Fabric as a Backend – Ep.554

Rayfin made this story loud, but Mike has been running Fabric as an application backend since well before it shipped. This episode is his argument for why that works — and Tommy pressure-testing it with the question that matters: does it do anything better than a traditional backend, or is it just convenient because you’re already here?

News & Announcements

• Build agents with AI for Data Agent (preview) — Configuring a data agent was never hard, just time-consuming and blind. This explores the schema, learns from query history, generates instructions, validates, and repeats. It covers lakehouse, SQL endpoint, warehouse, mirrored database, SQL database, and eventhouse — not semantic models yet. Tommy’s read: people struggled with data agents because they didn’t know what instructions to write, and previously every attempt happened in a vacuum with no history to learn from.

• Automate recurring business logic with scheduled User Data Functions — Schedule a UDF directly from its interface to synchronize external systems into Fabric, send notifications, or archive stale records. Mike’s framing is that this is basic functionality catching up — Azure Functions are cheap enough to run by the million, and Fabric’s UDFs should be too. He suspects the scheduling push comes from low adoption of the button-driven UI side.

• Direct Lake calculated columns (preview) — Calculated columns arriving in Direct Lake models.

• Agent skills are the new theme files — Tommy’s post, and the written version of the Ep. 538 argument. Skills and theme files face identical governance problems: distribution across teams, silent drift into multiple versions, no single source of truth. Theme files spawned theme_FINAL_v3; skills will fragment the same way. His recommendation is treating skills as versioned assets in a central Skill Vault, governing only what needs a consistent standard and leaving room for personal craft.

Main Discussion

Topic: Fabric as an application backend — what it’s actually better at

• Mike is calling it FAB. Fabric as a Backend. He notes Microsoft is communicating the idea through Rayfin without naming it, and dares the internet to adopt the acronym.

• The Steve Jobs iPad test again. If it isn’t better at something than what you already own, it has no reason to exist. Mike’s answer is the wiring: in the Azure portal you can build the function, the web app, and the SQL database, but you need to know how to make them talk. In one Fabric workspace they share an identity and the friction disappears.

• The SQL database was the unlock. Before it, Fabric was lakehouse and warehouse — batch, with expensive always-on streaming if you needed live data. A SQL database inside Fabric made a real application backend possible, and that’s when Mike started building in earnest.

• CI/CD comes largely for free. Most Fabric items now sync with git, that repo moves between workspaces, and deployment pipelines exist on top. Mike’s point is that most people never built this themselves because it’s genuinely hard.

• Tommy’s reframe, and his concession. His first instinct was that Mike was forcing an analytical platform into an operational role. What changed his mind is that the value isn’t proximity to Power BI — it’s that you can stand up an entire communicating infrastructure whether or not you ever open a report.

• The open question is cost at scale. Neither host has data on what CUs look like when a genuinely popular app runs on this. Mike’s counterweight: he rebuilt one client app to modern standards and it now runs for $50 a month, against a Fabric license they were already paying for, with autoscale for bursty SQL.

• Consolidation is the business case. Rather than paying a little here and a little there across many tools with separate billing, Mike is centralizing on Fabric and building the application layer himself.

• The creator agent, again. Mike doesn’t want an agent answering “what were my sales last month” every day. He wants it to build the report he uses every day — because a report can be tested, verified, and reviewed by other engineers, and the higher it goes in the organization the more eyes it needs.

You’re a system builder now

Mike’s response to complaints about agent output: it’s usually not the agent. If you can’t articulate the system you want — the tests it should have, the structure it should follow — you’ll get something shapeless. He’s using the same models everyone else is. The skill that separates results is describing the system precisely.

He offers two examples from his own week: an agent that built a paging download project with Playwright and a readme so he never visits those sites again, and a discovery agent that browsed a SQL schema directly from the terminal to find the right columns — no SSMS, no download.

Looking Forward

If you’re curious about FAB, start with a Fabric SQL database and one small app — the workspace identity doing the auth wiring for you is the part that’s hard to appreciate until you skip it.

Episode Transcript

0:01 Dance to the day to laugh in the mix. Fabric and AI get your fix. Explicit measures. Drop the beat now. Pumpkins can’t steal the crowd. Explicit measures. Hello and welcome back to the Explicit Measures podcast with Tommy and Mike. Tommy, hello and welcome back. what? I don’t say this often, but it’s great to see your face. Actually, I do see that a lot.

0:31 Actually, I do see that a lot. It’s funny for the listeners of this because the listeners have been getting episodes every Tuesday and Thursday, but Tommy and I have had some travel going on or I have and so I we’ve had to record a lot. So, even though you’ve been getting episodes every week on time, Tommy and I haven’t talked in quite a while here. We’ve been we’ve chatted a little bit on Teams, but yeah, we we still do some tech. Yeah. when you’ve been talking every two days for what, six years now? we’re we’re on five. Yeah, definitely five. So five. So five and a half right now. it it

1:03 five and a half right now. it it feels weird to be out of the routine. So anyways, that is just funny that we have that going on. And yeah, it feels good to be back and we’re we’re doing episodes again. This is another live episode back in again. today’s topic is going to really talk about fabric as a backend. Fab, I guess you if you want to call it that. You’re starting to see I think you’re starting to see on the internet a lot of people talking about Rayfin and building apps with fabric. I really like this topic. I’ve been building apps on top of fabric fabric as

1:34 building apps on top of fabric fabric as the backend service since basically January of this year a little bit earlier than when Microsoft was really communicating the Rayfin project which is really pushing this to the next level where the app itself is landed inside fabric., I have some patterns here. I think this is a strong topic. I I really do think fabric as a backend is becoming a very compelling story. story., and now with the adventation of agents, agents, agents are really helping accelerate this experience as well. Yeah.

2:04 Yeah. All right. Before we get into that main topic, Tommy, what else do you want to talk about? You got some news items that we should probably pull up here. We do. So the first one, my friend, is next week, next Thursday, a week from today actually on August 20th, we have the next Chicago Fabric and PowerBI user group. It’s going to be held 300 PM downtown Chicago, and we are going all through the Fabric Task Flow Studio. This is through Alex Powers and Microsoft’s Fabric Task Flow’s

2:35 Microsoft’s Fabric Task Flow’s user agentic solution to build your artifacts in real time. what I was able to do working with Alex is actually build a fork of it or a version which is a whole UI with some new features that actually can manage projects existing and new ones. So it’s not just creating projects but even your existing projects edit it see the phases that are actually occurring. And what we’re going to be doing during the user group is going from soup to nuts, going through a what would be completely nothing or just a set of instructions

3:05 nothing or just a set of instructions and then how does the actual artifacts get built, how you can actually add data to it in fabric task flow studio and really get everything up and running. So it’s really really cool. We hope to see you there. So that’s going to be this is a custom tool telling me that you built but it’s going to help you build I guess it’s really around executing the build of a workspace or patterns and workspaces right right using AI. So this is like you know helping you build prototypes helping you build more workspaces structures. Alex has derived this pattern of

3:37 Alex has derived this pattern of being able to use an agent to produce the documentation or produce the scripts that you need to build things and and this is a really interesting project. I really like where this is going. I’m shocked Microsoft didn’t really pick this up and actually add this to their products because I think this would have been a really nice product to see in inside fabric. But that leaves opportunities for us like developers for community. Yeah. To go build what we want to go build. I saw a very funny article the other day. someone on LinkedIn was complaining about when you open up fabric

4:08 up fabric and there’s all these little tabs across the top and if you close down fabric and come back all the tabs reappear. reappear. Yes. or you start a new tab, right? If you duplicate a tab or make a new tab with fabric in it, all the tabs reappear again and they’re all showing up., someone said on I guess it was LinkedIn is where I think I saw it. They said, “This is annoying to me. This feature doesn’t exist. I don’t like this feature.” And so they built with their agent a little mini app to automatically like open the browser, but you could immediately close all the tabs inside

4:39 immediately close all the tabs inside Fabric. So Oh, yeah. Yeah. Yeah. them and just land you on the home page, I guess. And I thought to myself, this is interesting to me because this is physically changing how we think about like this is this is what we’ve been talking about, Tommy. This is what everyone’s becoming a software like connoisseur, right? If the software doesn’t do what you want, you can go give an AI agent some some code and say, “Hey, I need some help here with this. I need to build something. Execute this thing.”, I was just working on a

5:10 thing.”, I was just working on a website and you may have maybe maybe have done this too, Tommy. Whenever you use a website, doesn’t matter what it is, there’s always some like API call or something that’s happening in the back end. end. You can let your agents have access to like a browser window. I was going to VS Code. I open window and I said I’m going to I signed into the application and I said I took a again I gave agent control access to my window my browser window and I said when I click

5:40 browser window and I said when I click this button it downloads records of this our podcast right and it only downloads 250. I want all 500 episodes. Can can you go do that? And what it did is it said let me let me try this out. And it went through and it clicked the button a couple times on its own. It watched the clicking of the button. It watched the API calls inside the website and it said, “Hey, I found some APIs that are on this website that uses your in token

6:10 website that uses your in token that make calls to this thing to go get the data that you need.” So, we made a bulk request. I can I can page through all the requests and produce a little file for you. And I said, great. So, I I asked it to do it. It did the project. I said, I want this to become a project. And so then it what it did, it added playright. It added a little readme file. It added all these project pieces. And so now when I want to go download things, I don’t actually need to go to websites anymore and use their APIs. I can go to

6:43 anymore and use their APIs. I can go to their front-end website, tell my agent, this isn’t this is boggling my mind, Tommy. likeing the existing APIs that the company did not document, but it’s still using my credentials. I still to go to the website, I still have to sign in, in, but the agent is smart enough to figure out what API calls are happening and which ones are getting the data for you. And it can build a little mini program that says, I’m going to go build all,, page through all the requests, get all the data that it needs, and then

7:13 get all the data that it needs, and then return it out as a CSV file. I thought this is insane what’s happening right now. And I really thought this was so cool that agents are now able to build standalone mini applications on top of this. There’s just so much there’s so much going on right now, Tom. It’s so cool. cool. I know. I know. And it’s it’s funny too because Mike, a lot of times you may have to nudge an agent to try again., for example, I was doing a new discovery and they gave us a database with tens of thousands of tables tables and they’re like, “We’re not sure which

7:44 and they’re like, “We’re not sure which ones are the right ones for X, Y, and Z.” Z.” So, I went to Claude, I went to Co-work, I went, “Hey,, I want I’m going to be,, semantic models, my fabric project.” I’m like, “Can’t you run SQL on on the back end on your terminal?” I said, “Yeah.” I said, “Go create a discovery agent that’s going to start just schema browsing everything for me.” So, it connected to SQL. I didn’t I don’t need SMS. I don’t need,, to download this. It can pull all this and it’s like, yeah, I found actually the columns

8:14 like, yeah, I found actually the columns that you’re looking for and the tables that are probably obtaining to this project that you’re working on. I said, cool. Well, let’s create a skill here. So anytime I need to do a SK SQL browsing for discovery that we have this right back we have the discovery and again both whether it’s the browser it’s amazing it’s really what you limit yourself is what the agent’s limited to I’m feeling at this point there’s another topic in here that is a really around this one which is what happens when the agent is actually smarter than you

8:45 smarter than you like h like how do you how do you as a less less sharp person than the agent that you’re working with. How do you do like it’s this idea of like if you don’t know what you don’t know, you don’t know what it can do. can do. Right. Right. Right. I It probably can do a lot more of what I’m thinking of. And I’ve had this aha moment, Tommy, over the last couple weeks of working with agents is sometimes we say, “Well, you what if the agent’s not secure? What if it doesn’t have security, right? What what’s all these what if questions?” And I look at this going, I’m like, I don’t

9:16 look at this going, I’m like, I don’t think this is a problem with the agents now. I think this is more of a problem with the users using the agents. Yeah. Yeah. Like it’s you not doing a good job saying I want you to build this system. I want you to have these kinds of tests. I want you to have these systems. So it’s it’s more of the issue of now you’re a system builder. Mhm. And if you don’t understand how to correctly articulate the system because I can use opus the same way that anthropic or any other developer or AI

9:48 anthropic or any other developer or AI engineer can use opus and they’re getting outputs that they like that they want. So what’s the only difference in this variable is how I’m talking to the agent or the skills and and things I have in place in front of it. The only difference here is me. It’s not the agent or its capabilities. It’s the conductor’s era, my friend. This is so permanent because a conductor may not know how an expert on the trumpet, but they know what sounds the trumpet can make. They know the limits and the ranges of it, and they know what sounds good in an

10:18 and they know what sounds good in an orchestra. And that I think our wisdom towards using it does not necessarily mean if we’re smarter or dumber than the agent. It’s the wisdom that we have and what’s capable, what the parameters are. 100%, Mike. So, yeah, really cool. So definitely join us as we walk through some of that wisdom on the

10:37 through some of that wisdom on the fabric task flow studios. Again, register on meetup. The link is in the episode and intubes. All right. Then link is in the chat is also here as well. I put that there as well. what else do you have for us, Tommy? So that’s Meetup. we’ve got a couple announcements here. Yeah. So we’ll see how fast or slow we run through these, but cool one coming up is from the fabric updates blog. build agents with AI for data agents. And what this actually allows us to do is con trying to configure a data agent is not difficult but can be time

11:09 is not difficult but can be time consuming., you need to understand the schema, write the instructions, craft the examples and repeated testing and usually the errors or the undesired results come from unclear instructions or mismatched queries or just misunderstood relationships. Well, they now have updated the data agent. So, you can build an agent with AI. It’s actually just like you would use the skill creator skill to help write a skill. An AI powered assistant that directly in the data agent experience

11:39 directly in the data agent experience that helps build, configure, refine, and validate data agents conversationally. So, it’s a chatbot to build an data agent. It’s literally the thing that builds the thing. Quickly understand your data model with schema exploration. Identify join paths. highlight important columns learn from previously extracted queries of service common joins or filters generate instructions and example queries for the agent for the data source instructions for the data source descriptions and execute and validate. So it really allows you to explore the schema learn

12:10 allows you to explore the schema learn from the query history generate the instructions from that validate it and apply and repeat as needed. This works in the lakehouse as SQL endpoint warehouse mirror database SQL database and event house not semantic models yet. So Mike, this is really cool because I think a lot of people had a lot of problems with the data agents because they were not having the desired results because you didn’t know what instructions to write,, or how the instructions should be. Again, you don’t back to my point earlier. earlier. Yeah. Yeah. You don’t know what you don’t know. Okay, great. You gave me this tool

12:41 Okay, great. You gave me this tool called a data agent, but what do you do with it? What should I be like there’s a large learning curve of like what information should be provided to the data agent? What does it need to have in order for it to actually produce real results to the end users? The one feature this one Tommy I’m really excited about is two things schema exploration. I think Microsoft has done a very poor job in pageionate report builder and PowerBI report building in the service even in desktop for that matter. it’s there’s no

13:12 for that matter. it’s there’s no way to limit the amount of viewer information that’s inside your view in those building experiences. So the fact that there’s no perspectives there is a big miss in my opinion. And so as I’m looking at companies now, models are pretty large. They have a number of fact tables. There’s a lot of dimensions in them. Not every fact table works with every dimension in all situations. So there’s this really a neat way of like I need the agent to explore and understand how the model is comprised and look at the schema def tell me

13:43 and look at the schema def tell me what’s what’s available. I’m typically I’m going to a model with a specific question question in this domain Tommy we’ve talked about this a lot in the podcast in this domain base of hey we have many domains of semantic models right we’re having the ops model the sales model the marketing model these are it may have the same dimensions in them but it’s that data set is is catered to that area or domain of information well how do which domain to go

14:14 how do which domain to go access access with what with what questions, right? right? It’s like it’s like you need like a router at the top level of your organization that helps you route your requests into different domains. And so none of this exists, but the data agent seems like it’s starting to fulfill some of that with the schema capabilities. the other one that I really like here is this query history. So,, if this is, and this is,, this is what I’ll have to be careful here because

14:44 have to be careful here because just because a query is executed doesn’t mean it’s a best p best practice or a best pattern. But But having the agent actually go look at the information like the log analytics of the model,, makes sense, right? This what are the most recent queries running against this model? that that would be a great area for the agent to give information about a model. What are your thoughts? thoughts? No, I I think it goes back to context, right? Where before the data agents especially, we’re living in a vacuum. Every time you would update the

15:15 vacuum. Every time you would update the instructions or try a new prompt, it really didn’t have all that history. And the fact now it’s like, hey, let’s look at the last 10. These two were good. This is how I want you to be. I like these four. This is the four that you need to really focus on. I was like, okay, refine itself that way where because this has context. It’s not just your prompting ability at that point. You have to have context for this to work. work. And and this is I think where we’re going back to Tommy is this whole idea of like there’s this business context, right? It’s it’s where does that live currently? That lives in like random

15:46 currently? That lives in like random documents. It’s in in maybe like a number of semantic models. It might be in someone’s head. and there there’s just a lot of this information that doesn’t really centralize anywhere., and I’ve been hearing a lot of people I’ve been doing a lot of shorts from our shows., so from those shorts, I’m getting a lot of people pushing back around like this., there’s a lot of conversation happening around what what does the central system look like? What does, you central system look like? What does,, a relationship and the definition know, a relationship and the definition of these tables come from? Where does

16:17 of these tables come from? Where does that need to actually live? And I think the question here or I think the answer to this one is much more broad than people think. Mhm. When you think about how you generate like just let’s talk a very simple example of a model, the table, the relationship, the dimension, and maybe a measure or two. If you just take those couple items, depending on what those things are, they could be built in multiple different places. The table could be built in Power Query, it could built be built in Fabric, it could be

16:47 built be built in Fabric, it could be built in data bricks, or it could be built in Snowflake. So where you build the thing the thing is the story of truth for that thing. So you need to be able to say look I want to continue building in data bricks and have that table exist as a single item item like that’s that is that system controls the definition of that table. You want to add a column you go back to data bricks you add a column. But if I add relationships where does the relationship need to live? Does that live in the semantic model? Do we have multiple different relationships for how

17:17 multiple different relationships for how different tables live? Should that be defined in metrics views in data bricks or should we divide this semantic model? So I think I think again you want to look at this holistic view of like where does the source of truth live on every single item of the semantics elements and that’s part of like what’s going on here. I think we’re starting to get a little bit more insight as to what’s happening here. there’s still not this centrally governed area right that I think we need and I I I really think there’s a story around the enterprise semantic model no data

17:48 enterprise semantic model no data included it’s just the semantics the definition of everything that needs to live somewhere and no matter what tool you use data bricks snowflake semantic models in PowerBI there should be different areas where different pieces of the story of the truth lives where it comes from I really feel like one lake has the potential to be that place. I just don’t think we’re there yet. Both from a user mostly from a user interface point of view and then two from the actual backend how it actually work. But one lake to me is a great candidate for

18:21 lake to me is a great candidate for where that can be. So I agree with you there’s no master universal universally accepted platform or in a sense location architecture for us to do that. But I think it exists. I think the the foundation’s there. Just build the house. house. So yeah. yeah, but I think data agents are good. We have two more. Mike, I think this Yeah. So next one we got is hey, you can automate your recurring

18:51 hey, you can automate your recurring business logic with scheduled user data functions in Microsoft Fabric. And what this is is user data functions as we know let developers go through custom business logic and run it in close to their data in powerba reports or in other ways using Python with the new thing that’s updated is scheduling which allows the functions that can run automatically on a reoccurring cadence instead of requiring a manual execution or some external orchestration from other toolings. So teams can rely on

19:23 other toolings. So teams can rely on simpler automation. We don’t have to worry about again workflows. The workspace centered operations you can configure, run and monitor within fabric. You can use reusable logic. You can track the execution history and outcomes and faster issue detection. So for example, things that Microsoft said you can automate, run data quality checks. This goes back to Claw Anthropic’s article about how they use self-service analytics. Analyze customer feedback on a regular cadence. Evaluate business rules and generating events. synchronize external systems data to

19:54 synchronize external systems data to fabric, send notifications and archive stale records. And again, this is you can configure the schedule directly from the user data to functions interface. Mike, I think this is a big deal. I think this is going to be one of the things that, dare I say maybe used more with user data functions than in a PowerBI report itself. Well, one thing I’ll just note here is this is just now covering some basic functionality that does not exist in

20:25 functionality that does not exist in well, it did exist in Azure functions, but this never really exists here, right? right now, Tommy, I have a number of notebooks that are making some Python calls and doing some things. So, fabric, this actually dubtales really well into our topic today. Fabric backend as a service. Yeah. right? Fabric as a backend. This is essential for actually having that become true because you do need to have regularly scheduled tasks that run that hydrate some data that load some

20:56 hydrate some data that load some information. So not having an application to be able to like regularly schedule the functions. Now you could have run a a pipeline, you could schedule a notebook, you can still run Python over there, but that’s not efficient and cheap like a fabric userdefined function. Fabric’s UDFs are

21:14 userdefined function. Fabric’s UDFs are like pretty cheap to run and because they’re just a function and right if you’re going to run something in Azure user fabric Azure functions are extremely cheap to run. They’re like nothing. You can run millions of them and they just run on hardware. They just go they just do a lot of great things. So having that same functionality here is great. is great. My one gripe with this one, Tommy, boy, boy, this is good. Oh boy. My one my one gripe with this one. This is good. We definitely need this feature.

21:46 definitely need this feature. Where does this thing link into my fabric Ray projects where my fabric apps, right? So, I really need these functions, these userdefined functions to connect to Rayfin. So, scheduling things, great. Love it. Not having to use a pipeline or a notebook. Love it. Great. It keeps my cost down. It makes it run efficiently. And I can do my testing in a notebook, run the test there, configure what I need to configure, but the function should be able to run all on its own. So I I

22:17 able to run all on its own. So I I absolutely love this feature. Let me don’t get me wrong, I love this feature. I do think this is a preparation step for more Rayfin usefulness. Yeah, Yeah, but I need these dogon functions to work with a Rayfin project. I need Rayfen to have a front end, a SQL database, a semantic model, and a whole fabric userdefined functions layer that I can work with. All that needs to work together for it to be like really effective for me because right now it’s just a novelty. Yeah, I think this is one of the first

22:48 Yeah, I think this is one of the first steps to get there. because yeah, I’m trying to think could ra actually use user data functions since Python, but no, I think there needs to be more native integration there because I think we’re going to see more with this backend with user data functions. I don’t and they’re probably doing this to me because I don’t think there’s been a lot of teams really widely adopting the button or user interface part of user data functions. Like I think some teams use it, but I have not seen this wide adoption. So they’re like, “Okay, maybe

23:18 adoption. So they’re like, “Okay, maybe UDFs work better from a backend point of view.” So I that’s probably where we’re going to see it. I think you’re on the right track there. We’ll see. Again, I think there what from what I see effort is being put into the appropriate places and it feels like we’re getting the right mix of features that we need in order for this to like become effective and useful. So very happy about that. Okay, awesome. Last topic here, Tommy, before we get to the main topic here. What’s your other

23:48 the main topic here. What’s your other news item? Quick. So, Mike, I since you’ve been gone, since you’ve been gone., since you’ve been gone. Yeah. Right. I have been really really into my blog,, prompting bi. com and trying to do two or three articles a week because all the things we talk about, I’ve been thinking for a long time like, oh, this is I want to expand on this. Well, what? I said, I’m going to put my money where my mouth is and actually start writing that. So I have before you go into the article itself I I want to ask it Tommy I one I’ve been enjoying your reads and your articles.

24:18 enjoying your reads and your articles. So one I’ve I’ve been enjoying those. The question I have though is Tommy are I feel like this is a thing that Tommy you would do in a morning where you would get up I’m going to project something here and you can tell how wrong I am on on what you’re doing here. Tommy, you have a lot of good thoughts around these areas. Like again, the one that you just published here, agent skills or the new theme files, right? This is I’m making a parallel and an analogy here. Makes a ton of sense to me., also have been loving the blog. I’m liking

24:49 have been loving the blog. I’m liking the the post article. I was retweeting some yesterday. I thought these are really good. Yeah. Yeah. I feel like this is a thing where Tommy wakes up in the morning, you’ve got ideas on the head, and you go outside and you start talking with Notion or your agent. You say, “Hey, I want to just can talk to you, just chat with you about this conversation.” And so like refining the idea with an agent in in place. Hey, I want to talk about this topic. This is interesting to me and then you’re just like working through the idea with the agent and having the agent like you and it build out the structure of the article

25:20 build out the structure of the article and have it help you write it. Is that what you’re doing with this, Tom? Because it feels you are spot on, my friend. So, no. And I’ll explain that workflow because understand your process on this. not the greatest writer. I can talk pretty well,, like I I feel like I have great ideas. Yeah. Play to my strengths, but writing has always been a very a struggle in terms of my mind’s always going faster usually than my keyboard. And a lot of times too, it’s remembering a lot of things that I’ve said on the podcast and I want to make sure I get

25:50 podcast and I want to make sure I get it. So, I actually created a few skills. one that actually looks at our transcript from previous chats and I go all right I remember like for example I know this is actually something I brought up before where I made this argument about theme files and skills and I’m like I want to expand on this so looked at some of my points that I said on the podcast and then this skill isn’t very interactive where it’s like okay how do we want to sort this here’s some ideas and then as I go through I’m like listen this is based on how teams use PowerBI this is how going

26:21 PowerBI this is how going through like it’s not just a markdown file. One of the big things also one of the big things we talk about is drift because theme files always got wrong I feel like the same with skills and then it walks through and then it helps me really write out from playbook to what my I want my takeaways to be because it’s allowing me to really expand on the thoughts that I’ve had. that’s what I did a lot with my harness style. I actually did a three-part series on the like the project hub and how that all works.

26:52 all works. And that was a great little walkthrough. Again, it’s like talking about the project hub, it talks about the assistance and then about really doing an MCP execution, all three of those working together. and it’s been great. But yeah, so most times when I do what I do Monday, Wednesday, Fridays, I bike in the morning. Yep. take my kids out in the morning to church, come back, and then it’s like, “All right, now I’m fresh. Let’s go through. Here’s my drafts.” And again, it’s all in cursor because I It’s a GitHub website. Really, you couldn’t

27:23 a GitHub website. Really, you couldn’t tell,? I love you can’t tell you you can’t tell that it’s a GitHub website, but the fact that it’s GitHub, it’s it now means agent has full control. You can talk to it. You can have it pull in things from notion and like, “Hey, we’re writing a new blog post. Here’s what it’s going to look like. Formula formulate all this. I I was just reading listening to another the CEO of Versell was talking about this exact thing not not necessarily websites and what you’re doing here Tommy but more about the idea that that you are building like Tommy you have this desire to build a blog post content

27:54 this desire to build a blog post content regular content coming out from here what you’re able to do is you’re you’re now building systems and processes around this that incorporate agents but still Tommy’s voice it’s still Tommy’s thing but the steps of like I want to go do these things. I need to get my ideas out on this. Yeah. Yeah. Notion thing. So now what we’re seeing, I think, is this this era of customizing workflows specifically for what we want to build inside our businesses and custom person. Yeah. And and it it that is but that’s

28:25 Yeah. And and it it that is but that’s our secret sauce, right? right? Our next era of business development is taking your knowledge of your business and what you do and how you do things. Like Tommy, you said, I like talking. I like chitchatting. And like that’s more of my medium. I’m better at that. Well, great. Lean into that. Do that more. And I’m finding myself I’m talking to my agent more and more. And I’m finding I want to have conversations with about what I want to get done as opposed to like typing things into a keyboard or typing things into a prompt. I’m even shifting how I like to do things locally. I’m doing a lot more talking with my agent. so anyways, didn’t mean to disrupt

28:55 so anyways, didn’t mean to disrupt you. I thought this is a really cool idea. I like what you’re doing here. Let’s go talk about agent skills are the new theme files. Yeah. And one thing we’re just basically the biggest things that I kind basically the biggest things that I mentioned was of mentioned was when you remember theme files the biggest part was hey I have a theme file for the team or the department. One of the things I did back in the day long time ago was every department and app in PowerBI had a different theme file because sales was green more of a green red was operations etc. But the problem was when something updated or we want to change

29:26 something updated or we want to change the theme file, it was so hard because my theme file may be changed in the template but existing reports were not. So I think there’s such an equivalent there with when you think of agent skills, especially when you’re working with the team. How do you actually understand what is the right one to use? If there’s a change, what do you actually do to make an update? Even if you have different computers, Mike, like I have claude desktop and then any other like copilot or claude code don’t share the same folder.

29:56 share the same folder. So how do you make sure those stay in sync? So we basically just talking over how do you actually what does governance look like which I equated to using theme files in PowerBI and then how do you inventory how do you locate or centralize where those themes are and how do you actually begin to start building them out. so I we talked about testing drift and just really how do you actually even have time for creativity which I called garage time. yeah it’s been really fun. Mike, one

30:26 yeah it’s been really fun. Mike, one of the things you’ll love, I just want to mention in the blog article is we did a training fabric series and I wanted to use baseball stats. You’re like, Mike, you’re like, Tommy,, that doesn’t really jive because baseball stats are pretty specific, right? They’re not sales or marketing. So, I took all my existing projects, went through fabric task flow studio, and built an entire architecture of what if you ran baseball, not the baseball team. So, concessions, season tickets, accounts, CRM. So, all my demos now are

30:59 accounts, CRM. So, all my demos now are this giant database of food vendors, concessions, things being sold, ticket seats, events that actually happen, and updates every day. So, it’s actually business data, just happen to be a baseball team. So, again, I thought you’d be proud of that. Fitting your area really, really well there on that side of things. So, anyways, yeah. So, all right, Mike. I think we did enough updates for the last two and a half weeks. I know there’s a lot more, but we got a good one. This is your idea. And Mike, I may grill you on some of this, but let’s

31:30 grill you on some of this, but let’s talk about fabric as a backend or a fob because because fob. Yeah. Right. So, I don’t know the best way to say it. So, Mike, walk me through this. So, this was a journey that I started having. I started having some aha moments. For me, the light bulb moment

31:48 moments. For me, the light bulb moment for me when this was really becoming a reality was when I could start using a fabric SQL server inside fabric. So there’s like this you for most of the time most of this has been like lakehouse and there’s a data warehouse but a lot of these things were like batch there’s a little bit of streaming things where you could stream data in but those were expensive just because there was an always on machine that needed to be always there to catch information and always stream it through when Microsoft introduced the SQL server

32:18 when Microsoft introduced the SQL server and I was like a very actually I was quite skeptical I was like why would I want SQL server inside fabric this doesn’t make sense and Maybe there was a vision in Microsoft where fabric was going to become a backend but it wasn’t communicated. None of the MVPs were being told about this. It just like SQL Server just showed up and I thought okay this is interesting to see where this is going to go and I started playing with it and I built an application for for me personally. I was again this is Tommy you and I were so the need was

32:49 is Tommy you and I were so the need was I I’ll give you the story here by the back end. Tommy and I were have been doing the podcast for a long time and one of those things has been well we want to be able to like track our usage and like what’s our impact and what’s our scores this is every year around March we go figure out MVP stuff we want to size things up size the market up what are we actually doing how do we align to our media versus other people’s media or whatever so I built this thing called content nudge and I was like what I’m going to I could have used air table

33:20 going to I could have used air table right I was building something agentically. I was just starting my agentic space and I was just trying to build things. A and at the time in January, February time frame, agents were starting to get pretty capable., I really had my aha moment in like end of November, beginning of December where things really started clicking for me. And I was like, okay, this this is going to really change how we do business., and so I started around there and I I started thinking, well, how could I connect a SQL database to my agent to my

33:50 connect a SQL database to my agent to my app? Can I actually make an app registration? I said, I’ll just spend like a a weekend figure. It literally took me like 30 minutes, an hour maybe max. And I was able to tell the agent, here’s my SQL server. Here’s the here’s the connection pattern I want to use. Here’s the app registration. Here’s, you Here’s the app registration. Here’s,, I’m going to make an environment know, I’m going to make an environment file. We’ll put the app partition there. And very quickly, I had a full SQL database made with tables that I needed to build what we’re calling content nudge, a website that’s going to allow

34:20 nudge, a website that’s going to allow us to be able to track all of our website data. And so I made API calls to like YouTube to X, Twitter,, and and LinkedIn and and get us and absorb data. And now I’ve got it so that I can export files from each of these systems and automatically upload them. And then I can measure my feed versus every other’s feed. So I have a number of like our YouTube channel and then a lot of other YouTube channels and I can monitor them. I can see how I could perform against them. What are they doing? I now have enough data to really observe well I know which YouTube channels that

34:51 know which YouTube channels that are in the PowerBI space and which ones are using are paying for advertising because I can see it in the data a little bit more. I I observe this now. So all of this to help me be better and more narrowed around, okay, do we do more shorts? Is that an impactful thing? do we get more views because of them? And so using all these tools and techniques, I was able to really dial in what is working well for our system, our team as as we build these things. So that was my first entry into this. And since

35:24 my first entry into this. And since January, February, this backend system has just gotten stronger. It just makes it more relevant. But for me, a lot of this back end as a service was the SQL database needed to be there. It’s an always on. It’s assid transactions in a database that can be edit update records inside a database correctly as if it was a database. And I’m now moving forward from just using SQL database endpoint to now actually using SQL database with the GraphQL. So using GraphQL on top of the SQL database gives

35:54 GraphQL on top of the SQL database gives me a pure API endpoint that I can just talk to. So I’ll just pause right there, Tommy. That’s my exploration. And let’s be clear here too. this fob or foss this fabric as a backend it’s not an official Microsoft pattern right they’ve not coined that term this is more communitydriven or rather m the carlo driven correct no I think with yes and no I’ll say I don’t think I’ve heard that at all from Microsoft you you’re now starting to hear it with the with rafin they’re starting to to

36:27 the with rafin they’re starting to to cover that right they’re starting to communicate fabric as a backend but they’re not using Bob specifically in this language, but you heard it here first. I I guarantee you’re going to say it now and now everyone else on the internet is going to pick up fabric as a back end. So I’ll make sure I definitely put the acronym FAB. We’ll know how well people listen. Yeah. Yeah. We’ll see how many the test the true test will be here is how many people are actually going to start talking about FAB now as a blog post and other things. So anything that comes out after this date is now marked. We we

36:58 after this date is now marked. We we have the official release date here today. today. And Mike, I think I’m going to go back here to everything you said. I you’ve shown me content nudge. You’ve shown me how it works. You show me the UI to the data actually coming in. And it’s really incredible. It absolutely meets the needs that you have and the workflow that you’re trying to do. Saving time, but also centralizing your data, right? Like making your data work for you at the end of the day. But let me ask you here though too. I’m going to go back

37:29 here though too. I’m going to go back always to my philosophy around technology around the iPad because I feel like this is such an appropriate one here and I need to ask you again. Why am I bringing up the iPad? When Steve Jobs came out with the iPad before he actually introduced it, he said, “If we’re going to ever create a tablet, it has to do things better than the other technology we have. It needs to be better at watching movies than a phone and it should be better better at browsing the web than a computer. If it doesn’t do anything better, it has no

37:59 doesn’t do anything better, it has no reason to exist. And I remember watching that as a kid or a teenager adult whenever they came out 29 2009 and it’s really stuck with me. So let’s take that analogy and if fabric as a backend is going to be a default way at this point in time, does it do anything better? What is it better at than your traditional back end? Yeah. Yeah. Yeah. So, so this I think there’s this idea of like the ease of getting things wired together. So,

38:29 wired together. So, one of the things that’s always been good about fabric has been if you just bring it to the lakehouse, once it’s in the lakehouse, I have it’s it’s inside fabric, right? So, getting the data into fabric has been like the tricky part. What happens when you generate your data in fabric? What happens when your data starts in fabric? the whole down stream reporting system thing gets much easier once you have a SQL database that has the data in it and it just works. I’m not worried about mirroring. I’m not worried about getting things over to

38:59 worried about getting things over to lakehouses. I’m not worried about pipelines. There’s a lot of like extra effort that you have to use to get data out of systems. It just works. It’s just inside fabric. So for me, the story has always been about fabric is there’s so many different tools at your disposal. It’s like a Swiss Army knife, but there’s so there’s a such a large repertoire of tools. I can just pick off the shelf which ones I need and just build things. And this is actually really good for your task flow studio, Tommy, because you can describe what you want to build and give it to Task Flow Studio and have it stub out most

39:30 Studio and have it stub out most of the infrastructure for you. So that that to me that’s one area that’s been really helpful. When you go into Azure portal data platform, right, you could do all the same things. You could build functions there. You could have a web app over there. You could have a SQL database over there. It really requires knowledge of how to bring those items together and have them all communicate together. There’s a there’s a frictionless part around, you there’s a frictionless part around,, if they all live in the same know, if they all live in the same workspace, the workspace has an identity

40:00 workspace, the workspace has an identity and you could just share that across all the items and artifacts so that it just works. And and that’s something that I find that’s very refreshing. all my management and authorization becomes easier to manage. I’m not building a bunch of app registrations. It’s all managed in one portal. So for me, the ease of clicking the button, getting the SQL database, the ease of configuring a workspace identity and adding it to the database and also adding that to functions and other things that I want to do. You other things that I want to do., the fact that the SQL database

40:31 know, the fact that the SQL database lives right next to my notebooks. I can easily go access those notebooks. There’s sample code. How do I connect to this SQL database? you could just drop it in like the code in the notebook just shows up and I’m connecting to the SQL database without firewalls and all these other stuff. So there’s this whole idea of like Microsoft is owning the management of the security and the integration of those different services together in fabric that just feels so much less friction than if I was doing it in the Azure portal versus fabric. Does it feel less friction because

41:01 Does it feel less friction because you’re already in the Microsoft playground already using PowerBI or is it really just that alone? So, one of the things I’m writing down as you’re talking is if fabric has a backend. FOB f are we are we going to land on fab or fab here. So, just because I’m not I think fab is probably going to be what people are going to say the most. Okay. So, let’s say fab. So, if So, if you’re saying you’re saying you say SAS software as a service, you say SAS,

41:32 software as a service, you say SAS, right? right? Right. Right. reporting as a service. Raz, Raz, but I keep wanting to go with fabric. So, So, yeah. So, so fab. Let’s go fab. Yeah., it fits then a little bit. Okay. So, let’s say we’re gonna for now on everyone knows fab as Microsoft adopts this. Say fab. That is the correct prun. Well, what? I’m just going to keep moving on here if I can’t even pronounce pronunciate., cool. So, let me back up here. Mike, if FAB is

42:03 let me back up here. Mike, if FAB is going to be what you’re saying it is, and it can be a backend, theoretically, someone can come in without having any PowerBI infrastructure, any semantic model needs. They could be using Salesforce and Tableau. And what I’m hearing is Fab could be a great solution to build your application on,

42:24 solution to build your application on, right? where it there’s no business intelligence required to get started with fab or to say we want to recommend fab as your application layer to me if that is the case and again if this is what I’m hearing from you that it’s not that fab works because I’m already using fabric is fab works because I already I can easily set up my entire infrastructure for applications talking to each other regardless if I ever use powerbi or notebooks or not is where I

42:55 powerbi or notebooks or not is where I think you’re saying this is potential. So I guess the question for you and what I’m trying to talk through myself here is can fab stand alone from everything else in in fabric. Say me again. Can fab fabric as a backend stand alone than actually doing any business intelligence work? So can it stand alone? Okay, I think so. So, and I think this is a big deal if that’s the case.

43:25 big deal if that’s the case. I think I think yes, because it’s it’s bounded by So, there’s a couple things, right? Mhm. Mhm. The the the fab is bounded by a workspace. So, workspace is your security boundary inside Azure portal. Your security boundary is typically like a resource group. So, it’s similar concept just changed over to fabric which is a workspace as the item. So, anything you want to have as a backend, you could actually have a workspace that is a backend element.,

43:55 workspace that is a backend element., one of the other pieces I’ll note here as well, Tommy, is another ease of using fabric is there’s a little bit easier use of using fabric as CI/CD, continuous integration, continuous deployment. So, most of the items, not all, but most of the items now are able to be synchronized with git. That git repo can move between different workspaces. There’s different ways of deployments. We also have deployment pipelines. But because of that, you now automatically get this ability to make a

44:25 automatically get this ability to make a deployment pattern of I’m building something here for the developers. I want to go give it out and test it and I want to move it into production. So these kinds of patterns exist work are all being integrated into fabric. So you could have a series of workspaces that are just your application. And then the neat part about this is you could build a separate workspace that is your analytics workspace and you can mirror and copy and get data directly out of your existing data sets and it’s just it’s just less friction. It’s less stuff to

44:55 just less friction. It’s less stuff to worry about. When you build apps and you have an app registration involved, those tokens expire at some point in time in the future and so someone has to monitor that. there’s always a person involved with some of those pieces when you’re using app registrations unless you’re using like managed identities and you’re it again the setup just gets harder when you add more automation and authentication to everything across the system. It’s just you just got to know how to do it and most people haven’t.

45:25 how to do it and most people haven’t. And I think this is where fabric is taking the load of some of these more complicated concepts and just giving you a simpler easy to use user interface that makes it useful to us. And I think this is a big point here because when you initially brought up this concept of fab my first thought is like are is Mike trying to force an analytical platform into an operational role where something’s like basically hey we have PowerBI but it also you can build an app with it which sounds odd where the only benefit would be because of the work with PowerBI but I think what I’m

45:57 work with PowerBI but I think what I’m hearing and what I’ve seen myself too is not only could fab stand alone without doing any of the nor your the standard things you would think of doing in fabric, right? When I think I’m going to have a company adopt fabric, why would you do that normally? You’re going to use PowerBI. You’re going to do semantic modeling. You’re going to integrate your data. data. Well, we’re talking about this whole other funnel, this whole other process and infrastructure for my applications where in a sense those two don’t have to

46:27 where in a sense those two don’t have to talk or ever overlap be right. So there’s one thing to say it’s like well fabric as a backend works really well if you already have fabric in PowerBI because that great integration but then there’s some trade-offs right but I think you also said a big point here too normally if you’re building an application there’s API calls you have to look for your database we’re going to store this there’s a lot of different areas and services that you need to purchase and store and usually they’re not super centralized but that’s not the

46:59 not super centralized but that’s not the case with fabric as a back end. And I think the other selling point here, Mike, is with fabric task flow studio or or fabric task flows with cloud agents and the API integration, Mike, and with Rafen, you can build that infrastructure without in a sense building it like the integration with AI is so important here because I can have that entire and again widgets and everything running

47:30 widgets and everything running using community tools or using the Microsoft provided API to actually use the MCP and actually can build that infrastructure. It doesn’t have to be for building a report. This is incredible to me, Mike. I I I think this is where things are going to go and I think this is where it’s going to get better. The same what I see here is happening is the same trend in democratizing building reports is now happening to building apps with agents. This is the same pattern

48:01 agents. This is the same pattern right we need a system in the back end. You can go talk to your agent and say hey go build me an app and it’s going to ask it’s going to recommend hey we’re going to go use air tableable for our backend. We’re going to use this API layer to serve APIs. We’re going to use this we’re going to host it in this kind this we’re going to host it in this app service. So the the agents are of app service. So the the agents are already understanding there are systems in place, but those are all like separately separate tools that are all being stitched together that makes it easy for that agent to be able to build agentic experiences on top of things. Once Microsoft builds a bit more of this

48:32 Once Microsoft builds a bit more of this agentic experience building things with fabric as the back end, it makes it so much easier. I can just talk the APIs, the agents can go do things., one of the other aha moments I had in this early time frame, call this around February, March time frame, I started not wanting to write a bunch of Python and notebooks. I do love notebooks. I love using Spark and Pispark and just pure Python, but I also wanted the agent to write the notebook for me. I wanted to describe the action or the engineer that I wanted and then I

49:03 or the engineer that I wanted and then I could go let it run and then I can go test the output and see if it runs what I want. So, I was doing a lot more of that. I’m actually giving agents access to an API layer. Go talk to my fabric. Go create this notebook or I create the notebook, go pull it down, make the updates, push it back up. So, I was doing a lot more things like that as well where I’m I’m directly working with an agent to build these things. Now, one concept here that I think was is really relevant here. We’ve been talking about this for a while, Tommy. Agents are not to be used in a area of

49:39 are not to be used in a area of business that is already known. What do by this? by this? Hold on. Agents are not meant to be not meant to be or not to be used are not to be used in an area that’s already known and discovered. It’s let me explain here. Yeah, if like for example, I could have an agent autogenerate a website for me. Every time a user visits a website, I don’t want the website to be autogenerated every single time for the user. I want the agent to help me design and build agentically generate for me

50:11 and build agentically generate for me the website. I want to run the website on commodity hardware, cheap hardware, make it easy to run, right? So if if there is a something that is unknown, a problem, something in the data, an issue in the data, we want to give agents access to help us reason through, think through, build with that data or system, right? So fabric as a back end, right? I don’t have an app today. I need an app. Agent, help me figure out how we

50:42 an app. Agent, help me figure out how we would build this in fabric. What’s the design? Hey, agent, help me think through the design of these tables. Here’s what I want this. Here’s the relationships of my data and what I want to capture. Hey, I’m building a time card app. card app. I have customers. I have employees. I have rates by customer. Like, I need all this information to exist together. Agent, come up. Help me design a star schema so that this will work. Give it the constraints. Give it the things that you don’t understand or you need help reasoning through and thinking through. Let the agent plan with you. Build that

51:13 Let the agent plan with you. Build that stuff with you. These are unknown aspects of the development cycle. That’s where the agent’s really effective and super useful and very good. The other part of this is once what those requirements are, then you have the agent help you build it and then run everything on commodity hardware. And what by commodity hardware is build deterministic systems. Build systems that we’re going to run the same way every single time without the requirements of an agent. Agents are there. And this is what we’ve been talking about in the podcast for months

51:44 talking about in the podcast for months now now is the concept of the creator agent. Use agents to help you create things and from there you’re going to get out really good useful outputs. So tie this back for me Mike because you said your statement was agents are not to be used in things that are already known which correct if you if if you if you for example a lot of times I see Microsoft messaging what’s my sales for last month

52:14 what’s my sales for last month wrong use for an agent don’t use that that’s not the right you can ask that once but a better question would be is help me build a report or a dashboard that illustrates my scales, my sales for the last month, and I want to slice it by this, this, and this. Right? That’s an appropriate request. The the request is help me build the report that regularly gives me the same answer over and over and over again. That’s what the analysts should be using agents for,

52:44 analysts should be using agents for, right? That’s what our business users should be using agents for. Instead, all the messaging I’m hearing from Microsoft is, and I think that the reason they’re messaging this way is because they don’t know how to do it well yet. They’re not there yet. It’s still a work in progress. So, that’s that’s the

53:00 progress. So, that’s that’s the questions that we want. I want the agent to build me the report that I go use every day. I don’t want the agent resolving the answer to me every day because then I can test it. I can verify it. I can look at it. I can have other engineers look at the the same thing to make sure that it’s right., the higher up that report goes, the more the larger the number of eyeballs on that report. The more rigor we’re gonna have to put into that report to make sure it’s right. I will modify your statement here because I see what you’re saying here,

53:30 because I see what you’re saying here, but I I would disagree that it should not be used for,, things that are already known. I would argue and modify this to say agents are not to be used for ambiguous one-time queries, right? Because that’s what you’re saying with why would I not ask it to do hey just show me sales or do a one-time thing that I’m not clear on yet. That’s where agents don’t really shine. It needs clear instructions and a clear workflow. When I just give the world to it to say, “Hey, you have access to a model. Just answer questions.” You’re not going to get

54:01 questions.” You’re not going to get great great answers here. But Mike, let’s go back. Well, hold on. Like, yeah, hold on. I agree with you. And that’s what I’m that’s my point though. My point saying I’m modifying a statement. Yeah. Yeah. Yeah. It’s it’s this anything ambiguous anything that’s one time time that that question it’s the difference. It’s also it’s also educating our users of these systems right our users of these systems also need to be more educated in in the idea that what questions should I be asking to my

54:32 questions should I be asking to my agent? And again, I’m gonna lean on the the the miss of Microsoft here is they’ve been touting, well,, ask it these questions. Ask it what you want it to do or ask it the that’s the wrong questions we should be asking. That’s not what we need. And this is why I really like this AI for data agents because it’s now starting to move away from these known aspects. Right, Tommy? You and I built a semantic model. We know what’s going on. We know the relationships and how it’s working. the lack of tooling today inside fabric and other tools don’t tell you how the

55:03 and other tools don’t tell you how the model is stuck together, how it’s shaped, how it’s pulled together, right? So, because of that lack of pulled together of the model, I now need to be able to like have an agent help me build these things. So, that that’s kind build these things. So, that that’s where I really see this as going is of where I really see this as going is if you even extrapolate further and say, okay, I want agent to help me build the back end of things. I want agents to build an application for me. I want to run it in fabric. agents should help me create all these things. Donald’s is spot on point right here. Donald in the chat, he said the end goal should be

55:33 chat, he said the end goal should be code whenever possible, right? The end go, the output is code. It’s testable. It’s repeatable. It can run over and over again. The code is really useful. Now, who’s going to be doing the testing and evaluation? You can argue there. It’s it’s only people. It’s a mix of people and agents. it’s a different agent, right? So you use claw to create and you maybe use GitHub copilot to test or something some mix thereof. You use opus to build and you use fable to test it. Like you can do different combinations there. But at the end of

56:03 combinations there. But at the end of the day when you produce code you get this really useful regular output that can be used every single time. And so I think that’s really where I’m going with this one is I want fabric to support more of my backend because it’s easy to use. I can build apps with it now and I get a really reliable structure of different tools that I can easily bring together and I’m not I’m not overly managing managing app registrations,

56:34 app registrations, secrets, other things that in the past are just it’s not impossible. You can totally do it. It’s just an extra step. So here before we just only plays the the laurels of fab here because I I do agree with everything you’re saying and honestly this all sounds great in theory. However, where I have some trepidation where I have some concern is if I’m deploying this out to the masses. Let’s say I go to sears doesn’t exist.

57:04 Let’s say I go to sears doesn’t exist. Let’s just use Sears. Let’s say Sears still exists. I’m like we’re going to create an application for you for everyone to access. It’s all going to run through fabric, right? It’s going to the fabric as a backend is going to run people going in the catalog for Sears. And for younger folks, Sears was actually a big deal, the biggest company in the world at one point. Now the concern I have here Mike is with thousands of requests coming to a fabric if you were to actually deploy something like whether it was Dillards or a giant

57:34 like whether it was Dillards or a giant company where people are accessing this and inputting and requesting what happens to your CUS at this point right what h what capacity are you spending for this compared to legacy systems that I don’t know if we’ve really discovered is the scalability From a cost point of view, can it build the infrastructure and handle the data? 100%. 100%. But the thing that I would be worried about is if I were to go to a customer today say, “We’re going to build your

58:04 today say, “We’re going to build your public facing app. All of it’s going to run on fabric as a backend on fab. That’s going to be the architecture that we’ve chosen for you.” Well, what are they expecting for a cost? And I think this to me is where I have trepidation with this before we’re going to say, “Yep, we are going to now go this direction.” And for me, there’s a lot of testing that I feel like needs to go into this. But Mike, convince me otherwise. I think I think your cost here is not just purely cost on running the assets themselves. There’s a cost to that. There’s also a

58:34 There’s a cost to that. There’s also a cost to how much time does it take your engineers to build something, create things, pull it all together, right? So there’s actually So there’s a tangible cost, which is okay, I’ve built the system, I’m going to run it. Now that application runs for 50, 7500 bucks a month. Currently in a project right now, now, the app has been overbuilt, substantially overbuilt. Running one environment is costing the customer thousands of dollars a month just to run an app that’s not even be used for many people. So we came in and

59:05 used for many people. So we came in and we rebuilt the app per modern standards, updating things. Now the app runs for 50 bucks a month. So we’ve substantially reduced the cost of the app. Now, we also know that we’re going to pay for fabric. So, we’re buying a fabric license and we can have autoscale turned on on that license. So, if we need bursty things for like the SQL databases or things that are on there, we can get those things. But what I’m finding is a lot of companies are already paying enough fabric to run these things. If you look at your margin

59:36 these things. If you look at your margin on your fabric experiences, there’s a lot of margin left there that that you’re leaving on the table. You’re paying for compute you’re not technically using all the time. Yeah. So I’m finding a lot of this SQL database area where it’s very useful there. and most of these application pieces functions GraphQL they don’t add a lot of cost to things in general. They they feel very reasonably. So I would argue the time you spend mucking around with Azure with people they’re people are probably your

60:07 people they’re people are probably your most expensive asset in building that new applications or things. the time of the developers, the engineers to do that, running the app is relatively low cost compared to those those other pieces. So, there’s this hidden cost of how much developer time do you need to get this thing to work? And so, that’s where I think when I look at this situation, I’m thinking, well, I have and Tommy to assuage some of your fears here, I already have applications in production today for customers that are using only

60:38 today for customers that are using only fabric as a backend. So, it’s it’s solid enough that I think I’ve moved and made made the effort to move into let’s pay for one thing. Let’s centralize our effort into just fabric. and so instead of buying multiple other tools that are all costing you a little bit here, a little bit there, billing, pricing, all the we’re now centralizing on just one fabric, which is taking away the a lot of the management need of the application side and we’re just building really good

61:08 and we’re just building really good backend as a services. And the added advantage is all that now exists in fabric now. And it’s very easy to go do analytics now. So I’ve actually saving money now because my analytics pipelines are getting much easier and I’m not burdening production systems. It’s just there. mirroring’s just turned on by default. It just happens. So, I really like this new system. I think we’re going to continue to see more of this. It’s continually getting it’s getting better and better over time. and I think we’re going to continue to see more investment from

61:38 continue to see more investment from Microsoft in this space because I think this is this is the future. The same way we democratized report building with PowerBI is the same thing we’re going to see with applications. We’re gonna we’re gonna democratize app building the same way. Anyways, Anyways, I I love it, Mike. I love this topic and I I think I can’t wait to see more what happens with UDFs and with Rayen because I think that’s going to be an integral part of this working. But Mike, love the conversations today. We’re back into it, man. man. With that being said, thank you very

62:08 With that being said, thank you very much. we really appreciate you listening to the podcast. We hope you’re getting some insight here. Thank you for jumping in and tuning in. For those of you who are in the chat, we really appreciate you making comments and letting us know what’s going on here. We love having the conversation as well. So, please engage with us on our social media platforms and channels., we are trying to be extra attentive to all of our posts on shorts and YouTube videos and everything else that we do across our media platforms. So, if you are using social media and you see our videos come up and you have a comment to say, please comment on them., we’d

62:38 say, please comment on them., we’d love to engage you engage with you where those things are happening and how they’re being used., so we’re a super big fan of Fabric and PowerBI and all these new Edent experiences. We hope you’ve enjoyed this conversation. That being said, Tommy, where else can you find the podcast? You can find us in Apple, Spotify, wherever you get your podcast. Make sure to subscribe and leave a rating. It helps us out a ton. Do you have a question, idea, or topic that you want us to talk about our future episode? Head over to powerbi. tipsodcast. Leave your name and a great question. And finally, join us live every Tuesday

63:09 And finally, join us live every Tuesday and Thursday, a. m. Central on all PowerBI. tips social media channels. Have a great day, everyone, 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 to laugh in the mix. Fabric and A. I get your fix. Explicit measures. Drop the beat now. Kings feel the crowd. Explicit measures.

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.