacy JSON format. now again coming back to what you said before about like making these changes to the report and stuff like this. I to the report and stuff like this. at the bottom line is you you got mean at the bottom line is you you got to do what you got to do right? So so if if you’re in a position where you got to make a report and there’s a specific

24:57 to make a report and there’s a specific visual requirement that you have to fulfill for stakeholders either because it is the best way to represent the data for them or because it’s just something that is required or even because you strongly feel yourself as the report author like this is going to produce a better result then you have to make that tradeoff of how much of this elastic bands am I going to wrap around this to to try to make it to look the way I want like with the SVG measures and the HTML. even the configuration

25:27 and the HTML. even the configuration like using the error bars as a target line is a very common one. things like that. how how much are you willing to compromise and where is the threshold at which this becomes unacceptable and I think every report author like has a gut feeling of where that is and it’s different for everybody., and there’s nothing inherently wrong with it, but it is like the fact that you have to deal with that is is very difficult. And if that person leaves,, then then someone has to take over. It’s it’s it’s really horrible to try

25:58 really horrible to try to reconcile. Oh, man. You just you just laid a you laid a term here that we’re going to governance. The G-word. You said the G word. word. We We we will get to governance, but Mike, I already have here a litany of things that you said that I

26:14 that you said that I off with you on. So I I love to get back. I I got to push. So first, can you just clarify for me? Maybe that will resolve some of this too. Okay. You mentioned the maturity of someone going from PowerBI, the the skill that they have designing reports and then eventually to fabric apps. Sure. Sure. Right. Right. Yeah. Maturity. I I but I don’t see so if I were to build a this maturity curve to me I don’t see that as I become better at PowerBI I’m really good at

26:44 better at PowerBI I’m really good at designing reports the next level like level the next level Mario fabric app right I made level seven in PowerBI level eight fabric fab exactly and for me because I think there are different journeys they might branch off but it’s not that I am so good at building to your point the thousand clicks yeah but Tommy like you Still there is still a maturity gap. No, like because there’s a certain there’s a certain like systems level of thinking that you have to take when you’re approaching a web app that is fundamentally code

27:16 app that is fundamentally code and you like you can have a supply chain attack in a fabric app. You you cannot have that in a PowerBI report. you you you need to have a certain like you can’t just rock into a fabric app and yellow the whole thing like you can for a LinkedIn screenshot certainly but not for something that you want to put in front of people yes I think I think there’s maybe when I say the maturity of the user changes slightly right and I think this comes

27:46 slightly right and I think this comes with experience over time I also think this comes with I I do feel like there’s this idea of like a framework And one of the one of the biggest challenges I think I faced when I started showing fabric apps was how do I get a consistent look and feel yeah across fabric and PowerBI together. And I think Kurt I think you’re you’re touching on this a little bit which is there’s an introduction of systems thinking that come with the the fabric apps piece which I do really resonate with. And so what maybe more my maturity level

28:17 what maybe more my maturity level Tommy is have you guys I’m sure you have. Have you guys seen NJ Park and what he has done with some of his fabric apps? apps? Yeah. Yeah. Okay. Okay. I’m thoroughly impressed with the fabric app. He built some like human biology diagrammy thing. And again, I’m not a scientist. I’m not that I can’t complain those things. But the fact that he could like click on different elements in the report page or the fabric app page and it was highlighting like a 3D body that was moving around

28:48 like a 3D body that was moving around had different parts of the body being highlighted and all these really in my head I may need some reporting or information that is being conveyed in that way. Right? I think a lot of these are in my mind there is this vision for what the report or what I want to convey visually about the information. Sure. The reason we use visuals instead of just tables of data all the time is because it helps us interact, have emotional connection, I think a bit more with the data. And we’re looking for patterns and we’re looking for trends. There are the skills that you need to

29:19 There are the skills that you need to get to that level. Not just making the thing, but also,, Tommy, when you give me requirements, what are you trying to do? Oh, I’m I’m a doctor and I’m doing these kinds of things and this is how I do this analysis. Oh, and I have to explain things to my patients, but my patients don’t like,, I have there’s this whole story that you’re teasing out of those individuals to get to the point where you say, “Here’s a report or an app that now can help you assist you with that story, that data

29:50 assist you with that story, that data story.” And I think that’s the part of maturity that I’m really focusing on here is when we’re talking about maturity levels, there’s a a storytelling and then how do you articulate that story down into code and then ultimately into reports such that it feels part of the suite of of PowerBI and and or reports we’re doing, but also enhances the experience so they they’re actually getting value from it. I could make the coolest 3D building for a construction company, but if it adds zero to no value, it doesn’t relate to what they’re doing in the day-to-day

30:20 what they’re doing in the day-to-day business. I’ve just spent a lot of money in tokens just building something flashy and shiny, adding no ultimate value to the business. And so, I think that’s the balancing gauge I’m trying to evaluate here a little bit is that user needs to understand the lemon is must be worth that squeeze. Does that help Tommy with your your point around No, it does because I I think it’s important to say this like, “Hey, Tommy, congratulations. You’ve been working in PowerBI for five years. Your own now go learn fabric apps.” Yeah. Now, yeah. Now you’re doing fabric now. now. I may not want to though., there

30:50 I may not want to though., there may be people who are not interested in going in and like, well, do I want to work on like web design? Like, how do I make this button consistent on all pages? To your point earlier, Kurt, it’s wide open. You want to use D3. js, JS, you want to use Plotley, you want to use Danb or Vegaite, Vega and Vegaite, you can the framework, you now have choices to pick the framework. Does that user want to spend the time to research the differences between each of the different languages to figure out which one they should be using in their

31:20 which one they should be using in their report? And that’s where you you have to be a studier of this of the design system or the studier of the reporting so you can actually really have a good opinion about where to where to apply your effort. Yeah. And I I think too it’s it’s it’s it’s not that like the choice of which visual framework I’m going to use is is where that weight comes in but it’s more like like to your point right like what you’re describing ultimately to me sounds a lot Mike like visual

31:50 sounds a lot Mike like visual literacy right like or or in context of data literacy like yep how well do you understand conceptually data visualization and its value and how well is that propagated throughout your organization like Yeah, I may understand it, but if I can’t build something that someone else can understand, I might as well just not build it. build it. Yeah. like if if you if you just like you’re struggling to be able to help people to understand more sophisticated visualizations even just in the PowerBI

32:20 visualizations even just in the PowerBI report like scatter plots and things like this then if you’re doing something that’s a lot more complicated especially when it’s introducing custom interactions and this is an entirely new experience like it’s going to be very hard for them to be able to get value out of it and it’s it’s one of the things I struggle with a lot with like demos of fabric apps and stuff like this is it’s very easy. We live in an age where it’s extremely easy to make something that’s very cool and interesting like like but there’s a difference between something that is interesting and

32:50 something that is interesting and something that is useful and and I’m not saying like the things people are doing are not useful people are doing a lot of very valuable things and ideas on their own are useful but it’s very important to keep in mind that we are not making these things for us and it was the same with reports we’re making them for the users and how can we make sure that to your point talking about consistency that we’re providing a consistent perspective, a consistent user experience and we’re empathetic toward

33:20 experience and we’re empathetic toward the users. So we’re not telling them like hey you just learned PowerBI here’s this custom thing I made and it’s different from that custom thing that he made but look it’s so cool. Mhm. Mhm. I I I want to touch on this because I always love when I write something down before it’s before it’s said, but because this is my biggest concern, I would change your phrase a little, Kurt, where it’s it’s easy to make something cool, but you also have to make it not just useful, but I would use the word consumable because to your point, if I’m

33:50 consumable because to your point, if I’m creating a bunch of fabric apps that hopefully have a similar design, Mike and I did a while back a series on what we called cognitive load And that’s the Yeah. And I think a big thing here is fabric apps are so custom, but it’s going to be very difficult if to your point if I’m giving out all these fabric apps to users or my team is and I it’s a large organization that’s going to overwhelm right off the bat. Even if I am working at the doctor, it’s really

34:20 am working at the doctor, it’s really cool. I can,, rotate an object around and they can show the different data. But if that is different than anything else that they’re using, well, I think to me that’s a problem. And I think fabric apps, there’s been a conversation Mike and I have been having on are fabric apps going to be the default way of building things. And I don’t think that’s the case. I think it is going to be these ad hoc moments, but I think the harder question is going to be to when to actually use them. And this is hard to

34:51 actually use them. And this is hard to have. This is a hard conversation to have because fabric apps are that potential is so large, right? In terms of you really can build whatever solution that you’re looking for. Yeah. Yeah. But it’s almost like you have to inense selfch checkck yourself like going to a buffet like I’m not going to have that other roll of sushi or everything at the table because you have all the options available to you. You have to be very selective. Yeah. And like when when we’re

35:21 Yeah. And like when when we’re thinking about when do you use them and stuff like this like it’s it’s it’s about matching the scenarios and the problems and the questions to the specific ways of addressing them but I think there’s also certain prerequisites that you have to have like we’ve been bypassing it a little bit but there’s the assumption that you’re using AI tools to make a fabric app that’s yes yes I wanted to touch on this so And like there’s to be frank I think that there’s going to be very few

35:51 think that there’s going to be very few data teams that have the skills and capabilities to be able to build a fabric app without those tools. And it does it does add the potential to say like okay here are these developers who can now participate in the fabric ecosystem. I think that’s great., but if if you’re a BI team and you’re like, “Okay, we’re going to leverage this to deliver these custom experiences for the business, you’re probably going to be leaning on on AI tools like,, is it something that people allow? Is it something that is has is

36:22 allow? Is it something that is has is there any governance in place?” Because of course we have this situation where people are chomping at the bit to get started with agents and stuff like this and they start trying things and then you talked about governance but then there’s also cost monitoring which is extremely dangerous for people right now. this is something I get really really really frustrated about right now is because you have a lot of people who are producing like content and stuff about AI and using AI but they themselves are using subscriptions., and in

36:52 using subscriptions., and in enterprises there’s these like per token costs. And so the workflows that these individuals are proposing who are relying on these subscriptions are completely unrealistic in an enterprise scenario where you have to be alone from a monetary standpoint, right? Yeah. Yeah. We’re just working at Divia and as Jen Hung said was basically like I’m giving people half a million dollars and if they don’t use it, I’m going to fire them. That’s the company I want to work at. Yeah. He’s giving $500, 000 of tokens and wanting them to use it. So, he’s also paying you a lot of money,

37:22 he’s also paying you a lot of money, too, to make sure you you have a lot. He was saying, “If I have an employee who’s getting paid $250, 000 a year, they better be heavily using tokens to do like to to get stuff done and build things, things, right?” But I like your I like your per Kurt, keep going on your point. I think you had a couple more things to say. So, so we talked about like I like the systems thinking and the maturity and I think that’s something I want to come back to because I think that’s also a prerequisite. Like if you say okay if you identify a use case where there’s a

37:52 you identify a use case where there’s a difference again between like a prototype case where it’s just like let’s just explore this and see what’s possible. fine, it’s ephemeral, it’s throwaway. Even if it’s just something like we’re doing some some like we’re doing a team building and we need to create a dashboard for the team building and you just you whip something up, it looks cool, it’s interesting, it catches people’s interest and sparks a conversation, you throw it away, fine. But if it’s like, okay, if we’re going to we’re on our our month-to-ate sales dashboard where we’re reporting against like the budget and forecast, we’re

38:23 like the budget and forecast, we’re going to make that in a fabric app. Like in order to make that decision, I think you need to take a step back and okay, what what are the ingredients we need to bake this cake because we’re going to have to we’re going to have to set up systems that simply,, we didn’t need to set up if we were doing PowerBI,, but are going to be necessary to ensure consistency like design systems or,, being able to,, being able to create certain rules about,, how they should look

38:54 about,, how they should look and deal and the things like this., and that that does take a lot of effort and it’s not something you can offload to AI either., and it it also takes a certain expertise that and maturity on its own that your team might not have., so that’s something you need to think about before you start to consider that broader adoption., is how can we ensure that this is going

39:18 is how can we ensure that this is going to not just be something that gives us some initial value, but that can can scale in a systemic way. I want to I want to jump on these points here. I just real quick, Tommy, I think you’re going to jump in here as well. I want to steer the conversation just slightly a little bit here. We’ve talked a lot about like,, where does it fit and some of the concepts pieces here. I’d really like to just I’m gonna ask a very I want to pull back a little bit in the conversation, but I just want to understand I think prerequisites is a really good call out here. here. just talking about what you need to

39:49 just talking about what you need to have in place to even bring fabric apps into your organization and I think visual thinking we talked about that earlier. earlier. Absolutely agree with that. there needs to be some capability around being able to visually think through your users and what they need and how to best serve them through visuals in that in that way. I want to talk on what harness are you guys using. So there’s this there’s a couple things that we you talked about. One of the assumptions I’m clearly not articulating well is if

40:22 clearly not articulating well is if you’re going to do fabric apps, you’re going to use an agent, an AI, something AI. AI. So I want to actually ask you as you are building things, what is your current chosen model? What are you using right now? And maybe what is your harness that you’re looking at? And maybe K, I’ll start with you and then Tommy, we’ll come back to you. But Kurt, when you’re building fabric apps, what harness and models are you finding the most success with when you’re experimenting or building things with fabric apps? It depends. So like I use a combination

40:52 It depends. So like I use a combination of cloud code and codeex and open code. and open code I only started using more recently because I’m trying to leverage local models more and more cool cool and so I know that a lot of people like you mentioned inj are sharing also what they’re using and what they’re doing so I also think that that’s really valuable to check out what they’re doing but the the from my point of view like it’s it’s it’s not so much like about like what harness you’re using but

41:22 like what harness you’re using but thinking about like what context you’re providing, and I think like it’s it’s good to be able to in general like discuss a plan with like one of the more intelligent models like Astra or Fable., and then to to execute that plan with something that’s more like Opus or Soul., and then to do smaller iterations on something that’s even like less than that like Luna Luna at maximum effort., and that’s that’s typically the approach I take, not just with fabric apps, but with a lot of

41:53 with fabric apps, but with a lot of different implementations. So, and the reason why I mention other people is I know that this tends to vary depending on your preferences and on your workflows. So, Tommy, what do you think? What do you how are you stitching things together?, what what models, what harnesses, what’s your com what’s your sweet spot right now you’re playing with? with? And it’s funny that you say though too with the different approaches where I don’t think there is a tried andrue approach yet. And I don’t know, I’m still debating whether that’s over the project or over the user, right? In

42:25 project or over the user, right? In terms of what the best approach is. I I’ll I’ll focus on the context side for because I agree with you. How I get my context into a project is the most essential ingredient. And to be honest, I know it’s not going to sound I’m not building it, but what I’m going to say, Mike. It’s okay. Notions, baby. First, I can share all my skills, which I do sync those skills. So it knows like if I’m going to build context out or instructions I write cloud instructions in notion and using agents that I built and like

42:56 and using agents that I built and like it knows the rafin skills it knows the fabric skills. So like hey based on the project the project day we just had a meeting this is the scope let’s start building out what those instructions would be for Ray and for this fabric app. That’s what I’ll feed into. And honestly I’ve tested out with codework. I’ve tested out with cloud cloud code and the fact that it can sync so well with reading what’s in notion I can simply say look at the master sequence or this idea that I have called master

43:27 or this idea that I have called master sequence of different iterations of instructions that I have like built you instructions that I have like built data discovery visualization side know data discovery visualization side of it each of these are different cla pages in notion and there’s this ma major master sequence that will tell claude hey focus on this this is our first phase. We’re going to only work on this right now. Once that’s done, it will update notion to say the next phase is the design phase or the,, the system phase. That has been the workflow for me because again, it’s not

43:59 workflow for me because again, it’s not just about the output going into fabric, but that whole and I I’m trying to think of a better word than synergy ecosystem of my context, my instructions, my execution. Now that being said, I I am absolutely a fan of Claude and Anthropic Opus Babel are what I go to. because of co-work and its ability to sync with notion, I haven’t been using as much cloud code CLI just because they have blurred the lines a

44:30 because they have blurred the lines a lot on their capabilities. Okay. Okay. So, but that’s been a worry for me. Yeah. Yeah. I want to kick that back to you Kurt as well. Are you using? So you said cur you said you were using o open code and then I think you said cursor was the hardness other one you no codeex and code my apologies cloud code and and codeex I I do like or I did like cursor I haven’t used it in quite some time but cursor has a very nice vision in terms of like what user interface and user

45:00 of like what user interface and user experience can look like in when you’re managing agents. I really appreciate that. But but I I I have primarily used cloud code and in tabular editor we also use cloud code. but I I am very into the terminal. So I fell in love with the terminal last year. year. And I I just I love to death working in the terminal. and so I use Herder as to be able to manage my agents so I can have multiple agents running at the

45:31 can have multiple agents running at the same time and they can communicate to each other and something like this. like for me that’s my preferred tool. things like that I need to look into that. that. I I appreciate it also that in in the terminal like things are a lot it’s a lot easier to be able to customize and extend things. and it’s it’s at least compared to in the past it was a lot easier at least initially to be able to give it something like a CLI and to be able to use it or to be able to kind able to use it or to be able to like pivot and change things on the

46:02 of like pivot and change things on the fly., and there’s just something so elegant about a user interface in the terminal. Like it’s just stripping away a lot of the excess and I don’t know I just grew like I’ve there’s a few people who I follow who design terminal user interfaces and it’s given me an appreciation for that aesthetic and I like it a lot., and I know it’s not for everybody. Like a lot of the people who I show demos in the terminal get very intimidated, but I would definitely

46:32 very intimidated, but I would definitely encourage people like to to not be because because it it’s just a chat window at the end of the day if you’re working with the agent and a terminal multiplexer is just a way to manage windows and tabs., and, yeah, it’s it’s it’s a wonderful workflow. It’s it’s very very nice. nice. Have you heard of Clink? CL L I C L I N K it’s an autocomplete for Windows terminal and you can actually feed it certain like for example your P by

47:03 certain like for example your P by your CLI tool you could actually in a sense integrate that with clink and when you’re starting to write a command out it will say the suggested command for a certain tool so for example if I was so I will look into her but I wanted to give you clink because I love that if you’re working in Python it’ll say oh you could say pi and or pip and it’ll show you all the available subcomands based on what you’re using. So that’s essential for me. Oh, cool. Yeah, I don’t know that one. I primarily don’t work in Windows

47:34 primarily don’t work in Windows especially since the beginning of this year at the beginning. Here we go. I I I hear a rabbit trail going. Go ahead. What are you working in, Kurt? in, Kurt? Yeah. right now right now working in Linux and omachi or so yeah which I’m I’m I think is is oh man this is a deep rabbit hole though to explain to people I know I know but it’s it’s it’s it’s a Linux distribution where it’s it’s very malleable it’s also very easy and everything looks very nice and

48:06 everything looks very nice and there’s something special about being able to turn on your computer and being able to turn on cloud cloud code and to be able to manifest literally anything you want in your operating system to support what you do and why you do it. Like there it is it is fun but also empowering at a level that if you like being on the computer, if you appreciate,, doing computational work,, Omachi is probably something you’re going to like., and especially if you’re really

48:36 , and especially if you’re really quite comfortable with agents at this point, like it’s, it’s it’s just incredible. Like when I when I first started using Omachi a few months ago, it was comparable to the first time I used Cloud Code where I really felt like I just I discovered an Eldrich wand from a timetraveling wizard that I’m not supposed to have and I suddenly like became Oh, I love it. Great analogy. Great analogy. This dusting off this wand that you’re like, where is this? And then all

49:07 you’re like, where is this? And then all of a sudden you’re like, oh, I can turn things into gold now. It’s but it’s it’s really it’s really crazy though because it’s so intoxicating like when I’m when I’m giving like seminars or workshops or lectures about AI and and coding agents to people who primarily haven’t haven’t started experimenting with them yet. I always by the end really try to tell them to be careful because there there comes a time where it’s just you get you get drunk with the

49:37 it’s just you get you get drunk with the power almost like it’s I I call it the token cocaine where you suddenly think like everything is possible and I have like you get these big crazy ideas and you’re just like you you feel almost inebriated and it is it is a very dangerous state to be in., and it it’s also not particularly healthy either, but, it’s it’s it’s also it is very satisfying. But yeah,

50:08 satisfying. But yeah,, Kurt, I really love this conversation. We’re going to have to touch on more of this as we go along because I I do think there’s a lot going here. And I have Okay. Okay. Yeah. Well, there’s a lot there’s a lot I I wrote a very AI song around tokens that I think are very relevant here because just give me more tokens. I just want to use more tokens., and I do want to unpack at some point, Kurt, some more of your experience around this operating system. One little question I do want to tag in here is if I’m going Linux, if I’m going I think I say Omari.

50:42 Linux, if I’m going I think I say Omari. Yeah, Yeah, if I’m going I’m very excited about it, but I’m hesitant in the fact that I can’t necessarily put PowerBI desktop on there. Is that a true Oh, okay. Yeah. Yeah. But you can you say so you can put PowerBI desktop on that. There’s an emulator for Windows on there. Is that what you’re what I’m hearing you say? Yeah, indeed. So I Or you just coded an entire new desktop. Yeah, he just he agently

51:12 Yeah, he just he agently built PowerBI desktop. The token cocaine can be powerful with some people. Make PowerBI desktop make no mistakes. Must go fast. Low click version 2017. Yeah. Oh man. But but but seriously though, like when you do try these kind though, like when you do try these things, like I I did experiment with of things, like I I did experiment with this back in like December and this was the moment where I decided, okay, I’m gonna stop like hyperfocusing on just the PowerBI and fabric ecosystem when it comes to working with AI. was when I

51:42 comes to working with AI. was when I I started to make like a a dashboarding tool with a drag and drop user interface. And you start to appreciate the care and the maturity that exists in a tool like PowerBI. And, like we talked in the beginning, like it can sound very flippant and to be honest, like sometimes disrespectful when we’re talking about how,, all the limitations and challenges of the tool, but the reality is that the tool is built in a way that’s extremely robust and capable of doing a tremendous amount

52:12 and capable of doing a tremendous amount of things. And when you when you try to build something like that yourself or even a custom visual for yourself, like I think I saw I don’t remember who it was, someone made a post about this on LinkedIn as well. you appreciate how

52:24 LinkedIn as well. you appreciate how hard it is to be able to get it to work right especially for an audience. but yeah anyway so on on Linux so you can have like PowerBI desktop working in this like little Windows VM and it it’s basically indistinguishable from the actual experience. Of course, it’s containerized, but even even the other day, day, I I was using Tabular Editor and I was able to even have the agent on Linux

52:54 was able to even have the agent on Linux be able to communicate with and control Tabular Editor in the Windows virtual machine through an encrypted tunnel. and that was that was quite straightforward. Daniel did facilitate that. but at my at my I was very interested in facilitating that so but it is it is possible so it is possible entirely like since since January I’ve been primarily working on Mac and more and more now I’m doing work on Linux but it is entirely

53:26 work on Linux but it is entirely possible possible incredible so so okay at this point my every is tearing their hair out no no no So wait wait wait wait wait wait. So quick disclaimer is early. Yeah. So I’m not saying like for everyone listening like don’t don’t go to any decision maker in your organization and say we’re doing but no but I encourage you for a personal computing experience at home. Go take an old laptop and just

53:56 home. Go take an old laptop and just install it and experiment with it. And especially with what I said like the fact that you can change and tweak and control literally anything on the operating system is incredible. But and and maybe down the road, like inside of an organizational or enterprise context, this is something that can be tremendously powerful. But it is simply too early. Way too early. Well, I want to touch one little piece of this, too. Love what you’re thinking there. This feels to me like the beginning of agents help you build

54:26 beginning of agents help you build tools. tools. tools wrap your business knowledge and process into a system. And I think this is when I look at AI and where I see it fitting the best for me in my workflows. I don’t use AI to go get asked questions about my data. However, I use AI to help me build complicated things or build things faster that would what I would have done manually myself. And everything I’m building is

54:56 building is agents help me build deterministic tools and harnesses and processes, things that make up the uniqueness of my business. And I think this is this is the the nugget that people need to take away from where does AI fit? Where does AI fit in either PowerBI desktop? Where does AI fit in fabric apps? Where does it fit in how we build visual and compelling data stories for organization? How do we answer questions? At the end of the day, the agents should not be giving your

55:26 agents should not be giving your answers. At the end of the day, I really enjoy and think we get much more consistent. Agents build deterministic tools that same inputs in, same inputs out over and over and over again. It’s a repeatable system and I really like that aspect of where I apply agents. And I think think what we’re seeing here is Omari omachi is an evolution in building more complicated stepby-step processes

55:56 stepby-step processes show up. I want a calculator that does this. Well, the normal calculator doesn’t have these kinds of extra functions. Well, I’m in a specific field of math that I need extra buttons. Okay, go research these functions and these buttons. Put them on my thing. Let me use them inside my calculator. Right? It’s it’s hyperpersonalization describing what you want. The agent then builds it for you and then you use that C, Python, scripting, whatever the thing is over and over again. And that produces the same input, the same output every single time. That’s that’s the key. I think

56:27 That’s that’s the key. I think it is it can’t but I I think there’s a butt there too. But I think well two butts. Two butts. so so the I agree and I definitely like that is definitely the Pareto take. I think for sure. and I think however it’s like fabric apps for example is could be on its own like a tool that you’re using. You could see it from that point of view like it’s a tool that’s deterministically facilitating your use of the data. but number one I think we need to avoid

56:57 but number one I think we need to avoid that we are using a power drill to create better screwdrivers. I think I think we have to try to avoid that. because I’ve seen this quite a lot that people are thinking like, “Oh, I’ve had this tool that I’ve used so much and it’s but I wish it did that.” And then so they like make their own version of it and I and then when we discuss it, it’s like, “Okay, but guilty.” guilty.” We’ve all done it. We’ve all done it. It’s part of the token cocaine. Like it’s it’s like a right of passage. We’ve all done it.

57:27 all done it. so, but then you you you pause and you realize and you’re like, man, what am I doing? Like I had that myself to an extent. Like when I was discussing back in December with Maxim and we were doing our big experiment with PBR CLI and the bottom line is we were trying to address like can you get pixel perfect control of PowerBI visuals programmatically and by the end of the experiment like it was like you can do you can control PowerBI reports in entirely programmatically you cannot get pixel perfect control unless you’re resorting to other things like

57:58 resorting to other things like custom visuals. Yeah., Yeah., okay. okay. But,, but then at a certain point I was thinking like what do I actually want to do, right? Like what like I have these scenarios where I want to be able to portray like a custom bespoke visualization that best fits this domain and this specific question in that domain. This already has worked for over a decade very well in D3. js. Why don’t I just use D3. js? like why am I making a better screwdriver when I

58:28 I making a better screwdriver when I have the power drill as an example. So the second the second butt is I think that there are small exceptions in there for non-deterministic problems like there there can be cases where having an agent can be still helpful and the AI can do its thing on its own if the the problem in the question you’re trying to address is something that you cannot easily deterministically tackle with a script or a tool., and the example

58:58 script or a tool., and the example that I tend to give is like for example, let’s say you have data that is for whatever reason it’s unstructured data and it’s very unstable and you you tend to have issues with the ETL of that particular data load through a notebook or whatever., and it’s early on also in the deployment and so you’re having these issues almost every day. You could put an cloud agent that you deploy that watches that ETL pipeline and that it attempts to fix it on its own and then comes to you with the approval. And so

59:28 comes to you with the approval. And so that’s it’s tackling the problem non-deterministically because it is inherently a non-deterministic problem. problem. Yeah. Yeah. Yeah. Yeah. Oh, I like this. So your example there I think is a very good example of where to potentially apply agents. So I don’t want to cut this conversation off. I’m thoroughly enjoying this. Kurt, I also know that you’ve had an incredibly long day taking care of family and work and all the other things here as well. So, it’s okay. I could keep going. Okay. I don’t want to I don’t want to I don’t want to slow it down too much here. loving this conversation.

59:59 much here. loving this conversation. Really like your points here. Do we want to go a little bit longer and and couple Okay. I don’t want to shut down too much too early. Okay. get in, lads. Hold on. Buckle up, baby. Stop one is fabric apps. Stop three is nobody’s getting out. The doors are locked. We’re going. Get your tokens out. We’re going to go have fun. have fun. This is the fabric wedding. Yeah. So So sometimes sometimes when I’m like

60:29 So sometimes sometimes when I’m like talking to my friends or like to to to even in the tabular editor team, I’m working with Eugene and Ruben who are amazing people by the way. And like where we start talking and I just like start rambling about Omachi and I’m like, man, I’m sorry guys. I realized But well, I blacked out. I’m sorry. Yeah, there’s a there’s a comment in the chat thread. So, I thought this was relevant and very appropo at this point. I’m trying to bring it back up, Mike. So, yeah, So, yeah, I’m not going to go too too far in the weeds. So, I think this is more conceptually here, right? Someone is

60:59 conceptually here, right? Someone is commenting here. He says, “Your scientists or your explorers around this AI thing are so preoccupied on whether they can or they could build something, they’re not ever stopping to consider if they should do something. Right. So it’s so and to your I think this is also the feeling we’re all having to some degree is right it’s exciting to do this. There’s something weird psychologically happening with agents. Before when I wrote code I I would I would build some code. I would

61:31 I would I would build some code. I would get really stuck. I’d be angry for two or three hours and then I would figure it out and I’d have this really big euphoria moment of like I got it. It works. I created something from nothing and it and I built it. Right? It’s there’s this huge excitement when you deliver something. I feel like for me is the agent has shortened that cycle from like hours or days of struggle getting through potential problems to now down to like minutes on things and and so I’m constantly getting these regular like success success success

62:04 regular like success success success which I think breeds to your point Kurt which is as we get these regular successes more rapidly in close succession we’re enticed to go build more ambitious things. things. Yeah. Yeah. Bigger stuff, things that are more out of my normal reach. So, I think personally, and I have like I have a lot of like controversial opinions on this, but I think that that I think that we need to be more ambitious right now. And I think that I

62:35 ambitious right now. And I think that I think that that’s something that we need to start doing. because the possibilities, the truth is like whatever people’s personal feelings are about AI and I still am very critical about AI. Like I’ve had a lot of people in the last like months and stuff talking about how I used to be very critical and now I’m not and stuff like this. Like it’s I still am very critical about it’s very important to do. I think you need like I usually say like the three C’s like you need to be critical, curious and creative and and these are things you really need to value at the moment or at least that’s

63:06 value at the moment or at least that’s what I feel like personally. But it’s it is important I think to be ambitious because the possibilities are higher the bar is higher and it’s raising. That is, I think, the truth. But, but, but so the the implementation cycles, yes, they’re getting shorter, but that doesn’t lower the bar for quality. And that doesn’t lower the bar that we need to meet to ensure that what we’re making is going to help people and is also going to be something that people can and want to use in a way that’s going to

63:36 and want to use in a way that’s going to be meaningful. And so, yeah. Go ahead. Yeah. Although, I’ll I’ll try to I’m being verbose. I’m sorry. And I think like the effort we need to redirect some of our attention and effort. Okay, the implementation cycles are shorter. Let’s focus more of our effort on the planning part and on the design and make sure that we add the friction there so that we’re still able to make sure that we’re going to design the right thing. We’re going to make the right thing. And you can have rapid prototyping inside of that process, but

64:06 prototyping inside of that process, but understand the difference between a prototype and a product. So you can throw the prototype away and then graduate to something that’s more robust when you need to., but that’s some of my feelings on that right now. And I I think it’s ambitious and I I want to just drop this on both of you here because I think it’s ambitious with the right environment because as I was writing down here just hearing you guys go back and forth the word build build. Mike, you and I know how much I have

64:37 Mike, you and I know how much I have gone down my own rabbit holes and Tommy has built has built an entire thing. For what? For what? For what? Yeah. No, but honestly though too in taking this in the context of fabric apps, my call concern is we’re just going to just build these apps when needed because it’s going to be easy to do in PowerBI. But none of we need discipline because you need that for the organization for the consumers. This not that you guys did not say this but the concern would be that we are

65:07 but the concern would be that we are building for the sake of ourselves. Yeah. Yeah. Right. And I think this is the biggest problem problem and this this has always been like a challenge too with like visualization to some extent because it is something that provides this certain level of subjective satisfaction. And I I’ve always felt like it’s one of the reasons why PowerBI had the bottom up success it did is because people get the five minutes to wow experience where they

65:30 minutes to wow experience where they create something like Jack or Janet from finance create their first report and okay like from an expert’s point of view maybe it’s not a very good report but from J Jack and Janet’s point of view and their colleagues it’s amazing go it is amazing it is an amazing report and they should feel proud and that will I think that can also happen with like fabric apps but on a different level. but we do need to still follow like a certain process to be able to mature it but also to gate it

66:00 be able to mature it but also to gate it so that we don’t just have like people sharing their creations proud as they may be with everybody and that we’re sure that it’s it is purpose driven. Sorry Mike I interrupted you. No no not at all. I’m I’m happy. Okay. The amount of t-shirts you just spoke into existence is is like phenomenal. Right., there’s there’s a shirt that should say to me to me and I will send them to your house. So, like,, I feel like there should be a shirt that just says critical, curious, creative. Like that that’s

66:30 curious, creative. Like that that’s that’s a phrase that that needs to exist. That’s my that’s like my whole thing right now is I think like if if you value those things if you value those three things like to your core I was going to put an adjective in there but that if you value those things to your core like deeply while still embedding it in I’m going to do the right things. I’m going to move the needle on things that matter like you can do oh man you can do amazing things right now. Like seriously amazing

67:00 things right now. Like seriously amazing things. So,, and I really want everybody to believe that and and I all these things I’m re I’m literally I was feverishly typing things down as you were describing them. I think you’re hitting the nail on the head. You’re resonating with so much of my experience as I’m interacting with these tools and working with them. You these tools and working with them., the visual is subjective know, the visual is subjective satisfaction, right? It’s it’s how the beholder looks at that information. I can look at a visual and understand how it’s written and how it’s built. others can look at it and not

67:31 built. others can look at it and not understand a clue about what I’m trying to convey in the message. Yeah. Go ahead, Kurt. Have you guys had the experience before where you made I think we all have, but have you had the experience before where you made a visual and you were like you you like slaved on like every aspect of that visual and you really thought like this is amazing, this is perfect. And you put it down in front of someone and they’re just like what? what? I don’t get it. Yeah. Yeah. Can I just get a can you just can you just make it a table? Yeah. Yeah. You’re just like, “This is a little much.” You’re like, “No.” Yeah.

68:02 little much.” You’re like, “No.” Yeah. Yeah. And your your initial reaction is like, like, “Why is it blue? Wait a minute. What?” You you you you feel like personally attacked. And I think that that’s pretty relatable for a lot of people because you put you put part of your identity in that. And I know we’ll talk about this in like a later,, thing, but you you really did put some of your identity and your your subjective taste in that and this kind subjective taste in that and this thing. So, it hurts when people say of thing. So, it hurts when people say like like this isn’t what I need or want or I just don’t like it., but I think it’s important to and and

68:33 but I think it’s important to and and this is with anything that you make, not just a visual, but something you make with AI or in a fabric app or in general. It could also be an engineering artifact like this is amazing code in this notebook and someone looks at it, they’re like, “What?” But,,, it you you have to be able to let that go and to be able to say like, “Okay, like, I I clearly didn’t understand this person’s point of view. how can I better understand it? And you make something that maybe subjectively to you is much inferior, but they’re like, “This is perfect. This is amazing.” amazing.”, , and and that’s that’s really nice.

69:05 and and that’s that’s really nice. So, So, I I want to have a question for both of you or Mike, did you want I was say, Tommy, let’s end on this question. Hopefully, it’s a good one. Yeah. We’ll ask you on this question and then we’ll The doors are locked, boys. The doors are locked. Nobody’s leaving. No way out. I I still have to show you all of this stuff that I made in Omachi. We’re not leaving quote one of my favorite movies. I got to show you my DVD operating system. system. Now you can’t leave

69:35 Now you can’t leave thing. So thing. So what is what is it? You’re you’re I’m not locked in here with you. You’re locked in here with me. Me? Yes. Me? Yes. [gasps] So I I want give me the proper environment or the most successful environment both from a developer and a consumer point of view for building fabric apps in this ability to do this rapid innovation and rapid development. What does a successful environment look like in terms of what potential guard rails? Are we building everything they

70:06 rails? Are we building everything they ask? What are what’s part of the environment that you’re setting up both the developers and the organization up for success? I’ll let you guys. Yeah. So, I my answer is maybe going to be disappointing, but I think I have ideas, but I think the simple truth is I I don’t know because it’s still early., , and I I have ideas on what I think makes sense. sense., but I can’t back that up with any specific evidence or anything. So, I don’t know. I don’t think anyone can right now. Yeah. So, indeed, I’m going to lean on Tommy. I like when

70:37 I’m going to lean on Tommy. I like when you ask these questions because it does make me think. I think the idea here is when something new like this appears, I look to the community to start teasing the boundaries and the edges and the patterns of what makes sense here. So I I look to the the indiv. So indiv. So the answer to this, Katami, what does a successful look environment look like? I’m going to challenge you the audience. This is your job. This is this is blog about about way to pone it off. Seriously though, I to to Kurt’s point, I have

71:08 mean I to to Kurt’s point, I have opinions around things that I think will work well, right? I think there’s there’s use cases. I think there’s things that are already materialized in my head, right? You don’t get to really get a a feel for this. So, NJ Park, this challenge goes to you. Tease out those things. This Kurt, this goes directly to you. Challenge you. Write about them. What’s working? What’s not working? I think the community needs to to take this experience and push on the edges of it., this is a challenge to Marco Russo. This is a challenge to Oh,

71:38 Russo. This is a challenge to Oh, I forgot the other anyone else who’s in the community. This goes out to you guys guys and gals and gals because you’re describing, you’re figuring out the boundaries of what works. And so, as us as leaders in this, and again, I’m going to really lean on the curious and creative side of this and maybe even some of the critical side, right? I want to hear you, the community, being critical about the boundaries of where this fits. I built this app, it worked really well, but we ran into these challenges. It was difficult to maintain or it wasn’t

72:08 difficult to maintain or it wasn’t difficult to maintain. I don’t know yet. Me personally, my personal experience right now, I like building apps. I really love hooking them up to SQL databases. So any for me fabric as fabric apps are a fabric as a backend and I love bringing net most net new net data. I love bringing the ease of bringing data directly into fabric because once it’s inside that fabric SQL database or once it’s inside a semantic

72:38 database or once it’s inside a semantic model I can really push it around when I need to inside fabric pretty easily. So for me that’s a win. I’m liking that part of it. Does this mean you build an app for your ops team in a in a warehouse that’s shipping product? Maybe. But I think you need to try lots of things and figure out what works and the community at large is going to help us build that. So, I’m going to answer your question with a,, I don’t know either, but I’m finding success with it. I want to continue learning and pushing into it, but I want the community to continue investing and blogging and teaching us

73:09 investing and blogging and teaching us about what the boundaries of this looks like. Tommy, what’s your answer? What do you say? I I think there’s never been a more critical point to have that environment to have a research lab for this, right? That be able to do I think everything Kurt that you were talking about when it comes to let’s honestly let’s just build rapidly. Let’s those ideas that we have but that needs to be not so much siloed off but in a research garage, right? to make sure that those things are not just obviously out just getting out into the wild

73:39 getting out into the wild and be critical of it but just don’t be afraid because to your point Kurt I can have this idea I can have it come to life in a fabric app and we can start iterating on what designed how those things worked have some creativity this other part of that department or environment is then the intentional side of it okay let’s start going through a process of a request that came in. How would we actually tackle that with the fabric app? What requests do we need? Because again, I cannot highlight more

74:12 Because again, I cannot highlight more to what you said. We’re so new to this. Everyone’s so new to this that how in the world Yeah. We don’t know. There’s that’s something that I also try to tell people and I try to be very empathetic that a lot of people haven’t gone through their coding agent journey. They haven’t huffed their token cocaine yet. And and so I I I I I try to tell people like look I don’t consider myself an expert in this. This is this is too new to say like that that you’re an expert in this like and I have a lot of ideas. I have a lot of thoughts

74:42 a lot of ideas. I have a lot of thoughts but it’s also why I don’t really share a lot of content about what I’m doing specifically and personally is because I I I don’t want to say like this is a recommended or best way or whatever. But but yeah, I think I think we do need to have that and I think your idea of the research lab is very correct. Like we embody this in tabular editor. So Yirun who joined tabular editor earlier this year. So he piloted the idea of having we have an innovation team who are dedicated to rapid prototyping and experimentation to find

75:13 prototyping and experimentation to find value. And so that’s something that served us extremely well and that’s been extremely valuable. And it’s a nice way to compartmentalize the innovation pipeline while still making sure that it can happen in a in a healthy way without shackles, let’s say. and so that was that was something that really unlocked a lot of potential in a lot of different things. And you can create experiments that are really outlandish, but you just have to know when to kill them. Yeah., and so I think I think that’s

75:45 Yeah., and so I think I think that’s actually a good takeaway for if if for for BI teams is to consider whether one or two people could be responsible for innovation or R&D in your team and maybe they’re responsible for trying stuff and then every week surfacing it to the rest of your team or if you have a center of excellence that you’re already operating or even if you’re just a business team using PowerBI that you have one person or a few more people who are responsible for that innovation

76:15 are responsible for that innovation but more importantly disseminating it or sharing it with everybody else great in an open way that that people can have an open discussion about is this valuable actually. Yeah. Oh, I absolutely I think that’s a perfect ending note right there. So Kurt love that love that love that off. Yeah. Closed the door is loud. I’m going to have to bar the door open here a little bit because we’re we’re have to run out of time here today because we do have to get get we got we got meetings we got to get to. Kurt, thank you so much. We really

76:46 Kurt, thank you so much. We really appreciate your your thinking, your capabilities, your your mind on this is so rich. I love picking your brain on this one. So, this is going to be a fun week. week. We’re going to spend two weeks with Kurt discussing, unpacking. I think we’re going to get into some really interesting topics here. there’s a lot of things around like artistic and how creative things are and we’ll even probably touch on a little bit of like where music and AI intersect a little bit here too. I think we have opinions there that would be worthwhile teasing out. perform. perform. I will not be bringing that one up. I’ll let you.

77:17 let you. So that being said, Kurt, thank you again a million times. I every conversation I have with you always leaves me thinking and and being thoughtful around what I do and how I do it. So, I really appreciate you bringing that knowledge and your expertise to the to the audience to the to the community here. That’s why we do what we do here and absolutely love your engagement with this one. So, thank you very much, Tommy. With 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 helps

77:47 subscribe and leave a rating. It helps us out a ton. You have a question, idea, or topic that you want us to talk about in a future episode. or if you submit them in this week, which I’ll check the mailbag, head over to powerbi and you want to ask Kurt, head over to powerbi. tipsodcast. I’ll just put you on the spot., leave your name and a great question. And finally, join us live every Tuesday and Thursday, a. m. Central on all powerb. tips social media channels. Thank you all so much and we’ll see you next time. Take care. Explicit measures. Pump it up.

78:17 Explicit measures. Pump it up. Be it high. Tommy and Mike lighting up the sky. Dance to the day. The laughs in the mix. Fabric and A. I get your fix. Explicit measures. Drop the beat now. Podcast the crowd. Explicit measures. Explicit.

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.