PBI Desktop Bridge Tips – Ep.556

Two weeks after the August 6 episode, Power BI Desktop Bridge has moved from an announcement into the way Tommy Puglia and Mike Carlo are actually building reports. This conversation is the part they learned in daily use: which piece is the API, which piece is the CLI, and why the same prompt keeps designing a different page.

News & Announcements

• Enhanced Data Agent Visualizations with Fabric Visuals (Preview) — Charts a Fabric data agent renders now use Fabric visuals, the same system as Fabric apps, and you still trigger them with the verb “show.” Tommy Puglia’s objection is the second standard library: Fabric apps and data agents do not share Power BI visuals, Deneb is already a separate look, and he would pick one library for the reports he ships. Mike Carlo has not tried the preview; he wants a checkable SQL or DAX table before any chart, reads the supported types as reused Vega-Lite from the Fabric apps skill, and would rather see Microsoft’s Flint grammar in Power BI Desktop — one JSON description that can emit Vega-Lite, Plotly, or a third output — though he is not sure this preview is using Flint.

• Connectivity Patterns in Microsoft Fabric: A Guide for Data Integration Workloads — The post is a flowchart for how a workload reaches its source: a VNet data gateway, an on-premises data gateway, or a direct cloud connection, based on where the data lives and whether Microsoft Fabric must cross a private network. It covers Gen2 pipelines, the copy job, mirroring, semantic models, and paginated reports. Mike Carlo and Tommy Puglia still treat a gateway as a hard stop when data sits behind a firewall, such as AWS into a lakehouse, and the pattern in their consulting over the last few years is to move that data into Fabric, where Fabric as a backend (FAB) needs no gateway because access is already there, and the VNet data gateway is still the option they see people struggle to set up.

• Design the Report From the Meeting You Already Had — Tommy Puglia’s write-up of the workflow behind this episode. Handing an agent a finished semantic model and a measure list produced a different page on every run of the same request. The run that shipped used the stakeholder meeting transcript, screenshots of a page to echo, and the metrics that conversation named, frozen into a brief and run through Microsoft’s three report skills: planning locks the spec, design chooses the pages and charts, and authoring builds the file and verifies what rendered.

• What is the Power BI Desktop Bridge? — Microsoft’s overview of the local server inside the Power BI Desktop process: a named pipe on the machine, JSON-RPC, no remote access, and a separate pipe for each open Desktop window. The easiest start is the Desktop Bridge CLI, which wraps the bridge so you can use it without a custom integration. The methods the page and the hosts both walk through are the manifest, application state (open file and unsaved changes), a PNG of one report page, and a reload from disk.

Main Discussion

Topic: How to use Power BI Desktop Bridge without getting a different report every time

The modeling MCP server answered how an agent changes the semantic model. Power BI Desktop Bridge is the follow-up: how that agent builds the report. Tommy Puglia is writing the instructions from stakeholder meetings, Mike Carlo is using the bridge as his primary way to build, and the tips below are the API-versus-CLI split, the unsaved-state trap, and the spec that has to exist before either of them lets an agent touch a page.

• The bridge is the API inside Desktop. The CLI is the package the agent calls. The bridge is a local server running in the Power BI Desktop process. Mike Carlo’s read of the product UI is that you enable external tool access through local APIs, and the phrase “Desktop Bridge” does not appear on that toggle. The CLI is an npm package that tells an agent how to call the methods. It is the straightforward start, and it is what the documentation points to, so Mike’s bet is that Claude will install the CLI when you ask it to set the bridge up. The protocol is methods with inputs and outputs, so another client could speak to the same server, including a Python one on a machine without Node. Tommy’s point, once they untangled the words, is that almost nobody should be typing those commands by hand. The CLI exists so the agent can call them.

• Four methods, and save before you reload. The surface they walk through is the manifest, application state, a page screenshot, and file reload. The docs place this under agent skills for report authoring, which is how the hosts pair the tools: the modeling MCP server for the semantic model, Power BI Desktop Bridge for the report. Desktop keeps your clicks in memory. The PBIR files on disk stay as they were until you save, so a screenshot can show visuals the files do not contain yet. Each open Desktop window runs its own bridge, and state is how you learn which file is open and whether it is saved. When the agent edits PBIR and reloads while you still have unsaved work, that work is gone. Tommy Puglia had Claude write a batch file that stops and tells him to save before it reloads and takes screenshots.

• A measure list is an inventory of what the model can show. Tommy’s experiment is to send the same prompt, against the same data, twice. Even with skills, he gets a different design each time: a full page, often a nice one, and a different page on the next run. Forty measures are the inventory. The requirement is which three the meeting cared about, and why. His lead-in is the stakeholder transcript, refined with a grill-me skill and often staged in Notion, then Microsoft’s three report skills. Planning gathers requirements and locks a spec before anything is built. Design handles page archetypes, chart choice, and palette. Authoring does the mechanics, then validates and verifies what rendered. That brief is the prompt he sends.

• Mike starts the spec outside Desktop. For a family of reports that should share a header, type sizes, and branding, he builds the style guide first. An agent produces a low-data HTML mock of the page background, size, headers, and footers, and he turns that mock into a reusable standard for the agent that talks to the bridge. Fabric data apps are his example of the skip: “build me a report” comes back with no spec and no design pass. He has seen the same scattered output Tommy describes, and he treats it as a process gap. Engineers using Claude Code and GitHub Copilot already get repeatable results when the brief is doing the work. He has been building theme generators since 2017, so the aesthetic half of this argument is the half he refuses to leave to a fresh prompt.

• End the session when the result is right, and keep the lesson as a skill. Mike’s sessions have one job: a page, a filter pane, a fly-out panel. When the result matches what he wanted, he asks the agent what it learned, shapes that answer into a skill, and opens a fresh session so the context window starts clean. The skill is how the lesson moves to the next chat. He used to keep a reflect skill. These days the line is just “reflect on what you built,” and he edits the answer into the skill he actually wants. Tommy has seen Claude’s cowork flow offer to save a skill at the end of a task. Large opening prompts are what you write for net-new work before those skills exist. A later tweak can be a short correction. Both hosts say to start from skills that already exist: Microsoft’s skills for Fabric, and Kurt Buhler’s skills for semantic modeling and report design. Skip them and the page is worse.

• The hours you get back are the clicking. One spoken instruction can put every visual on its own hidden page, so the agent styles a single visual with the rest of the report out of the way. Doing that by hand was an afternoon. With the agent it is a minute or two. Bookmarks, which Mike has told people to leave out of reports for years, become readable: the agent can explain what a bookmark touches and build a new one, including the fly-out panels and filter panes showing up in the community. He is torn about that pattern, because bookmarks were a maintenance nightmare, and he is also watching agents make them manageable. Reordering pages and hiding the rest is one sentence instead of days of dragging tabs. That is why the bridge is how he wants to build most of the time. Tommy’s pushback is about the first instruction of a new report: that prompt still comes from a prepared brief, often a Notion page that already holds the sequence, so the agent spends its tokens executing requirements that already exist.

• Seeing the report requires Desktop open. The agent changes the PBIR files on disk. The bridge is how the open Desktop window reloads those files and screenshots the page. Tommy’s side of their argument is that editing PBIR is file work, and a Git-connected project in the service can edit those files too. Screenshot and reload are the Desktop half, which is why this feature still needs the desktop app running. Sarah Rodriguez’s post, which Mike dropped in chat, adds a button in Desktop that opens the project in VS Code. The reason to open those PBIR files there is the agent. Mike reads the instant reload in that post as the same bridge.

Looking Forward

Before the next report, enable external tool access through local APIs, ask the agent to install the Desktop Bridge CLI, and load Microsoft’s report planning, design, and authoring skills. Take the transcript from the last stakeholder meeting, write down the decision the page has to answer, and save the file in Desktop before you let the agent reload anything.

Episode Transcript

0:07 Tommy lighting up the sky. Dance to the day to laugh in the mix. Fabric and A. I get your feels. Explicit measures. Drop the beat now. Feel the crowd. Explicit Drop it loud. Hello Tommy and good morning. Welcome back to the explicit measures podcast with Tommy and Mike. We’re all here ready to go ready to talk about some new

0:38 information today around PowerBI. How you doing? Oh yeah. Specifically more on PowerBI desktop bridge. You and I I think have been having such a great time with it that yeah I think that more conversation needs to be had. It now deserves a full new episode. And with the evolution of AIS and agents and things that are happening now, it it’s quite spectacular what we can do now with AIS and the agents, which if you build better processes or take your existing processes, it really changes what you

1:09 work on things. Actually, I was thinking about this the other day, Tommy. some people are very hesitant around the AI and they’re like, “Oh, it’s going to take our jobs. It’s going to do all these things. It feels like AI has now become a co-orker or or a another employee in your stack of things that you use every day. And I feel like it’s it’s turned everyone into a manager to some degree. Every

1:39 everyone is now a manager of things. I I would completely agree with that. just what I I actually had a blog post about meet my assistants and I went through my operations one I went through my consulting consultants and I have a morning brief one now and it’s so much more than just giving the information in a chatbot right yeah it’s helping things out and now to your point it’s also the executioner as well yes it gets things done it’s getting things completed I’m doing a little experiment every time I I find a new

2:10 agent or a new experience agentically I I it’s been slowly increasing the complexity of what I build with them. So, a lot of what I do is software development with our company and our team. So, I’m I’m slowly like turning up the dial and every new side project that I’m doing or project I’m doing for my business, I’m trying to think of, okay, what what do I what is causing me friction in my business? What would I need to build that would solve that? And now I’m starting to build a

2:41 full process around this one. So Tommy, , before we get to the main topic here, actually I should probably throw the main topic because I feel like we’re starting to get into the real conversation here. So let me throw the main topic here and then we’ll come back to this topic., the main topic for today is PowerBI desktop bridge. This is talking about the system that allows you to have your AIS connect directly to PowerBI desktop. And there is a number of API methods that are be able to be used. It’s called the bridge. mmanifest.

3:11 it’s it’s allowing you to like communicate directly to your desktop application. So that’s what we’re going to discuss. We’re maybe talk a little bit about the mechanics of it and then also how we’re using it with our AI agents so you can get some best tips and tricks on how Tommy and I have been using this desktop bridge to build reports which is arguably I think the missing point here. we got the PowerBI modeling MCP server first that came out initially. Great. Loved it. thought it was really good. The immediate question

3:43 from people was how come I can’t build reports with an agent? That’s what I want to do. I’m like, wholly agree. Yes, agreed there. And so, this is one of the features that are closing that gap. And I feel like Microsoft is finally taking note at least with some of their tooling on what on where we actually want to use AIS. We’ve been saying this for a while on the podcast, Tommy. We don’t want AIS to just give me the answer of the sales numbers. I want the AIS to help us build the report, the dashboard, the thing

4:15 that gives us the answer. That’s what we want. Which is a great segue into the first article, which completely flies in the face of what you just said. So, yeah, go ahead. Mike, our first news article here is literally getting your chatbot to answer questions for you., just with enhanced visuals. Now, to be clear, this is coming directly from Microsoft. This is the Microsoft blog. And again, they keep hounding this. I want AIS to answer my questions. I don’t want

4:46 you to give me the simple answer. I want you to give me the full report that does answer the question. That’s different. The statements on the explicit measured podcast may not necessarily assimilate with Microsoft’s actual official stance. Sorry, battery is not included. Yeah. these are not this is not real financial advice. Yeah. You’re not necessarily in line with Microsoft policy. We are not in line in Microsoft on this one. So So Virginia Microsoft employee,

5:16 thank you for writing the article. You’re wrong. Jeez. All right. Well, let’s dive in why she’s wrong. No. I like this. So basically big takeaway here is there’s enhanced data agent visualizations with fabric visuals. So previously if you ask a data agent fabric a question like what are my sales over the last 12 months it provided visuals but it wasn’t really in any standard that is previously whether it’s in PowerBI standard visuals

5:47 or in fabric app visualizations. Now any chart that’s rendered by the data agent uses fabric visuals. It’s same visualization system across fabric apps. So again this replaces the older standalone version. It helps also update ensures the data and responses match the look feel and behavior of other fabric experiences. And again so really consistent experience however this is where I have a problem. Mike the visuals are triggered the normal way. So

6:19 show a verb just use a verb show and it’s going to prompt for a visual. But Mike, it’s not a similar experience across all experiences because fabric apps don’t share the same visual library as PowerBI visuals. So you right now we are right now and this is where I wanted to focus more. We are right now living in a world where we have two standard visual libraries in Microsoft’s data platform ecosystem. Fabric apps and data agents and PowerBI.

6:51 And this is something I I’m not a fan of. And you’ve all also talked a lot about DAB and that’s a whole other standard visual that is third-party but pretty awesome. So Mike, this is frustrating me because what should I use if I’m going to I this is what p I I take pause to really deploying fabric apps because it’s going to be a completely different look and feel of all my reports. So, a hot take

7:24 would be use one or the other. Mike, what’s your thoughts here? What’s your take on this? [snorts] Yeah, I I don’t I’m trying to read it a little bit between the lines on here. I haven’t played with this, Tommy, nor have I been. So, be disclaimer approved. Right. So, I I haven’t gone in deep on this exact feature., I understand why consistency and high visual quality matters. I agree with this. All all the principles here. I agree with if I’m going to talk to an AI agent, if I’m going to have it build visuals for me inside a chat experience, right? The

7:55 first thing I want the agent to do is write SQL or or write DAX for me. The first that’s exactly what I want to do first. Give me a table of data that I can come out and verify that it looks right. So that’s that’s the first place I want to get to because even when I’m in PowerBI desktop, I’m likely doing a mix of building tables first and then from those tables then building the visuals and things on top of them after I understand the data. Right? So when we’re talking like data exploration trying to get answers about things we need a lot more of a table experience. Okay. So that’s

8:25 number one. when they when they talk about this supported chart types and they’re talking about how this links closely with fabric apps I don’t understand and they didn’t directly call it out fabric apps does not use standard v a visual system to be clear when you use fabric apps there’s this thing called skills for fabric and when you go into fabric apps and you say build me a a visual report

8:55 it builds vega Vega light visuals by default and from those Vegaite visuals there’s a whole skills library that describes what visuals it can create. So this feels like it’s borrowing from the work they’ve already done in fabric apps which is Vegaite and a skill. They didn’t call that out in this article anywhere. And I look at this going why didn’t you call this out? Because if you’ve already built the technology somewhere else and you’re reusing it here, at least tell us you’re reusing

9:25 it. It they dance around the topic. So I don’t I don’t like this topic for that reason because I think what you’re doing is you’re borrowing Vegaite which is a good standard which does do a lot of things which you can use db with in making those visuals inside PowerBI. So I’m looking at this going there’s nothing rocket there’s no rocket science here. This is not a revolutionary move from Microsoft. This is just reusing a very handy skill and an existing spec that’s been around for years.

9:56 Pause here for just another moment. There is another project that people are not aware of yet. There’s another project called Flint Visuals. Flint Visuals is a it’s a derivative of Vegaite. It’s a derivative of what’s happening there. and and that visual language is a way for you to again describe a grammar language about what the visual should do in JSON form and it produces or outputs

10:26 a a Vega light visual. It outputs another plotly visual and like one other visual type. So one one visual language actually outputs three different kinds of visuals with the same language. You could switch the the syntax of it. So, the reason I think this is important is because I think maybe this is using Flint visuals. I’m not sure. The Flint visual spec is from Microsoft. It’s being built. and I I think it’s actually a really

10:58 I think the Flint visual language is actually what you want to use and that was what should be added to PowerBI desktop and that way you can get consistency across both platforms. I am very pragmatic when it comes to things like this, Mike, because as as much as I like the the aesthetics of fabric app visuals and the the if that’s the actual backend for what we’re we’re talking about now is Flint, I like the aesthetics better. However, Mike, and again, let me preface this because we’re

11:28 going to talk about PowerBI Desktop Bridge. From a human point of view, if it’s going to take me 3x the amount of ti time to build the same visuals that I can do in PowerBI, right? Again, not saying agents are involved in the design of this. Well, if you incorporate Flint

11:46 into PowerBI, but it’s going to take me 3x amount of time to then build the same visuals, the same bar chart. It just looks better. That’s not worth it to me because again a bar chart can still convey the information whether it looks super pretty or it looks still pretty good in PowerBI desktop and I think that’s my hangup here because I hear what you’re saying but to me and I I am giving a little leeway. Anything anything I build with agents takes less time than doing it manually. That’s what I understand. I understand

12:17 that’s why I’m saying you got to take that out just for a second. But I think I think the derivation of your question isn’t even real. No, because not everyone is doing Okay, Mike. Until we’re at the point right now it’s not real to assume that everyone’s building agentic visualizations right now. No, but this is an agent experience. This is talking to chat with an agent about making visuals. Like that’s literally what they’re saying in the article. So like but you can deny you can design yourself. You don’t need an agent to do that.

12:47 It’s designed for agents. Okay. Iintlyn Flint visuals is a language that’s made specifically for agents to understand how to build visuals. Yeah. Yeah. So, so like it’s not just another db or no no it’s an agent [groaning] first design language for visuals. So like so so my my my gut feel here is in all these situations like every time I start adding agents to my workflows things that would take a long time are not taking as much. Actually I commented

13:18 on Okay. Sure. 100%. Yeah. So, something someone I commented recently on I’ll try to see if I can find the post here. The the community is starting to wake up about Desktop Bridge. And the reason I know this is because I’m starting to see real people commenting and making specific comments around elements that are saying, “Wow, this is so much better. This is so faster. This is a a

13:51 way better thinking.” And I literally commented on this post here. Let’s see if I can find it here. going through my my history. I post so many things on social media nowadays. It’s just it’s it’s hard to keep up with everything. ah shoot, I can’t find I I’ll scroll till I find it. Yeah, but no, I 100% because Kurt K current viewer Yeah, Kimberly Kimberly was the person on LinkedIn who who she said, “Stop the presses. I fundamentally changed how I

14:23 build reports. This morning I asked chat GPT, can I actually create a PowerBI report with a desktop file? And she says in the in the pose, this is where it’s get this is where it gets really wild. I didn’t write a lick of DAX. I did not suggest a visual. I did not tell it where to click. It just built the report. It just worked. And I’m thinking to myself, it’s not going to take like when we had AI experiences before, Tommy, like chat with your data or ask a question. It was always very underwhelming. We’d ask a question to like we’ll give it a

14:53 try. We’ll see what it does. And then we’re like in the first three minutes we’re like that didn’t really do what I wanted. That that wasn’t very helpful. You’re like I Yes. Yeah. And so you’re like okay like minimal value added. So it moved the needle maybe a little bit above but it wasn’t enough to be like this is really useful. Nothing you’re going to package and ship. Yes. these agents experiences huge bounds in like people are like wow

15:24 stop the presses hello pay attention this is so much better than what it was so the fact that people are going to be like I Tommy we said this a while ago I said the data engineering and the BI space is going to get it’s it socks rocked it it it’s going to be like there is a revolution coming and something is happening here that is going to really fundamentally change how we build things and It’s all going to be done with AI and agents. It’s going it’s going to fundamentally change how we do things. And I I want to also make sure that we

15:55 talk about the positive here, too. I know you’re mentioning that, but we always mention on the podcast features that come out like we don’t hear anyone talking about it. Therefore, like there’s not a lot of buzz. That’s I think you and I mentioned that a lot. We’re like there’s not really a lot of buzz here. So, are we going to spend that time if that’s not being used? Well, to your point, Mike, PowerBI Desktop Bridge does get a lot of buzz. Kurt Beller has developed skills for semantic modeling but also for report design as well. So people are taking the time not just to say this is really

16:25 cool. Yeah. But to build supplementary things and again you want to know where the wind’s blowing or which read the tea leaves. This is a great way to look at the community. And Mike just one thing how great is it too how that we early in the game were like I think this AI thing is going to be good with B with PowerBI and fabric. I we’ve been talking about I think I mean half our podcast now is about the agentic experience but you and I were a bit early in that because I think we saw the value but yeah I think pretty cool here but before

16:56 we actually get to bridge actually you want to just dive right into it. Well I think we have do we have any other news articles? We got one I have some and I have some questions on here. we have one about Yeah. Let’s just talk about let’s do the last let’s do the last news article and then we’ll jump in the main bits. So this is a connect connectivity patterns in Microsoft fabric a guide for data integration workloads and the title is misleading because it’s actually really providing a framework or decision framework for what

17:26 gateway to use or really are you going to use V-Net are you going to use onrem or you going to use direct cloud connections and it’s a Microsoft it’s actually on the fabric Microsoft blog talking about which choice matters because again the workloads gen two pipelines copy job mirroring specific models pageionated and these are obviously a secure path to reach the data sources. So they tell you and I’m feeling this is pretty fundamental. So I’m not sure where this article is coming from. What question you start

17:56 with? Where does the data source lives and what does a fabric need to cross a private network boundary to reach it? So again there’s the V-Net gateway where it’s reachable from the Azural virtual network provides private connectivity. There’s the on premise gateway where it’s an on- premise or private network and it’s not accessible from Azure or is a publicly accessible and cloudbased connectivity point and Mike as I look at this more and more how important let me just ask you here because again

18:26 nothing in here is groundbreaking how much are gateways how important are they still because if I were to look at my feature pyramid of all the things in in fabric we’re gateways seven years ago or even five years ago was,, maybe in my level three on my feature pyramid. Yeah. I feel like it is beginning to slip. Like it’s like a allstar baseball player who’s really great and it’s just getting in the later years of his life. Like you can still hit a little but it’s hitting lower in the lineup., it’s not

18:57 going to make the all-star team anymore. What do you think about where gateways actually land in terms of the importance? Hm, that’s an interesting question. I’m going to probably maybe lean on gateways aren’t important until they’re important. Sure. Yeah. Right. So, if you have data sources that are in the cloud or already in Azure, gateways probably aren’t that big a deal. But the s the first time you want to get data from AWS behind a firewall into into your reports

19:28 or into the lakehouse or something along those lines, the first moment you do that, you’re like, “Oh, shoot. I need a gateway.” And like it’s it’s a hard stop if you don’t have one. Yeah. So I I also think it’s a very industry specific thing. There’s certain industries that are going to require much tighter controls around gateways or not having gateways in there. you know most of them will most organizations are acknowledge that hey look we’re going if we have data that’s on prem we get there’s a gateway it has

20:00 to be established for us to pass data into the cloud. So I I I’ll look at this Tommy going yeah it’s probably not high on the feature list of things that I use now. I think more of my effort these days are focusing on let’s actually migrate away from the onrem from the v-netss from like let’s let’s move the data directly to where it needs to live. And one thing I’ll I’ll maybe point out here as we talk about our main topic today,

20:31 fabric apps and fabric as a backend. Fab as we’ve been coining it. F aab fab. so the fab framework is fabric as a backend. Zero gateways needed. Data is already there. So, so one of the things is just rebuild or build your systems in a place where the data already exists and has secure access against itself. Then you don’t need any gateways. It’s all it’s easier to get access to it. It’s one less thing you got to manage.

21:01 It’s less hardware. Everything becomes easier when you start really thinking well we’re going to directly use fabric backend as a surface. Right. And I think to your point if you have on premise data yeah gateway is required right but it’s no longer such a I remember when I was writing statement works in the past they were pretty it was a pretty integrated the area about how we’re going to make the gateway work and I’m just not doing that as much because again to your point if we are using onrem there’s always a conversation and a lot of times by the

21:32 clients on how do we actually get off of this too moving over so it’s still important but it doesn’t hold the same pedestal tool that it did. No, I I think again in my consulting years here, the last maybe 3 to 5 years of doing this, more of the conversation is let’s move the data away from places where I need to use gateways. Let’s just put the data in a place that works without the need for gateways cuz it’s just again it’s just one more thing to manage. It’s one more thing to break. It’s one more computer. It’s it’s

22:02 more everything is just more cost and more headache by adding gateways in. Again, if you need them, you have to have them. There’s no questions about it. I’m not saying don’t use them. But I’m just saying the trend I feel like is people are physically trying to move away from having the gateways just be part of their ecosystem anymore. So, I like this. The connectivity flowchart is good. It helps you understand where these decision points are being made because there is options. There’s many different options here. So, the one that

22:32 I think confuses the most people is the V-Net gateway. I think that’s the one that’s a little bit harder to set up. the V-Net data gateway is the interesting one. Yeah. All right. Good. Well, let’s get into the let’s get into the meat and potatoes for today. Mike, we already we’re talking about PowerBI desktop bridge and we did do another episode or previous episode about desktop bridge. I believe few weeks ago, two weeks ago, was not too long. on August 6 actually. So, yeah, I think it was more about the

23:02 announcement of it, what is it, and how to do it, use it. And like now that it’s been some time, it’s been probably two weeks. This has become and I’m just going to start dive right in with both feet. This has become such an integral part of my workflow both from how I write my instructions to actually building that I find this to be speaking of where it’s

23:29 on my feature list. First, I get to finally put something in your face because this requires PowerBI desktop and something on your computer to do. You cannot do this on the web. And well, no, yes and no. No, no, no. Yeah, yes and yes. What do you mean yes? No, yes and no. It’s it’s using the same mechanics. You’re just editing timal or you’re editing the PBIR. PowerBI Desktop Bridge is a Windows terminal tool that edits the PBIR. It’s not editing

24:01 the actual desktop files. I understand, Mike. Which means which means it that means if you’re editing the PBIR, it doesn’t mean this. So the desktop bridge is a hook between your agent and your desktop application. Yes. But the edits are still being made at the PBIR level. That doesn’t mean doing this whole system is only desktop ccentric and needing to have desktop around. This just means this is just

24:31 using desktop to visualize those files. You can do the exact same thing through the service with a GitHub connected file. You cannot visualize the report. It’s taking screenshots of existing report that you need to have open if you want to use this. No. Okay. It’s there’s a different there’s a different system, but this is this is just a way of working with desktop. You’re not using desktop bridge with the service to be clear. Yes. Obviously, this is called PowerBI

25:01 desktop bridge. But what I’m saying though is words have meaning using agents to build reports. This is not why desktop needs to still exist. I’m still saying kill desktop. Okay. No, I know you’re still saying that. This tool, not agents or all the other No, it’s a feature of desktop. Sure. It’s not a tool. It’s a feature of PowerBI desktop. you. I am not going to curse again on the podcast. I’m not going to let you get me like

25:32 that again because I feel like you were just beginning to troll me. Anywh who, PowerBI desktop bridge is a Windows terminal tool right now as of today that allows you to screenshot inventory and help build out your report pages and from all the visuals to building pages to editing the what is actually displayed uses a set of commands that’s terminal based. But again, hey, if you got a ter if you got an agent, it

26:02 can also run the terminal tool. So everything in the instructions for PowerBI desktop bridge is very developer based like hey you need a console or Windows terminal here are all the commands. Let’s let’s be upfront right off the bat. You don’t have to know them to make this work right because you can ask an a I literally on my surface asked an agent to install PowerBI desktop bridge and just set everything up and so it can run everything. So what’s neat about this is you can ask install desktop. So again, I’m going to

26:33 pick on your words here a little bit, Tommy. Install desktop bridge. You need it installed. It’s an MPM. Mike, it’s a package. It’s it’s it’s a I’m getting confused then. No, you go you go into PowerBI desktop. No, I think I I think we’re mixing terms. I think we’re missing terms. It’s it is a package for your agent. It’s actually a CLI. See, it’s a command line interface first, but it allow

27:05 obviously an agent can talk to it, but you need something installed on your computer. I’ll bet you a stake on this. No, the desktop. No, the desktop bridge. Hold on. I think we’re mixing terms here. Okay. I think we’re mixing terms here. We’re conf we’re conflating what is PowerBI desktop bridge and what is tools that interact with this with the desktop bridge, right? So let me let me just slow down here a second. The desktop bridge is a API surface

27:36 layer that lives inside PowerBI desktop. Yeah. And you have to go into PowerBI desktop to go turn on the thing called preview feature and you say enable external tool access through local APIs. It doesn’t even say the word. So literally the phrase desktop bridge does not appear inside PowerBI desktop. So when I’m thinking desktop bridge, I’m thinking about the feature of PowerBI desktop that is external tool

28:07 access via local APIs. That’s what I think we’re saying. Okay. Yeah. And I think this is this is the confusing part. So the bridge is a local server running inside PowerBI desktop process. Okay. Now I would agree with that statement. That’s it’s a it’s something from desktop. That’s the bridge, right? Now the CLI is code that you give to your agent so that it has the tools and the APIs that are being given to it from your agent. Is that is that clear?

28:38 The most direct way. So the easiest way to start with PowerBI desktop bridge is with the CLI. It lets you leverage the bridge in a simple practical way without building your own custom integration. Okay. built on top of the bridge. It includes a set of intuitive commands ready to use. Okay, you’re going to use these external tools the best way and most straightforward ways using the CLI. Okay, that’s that’s the distinction that I want to call out here. So, I want to call out here the the desktop bridge is

29:08 something that runs on your computer. It’s built into the desktop and the CLI command line interface is a package like you said. I’m agreeing with you there. It’s a package but all that is that is not the bridge the that is just a interface layer that says I’m going to get all the definitions of these different API. The methods we have are bridge. mmanifest application get application report report screens capture and then file reload. So there’s basically four

29:39 main commands that we can use at this time with desktop bridge. And all the CLI is doing is saying, “Hey, agent AI thing. This is how you make the command calls for these different things.” Right. To be clear, yeah, CLI doesn’t. Yeah. The CLI also, just to make that, sorry, you were confusing me because you were blurring the line between the CLI and the definition of what the bridge is a little bit too fast for me. And I want to make the distinction that there are two different pieces. And you don’t need like it’s not required that

30:09 you have the CLI. It’s just a method by which you can do this which means other people could create other tools exactly that do the same thing or mimic what’s happening inside the CLI as well. With the CLI you don’t need an agent or any AI to run it. You can run the commands just in your CLI tool to reload. But to your point, yeah, you can you can build your own customization. But PowerBI desktop CLI br the bridge CLI was built specifically to interact

30:39 with that local server that’s running. Yes, agreed with that. So, and then now the agent on top of that is the agent’s going to say like, hey, I have PowerBI CLI. Use this to do that connection rather. And I bet you too if you told Claude, I want to run PowerBI Desktop Bridge,, let’s let’s build it out. It’s going to probably by default install the CLI tool because it again is the most direct way to communicate. That’s the documentation that show like that’s in the documentation. So it’s not going to go I guess you could maybe tell

31:10 Claude like hey I don’t want to use the C line I want to build my own right and and give it the documentation cuz it’s just methods it’s just and it has here’s the input here’s the output here’s the bridge manifest what it gives you. So, I think you could do this without having the CLI be there, but why would you? It already it’s like it’s just going to be easier. So, you’re right, Tommy. I think the CLI will be the main thing. , and again, it’s just a command. It’s an MPM package, correct? Yeah., yeah. Yeah. I don’t know. The reason I’m

31:41 asking the specifics around the mpm package because mpm is pretty transferable to all machines but it does require some like back I think mpm is node and I think you need like node running on your computer to do this. So like for example if you didn’t have node and you only have python you could probably still build a package that does the same thing. This would be a python driven package as opposed to this mpm package. So I just I’m just trying to contextualize this Yeah. Yeah. Yeah. Yeah. Yeah. There are a lot of this. There’s a lot of So sorry I went very

32:12 deep in the weeds on the technical, you know, developer experience of this because I think these details matter, especially if you’re trying to build on top of something like this system. Yeah. And I think moving forward from the agentic side, the CLI tool, not many people, Mike, there’s not a lot I think of reasons for a user to just use the terminal, right? Like that terminal just to you’re talking about Flint where Flint’s base for an agentic experience for an AI to read

32:44 the CLI tools built for an agent to use those commands. I like there’s no there’s no reason unless you just want to Yeah. so all the tools I don’t know why I don’t know why I’d be wanting to get all that stuff. But when I look at these things though if if you look at the you know get application state or you you bridge manifest right it’s these things are only there to give you like the timle or the sorry not the timal the PBIR format to the agent so the agent can like see and read a little

33:14 bit of that and it’s just passing PBI files of definitions of things back and forth between your agent and your file system on your computer. So,, that’s already built in. And one of the first recommendations they talk about in this feature set is you got to use the report authoring skill. So, that same report authoring skill, I think, is the same report authoring skill or visual authoring skill that they’re using in this earlier

33:44 data agent visualizations with fabric visuals. I think that’s that’s another what they’re I think they’re reusing that skill again in that other article we talked about right prior prior today right and there’s something called the application state where it actually sees if the application’s open what files open if it is it saved so that does m yeah so savings interesting we got to talk about that one keep going okay yeah so because that’s a big part because I actually have created something with claud that we’ll talk about where I’m running a bat file that

34:16 micros or that Claude created for me. And it says, “Hey, we were going to do these things that you said to do, but you need to save your file first. We’re got to reload it before we’re actually going to take screenshots or update the visual side of it because again,, you don’t want to it will detect if there are unsaved changes before running operations that might overwrite content. So, the application state matters especially when it does the file reload. , useful after you make manual edits to the report file. So these are all

34:47 pretty neat and again it’s interesting Mike because in the documentation PowerBI desktop bridge is underneath the agent skills for report authoring. So very interesting here as MCP and PowerBI desktop bridge are the two tools right now for agentic. MCP is for semantic modeling. PowerBI desktop bridge for report authoring and Mike. So yeah, what were you going to mention

35:15 about the savings side? Oh, savings. Interesting. So there’s a there’s a concept or conceptual things here that’s happening inside the desktop bridge that I’ve learned that you have to be aware of. So one of the desktop bridge items is a status, right? So let’s imagine you have two PowerB desktops open, right? There are two gateways now running, right? One for each desktop if you have the the setting turned on. So one you have to figure out which what number of gateways are open and what are you trying to connect to. the one

35:45 thing here that’s also interesting is these unsaved state changes. So when you have a PowerBI file open and you’re using the PBI format PBI R format report format when you make changes in desktop the changes are like in memory of desktop. It it’s like it’s like a they’re not physically changing the files at the time of clicking the items. So it’s almost like there’s a cache or a local

36:16 version or like a unsaved state of everything you’re doing. Desktop is aware of those things. It’s almost like there’s a little bit of a little mini branching strategy, right? The files are on disk. Those files are there. PowerBI desktop reads those files and then the report opens. as you edit things in the report, the files are not changing. And so the status here, the state is really interesting because if you ask this CLI to take a screenshot,

36:46 you may be getting screenshots of things that are not representative on the state of the PBI files. Right. Right. This is happening to me already. Yeah. Because the screenshot is like you’ve added three visuals. Great. That is fine. It just means that is going to be a different version than what you have on disk. And so when the agent is looking at the agent’s not physically looking at the report page, it it this is why how do I say this? It’s like things get out of sync. And I you said Tommy, you’ve been burned by

37:16 this. I’ve been burned by this as well. You make edits in the report cuz we’re I mean me and my agent are working together. Maybe it’s working on something. I’m building something as well. I’m asking you to change something. I’m also modifying things as well. What happens is the agent is modifying the PBR files. Those are the files that are on your machine and then what it’s doing is hey the agent says I’m done editing refresh desktop get the latest version and any work that I would have done gone because it’s hallucination here too I’ve noticed.

37:47 Well, yeah. And this is where the sk the skills I think are trying to help remove some of that. But yeah, you’re right. Right. It So, all of this to say is there’s a lot of things here that make this easier. And I’m seeing other people like NJ Park NJ is talking about doing fancy slide out menu pad panels and stuff like that. And so, as much as my heart grieves this a little bit, you know, Tommy, when we were back in the day when we got bookmarks the first time, we were like, “Look at this. pop up flyyou panels and panes we’re

38:18 like this is cool like look how cool we can and then it became this nightmare of managing bookmarks and everything else that’s going on because I couldn’t see all the bookmarks and what was interacting with things now the agents can actually handle all the bookmarks so teaching an agent a skill on how to make a flyyou panel right and how to make a filter pane and you can now teach your agent how to do it and then it’s really good at reading the code and understanding what needs to be changed at the code level.

38:49 So, so that is a really advantage. But what this is doing is it’s making it less friction for people to now build menus inside PowerB Reports, which again I’m like, I’m not sure I really like this, but the agents are making it easier to manage, so maybe it’s okay. I’m a little undecided on this new feeling. I’m a little torn about this new pattern that I’m seeing emerge in the community. Well, let’s talk about the agentic experience here because this is all well and good to me, Mike. And I again, I wrote an article just as I was thinking

39:20 about this more and also my own experience. You can now again you use an agent to help build design your visuals and then have that input into the report. You don’t have to even open up the visual pane. PowerBI can actually design so in so many words the report for you. However, just because it can doesn’t mean it’s good. Doesn’t mean, , is it worth it here? And this we is where two really main areas

39:50 come to when we when I think about my experience here and how do you actually make this work, right? Because Mike, do I want you to test you probably have done this test already, but for those listening, do yourself a favor, do a little experiment for me. My kids are beginning to do science experiments. So let’s do an AB test. I want you to write two the same prompt to PowerBI desktop bridge on whatever agent tool you want and have the same data. So you have your controls there.

40:21 See if it comes out with the same design in both experiences even if you’re using skills here. And this is where the problem lies because general prompts with general context you are going to get completely radically different designs and you got a report you got visuals it does look pretty nice but the thing is it’s sporadic to me because it it really honestly Mike every single time I’ve tried this experiment I get radically different designs for the report. I’m

40:51 not necessarily saying again I’m saying hey I have this report I want to show these three measures a measure list so to speak like these are the measures we want to show we want to show it over time cool it will design it will build a full page but every time you do that same prompt you’re going to get different designs and that for me does not work that does not jive that is not to me a really great experience because I can’t rely on any consistency it’s interesting that you’re making that analogy Okay. So,

41:24 I’m going to analogy. Well, not analogy like that that phrase. Okay. It’s interesting you’re talking about this this this let me call this a feature or pattern that you’re you’re seeing here. Like so I would one thing is I’ve observed the same thing Tommy. I’ve I but I think I’m coming to a different outcome than what you are. So, I’ve observed the same pattern of like doing giving an agent a prompt and then getting like wildly different or not wildly different, but, different enough outputs that you’re like, hm, that was it wasn’t really wasn’t expecting, but like I’ll go with

41:55 it. Like, it’s fine. I I’ll use it., so on one hand, I I agree with you there. However, I I also look at let me let me let me parallel this with another analogy, right? I hear a lot of people saying, well,, we can’t trust the AI and it’s going to hallucinate and all these things and it’s going to build things that we don’t want and it’s not right. Okay, it will. But who but there are engineers out there and there are people that are using agents and claude code and GitHub

42:26 copilot and they are getting good outputs and they are trusting it. So it what this tells me is there is a place in the world here. There’s a there is a a methodology that works that does get you consistent outputs and good results. Okay. So when I look at that I think okay when when I hear people say the agent won’t do what you want. It’s going to give you these wild results and these different answers. My answer my response to them is is it is it your process? Is

43:00 it your prompts? Do you just not understand how to give the AI information and how to generate what you want? And I think it’s really the latter here on these questions, right? I don’t think this is a problem with the AI or the or the harnesses that the AIS are being used with. I think this is more of a prompt problem. I think this is more of a user lack of knowledge issue. I think this is a lack of process issue. So if you really want reports with very

43:30 consistent outputs, let’s say I’m making a family of reports and I want them to have all the same style, same header, same information on them, I likely should not be starting even inside PowerBI desktop. I likely should be starting in some sort of design application and starting to figure out like what is my style that I want to use? What is the font sizes? Do I have a header? Is there a footer? like what is the branding requirements that I need to design. I’ll

44:00 design those brand requirements with an agent and get those together. And this is one of my big complaints around the data apps, fabric data apps. They just let you like build me a report that does this and outspits something. There’s no attention to specs. There’s no attention to building designs. anything. And so what I’m doing a lot more now, Tommy, is before I like I really want to see renders of what I’m building way earlier in my process. And

44:30 now with agents, I can make those renders really easy. And what I can do with those renderings is I can go to cloud code design. I actually love cloud code design. And it makes a lot less it it makes a lot less AI slop. So I’ll start with an HTML document. I’ll go to cloud code. I’ll say, “Go build me a style guide. Go build me what I want. What does this thing look like?” And I’ll actually ask it to build generally like page background, page sizing, all the headers, the footers. Let me let me

45:02 with a very low non-data system build out what I want. Let me let me back at the napkin math, right? The system. Once I have that, then I go back to Cloudco and say, “Take this design, make this a standard, make this something that’s reusable that I can give to my other agents that now talk directly to desktop bridge.” And now you get a much more consistent output. So, it’s these kinds of patterns that people need to learn because

45:33 most people, Tommy, I’d argue, who build reports do not start with a design spec. So, it’s it’s funny because I think there’s two sides of the coin here. And I love that you this is why this is why this podcast has worked for 556 episodes. You are looking at this and your biggest price is right on this on this the price is right for this podcast. It’s this is all free knowledge you get from this one for free. Well, honestly though I and I love the way you think on this because

46:03 you completely took this from the pure design, the aesthetic side of this, making sure that we from from,, top to bottom have consistent visual aesthetics, design patterns, right? And dude, I’m I’m the theme guy, Tommy. Like I’m the guy like it consistency in design is like my jam. Like I’ve been building 10 years now. Really? I’ve been building theme generators since 2017. Like, I really think I know what I’m talking almost 10 years have I been building

46:33 consistent design tools to use with PowerBI. No one’s got a better tool than what I’ve got out there right now because I’ve been doing it for so much longer than everyone else has. Well, you could say I’ve been doing this before AI. I built this before AI, which almost sounds like, wow, they really cared about this. Yeah. So, that just tells you how old I am and how long I’ve been doing it. Yeah, I remember V1. Not something I’m proud about.

47:01 but I have taken this from the other side of the coin on context because to your point everything you said one 100% correct validated absolutely need to do. However, if you just stop there for me, well, you just have a really nice looking report that I worry is not going to actually be helpful, right? is actually not going to really showcase what you need to show. It will show data and it will look great, but again, you may still have different designs or

47:32 visuals in different places or how it’s actually showcasing the report. And for me, Mike, you have to now add two elements to this recipe for really PowerBI Desktop Bridge to shine, to actually run the way you desire to. The first one I’m going to say is the something that’s just available that you should already know about and that’s the report skills. There are three of them. This is part of the PowerBI’s authoring plug-in that Microsoft has. And there are three there are three skills here.

48:03 There’s PowerBI report planning which gathers and these are all agent skills. I want to make sure there’s note there’s a report planning skill that gathers requirements and locks in a spec before anything gets built. There’s PowerBI report design that handles open-ended visual decisions, page arch types, chart to selection, and the pallet if you didn’t have that already. And the report authoring, which actually does the mechanics, validates and verifies what’s rendered. And it does a lot of this in in J in JSON as well, but again, it’s

48:34 meant for the agent to do. And Mike, you I would not even dream of building anything in PowerBI desktop bridge if I’m not using the skills off the first place. Like like I’m just really at this point now where I don’t do a lot of things without skills in my daily workflow. I have a skill for most everything and if not I have a agent that researches new skills. Anyways, so the skills right here are so important because

49:04 your measures and if you tell the prompt,, that tells the agent what’s possible, but not necessarily what you want. And that’s that first side of the coin here when it comes to let’s say your model has 40 measures, right? How does it know which ones to choose or in what context to choose them in? And then the other side here, so that’s s part one of getting your right proper context. The other side is your prompt and my article and I think things we’ve talked about too rather than

49:36 having a simple prompt Mike I would even argue your next stakeholder meeting or the next meeting that you have use that transcript to design the report planning. So Mike my workflow right now anytime I use power bay desktop bridge I have a stakeholder meeting we do our sandbox meetings where maybe it’s the first discovery I have import in incorporated those skills into notion to help me build powerbi desktop instructions or bridge instructions for claude so it has that proper context etc

50:07 and knows the meaning and everything we talked about I use the grill me skill to refine that so there is that time that’s taken and that’s what I feed into claude or my agent or cursor whatever the tool may be to actually then begin to design the report utilizing those three skills but the whole actual instructions is very large in terms of what I’m actually feeding for bridge this to me is the way and it’s very hard I am not ever going to use PowerBI

50:38 desktop bridge with an agent with a two sentence prompt it’s not worth it so for me like you I feel like I have to do this amount of work leading up to it if I actually want to get what I want. So, I outlined a a two-prong process there, both meeting notes and using the skills. So, Mike, what’s going on your process that you’re doing? What’s your take on what I said? Is that am I doing too much work here? No, but I mean, I think for the average user, there’s a little bit of like a learning curve here, a little bit of

51:08 experimentation and getting up to speed with this. Yes, there’s some things in the community that is going to make this easier for us. Some skills that are already out there that exist. You should definitely use those. That’s a that’s a mustave. Go look at skills for fabric. Yes, you have to have that. Go look at Kurt Buer’s skills that he’s using with reports and PBS. Have to have it. Like these are things, but you have to spend some time to slow down and digest what is happening with AI, what’s happening with these skills. If you don’t use these, you’re going to get a much less quality report output. It just

51:40 won’t be as good., one thing I’ll also just note out too, Tommy, another blog post that’s actually very tangential to this that you’re talking here now,, Sarah Rod I sorry her name is Sarra Sarah Rodriguez just posted I think a community item here recently which is talking about there’s another edit your PowerBI

52:10 desktop top projects anywhere and reload them instantly with PowerBI desktop, which I think is using the same bridge experience that you’re talking about here. I’ll put this here in the chat window as well, Tommy. So, this this just came out as well. There’s now a single button click in desktop that opens VS Code. Wow, that is awesome. So, super big win for people that are a bit more technical there. But you the reason you go to that PBIR file format and get that open in VS Code is so you can get access to your agent. That’s that’s why you do that, right? and so

52:41 that’s that’s the example here as the other edit reports with desktop with instant and it automatically refreshes. So as it edits it it’s know it’s using the CLI it’s probably already built in and then as it the files are being edited that desktop bridge is automatically adding and updating things as well. No. And I So, Mike, how would you recommend your own workflow though before you actually go to Claude or the agent to write your instructions? Like, are you writing the instructions yourself? Are you using a

53:12 tool to help write your instructions to Bridge? How what’s your step before you actually using Bridge? So, right now, it’s a lot of experimentation for me. It’s just talking to the agent and figuring out what it is building. I I find building skills is probably the most effective way for me to build things with agents at this point. And so I guess Tommy, what I what I would do is my pattern right now is go to desktop,

53:45 open a file up, use an agent, get desktop bridge working, right? All the things, right? And now now there’s a now there’s a button, right? Now there’s Siraj said now there’s a there’s a button you press you get your report open you you click open in VS Code boom runs awesome so so now you’ve got that whole experience like already wired up ready to go you can immediately start talking to the model this so that being said now that you have that I’m going to

54:15 experiment with things build this create that make this page build this panel this flyyou panel and describe what it is. When you’re done though, when the AI does something that you like, I pause and say, “What did you learn in this session? What did you learn to build here?” And then I ask it to reflect on that. It’ll give me some answers. Oh, we did this and this and this and I learned this and I I learned we did A and it didn’t work and then I did B and then it did work. So, the agent actually

54:45 explains this to itself to you like in the recent session you did. So, think about when you when you talk to agents. I’m not spending hours and hours of time on one prompt or one chat session. I’m starting a chat session with an intentional purpose. Build this visual. Create this page. Build a filter pane. At the end of that session, the temptation is to move on and start talking to it and starting a new session or or start

55:16 adding the next feature. Stop. Don’t keep going. Pause. Ask your agent to reflect. And then if you think you’re going to do that thing again, ask the agent to build you a skill around it. And then it becomes a skill. And I think that’s the trick here is as like it’s taking that that doing this little bit of step of like pause, reflect, make a skill means I can now reuse that same 15, 20 minutes of work

55:46 in another session. And then what I do is I go back up to VS Code. I click new session. Clear the context window. We start over again. Yeah. And we start fresh with a brand new thing. And this way I can then bring I can bring that knowledge I had previously over to my new session by using the skill that I just made. So the transfer of knowledge from one session to the next session is the skill in my opinion. Do you have a a reflect skill that

56:16 you use? I had created one in the past. I just don’t use it anymore. I just literally ask it. I just I just literally say it’s it’s not actually a skill. If it would if I did, it would be like a sentence or two. But it’s So does grill me. Yeah. But I but I just literally it’s fast enough that I’m just talking to my computer anyways. I’m just like reflect on what you built. And that’s literally all the line it is. Reflect on what you built or reflect on what you learned. Right? That’s it. And then usually what it so what happens though Tommy is it comes

56:46 back with some messages and then I shape the response into the skill that I want. So I wanted I wanted to respond first and then I read the response and say okay ignore that ignoring the the the gateway crashing issue, right? Make a skill around specifically this lesson that you learned, right? Hey, you learned how to build a visual the way I like it with the properties I want. Great. Make that a skill and now you can just call that skill when you

57:17 want and it just it makes it a bit easier to do there. Yeah, I think there’s actually I I think I just saw it on Claude at some point there’s actually now in their co-work at the end of it says, “Hey, do you want to create a skill for this task?” So like Yeah, but like as we look at f as we look at PowerBI desktop bridg and I love the reflect thing. I think there’s a I there’s a skill there to to do but is this publish and ship for you? So like I said for myself I look at what I’ve been able to do again incorporate

57:49 the workflow. I’m very cautious and I am very intentional about any agentic execution that I do or and whenever I apply it that it’s either got to work to a degree that meets a standard that I have or I’m not going to use it right just because it does something doesn’t mean it’s worth doing. and I think for me I look at PowerBI desktop bridge and this now has gravitated towards part being a part of the official workflow

58:20 where it is doing things better than I can from a design point of view with the proper context right so my work is now in the planning of what I want the report to be but to me like it’s going to be very hard for me to just not use PowerBI desktop bridge. So when you look at how important has it been to you, is this

58:44 something that you see as I’m doing it every time, most of the time, some of the time, or, how does it kind of match for you? For me, this is a at all times type of thing or close to that because of what I’ve been able to see the output to be and the quality of that. Hm. This what this is doing for me in my workflow, this is opening up brand new possibilities that I’ve never seen

59:16 before. Let me give you an example. Sometimes when I’m building things with PowerBI, I like doing like experimentation on a separate page. Yeah. Or,, one page has four or five visuals on it. I don’t know, Tommy, if I need to be one page at a time anymore. So, so I’ll just say it and I’ll we’ll unpack what here in a minute. Right. In a report that you had previously, you

59:46 maybe have three pages. On each page, you have maybe three or four visuals on each page. Great. Why not build a page per visual and hide those pages for development purposes? Oh, right. Okay. So, hang on with me here. Right. Yeah. What that does is if you make one visual per page, it allows you and the AI to name the page correctly. It allows you and the AI to work on the styling, the technique, the design of that

60:16 singular visual without getting all con, , messed up with everything else in the report. Right? So, one of the things I find with AI is getting noise out of its system makes it really easy to work with. So,,, if I’m doing things with an agent, asking a simple command that says, “Look, I got a three-page report. Go through this report, find every visual in it, make a brand new page, and put each visual on a brand new page. I’m going to work on stylizing or developing the single

60:47 visual one at a time with you. Previously, right, if I was doing this in desktop with clicking things, not useful at all. Would take me hours to do this. doing this now with an agent, what is it, 30 seconds, a minute, two minutes, maybe? It just builds it. Done. I made it for you. So,,, Tommy, for for months now, years now, I’ve been saying, “Ditch bookmarks. Don’t use them. Don’t put them in the report.

61:17 Don’t have them around.” The agent can now easily understand what bookmarks are there. And you can go to the agent and say, “Hey agent, make a new bookmark or explain to me how this bookmark is being used or tell me all the items this bookmark is touching.” These are previous experiences I was never able to get a hand on before. The agent just produces them. So th things like you know heavier bookmarks, custom filter panels, one visual per page for development purposes and hiding mass

61:47 pages, reorienting like just just dragging and dropping pages around. Tommy, how how many how many hours of your life have you spent? More than I want to admit. Dragging pages into the appropriate order for the report. Hours, Tommy. We’ve spent we’ve spent days of our life just dragging tabs around because there is no feature to mass move them, right? Like like the famous soap opera or whatever. Yes. Days of our lives.

62:18 So, so the reason I say this is because I can now with a single sentence say look at the report and say okay agent and I can talk to it. I don’t have to type this. I talk to the agent and say I want the page order of reports to be this this this and this. Done. All other pages, hide them, move them to the end. One one line of sentence now does hundreds of clicking and dragging and moving things around. Easy. So this is where things are going. Things are going into this direction. So for that reason,

62:48 Tommy, it’s this experience of using desktop bridge is becoming my my primary method by which I want to build reports with. I like what this does because an average user not doing an agent to do this is going to spend hours of time doing what I can do in minutes now. And it’s very evident to me that this is going to be the future. I I love that, Mike. And I I the only thing I would push back on slightly is

63:19 the fact that for me I think that really any bridge instructions needs something more than just a simple like a sentence or two. Like I I’m very I’m very intentional on most anything I do right now is with a prompt from somewhere else just because that’s been the best. Honestly, I I can’t imagine a way where I’m just writing a sentence. Now, refining things are different where I’m

63:50 going, “Hey, we want to actually do this, not this.” But when I start a session, like for me at least, it’s coming from another tool, another harness that I’ve actually built out everything needed from the instructions point of view. Actually, I Yeah, like in a sentence will be like, “Use this notion page, which I have like a master sequence like no cloud instructions. Let’s get started. we’re going to focus on the bridge or we’re going to focus semantic model and it has the updates and everything for because it’s just so much better. It is so much more helpful for

64:20 the agent and much more efficient to get what I want to use the right tokens and we don’t have to build that out together. Well, the distinction here that you I think you’re building here Tommy is it’s a difference between building net new versus refining and tweaking. Mhm. And I think I think you’re right, Tommy, but I would also argue the reason you’re using really large prompts at the very beginning is because those aren’t technically skills yet that you’ve designed at this point. And so you’re giving more information and context at the beginning part. And I would agree with you, Tommy. What you’re describing

64:52 to me sounds like what all the papers support around AI. Give AI as much context as you can at the very beginning of what you’re doing. Do more planning mode. do more thinking through with what you want to build and then ask the agent to build a prompt that does that for you and you’re going to get a much better result on the output because you’ve you’ve planned it with the AI before it happens. So I I agree especially with bridge and the visual side of things, man. Like it’s there’s a lot more to get

65:22 wrong. You got to have a lot more you got to be a lot more descriptive wrong because you can show you can build visuals any way you want. There’s a lot less definition of what’s wrong, right? Because you can do code that’s wrong, but a visual this is why that’s so important to me. So I think I think there’s something there Tommy around like the developing the process,, your your initial parts of the problem. So when you’re starting net new, when you’re starting with with brand new things, your prompts will probably be a lot larger until you get to the place where you’re actually building some skills, you have some things to shortcut

65:53 those things. One thing I really dislike is I like I dislike repeating myself. I don’t like having those really large prompts and having to send it again and again and again on every first session. So that’s why a lot of times I’m building skills. I’m having it remember things. That’s helping me keep those initial prompts down. and making the project much easier to use and read. Anyways, love it. Great topic. Love it. Yeah. Yeah. I think I think this new desktop bridge is going to physically change Tommy how we build desktop and what we

66:23 build with desktop and what’s possible with desktop. Remember do you remember the I don’t know if you were there with me or if you seen one of these. Have you seen the world championships in any one of the Fabcons recently? No, I have not. Okay. So, every FabCon the last two years running or last two events that I’ve been at, they’ve been doing this world championships, the winners of the championships are typically using HTML visuals and they’re using agents to build HTML that’s built as measures that go into

66:54 the report. And there’s a lot of copy paste that’s occurring when this happens. Okay, I get it. But this is the precursor to everyone wants to use agents to build really good reports. And so the next world championships, are we going to allow AI agents to help us build those reports? Is our report building going to be more prompt driven than anything else? Now I think they should allow it. I think it’s I think it’s I think people need to

67:24 see how fast it is to build really good stuff with AI. You need to show your work. I think they I think it’s just like when you in math where it’s not just the results like you need to have that page like my dad would anytime I help me move for my homework it’s like I don’t want to see your results I want to see the notepad you had to show how you got to that answer so yes agree that is the sauce that’s the secret well you once you learn how to do the planning and the designing and the pro all of that has to happen outside of this and so that’s that’s really where

67:55 this starts and who knows Tommy in in this next round of world champions of world visual champions. Maybe there’s going to be someone who brings in cloud code and some other code agent and say, “Look, I want to prompt it like this, this, and this.” And you’re going to actually watch them build prompts on stage and go through the whole design flow with a designer and then from there build out the report. I would be very impressed if you could build a goodlook or decent report within an hour of

68:25 time that they give you. But that would be very impressive to go from nothing into something within an hour. So that would be really I’d like to see that actually. I’d like to see someone do a full design, agentic experience and report build all inside one session inside fabric. Yeah. Well, hey We we had a great time on this, but unfortunately I got to call it quits for today. As much as we we’re probably going to talk about this more again, but

68:56 Mike, I think you’re right, Tommy. This has been a great topic. Thank you all very much for listening and hanging out here with this this conversation. This is an incredible new feature and I’m going to say it’s substantially in that game changer arena. Desktop Bridge will change how you build reports and a lot of that click tax that you had in the past will go away. This this is the solution for a lot of that. So anyways, looking forward to it. building reports with code and with

69:26 agents is the way to go. I think you everyone needs to learn this. hopefully you learned a couple tips and tricks and techniques from Tommy and I on how we’re using it, where we’re finding value, and you know it’s going to change your workflow. It’s going to be different. Tommy, where else can you find the podcast? You can find us on Apple, Spotify, wherever you get your podcast. Make sure to subscribe and leave a rating. Helps us out a ton. You have a question, idea, or topic that you want us to talk about in a future episode, head over to Powerat Tips/mpodcast. Leave your name and a great question. And finally, join us live every Tuesday and Thursday,

69:57 a. m. Central, and join the conversation all powertips social media channels. Excellent. Thank you all so much, and we’ll see you next time. 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 feels. Explicit measures. Drop the beat. Now, kings feel the crowd.

70:29 Drop it loud.

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.