The Hype of Rayfin – Ep.547

Six weeks after the Rayfin episode, it’s time for the harder conversation. Mike applies the Steve Jobs iPad test — a new product has to be better at something than the things you already own — and mostly passes it. Tommy brings the enterprise objections: security review, maintenance ownership, and what happens when the person who built the app leaves.

News & Announcements

• Stop re-prompting: second brain agent instructions — Tommy’s own write-up of the second-brain workflow from Ep. 544, expanded with the part he thinks matters most: write-back. Doing the work isn’t enough — the agent has to update the second brain with what it did, and that loop is what makes the whole thing compound.

• Securing outbound AI connectivity in SQL Server 2025 — Database engines can now call external AI services with security, permissions, and auditing intact. Keeping the AI call close to the data reduces unnecessary movement and keeps policy enforcement where it belongs.

• Controlling storage costs with OneLake lifecycle management — Mike ranks this as genuinely important: most organizations do a poor job at the file management level, and there’s usually a lot sitting in a tenant that nobody needs daily. Data you touch once a year shouldn’t cost what data you touch hourly does.

• Deep dive into Power BI reporting with the new date picker — Relative ranges that stay dynamic even when visuals are pinned to dashboards, rather than freezing. Tommy has a proof of concept next week that needed exactly this. Currently behind preview features.

• Rayfin on Microsoft Fabric — The main topic’s reference point.

Main Discussion

Topic: Is Rayfin as good as the hype, and what breaks when you try to use it at work

• The licensing discovery. Mike set up Claude to talk to the Power BI MCP server through a service principal, which reaches the semantic model over the XMLA endpoint — no Power BI Pro license needed. He calls it the epiphany of the week.

• It also moves the cost off your capacity. Four data agent queries on an F2 burned about 5,100 CUs in a few minutes. Running the same work through Claude against the XMLA endpoint cost nearly nothing on the Fabric side, shifting consumption to tokens he’s already paying for.

• The gap between idea and working thing is very short. That’s Mike’s core case for Rayfin, and why he compares it to playing a strategy game. Not that it’s easy — that the loop is fast.

• He’s deleting subscriptions. An $85-per-user Dynamics license, a $20 tool, a $90 tool. He only ever used maybe 30% of each, and Fabric is already a sunk cost — so the data lands in Fabric rather than in someone else’s platform.

• Tommy’s maintenance objection, and Mike’s answer. Who maintains the app? Mike’s position: you and the AI. Have it document how the app works and what features exist, and fixes come at the same speed the build did.

• Tommy’s security objection stands. A Power BI person introduces Rayfin to their company, and now IT is looking at a new public URL and a new auth path. Until this is seen in the wild in enterprise settings, he thinks you’re asking for trouble.

• Mike’s counter: you already bet online. If your organization is on SharePoint Online and publishing Power BI reports to the service, Rayfin isn’t a categorically new risk — and there’s a Microsoft security team behind it.

• Don’t lead with replacement. Both agree the wrong opening move is proposing Rayfin as a replacement for existing systems. Proof of concept is the entry point.

• Senior PMs get more powerful, not less. The team of engineers needed to build a large application shrank. The person who can articulate business requirements precisely can now get most of the way to a working thing.

Where it clearly beats a Power BI report

Custom interactions the visuals won’t do, and deployment. Getting a Power BI report embedded into a website or mobile device is fiddly. Rayfin produces a public URL, the user signs in with Fabric, and the app works — authenticated, no embed configuration. For an intake form, a vacation tracker, or a task list, that’s the whole difference.

Adding to the failure-mode library

A workflow worth stealing from the news segment: after an agent fixes something, ask it to reflect on what it learned, distill it, and append it to a documented library of failure modes. Otherwise you pay to solve the same problem again next month.

Looking Forward

Try the service principal plus XMLA route against a semantic model before you provision anything new — the licensing and cost implications are worth knowing before you plan a rollout.

Episode Transcript

0:01 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. Has kings feel the crowd. Explicit measures. Hello Tommy and good morning. Welcome back to the explicit measures podcast. It’s a beautiful summer day. Apparently the weather’s going to be nice today.

0:32 the weather’s going to be nice today. It’s exciting. it was actually Yeah, I agree. yesterday was actually a reprieve from this extreme heat that we’ve been going through here. our our family was I don’t know. It was in the 70s, maybe the 80s or something like that. 80s in the evening or 80s during the day and 70s in the evening or whatever. And one of my family members goes, “It’s cold.” I’m like, “What are you talking about? I’m This is perfect. This is great weather is I’ll take this over the 93 degree days that we’ve been having.

1:02 93 degree days that we’ve been having. So, it’s funny how your body acclimates to like different temperatures fairly quickly. quickly. The the hilarious thing is always in February when it goes from like -10 to 20. You’re like, “Wow, it’s warm out. You can feel it. You’re like, “Oh god, I can wear a t-shirt.” Yes. Exactly. Awesome. Today, our main topic, I’ll get into that and then we’ll jump into a little bit of news here and go through some other items. Our main topic is the hype of Ray Finn. So Ray Finn’s been all all over the the internet. If you haven’t heard

1:33 the the internet. If you haven’t heard about it yet, you should definitely explore it. if you want to, we also have a series of really good I think we did like six or seven. I think we’re up to eight videos right now talking about Rayan, how it works, what’s it what is it built like. So just a whole bunch of really useful knowledge about how to get started with Rayan. It’s actually pretty easy. Tommy and I just did a user group down up in Chicago, down in Chicago. I guess it’s down for me, up for some other people. but we did a a PowerBI user group in Chicago and there we did a deep dive into Rayfen. we

2:05 we did a deep dive into Rayfen. we did multiple demos, made multiple different applications while there, which was fun as well. Yeah. And shout out to John who drove all the way from Ohio just to come see us because he heard us. He listened to us on the podcast and then next thing he’s there for the user group which was awesome. Hopefully it was worth his money which was free. So exactly awesome. That being said, let’s just get into some news items. Tommy, what did you find for us for news? You had a number of items here. Yeah. So we’ll run through these. So three from Microsoft and actually one

2:36 three from Microsoft and actually one from yours truly here. Mike, I realized we talk about a lot of great things on the podcast and I always write a ton of notes, a lot of inspiration from it. So, I wanted to actually start my blog again and now that it’s not hacked and I moved it over to GitHub pages, which is kind it over to GitHub pages, which is neat. of neat. Amazing. By the way, amazing, amazing, amazing. And I realized that I want to always expand, not just summarize an episode, but really expand and help really people out on especially the I think some of the nuggets of information we talked about.

3:07 nuggets of information we talked about. So, I have an article here. It’s on promptingbi. com and it’s really just about that second brain, the harness idea. Again, this is very near and dear to me in terms of my own workflow and how much time and value time I’ve saved and value I’ve added I think for customers and also the projects I’m working on. So I really actually go through in terms of what are the elements you need to have a workflow around AI and it talks about skills, agents, organized t context and the

3:37 agents, organized t context and the ability to talk to other tooling and those are really and really talks about how those elements work together. How do you actually give context the prop or how do you actually do the proper context or at least which works for me talking about your project database, meeting database, milestone database. then talks about the different agents that I had. I talk about my two agents which again inspired by Michael my consultant consultant which I call Pulpete which is meatball which is always fun to write hey Poete I need

4:07 always fun to write hey Poete I need something and then my oper just makes the day a little better and then just turning of how it builds instructions the best way to write instructions and then what you do with other tooling one of the big things I expanded on the article too and it explained was you have to do rightback you need to update that second brain again with what’s been done and that workflow really helps. So yeah, Mike, I workflow really helps. So yeah, Mike, we talk about a lot of good things. mean we talk about a lot of good things. I always a lot write a lot of things on I’m like I should write about this and I just said I’m going to do that.

4:38 just said I’m going to do that. Awesome. Totally agree. there’s there’s a lot of things Tommy I think in here. I think one of the I read an article recently just I think it was either an article or something off of X. They were talking a lot around your knowledge compounds with agents, right? And and it’s again it’s a lot of this right now it’s very new. There’s a lot there’s not there’s starting to patterns to start evolving out of this but every so often it feels like not every so often every month it feels like there’s another new pattern. People are observing new things.

5:08 observing new things. There’s something new to learn and educate yourself on. And I think some of the notes here you take Tommy this is really good because this is born out of months of time working with agents and figuring out what works and what doesn’t work. I really like this idea of the second brain. I I right now I do a lot of my work with agents inside git repos and I’m finding myself building more like a folder that has a whole bunch of docs in them. You has a whole bunch of docs in them. how how this app is structured, how

5:38 know how how this app is structured, how this app works using this part of the app. app. One other area that I think is very useful here is often Tommy we’ll use an agent to help us debug something, right? the app doesn’t run, it won’t start up, something’s broken, authentication won’t work. And then you work with an agent through a session until the agent understands and figures out the resolution, right? right? And then we just like say, “Okay, it’s fixed and move on.” One thing I’ve been doing a lot recently is I’ve been asking the agent to reflect

6:08 is I’ve been asking the agent to reflect on what you just learned. Reflect on you what you just learned and summarize it. Distill it down to like,, a very simple information. And then also I’ll ask it to okay now take this information add a document in the docs folder for this failure mode or add this to a library of failure modes that we have observed in the past so you that we have observed in the past so how to fix them know how to fix them right right because the number of times we’ve spent like having an agent fix a problem and then we build something new and the

6:39 then we build something new and the agent again breaks it or something changes and that problem arises again. We go through the same cycle burning the same tokens again around exactly exactly how to fix that same authentication issue. How to fix that same UI bug every single time. And why not if you’ve already spent the tokens capture that information put it in the repo. Exactly. And it doesn’t just help you build things. One of the things that I I was explaining is how I used the cloud instructions for the chrome for

7:10 cloud instructions for the chrome for cloud chrome to go through postman and it found some blockers for what we were trying to do which when I was ready for my next meeting with the team was like hey this is something we need to talk about because you said this was going to be here we didn’t find it so it also it’s not just organizing and just building for you but again it really becomes your war room so to speak for everything that you’re trying to do in every conversation that you have. So Mike, to your point, it may be great just to say tell me what you

7:40 great just to say tell me what you learned, but if that information doesn’t go anywhere to enhance the current project you’re on, like, it it gets lost in the in the information sphere. sphere. Love it. Really good blog post, Tommy. This is really good. Extremely useful for those who are either getting started or need to jump into the compounding learnings of using agents, how to leverage them, and start getting patterns here. So there’s a lot of good patterns here that exist. Awesome. What else do you have? All right. So the next one is from the Microsoft blog and it’s about SQL Server

8:11 Microsoft blog and it’s about SQL Server is introducing govern secure outbound AI connectivity. And simply what this is is allowing databases engines to call external AI services keeping security permissions and auditing. So it’s really Mike we’re finding the SQL server is involving to really participate directly in AI enabled workflows. the integration is closer to the data so it reduces unnecessary movement and anything that you would have to do with security policies. So you can run AI scenarios like brag other other items

8:43 scenarios like brag other other items under a database governance in avoiding middleware patterns. So again the two basic things are an external rest invocation and then external mono objects with there’s actually a call that they’ve created in here. It works with shared security principles. And yeah, this is pretty neat, Mike, because SQL Server’s never gone away. And I know there’s been a lot of times to say, “What’s the best way?” But until we find a better system, SQL Server really is

9:13 a better system, SQL Server really is pretty incredible. And this is just making it more part of the workflow. I think this is really appropriate. I like this whole I have to dig into this REST API a little bit more, Tommy. My understanding has been in the past whenever you wanted to talk to SQL server there was like a series of drivers you always had to install right SQL SQL has the there’s a library there’s some ADO. NET net or something something some library that had to be installed

9:43 library that had to be installed that would allow you to like communicate with it and so having a straight REST call, right? So every time your agent wants to talk to SQL, you would need to like write a Python script. Oh yeah, Oh yeah, that would install some library that could then correctly talk to or or whatever that may be. Might be Python might be whatever you choose like whatever language it is, it’s not a big deal. Umnet, whatever. But the the concept is you have to have something installed for it to run. Now, when agents run in the cloud and you just

10:14 agents run in the cloud and you just want to like whip out a command, where do you install this stuff? What what machine is it being installed on? And so, I think this is a really good solution to like opening up more of the SQL experience. One of the things I’ve been playing with, Tommy, have you you’ve built a number of SQL servers inside your fabric environment? Yeah. Have you ever used the GraphQL to talk to your SQL server? I actually have not.

10:44 I actually have not. So it’s interesting. So SQL server again, you have a connection, you have the the libraries you can go use to to talk to the SQL server. The GraphQL removes the need for you to install any server stuff. And GraphQL has this concept of like a mutation, right? So you can read some data and mutate some data at the same time with one command. And it’s a little bit more flexible. But again, I know enough about it to be dangerous, but my agent knows how to write everything about it.

11:15 how to write everything about it. And so that so I can so I’ve been building little applications with particularly with Rafen or or building things with apps,, static web apps or things that I’m I’m interacting with and more and more I’m not using the libraries of SQL. Instead, I’m thinking, okay, stand up the SQL server, add a GraphQL, and even in the GraphQL, you can say you can expose these tables in the GraphQL to allow an agent to talk to a to and then it just

11:46 agent to talk to a to and then it just say, “Hey, I have a GraphQL endpoint that’s connected to SQL database. Here’s the endpoint. It authenticates. Boom.

11:51 the endpoint. It authenticates. Boom. It’s talking. There’s there’s no other code needed. It just works, which I really like.” No, I I think anything that’s new to you and I think this is your just a good point here too. You may not have background in database admin or the capabilities of SQL, but to your point, as long as you’re aware of what it can talk to, if you do have something like cloud code, well, you have a lot of options here. So, that’s pretty neat, Mike. So I think this is great because we’re seeing more around

12:22 great because we’re seeing more around if you’re going to use AI or and again what an ambiguous term that is in general right but if you’re going to use AI use AI in the most general sense you do need your data to run against it and right now again one of the biggest places SQL server so that is some of the updates for the SQL server Mike we got a cost one one as well so this is another article we’ll run through this this controlling storage costs with the one lake life cycle management and

12:53 lake life cycle management and Microsoft’s just announcing the GA of one lake storage tiers and life cycle management which is trying to help organizations reduce their long-term storage costs by moving data between hot cool and cold tiers based on usage patterns. So again the hot tier is the way for frequently asked data. Cold and cool tiers are for lower cost storage for infrequently accessed or long retained data and the access and transition cost increase in cooler tiers but the policies help optimize the data moves and they’re really customizable.

13:23 moves and they’re really customizable. It’s not necessarily all automatical. You can support multiple rules and filters and conditions and the policies can really reduce your cost. So this is really great. You can do predictable access patterns. Anything in bronze, it moves to the cold tier after 30 days if it’s really accessed. One lake reporting diagnostic logs find really accessed data to find hotspots and you can even do auto tiering as well. So Mike, this is again you’ve talked about this a ton and I really do enjoy it because one of our jobs and one of the conversations a

13:54 our jobs and one of the conversations a lot of clients have asked me lately too is about cost around fabric. They want to do proof of concepts. They want to see this. So this is a really great example. So I know you can really talk to this. So Mike, how do you see this playing in in terms of the importance of this? Where would you rank it? I think this is fairly important. a lot of organizations don’t really do good job at the file management level. you if you looked h at the hidden level behind your fabric or your environment, there’s probably a a good number of things in there that are just

14:25 number of things in there that are just consuming storage where you don’t need them to be. Things that you’ve accessed a couple times. Delta tables can quickly eat up a lot of storage initially as they’re hydrating and loading themselves up and then eventually they would start, you then eventually they would start,, tapping out or know, tapping out or plateauing a bit to a certain amount of storage level. But regardless, Tommy, if you if you’re not paying attention to the storage costs, those will creep up to you and after years of work, you have all this file storage laying around and

14:57 all this file storage laying around and you could potentially incur substantial monthly costs around just storage costs and that you’re not using. You’re not using them or the data is adding minimal value. Like these are some hard questions, Tommy. Like do we do we want like do we need to save every single edit on every single table that was done 5 years ago or can we distill all the changes and edits down to like a single month is that something we can we can get rid of right that’s the

15:28 can get rid of right that’s the question I think we we need to start answering here with some of this and also storage tiers don’t vary widely but like if you are not if you just want a copy of the data if you just want to hang on to it. I’ve backed up a SQL server on prem. I need to store the image. We don’t know when we’re going to use it again, but we just need to have it for record purposes. You don’t need it in hot storage, right? right? There’s other like warm, cold, you could put it somewhere else where it lives. It’s a more expensive to access it, but you’re not going to access it every day.

15:59 you’re not going to access it every day. It’s maybe one time a year, right? right? So, I think you just got to be more strategic about this one. And you have to ask yourself some real questions around why do we need this data? What’s the value of holding on to it? And it probably this gonna be a dumb question probably, but when you think about this with mirroring or shortcuts, you’re still accessing the data. But again, again, this is part of just your own approach, too. Because if you are building the right way, you’re really only using some of the bronze,,

16:29 only using some of the bronze,, for certain situations, right? You shouldn’t have to pull that every single time you’re going through a pipeline. So it makes a lot of sense but it goes back to how you plan how you architect this. Yes. Yeah. There’s a there’s a lot of again this is like good system design just having good system design around the data and the data storage. So I think this is really useful. I do think it’s really helpful. from what we’ve heard Tommy the one lake is just blob storage accounts. That’s what it is. It’s it’s another API layer wrapper on top of a blob storage

16:59 wrapper on top of a blob storage account. account. The fact that a blob storage account already has hot and cold storage. Mhm. Why we haven’t had this sooner, I don’t I don’t know, but it it feels like it feels like this was an an ample time to have this feature. This makes sense. I think this is in alignment with what we’re doing with our data. Speaking of something we should have had sooner or why didn’t we have this before? What a segue. Let’s go to PowerBI and you want to talk about something we should have had sooner. How

17:29 something we should have had sooner. How about a date picker slicer? How about that? How about this is the first time it came out in 2026. So what’s now available in PowerBI is a dynamic relative date range that actually moves forward as the data refreshes. So users can actually choose a relative range that looks like any other stupid date picker filter that you do on any other UI. And there it does support some more features because you can support the last date, the first date or today.

17:59 last date, the first date or today. and just simply allows you to stay dynamic. And again, how in the world was this not possible until I don’t know July of 2026? It’s this is great. Don’t get me wrong. I am so happy this is out, but these dynamics even stay the relative ranges stay dynamic when the visuals are pinned to dashboards instead of becoming static. So, this is also pretty neat, too. Shout out to dashboards. And you do have to go to preview features to get this done. Few of the scenarios they talked to show the last full month., useful when the

18:31 last full month., useful when the current month is incomplete. Handling future dates in a date table. Compact formatting and the viewer experience. They can choose that relative range. So, and this is all available rather than in the filter pane. Yep. Yep. But this is now available only existed before. before. Right. Right. This is now available as a visual that any person consuming the report can do in the canvas itself., this is this is a major win, I think, Tommy. One of the things I’m most excited about this particular post, Tommy,

19:01 Tommy, is the manual selection. Having a full calendar just appear and select start and end dates on things. things. Just the UI. What the heck, dude? Like, this should have been out a long time ago. This really just speaks to like I’m pretty sure this is on Zoe’s team and Zoe’s doing great things here and awesome. Kudos, Zoe. This is amazing feature. Glad we have it. This has been so needed. And I can’t tell you the number of times, Tommy, I’ve gotten calls or messages around, man, it’d be really

19:32 messages around, man, it’d be really nice if we could always have like the last 14 days, the last month, the most recent month, just something on the page that was like relative and it would just always update as you move along. The only thing we could ever solve that with was well, you can’t really do that on the actual page of the report. you had to build an additional date column that had like an index on it and then you’d have to be like okay the index is like this number and then they would always show but then does anyone understand what the index means when your report’s on minus14 like what does that mean?

20:03 what does that mean? I literally have a proof of concept coming up next week that this is going to be essential for and I was going to figure out what to do before. So no I I I’m so happy to see this. So again shout out to PowerBI Mike I’m going to make a mental note here. I think we need to do another completed or kill it episode. I think it’s about that time. I think it’s probably right there as well. And I I can I can vouch for this feature coming directly from Zoe because I’m looking at the screenshots and I’m happy on the page here. I have my cursor over it and it hovers the data tool tip

20:33 over it and it hovers the data tool tip and it’s the data tool tip on the image is data Zoe and then it has PNG at the end of it. So, so Zoe, thanks so much for the blog post and Zoe Douglas made the the blog post here as well as the author. So, Zoe, nice feature. This is great. We’ve been needing this for a long time. This is a huge creature comfort. This is just needed to be competitive with every other tool feature that’s out there. Right. Right., one beat from the street, Tommy, that I was thinking about this morning that I just want to bring out.

21:04 that I just want to bring out. Yeah. Yeah., I work a lot in the embedded space. I have a a solution called Intlexos. We do tons of embedding for things. so if you ever have any questions around embedding, hit me up. Let me know. So if you want to if you want a nice white labelled solution around embedding, we’ve got the product for you. you. If you want to get started in 20 minutes or less, talk to Mike. Yeah, it’s fast. Tommyy’s seen it. So one of the thing I got a question around was everyone’s trying to clamor around, well, how do I use agents? And one of the questions was well the assumption

21:35 the questions was well the assumption was AI agents need a PowerBI pro license to work to work right if I’m going to connect to a semantic model in the service if I’m going to use a data agent I need to have a user attached to that that’s not right Tommy there there’s been some recent improvements on the PowerBI MCP server and if you use a data agent a data agent also has the ability to be using a service principle name. So you

22:07 using a service principle name. So you can use a service principle. So I just did some testing the other day. Tommy, I know you’re a big Claude user. Mhm. Mhm. So you can set up your cloud to talk to the PowerBI MCP server and you can use a service principle to talk to that which the service principle can talk to the XMLA endpoint and that means you only need a PowerBI

22:37 oh yeah PowerBI F skew with no users attached to it. don’t need a pro license to get this to work. And I was having a huge aha moment. I have the I literally have the documentation up, Tommy. I was figuring this out at the end of yesterday. there is there is a a manual setup of the MCP server. And so I have Claude running with the MCP server, an Azure tenant ID, an Azure client ID, and an Azure client secret.

23:07 client ID, and an Azure client secret. So I can set up authorization mode service principle on the M fabric PowerBI modeling MCP server and talk to it directly with my cloud code. I found this to be incredibly interesting because because in the past you’ve always signed into things. Yeah. And this is an important thing too I think especially for us Mike as consultants because I actually created a fabric setup guide for when I kick off with new clients and it talks about actually creating a service principle

23:38 actually creating a service principle going through this to your point because

23:40 going through this to your point because I’m like listen do you want to be done in two weeks or two months more or less and a part of that too and a lot of the clients are totally fine with that especially when they understand what it’s going to be used for and how they got access but I think this is really important if you are internal working too where this is something service principles been around for a while just for the audit ability for to go through usage to go through the the scanner right but think about this implication Tommy like think about what this really means

24:11 like think about what this really means this means any AI agent you have Tommy’s GitHub co-pilot claude code right right this means doing this means I can have anyone in my organization set up I I don’t know if this is a this is good or bad I don’t know well I don’t know Tommy like it’s good and bad right so but the implications of this is if you want to talk to your model with

24:41 if you want to talk to your model with an agent an agent sure I see what you’re saying there’s zero licensing required for the user to show up to the model so I don’t know so Tommy what I’m what I’m what I don’t know is I don’t know if it’s a good idea to exclusively give your agent access directly to the model., I don’t necessarily want it to be able to edit the model, right? But I do want it to read the model only. So, I don’t know, I got to look a little bit more on the permissions on some things, but just just think about that for a second. If I can give everyone or a team of

25:11 If I can give everyone or a team of people access to Claude code and then from claude code, set up their app registration, everyone on that team can now directly talk to the model. Give me the answers for this. Show me the relationships of the model. Run this query. I want to understand this data point around this. And Tommy, if you use the data agent, the data agent is expensive, right? Data agents are using some heavy amounts of tokens. I literally was about to ask

25:42 I literally was about to ask I ran. So the data agent side of things I ran four queries. Okay. Four queries cost me 5, 100 CUS. And these were simple queries like I say dollars. No no 5, 100 CUS. But think about it. it. I’m on an F2 testing this out. I ran in probably about I don’t know five, six minutes, 10 minutes, four different data agent queries. I sent the query to the data agent. The data agent thought about like okay how am I going to query this? Use

26:13 okay how am I going to query this? Use some compute figure out what it was do natural language to SQL. Do natural language to DAX. Wrote me an answer and responded back. Boom. Answer comes back. Right? So I got answers out of the data. They they were right. They were they were actually grabbing data from the model. Okay great. But 5, 000 CUS, when you do the math on it, a 5, 000 CU job at an F2 level, an F2 gives you 120 C use per second or sorry, 120 C use per

26:43 use per second or sorry, 120 C use per minute because you get an F2 has two virtual cores, two cores or two units per second. That’s what an F2 does. So in 60 seconds, you get 120 CUS in a minute. So then when you do the math out that I don’t know 5 10 minutes of CU usage cost me 43 minutes of total capacity usage across my and that’s just one person right and so this is interesting

27:13 this is interesting one person with four queries imagine Tommy the amount of time that we prompt our agents over and over and over again we could spend hours of time doing this right this was like a five minute window of like me just testing something out out this could get apply back and forth. Yeah. And that’s,, if you’re asking to refine things and give you more details and more. So I looked at the data agent cost of this and thought, “Oh my word.” On the other hand, if you use the PowerBI MCP modeling server, that one ran all the tokens through my

27:45 that one ran all the tokens through my agent, my agent, it figured out how to talk to the XMLA endpoint and almost cost me nothing on the fabric side. So, it shifted all of my token cost away from the data agent and put all the token costs into claude that I’m already paying for that I have tokens with. So, I can shift the token usage out of data agents into Claude prepaid already have money already have tokens ready. They’re there. I’m either

28:15 tokens ready. They’re there. I’m either going to use them or not, right? And then I tell Claude, “Okay, now run these like you figure out the decks.” It’s way more economical to have the MCP than the than the data agent. I don’t think you’re I I agree with you to a point. I think the way with the service principle doesn’t scale for a lot of companies because Okay, you Max. I’m not I’m not willing to say yes or no on that statement right now. There are I think some substantial

28:45 There are I think some substantial security implications cuz like can you actually actually do I want people even editing the model? I don’t probably and well no what I’m saying is how are you do you have cloud max I have the the middle tier okay 100 so you would have to give that to each user who would be able to access this this because I think I’m trying to think right now Mike with what’s your intention for a organization using the service principle in cloud or another agent are you saying for people to in a sense circumvent the data agent

29:16 to in a sense circumvent the data agent or are you asking for everyone to edit it right because there’s One way you can look at this that I’m hearing from you where it’s like, well, this would be a great way for people to chat with their data. That’s could be a solution here. The other one is we want everyone to build with the service agent. To your point, point, your example with the data agent, right, that’s meant for a consumption point of view. I’m not modifying or building. I’m reading and I’m gathering or,, collecting information. I

29:46 or,, collecting information. I don’t think I would do that with the cloud service principle. I think you’re going to use up a lot of Claude tokens and then again you have to scale that to everyone. everyone. Who cares? Who cares? Who cares if you use up a lot of Claude tokens? Like if if I’m if if my company’s stance is I’m giving everyone in the company Claude Mhm. Mhm. and that’s what we’re going to use for an AI. You’re going to use some AI agent. If I have to spend compute units through data agents or I can go through Claude and get the answers through Claude, Claude, I’m already sp like it’s it’s like a

30:16 I’m already sp like it’s it’s like a sunk cost fallacy. Like Tommy to your I understand I may use a lot of cost there but I’ve already like the as soon as I commit to a claw subscription I already have $120 $20 $100 $200 whatever whatever the spend is I’ve already committed to consuming that usage usage you can make the service principle too in a group with certain permissions where they only have read permissions that way no one’s accidentally modifying anything if you went to the consumption

30:47 anything if you went to the consumption collector about this is what I don’t know exactly Tommy I don’t I don’t I don’t know exactly how like I’m not sure what level of permissions I need on the submit like this is got to do this is where I need to do more testing right so you have to give the give the MCP service principle membership privileges at the workspace level for it to work right I don’t know if I can go down to like a lower end level can I just give them access only to the the model. Can I

31:17 access only to the the model. Can I change their permissions on the model to just be a readon model? So, there’s there’s some security things here I’ve got to still work out. So, be cautious. What I’m saying here today, yeah,, you could try it, but don’t don’t roll it out today. How about that? That’s my point. Don’t don’t go in and be like, “Boss, I found a way to save some money by not using dats to the organization saying proud to announce.” Yes. Exactly. Right. So, so that’s that data agent side of things is just very interesting., anyways, I had a really big epiphany yesterday. Very pleased to

31:48 big epiphany yesterday. Very pleased to have agents talking directly to semantic models without any need for any PowerBI Pro licensing at all. As long as you have an XMLA endpoint and you have a an MCP server, boom, you’re up. You’re off to the races. Now, one thing I didn’t test, Tommy, is does this work in power u premium per user workspaces? So, that would be one I didn’t check out. It does have an so the premium per user workspace does have an XMLA endpoint and I didn’t test out the

32:19 endpoint and I didn’t test out the cloud code service principle access on a premium per user workspace because that would give you you don’t even need fabric at that point right you just you just land a premium per user workspace 100 gig models at your disposal tokens being used by claude to generate DAX and code and answers running against a larger model this could be Wild. Could be. Could be. Could be. Could be. Yeah. So, I I like I said, I will hold my breath until we actually I that would

32:49 my breath until we actually I that would be a great user group actually to test that out too to see what that looks like. like. Well, I wouldn’t do it in user group. I would test it out and then show people what I figured out. No, don’t test it on the We’re going to figure this out together because I don’t want to do it alone. This is this is an undefined user group. We’re going to just like figure things out as we go. If you want to see trial and error, join us for the next user group. So, No, that’s great. All right, Mike. We got a good one today. So, let’s go, I think, to the main topic. And Mike, yeah, we’ve done way too much introduction time. Good conversation, but let’s let’s migrate to our main

33:20 but let’s let’s migrate to our main topic. All right, speaking of AI and agents, Tommy, give us the give us the run here. What are we talking about for Ray Finn? Ray Finn? Oh man, Mike, so we have done a few episodes on Rayfin. Mike, you’ve already talked about Rayin a ton and also doing the demos. I love it. I absolutely love it. Love it. You’ve done a great user group for Chicago last week on Rathan just how easy it is to start building and not just building a report or visuals but what you showed was building an application a task list what was a

33:52 application a task list what was a booking you were saying for if you were people were out of office their vacation schedule schedule I saw someone on on LinkedIn said hey I built a little vacation tracker and I have a vacation tracker that I’m using and built with Rayen and I’m like that was a great idea go here, sign in with your company email address, go enter in your time, when you’re going to be out, bing, bing, bing, done. And then you can even show on that page, here’s everyone who’s out of the office in a very general way. So that’s incredibly useful., so I’m I’m looking at this

34:23 useful., so I’m I’m looking at this going, going, I really like the ability to have Rayfin be this read, write, access thing. And again, again, my big my big pain point here, Tommy, is every time I buy some application, every time I buy some system, always there’s some data in the back end that I can’t get my hands on. It’s just annoying to me. And Rayfin, I can just communicate, I need this, build me something that does what I want. And all the data lands in a nice, clean SQL

34:54 all the data lands in a nice, clean SQL database. I can read right right to it and it’s right next to all my other data that I care about. It’s so good, Tommy. So, and this goes to the other side and I want to take this too with what we call we’re going to call the hype of Rafe and Mike because to me I’m maybe going to play a little devil advocate here because I think there’s a hype here just because let’s again break up what we’re going to talk about there. I’m going to put the angel’s advocate then put the devil’s advocate. I I don’t know where I don’t really don’t know where I’m going to go with this in a

35:24 where I’m going to go with this in a few ways. There’s a few directions, but I think here’s the thing. Anytime there’s a shiny new toy, we’re always

35:30 there’s a shiny new toy, we’re always going to be obviously kneede into it and you are already seeing some incredible implications of it with the testing. However, However, is it perfect? And I think there’s a few things I always like to go through anytime we have a new product. What’s the intention? What does it do better than the other things that exist? Right? Because I’ll always go back to Steve Jobs when he made the iPad. If we’re going to make an iPad, it has to be better at some things that we create like the iPhone and the MacBook or has no reason to exist. And with Rafeen,

36:01 no reason to exist. And with Rafeen, yes, it’s incredibly easy to create and I can embed it in fabric. I can have people log in other places. I can create an application, but is it better than enterprise applications? And I’m not just talking about Power Apps here. I want to bring that story in for a reason. But I think you have we have to think about this is if I were to do introduce Rayen to an organization and create an application a way people are inputting and managing their information it has to

36:31 and managing their information it has to be better than the enterprise apps they use. This cannot just be for me and my team of small people. it can be but I’m saying when you’re going to scale it out does it work in that way? Before I go into that series of terrible negative questions, Mike, let’s just talk about the the real the real true hype of this is how easy it is to build an application connected to your fabric data. That is the winwin with fabric bar none. Mike, you showed three

37:01 none. Mike, you showed three applications in a series of 45 minutes, maybe an hour during our user group and that’s what you’ve done with the demos as well and expanded on that just with some simple prompts. So there’s a huge win here. Absolutely. However, it is a shiny new product. We have not seen this scaled in an enterprise yet. I know Microsoft has some case stories there and they are great, but how long this has not been in the wild for long. So necessary questions need to be asked. So

37:32 necessary questions need to be asked. So let’s start there when it comes to to what’s the simplest way that we can say why is favor RAN so effective but without overhyping it. I I think there’s this trend Tommy that I really enjoy and this is one of the reasons why I love working with Ray Fen. It’s this concept of I can come up with an idea and the distance between my idea and getting something that works is very short. It’s almost like playing a video game, right, Tommy? I I love strategic

38:02 game, right, Tommy? I I love strategic games, like strategy games, right? I one of my favorites is Minecraft 2, the very early versions of this, right? You build a little hut, you have a little bit of a these these humans or orgs, and you go have them go mine trees, and then you build these bases, and you you a compound thing. You build these strategic systems that work, right? It’s system building. And I feel like Rayfin is a video game to me in some at some level because I don’t need to know how to write HTML. I don’t

38:32 need to know how to write HTML. I don’t need to know how to write all these different items. I just need to know enough about the game is to describe. I need a component that does this. I want the page to have a header that does this. I need a menu that does these things. And so with relatively ease, I can describe what I want. Now Tommy, you made a comment earlier about like the enterprise large software packages and and corporate or enterprise Y packages. Nothing wrong with that. I think those are very good solutions here. But what I

39:03 are very good solutions here. But what I think I starting to see here is Rayfin is starting to very quickly eat into the usefulness of those large corporate pieces of software. Let me give you an example. One of the pieces of software that I’ve been using in the past was Microsoft Dynamics. Microsoft Dynamics tracks like my customers. I can get leads. I can bring them in. I can track opportunities. I can all this stuff. I don’t really need that anymore. My new I’m I’m not a huge company. So, I was

39:35 I’m I’m not a huge company. So, I was paying for the Dynamics license. I was paying $85 per user to have that run. And that was I was just That’s a lot of money per person. for Rayfen cost me zero added dollars and I can build out whatever I want inside Rayfen, right? I can build an intake sheet. I can build data loading. I can build and all and all of the data doesn’t sit in Dynamics or data versse or whatever you want to call it the whatever the wherever they store their stuff from.

40:06 wherever they store their stuff from. Again, it’s still Microsoft. I can still connect to I can still get access to the information. All that data now lives in my SQL server in my fabric exactly where I want to do analytics and reporting on it. it. Everything is there. And so fabric as a backend I think so let’s not talk about rayfin. rayfin. Yeah I was about to say that is a difference there. If I zoom out a little bit right is the proof of concept of fabric as a backend to your data systems. Period. End of story.

40:37 End of story. Okay. Because Rayan’s like this shiny wrapper that sits with the UI. It’s, you wrapper that sits with the UI. It’s,, but the the magic of it is I can know, but the the magic of it is I can talk to an AI agent and you can just say I need a database that does this. I need a lakehouse storage that does this. I need whatever whatever you want to describe. It can build it for you. And so right now the only two sources that are supported are semantic models and SQL databases. That’s all we’re doing right now. But even still with those two

41:07 right now. But even still with those two data sources, eventually hopefully we get like Cosmos DB in fabric. Hopefully we get lakehouse and storage accounts in fabric like all those other things. We need fabric userdefined functions to use inside Rayfen. So there’s a lot of other things that need to get bolted together to make this even more compelling. But at the end of the day, Tommy, I think my argument here is in those corporate pieces of software, you’re paying a lot of money for them to run and be secure or whatever, when I look at how much of those large

41:38 when I look at how much of those large corporate pieces of software that I use, I don’t use 100% of the software to get things done. In reality, I use some subset of that. And in certain pieces of software, I’m only using 30% of its capability. And then there’s just things I want to do that it just can’t handle. It doesn’t just do what I want. I can’t tweak it for the needs of my team. So your company is unique based on how you go to

42:08 is unique based on how you go to business, how you do your process, what your people do on the day-to-day, that is unique to your business. These large corporate softwares say, “Look, we’re smart. We’ve done enough investigation or research across hundreds if not thousands of companies and we see the same workflow pattern emerging customer relationship management right SEO optimization there’s common patterns across these large portions of organizations that companies come in and say look we know how to solve this

42:39 say look we know how to solve this problem really really well and we can solve it for hundreds if not thousands of companies what Ray Finn I think is doing is It’s shifting that mantra over to what is your workflow? What works for your business? What makes you unique compared to everyone else? And instead of having to have hiring a developer and all this extra cost around that, that, now you can just go hire an agent and your agent builds out what you want. And my agent cost me and my agent cost me $100 per month.

43:11 and my agent cost me $100 per month. You were doing so well, Mike. I was with you so long. And then you just left and then I Yeah. No. So I want to I do want to agree with you on a big part that you said here because I think this is fundamentally not hype with Rafen and fundamentally what separates Rathon apart is the real win here is showcasing fabric as a back end and I think let’s not understate this the fact that the ease of the management and the input of data can seamlessly work with fabric

43:43 data can seamlessly work with fabric right yes can you do this if you built it yourself. Absolutely right. But there’s a lot more steps to be taken and a lot more security. Again, the fact that the security, the policies, the integration work so seamlessly and being able to build a,, very straightforward app is incredible. So that’s where completely I agree with you where you said the proof of concept. But Mike, I wanna we need we need to have some real talk here because I need to ask you, have you actually built I know

44:14 ask you, have you actually built I know you’ve been building a lot of applications, but actually this is probably a better way to say it. I was doing a discovery call with a client and we were going through some of their data sources opposite a lot of applications. Okay. Okay. What I was looking at looked like it was from Windows 97. And you think about this and you’re like, why would they be using this, right? Why would you do that? And you the first thought like, well, they’re not dumb. Like, they’re a great company who’s doing well. So, why is this clunky

44:44 doing well. So, why is this clunky software out that everyone has to install still with a CD? It looks like,, with the gray title and everything, the gray menu bar. Sure. Well, you have to understand when you were where I could not follow you was when we’re talking about the enterprise application. You have to under we have to understand not just you but we have to understand the fact that there is a process inter ingrained and intertwined with any application that you have. Is it easy to build a fabric? Absolutely. However,

45:15 build a fabric? Absolutely. However, yep, yep, there is app management, there’s app ownership, there’s the app life cycle that you have to build. Mike, in my own experience building Power Apps for the enterprise. Well, the first three months we’re not building anything, right? If the first three months is understanding how people are inputting data, where everything’s coming, the flow of information, the approvals that have to go into place, and then you could talk about the UI. None of that was the UI in the beginning because what you do when

45:46 the beginning because what you do when you introduce something an application is you are going to drastically change everybody’s workflow where they go what they manage how they do it which is not it may be easy to build the app that’s not the point when you buy corporate software that’s what changes everyone’s workflow right whether that’s what changes everything and the number of times I walk into new technology that’s when you bring in corporate enterprise software, you have to adapt to how they decide to use the application like that

46:18 decide to use the application like that that that is how things work. So like I I totally disagree with that statement like like well no when I was saying power apps when I was building power apps we had a lot of customization with that enterprise software yes absolutely but this but this is I think my point though my point is like you want to do work like your company knows how to do work a certain way like there’s a certain process and a flow of things the number of time so I was working with a company that was bringing in a very large software package to run their plants for manufacturing

46:48 to run their plants for manufacturing Mhm. Mhm. Okay. SAP SAP is extremely expensive. Yeah. Yeah. And they had this homegrown built solution before that they were currently using using and the amount of millions of dollars that were spent to try to migrate into SAP and it was customization and this is how we do our work and this is what our workflow looks like. Again, t not not disagreeing Tommy on like the support and effort on those things like those are things that have to happen regardless. But I believe with

47:18 to happen regardless. But I believe with AI, the ability to maintain the apps,

47:21 AI, the ability to maintain the apps, the ability, so there’s always this question of like, well, I can build this app, but who’s going to maintain it? You and the AI will. It’s it’s the ease of which it was able to get the app to build itself initially. You have the AI document how the app works, what features are available, and if you have issues with it, you work with the AI to fix them. the the same speed it took you to build the app is the same speed it helps you to fix and update the app as well. And so I I

47:51 update the app as well. And so I I really do think like we can we can stop having to like push software into our hands and we can we can start moving into what works for our company and moving closer to that direction. Mike, you’ve worked with ITA teams before, right? , I’m No, I’m I’m being rhetoric here because of course you have you’re Yeah. I am the shadow of the IT. Like, you are

48:21 I am the shadow of the IT. Like, you are their biggest pain in the I’m their biggest pain in the butt because I don’t want their hands on anything. anything. Everything you just said, they’re like, “Yeah, like here’s and I I agree with you in maintainism. They they can’t they can’t it teams are unable to keep up with this.” Like again this is the same way that Power Query and PowerBI and Excel brought frontend central more data to the business user. The same things happening with apps and AIs as well. This is traditionally a

48:53 AIs as well. This is traditionally a role that was only owned by super technical people. We would hire very specific technical people inside the the IT group. Tommy, there’s companies now there’s there’s there’s like kids now like 19 year olds now building, you like 19 year olds now building,, 10K, 20K, 30K a month of revenue know, 10K, 20K, 30K a month of revenue from apps. They’re building on Twitter. Yeah. Yeah. Yeah. But Tommy, like this is real. Like it’s not it’s not like I’m I think people are building I honestly I

49:24 I think people are building I honestly I think people are building real things. There was a news article a kid built a food fitness app. Oh, Mike. He sold it for $30 million and he’s like 20. Tommy, we’re missing the boat here. Like, if we No, no. No, no. If we don’t think this is going to change the world, we better get our heads out of the sand and start figuring out like this is changing the world. Like there’s articles that are talking about like the one employee billion dollar software company. Like the systems we can put in place, the things

49:54 systems we can put in place, the things that we can do is is only the limitation of our imagination at this point. Yeah. When when did that ever not work out before where people were just building things and without without regard? What was it like 1999? Nothing wrong happened then. So, I’m not saying this is the internet bubble here, but I’m saying no. But I first off, I have an issue with the Twitter thing because anyone who shares how much money they made on Twitter, have you ever done that? I’ve never done that. I’ve never said I made $15, 000 with this CL. Anyways, I hate those. I do not like

50:24 CL. Anyways, I hate those. I do not like those people. I have I have a strong reaction to maybe maybe the numbers over inflated Tommy. It could be $75, but I know but I have to imagine some like we’re starting in this new world like there’s what you can’t ignore though is you can’t ignore people are building apps with zero coding knowledge. 100%. 100%. So I have to project a little bit forward then to saying look if people are making full-blown apps with zero coding experience that just

50:54 with zero coding experience that just means all the technical barrier that we would like land inside the IT team or the the very heavy technical team is just gone. It’s not there anymore. So so that’s what I’m saying. The technical barriers are dropping. We’re lifting a lot more of that tasky level stuff over to our agents. And I think you agree with all this stuff, but I’m just looking forward to this and saying it’s again, Ray Finn, the hype of this is I really enjoy

51:24 is I really enjoy talking to something and saying, “Here’s my idea. Help me reason through it. Let’s get an answer out.” And for me, so impactful. And I have never argued any of that. Like I said, I’ve been a huge proponent myself individually,, in parallel with you on this. Now the I think the the contention I have though Mike is let me provide a scenario for you because who’s going to know about Rathon it’s going to be someone working fabric or PowerBI right they introduce their company hey I’m

51:55 they introduce their company hey I’m going to take over we’re going to take all this input of data that we have and I’m going to build the application inside fabric and someone says well do you have experience app building no but I have an agent I have with a new thing called Rafen that’s in preview or whenever it becomes generally aail available. I I fail to see Mike here again without disagreeing at all with the ease of building because I’ve never argued that. But the problem we’re going to run into right now and again July 2026 and I

52:27 right now and again July 2026 and I think for quite a bit more is that ownership side because who’s going to own it, right? Is it going to be the PowerBI team or the fabric team who’s going to be building those enterprise apps on behalf of the organization? Are they building side apps? Are they overtaking over the enterprise applications that a business required to run on? That’s where I want to have the conversation here because I do not disagree at all on how powerful Ray can be. But I think the thing is when we’re

52:57 be. But I think the thing is when we’re talking about who’s going to be building it and again obviously the agents but in terms of who’s going to be owning that on if people in terms of the governance of those applications which is essential for businesses right does that come from the an application team developers you already have or are we just giving that to everyone who’s a fabric developer at this point and I think there lies one of the big questions one I failed to see a lot of organizations being totally fine with that when we already have issues

53:27 with that when we already have issues with security as it is, even if the security is great. And then two, again, you’re going to run into the same issues that I feel like we had in the internet bubble where there’s going to be a lot of things being built, even if I could have an agent built it. Until we actually see this in the wild in an enterprise setting, you to me, you’re asking for trouble.

53:57 I I I no and yes, you do need to have plans with it, right? I’m not going to say just do it and give everyone I’m not going to even go in and say I’m not going to say give everyone right then and go and and and go, right? You’re going to find there are certain people in your organization that have a much better capability, knowledge bent towards building these things. There’s going to be people with ideas. There’s tons. So what happens right now is every like not everyone but like the increase in ideas

54:29 increase in ideas that you can come up with. What if I had an app that did this? What if like even even challenging your team to say okay okay stop thinking of doing work the way you’re doing it? Start thinking about how to do your work more efficiently or more effective. Start focusing on that part. So I I think there’s something Tommy here around like this idea of we don’t even know how far along we are in the space and and we just need to have more people have access to the ideas and what this is doing. So to your point

55:00 what this is doing. So to your point Tommy the governance the security things I do think there’s some governance board BI central team like this there’s already things that need to be in place to have you discuss what is an appropriate app when when do we publish it out to the rest of the organization how should people access this application do we want to have certain permissions or not around application so all that that conversation still needs to happen to happen right right I’m not saying ignore for that. What I’m

55:30 I’m not saying ignore for that. What I’m saying though is the the the team of straight coding engineers engineers to build large applications to support the business has just shrank. Yeah. Yeah. And the the individuals that become much more empowered in your organization are the really senior PMs on projects. A PM can now articulate the requirements of the business and get 80 90 95% of a

56:03 the business and get 80 90 95% of a solution built without the need of a team beneath them. And so this is what’s this is the shift that’s happening the middle management layer of so when we talked Tommy about like Rob Kie and he was talking about all this stuff around like power pivot pro and one of the things he communicated was if you took an application and just wrote every line of software if you knew exactly every single line you needed to write it would be a book right if you just sat someone down and said type all this out line by line verbatim we already have the answer

56:34 line verbatim we already have the answer it would take you like two weeks to type it all out, right? But the reason projects take months and months and months to build is because of all the communication between the different teams and understanding. So if you take all of that and condense it down to one person with only needing one individual to iterate and build on things, things, we can build so many iterations on stuff so fast. We can figure out what works. We can figure out what what runs well for our business. And and just me

57:04 for our business. And and just me personally, I’ve said this many times on the podcast, left and right, I’m deleting subscriptions and software in my company all over the place. Everything I’m looking at now is how can I get my data faster to fabric? What do I need to do to land applications or things that I build and again the Rayin app is built with an agent? It’s not run by the agent, which is I think a huge distinction here. That’s really good. I’m building a deterministic input output system that

57:35 deterministic input output system that the agent creates with me. The output of my app is super deterministic. It has a data structure. It has tables. It has a very specific purpose. And the speed in which I can then help users get up on speed on board with what’s going on at my expectations, what you’re doing, all all these things very very quickly. And so I for that reason I really like this. I’m I’m continually deleting more and more software across my company thinking that I’m just going to rebuild all the fabric apps that I need and build it on top of SQL databases behind the scenes.

58:07 top of SQL databases behind the scenes. And I think that we are getting to this disrupting stage, Mike. But here’s where I’m landing here with where we are right now with Rafen because everything I said I absolutely agree with is except for the project manager can run it all because they’re going to cause I think more issues in the long term as of right now unless you have really that technical skill that really the knowledge of just how things work. where where I really want to point to is Mike there’s going to be a slow adoption here with Rayfin because

58:39 a slow adoption here with Rayfin because I’m working with a industry right now or a organization where they’re concerned about about how well fabric works in their industry even though we and we’re about to send them all the case studies and get a few references here which is crazy because even fabric right now they’re like well we don’t know anyone who uses fabric in our industry therefore or you fabric in our industry therefore or it’s not going to work. So for me, know it’s not going to work. So for me, you look at Rafen and I see this right now in terms of how best to utilize it.

59:09 now in terms of how best to utilize it.

59:09 now in terms of how best to utilize it. If I wanted to start utilizing it, I can’t can’t go in expecting I’m going to replace existing systems. I think that would be a high bar to reach. There are organizations out there, Mike, that I think are absolutely going to be for it. But if you were look at the majority in how slow they are to adopt technology, they are going to be very concerned because like this is new. We don’t know about the security. We don’t,, we have to replace everything. You kidding? Do we have this cost? How do we

59:39 kidding? Do we have this cost? How do we know it can do everything we need it to do? I see right now Rafen as the best proof of concept to get started in in a few days or a day to showcase how fast we can move with fabric and what is possible with fabric. The other side that I see Rayen without being hype and really the true,, implementation of it is for things that simply don’t exist systems that we can quickly integrate into like for example,, we we need to just do a check the the vacation thing. We don’t

60:10 the the vacation thing. We don’t actually have anything with that for our role or we want to build a quota quick dashboard so everyone can quickly put in what their quota is for the month and on their team and that which gets fed into fabric perfectly. So things that may not exist and make it easy to use. use. Outside of that right now, Mike, I think it’s hype. I think the hype because you have to deal with replacing exist existing systems, slow,, organizations weary of new technology and then simply again the scalability of

60:42 and then simply again the scalability of replacing an enterprise application. So that’s where I am. I think Rayund is going to expand. But right now I would start using right in right away as a proof of concept. So I know you’re going to disagree with me. I just think it’s just a matters of how much. But where do you stand in terms of how I think of where its current implement imple implementation of it is and its current use and then what I considering hype.

61:12 I I think we’re early to this game right now. I think the structure that organizations need to understand, but I’d also argue there’s an entire team at Microsoft working on security and doing things in a secure way that I can rely upon., I’ve already already we’ve already bet on SharePoint online, we’ve already bet on using PowerBI and reports online. So, if those two are if you’re already using those areas, my argument is getting to Rafin is that much bigger. you’re already online,

61:42 much bigger. you’re already online, you’re already in Microsoft’s cloud, you’re already using all their infrastructure, like that’s the part that you’re you’re betting into. And I think that’s a secure bet. I think Microsoft is one of the the easier applications or systems to use with Entra ID and authentication and getting access to the right things. A lot of organizations will use Google or AWS, but they’ll still use Enter ID because it’s so it’s so solid, right? So I look at this going like this is this is a

62:14 this is this is a a system where we’re catching up our governance what we need to build how we build it is catching up to the speed of which innovation is happening here. I Tommy I it’s a bold statement that I say like I’m you could replace all your applications and software with Rayfen apps cuz because I think there’s there is limitations to certain things Tommy. Mhm. Mhm. But we’re quickly getting to a place

62:44 But we’re quickly getting to a place like when Rayfin gets the ability to use a fabric userdefined function as part of the back end. Like one of the things we can’t do right now, Rayfin cannot call other services. Rafin cannot call other APIs. Rafin cannot do other things outside of SQL and semantic models. So if it if you’re like again the vacation app, vacation tracking app makes total sense, right? A time tracker, right? You run a consulting firm and everyone needs to track their time on certain things, you can build the UI however you want, right? That that makes a lot of sense.

63:14 right? That that makes a lot of sense. So doing those kinds of things makes it really easy for you to build a solution that works for your team. You solution that works for your team., I’m deleting an $85 subscription know, I’m deleting an $85 subscription here. I’m deleting a $20 subscription there. I’m deleting $90 here on this other subscription, right? And then all the data goes into fabric, right? Which I’m again, I still have to pay for fabric, but at least I’m paying I’m already it’s a sunk cost. I’ve already spent the money to have fabric run. So when I look at this and I’m looking at these situations and and and

63:45 these situations and and and extrapolating here, this is where we’re at now. I’m projecting three months, 6 months down the road. Yeah, Yeah, we’re going to continue to hear a lot more news around this. We’re going to continue to absorb more of this., and I really like what this is doing, which I feel like Rafein is proving the concept that your AI agents are assistants de co-developers,, technical leads

64:15 , technical leads that are able to build what you need for your company. And they’re not just building, you’re not building a system where AI is regularly used over and over and over again. Instead, you’re building a system where you have an application that is running on commodity level hardware. It’s a static web app. It runs on fabric things. These are these are wellknown developed connection patterns and paths. SQL server has been around forever, Tommy. Right. Right. Probably probably almost as old or older

64:45 Probably probably almost as old or older than us. than us. Definitely. Yeah. So, so when we look at that, we’re building on really foundational things. And I I think we’re not far away from this just taking off. One of the other things I’m really pro on is Replet. Replet is is partnering with Microsoft to build Ray Fin solutions fast, quickly. quickly. This is this is a game changer for companies. This this is as revolutionary as bringing PowerBI and replacing some

65:15 as bringing PowerBI and replacing some of the work that you used to do with Excel. Mike and I think you want to really wow your boss or wow your team just immediately you can literally you can have a call going man we’re we’re trying to do this and be great if we do it at fabric give me an hour give me an hour and I’ll showcase to you a full app that’s integrated with fabric you want to really show people what’s possible I think that’s really really works and I completely agree with you we’re in the early stages right now of what it can what it can be so I completely agree

65:45 what it can be so I completely agree with you That’s where we’re going. I I This is This is This is the future. And actually, someone here on two comments that came through in the chat here, Tommy. one user was talking about things inside our Twitch account and talking there and saying, “Hey,, I’m worried about AI replacing my job.” if you don’t learn AI, AI will replace your job. job. Your job will be remain consistent and valid if you step into AI and learn what

66:16 valid if you step into AI and learn what it can do. You need to understand the boundaries of what AI can and cannot do. So if you’re not exploring today, you need to start exploring. And to be honest, Tommy, the barrier to get started is very low. You can start with AI for $20 a month. Yeah. Yeah. Just go get a solution and start learning it. You don’t have to use it on your company stuff. Use it on side projects, whatever. You need to start understanding how it works and what it

66:46 understanding how it works and what it can and cannot do and where your value is added. With the with the AI, right, Tommy, we could be a we could be a 5x a 10x engineer. But with AI our disposal, we become a 100x engineer. We can do way more stuff. Right. Well, the and the best way I I’ve heard it is AI is not going to take your job. Someone who knows how to use AI is going to take your job. Exactly. Right. And then the other comment here is talking maybe specifically around Rayfin. Right? There are a lot of limitations in fabric today or PowerBI for that matter. The visuals

67:17 or PowerBI for that matter. The visuals don’t do exactly what you want. There’s some custom interactions that you need and you can’t easily deploy a PowerBI report into any mobile device or any website that you have today. It’s a little bit more complicated to get embed and things working. Rayfen has none of that issue. Rafen makes a public URL that you can go to and when you go to that URL you sign in with Fabric that opens a little signup window. You sign in, you get into the into the system and the app just works. You now are authenticated and then your

67:48 authenticated and then your authentication lets you talk to the semantic model and the SQL endpoint. So this again huge Tommy this means you’re building full websites that live outside of fabric. you don’t need to go into fabric to use them, but all of your fabric users can sign in with them. This is huge. I think this is a major pivot point here is this just has made the ability for anyone to create things, right? right? Extremely wide. Oh, Oh, and so this is really again this is why

68:20 and so this is really again this is why I’m so excited about this and so why this is so impactful. So if you want to check out my demo, I actually did a demo. I showed you how to get to the URL of the website and when you open that it makes you sign in and when you sign in you can you can see all the application and and data things from there. So just be mindful of that. That’s how this is working. there’s a lot of really rich things. Now to your point about governance Tommy this opens a huge can of worms around what does governance mean and who’s reviewing these apps and what does an approval process look like for these apps. Just because you can do

68:50 for these apps. Just because you can do it doesn’t mean you should. All right, that being said, it’s a good point. Yeah, dude. Great, great discussion today. Any final thoughts, Tommy, before we wrap here? No, I Mike, I think we’re hitting the nail on the head on what it can be. I’m glad we had this conversation, too, on again where does Ray fit. I always love those conversations around other products. So, really enjoy this day, Mike. Mike. Awesome. But that being said, we want you to encourage you to go check out Rafe., I will give everyone here a playlist., so right now on the

69:21 playlist., so right now on the PowerBI tips YouTube channel, we have a I believe it’s like an eight series video around all of the Rayfin things that we have going on. So,, I will give you the link here, the playlist in the description or in the in the video here. If you want to learn more about Rayfin, check out this playlist. It’s 8 hours of content. Tommy and I talking, Matias and I talking about it. There’s actually some real demos. You want to get started, you want to see the templates that are out there, we need you need to watch this series. It’s free. Go check it out. Yeah, Yeah, it gives you it gives you a lot of the

69:52 it gives you it gives you a lot of the background of like what can be built with Rayfen. How do you make things happen with it? what can you create with it? So, that being said, go check out the series. We’ve got a number of videos around Rayen and Replet and and Rayen fabric apps and how how all these things are stitching together. Tommy, where else can you find the podcast? You can find us on Apple, Spotify, wherever your podcast, make sure to subscribe and leave a rating. It helps us out a ton. Do you have a question, topic, or idea that you want us to talk about a future episode? Well, head over to PowerBI tipsodcast. Leave

70:22 head over to PowerBI tipsodcast. Leave your name and a great question. And finally, join us live every Tuesday and Thursday, a. m. Central to PowerBI tips social media channels. channels. There we go. We’ve only done this 540 times. times. Yeah, I know, right? I know, dude. Dude, I need to write it down again. All right, thank you everyone. We appreciate you listening. We’ll see you next time. Measures pump it up high. Tommy and Mike lighting up the sky. Dance to the day to laugh in the mix. Fabric and

70:52 the day to laugh in the mix. Fabric and I get your fix. Explicit measures. Drop the beat now. H 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.