Microsoft Fabric & Azure AI Foundry – Ep. 389

Mike and Tommy unpack what OneLake security and data access roles mean for real-world Fabric governance—especially when your data lives behind shortcuts. Then they zoom out to Microsoft’s Azure AI Foundry announcements and talk through what it could mean for building AI-enabled apps on top of trusted data.

News & Announcements

This episode’s news is all about tightening the loop between data governance and AI adoption: new OneLake security capabilities make it easier to control who sees what (even when you’re virtualizing data with shortcuts), and Microsoft continues to frame Azure as the end-to-end platform for building and operating AI solutions at scale.

• Define security on folders within a shortcut using OneLake data access roles — OneLake data access roles started as “secure the shortcut root,” but this update adds the missing piece: you can scope permissions to sub-folders within the shortcut. That matters when a single shortcut points at a shared lake location but different teams should only see specific tables/folders (for example, an HR table vs. general datasets). It’s currently in public preview and pairs nicely with patterns like lakehouse schemas when you’re organizing lots of shortcut-backed tables.

• Get started with OneLake security (preview) — Microsoft’s docs walk through the RBAC model for OneLake security: define roles, assign users/groups, and grant access at the folder/table level across supported Fabric items (like Lakehouses). A key callout: this is a preview feature that’s enabled per item, and once enabled it can’t be turned off, so it’s worth testing in a controlled workspace before you roll it into production governance.

• The next wave of Azure innovation: Azure AI Foundry, intelligent data, and more — This Ignite-era announcement positions Azure AI Foundry as a unified platform for designing, customizing, and managing AI solutions, with an emphasis on bridging the gap between experiments and production. The post highlights unifying tooling via an SDK and evolving Azure AI Studio into an enterprise management console—signals that Microsoft is aiming for “AI platform operations” to become as standard as DevOps.

Main Discussion: Microsoft Fabric & Azure AI Foundry

The conversation ties together two sides of the same challenge: you can’t responsibly build AI on top of data you can’t govern, and you can’t modernize data platforms without anticipating how teams will want to operationalize AI.

Fabric governance: shortcuts, folders, and the reality of shared data

Shortcuts are powerful because they let you virtualize data access without duplicating storage—but in real orgs, shared lake locations quickly become messy. The key nuance here is that security needs to work at the level people actually collaborate: not “this entire shortcut,” but this part of the shortcut.

Mike and Tommy talk through why sub-folder security is such a practical improvement:

• It supports “one shortcut, many consumers” scenarios without spinning up a maze of workspaces or duplicating data.
• It makes least-privilege access more realistic, especially when the same lake location contains both broadly shared and sensitive tables.
• It reinforces a bigger point: Fabric’s governance story is maturing from “admin controls” to “day-to-day data product controls” that practitioners can actually use.

Azure AI Foundry: what it implies for building AI-enabled solutions

On the AI side, the big takeaway is less about a single product name and more about Microsoft’s direction: organizations want to move beyond demos into repeatable, governed AI delivery.

The discussion hits a few practical implications:

• AI initiatives succeed or fail on the quality and accessibility of the underlying data—so investments in Fabric/OneLake security aren’t “just admin work,” they’re prerequisites.
• A unified toolchain (SDK + management console) is a signal that teams should expect more standardization around evaluation, governance, and operationalization for AI workloads.
• “Intelligent data” isn’t a buzzword if it forces you to formalize your data architecture: permissions, lineage, and semantics become the things that determine whether AI outputs are trustworthy.

Looking Forward

If you’re experimenting with shortcuts heavily, this is a good moment to revisit your governance model: decide when you want one shortcut vs. many, and define what “access” means at the folder/table level. And if your org is ramping up GenAI use cases, treat the AI platform conversation (like Azure AI Foundry) as a forcing function to harden the data layer—because the fastest way to lose trust in AI is to ship an app that can’t explain (or control) what data it used.

Episode Transcript

0:33 good morning and welcome back to the explicit MERS podcast with Tommy and Mike good morning everyone good morning and I I dare I say happy new year because I there’s a lot of chat going on on when is the cut off date for happy New Year think you run out of time here it’s already it’s a week past Tomy we’re halfway through January I think you’re done so I know Larry David has a three-day limit depends on if you seen people before right if I haven’t seen them yet I feel like I’m I’m a little bit more leeway to go a bit a little bit further into the new year if I text someone does that suffice though because

1:04 someone does that suffice though because I don’t want to text to say happy New Year say you you wish me a happy New Year I I only text it because I have an iPhone and I want to have a little graphic thing to show up and it like does the little firework stuff when you when you say it so true that is a nice little little feature I I would say yeah we’re you and I are well passed off we’ve seen each other but if I haven’t seen you let’s say I took a nice vacation and it’s the 14th now like I was on vacation you already had your work holiday it’s the 14th are we you pass the deadline no matter what I’ve seen you I haven’t I’ve

1:34 what I’ve seen you I haven’t I’ve seen you since then but if I haven’t seen you scenario i’ i’ I’d say happy New Year it’s a good conversation opener I think it’s way too far even if I haven’t seen anyone it’s like I gotta wait till next year man you missed it you missed it you missed it you gota wait till another year whole year let’s get into our main topic today so our main topic for today is going to be around Microsoft Fabric and Azure AI Foundry just thinking about where data lives where it exists inside Fab fac and it looks like it’s getting easier and easier to pull data out of fabric and use it in other parts of

2:06 fabric and use it in other parts of azure as well which is nice because then that makes sense you’re going to be bringing all this data to fabric you should be able to bolt on other elements of azure if you need them anyways that being said that’s going to be our main topic for today couple news articles coming out today so there there has been very it’s January most of the development stuff slows down from Microsoft side there’s typically not a lot of updates but there has been few updates on the fabric blog but not necessarily the Microsoft powerbi blog so let’s just I’d like to kick one off

2:36 so let’s just I’d like to kick one off here I’m really interested in this feature it’s actually in preview right now but it looks like Aaron is talking about defining new security on folders using shortcuts now this seems like a very nuanced topic but I believe this is the beginning the the beginning work around getting I think they called it one L security which is kind bad way of naming it because one link was always secure it was always using enter ID to secure who

3:06 was always using enter ID to secure who could have access to things so yeah I don’t really like the term they used one lake or one security for the fabric ecosystem but it sounds like they’re just adding more granular controls to one Lake and this is something that people have been asking for I think for a while now how do I give people access to single tables or how do I give people access to single folders inside the one the lake house experience and I would agree with mik this is more than just a nice little feature I I’ve talked to some clients we talked about fabric security there’s a long pause like yeah we don’t want that

3:37 long pause like yeah we don’t want that yet like we’re not ready for that or it didn’t have all the bells and whistles or really more the foundation from a security point of view that they want yes and they’re looking at it from a hey I have data bricks I have other systems I have other security mechanisms that I’ve been using right so you you’re kind I’ve been using right so you you’re always comparing fabric against what of always comparing fabric against what I know in Azure and what I know in like data bricks as these are the levels of security boundaries I could place now you come in FA oh by the way everything you put in lace house can be

4:07 everything you put in lace house can be seen by everyone in the workspace like that’s that was a bit heavy-handed I think you have to go there initially but we don’t the features that are coming out now are helping us get down to that that detailed view of security it is a good point for better or worse everything in fabric is going to be a comparison against the current products right now because nothing’s terribly new it’s just much more integrated but yeah man when I see this when I saw this come out I was like okay I already have like three scenarios where this is going to be incorporated in training and I have

4:38 be incorporated in training and I have that client I was talking about I’m like I’m gonna call them back and go hey just so that now a feature that you were asking for that was not available that really was your deal breaker is available but Mike it’s still it’s a preview feature so it’s at least it’s being worked on again one of these things there’s it’s really hard to know what features exist or not exist based on what’s preview and not preview again we’ll be very cautious here around preview features don’t use in production but definitely test it out

5:08 production but definitely test it out now I think it’s definitely something you want to go explore use if you have a proof of concept this is a great place to go try this out with so I do like this one I think it’s they’re talking in the article about being able to manage one Lake data access you can make a role name and then you can assign people to that role and then I think in that role you can then pick out the various folders or tables or things you want people to have access to so I think right now the security is at a object level item right so it’s

5:38 at a object level item right so it’s either a table or a folder that you can grant access to I think eventually would be nice if we can get down to like hey I have a table and here’s the roow level security I’m going to apply to that table so basically filter the table down to records this user can see I don’t know if they’re going to ever get there at that point I’m not sure if we’re going to get that sophisticated but it would be I I’d like to see it event so Mike this brings up a question that I have for you I think for us to discuss right off of the security feature to your point this is very nuanced this is

6:09 your point this is very nuanced this is not just say put a Security on your semantic model or on a workspace this is very this is like it’s it’s very into the weeds so and I think a lot of these security features that are going to come out are going to follow this route so my question to you who’s in charge of this at an organization if I am in charge of the lake housee if I’m in charge of that workspace am I doing this do I obviously this is not just something that you just go into yeah this is something I think

6:39 go into yeah this is something I think that actually aligns very well with the administration and the deployment and governance road map so the adoption roap from powerbi or now fabric I think they’ve adjusted it for fabric related things but I think this is actually very clearly in line with what you should be doing already in workspaces is really think about who is supposed to be owning and controlling things inside that workspace one of the things that we’ like to convey to our customers is you should only really have one or two admins on a workspace

7:09 have one or two admins on a workspace that those are the people that are in charge of all the administration the governance of those things typically this is a we’ see this is a person potentially from the central it organization so there’s like a central it person or Security Group that’s being attached at the admin level right so that way there the admins can come in and assist with anything that’s inside the work spaces that they need to if required but there should always be some like business lead person or persons attached to that the reason being is that way when something

7:39 reason being is that way when something goes wrong or you need to go talk to someone or there’s a transition plan if someone’s moving on having one admin in a workspace decides that there’s some ownership around that the majority of the people should be contributors and not even members I think for that matter so I think it it depends organization by Organization for sure the admin is responsible for if they were supposed to be security those that’s the reach out that’s a point of contact that’s how I would look at it I don’t I don’t take the approach of a centralized Authority on everything like this

8:09 Authority on everything like this because this is this is going to be there’s going to be some organizations that have centralized business intelligence teams that do this and there’s going to be other parts of the organization that just own and manage their own things so you have to have a plan I think to manage both of these and as your organization gets larger you’re going to have a wider diversity of how you have to manage this stuff the and there’s a bit of a catch22 there right where especially with security now in Fabric and I’m going to speak specifically to fabric powerbi I think we we talked about security if you

8:39 think we we talked about security if you want to Excel and go quickly and accelerate your adoption you can’t do that and at the same time expect people to understand the security and fabric without training without going through proper channels right so you can’t have both you can’t have this accelerated adoption and expect everything to run smoothly like we’re we’re looking at subfolders and shortcuts right you’re not just getting a person who’s getting the fabric day one and saying you can understand this

9:09 one and saying you can understand this you can under you can run with this if you want the right security if you want to feel secure about that there’s going to be that additional training that you’re GNA have to do with those admins or those I don’t want to say lay people but the the business to do but then you’re hampering adoption a little because you everyone would have to go through that route on the other I see what you’re saying there though I see what you’re saying about the the the the training around things although I disagree with how you’re approaching it I think the I think you’re assuming that people are going to show

9:39 assuming that people are going to show up with day one and just like here you go you shouldn’t do that that’s

9:43 go you shouldn’t do that that’s that’s not that’s not the right approach I think because fabric actually has some weighted cost to really step into course doing something in fabric right so there there’s an additional cost they’re going to have to Pro licenses they’re going to be paying for premium at some level some level of fabric so I don’t think you give everyone access to this right this is a there are probably a select few users in your company that are going to try things out and figure out what works and those users are going to be like expected to go learn about we

10:14 expected to go learn about we don’t know how to use fabric security for our organization yet we need to figure out what that looks like or we need to go get some training from some other place or read the docs or spend some proof of concept around does this actually work for our use case so I think there’s going to be some people that are bleeding edge that you’re going to give them access to it but knowing that they’re going to figure out the policies or the procedures that will make this work in your organization for the broader part of the organ the broad up part of the organization I don’t think you give access to this I think you have there needs to be a level of gatekeeping and a

10:44 needs to be a level of gatekeeping and a minimum amount of training to get people in the door right I think there is too too many dangerous not dangerous there’s too many opportunities to share things in an inappropriate way or give out too much information that would cause the organization to say whoa whoa whoa we didn’t hardily have our right the right policies in charge here I think we’ve talked about this in the podcast before I’ve been in an organization where a report got shared with everyone’s salary from the company to the entire organization what this was it wasn’t it

11:15 organization what this was it wasn’t it wasn’t a fault of these technology it was a fault of the training or the onboarding of the organization giving people access to PBI oh it’s parbi great they’re asking for it let’s give it to them there was no thought to how are these people going to learn how to share how these people are going to learn how to build semantic models what what do we as an organization think are the right boundaries of what parbi should be able to do because it’s it can do a lot and it’s it’s designed to make it easy for people to share things a th% and but

11:46 people to share things a th% and but you’re also talking about two people who eat sleep and breathe this stuff so obviously if you and I were rolling this out it’ be a bit smoother what was your background in engineering again medical engineering so I come from the idiot engine engineering school and that engineering specifies that everyone’s an idiot until proven otherwise you and I both know that there’s gonna be a lot of people who jump into fabric a lot of organizations who jump into fabric where they’re that it’s not going to be that SM the same roll out there’s not GNA be

12:16 SM the same roll out there’s not GNA be the same knowledge and you’re gonna still have to deal with this we’ve C we’ve covered that but I think the bigger thing here is if you’re an organization just starting or if you’re the person running with this I would absolutely push for the proper channels of if you’re going to be an admin of a workspace you are very much more needing the proper training for those people if you’re going to rely on people to do that much more than even there just a semantic model where it’s like hey I can build I can read or I can

12:46 like hey I can build I can read or I can share it in an app cool well we have so many other suble levels here and so many other tentacles that you got to be aware of right right off the bat I’m gonna take a little issue to your word of or phrase there you’re using for proper training I don’t know if that’s clearly defined for people and I don’t honestly I think that might be a term like the question would to my to my standback of this whatever you say like so what is proper training what does that mean to our organization and I think proper training I wouldn’t necessarily

13:17 proper training I wouldn’t necessarily use that I would probably use just the phrase of people have been trained or have been required to go through some gone to aing or something Microsoft actually has a lot of really good documentation and actually on their Microsoft learn platform which I’ll push people to there’s actually a lot of really good documentation there and actually I like it because it quantifies a person’s score to what they’ve learned so as long as you’re signed in you can actually get a score and we can verify that you’ve done the reading of the content and there’s a test or quiz at

13:47 content and there’s a test or quiz at the end that give you just a little bit of a mental check as to did you actually read the content and actually I really like that also Tommy I think to your point as well around those training pieces of this this might be an area where I in the past I have not hired based on certifications from Microsoft oh oh oh but now that we’re talking like security and we’re talking people that have more access to the things in fabric again there’s just there’s so much to do I think I might lean a little bit more

14:18 I think I might lean a little bit more on hey this might be a good opportunity for if we’re going to give people access to these things we would want them to have gone through a minimum of a certification as a way of learning and experiencing this platform before they actually get into using it so there may be something there too Tommy I don’t is there really a there’s a lot of like a a dp600 Ved security one but they all cover it in in in various ways well everyone put in your notebooks episode

14:48 everyone put in your notebooks episode 389 where Mike talks about the benefits of training you can put that in your AI title No but that’s a good point because when I say proper too maybe that’s because I’ve been talking to Gil for from Australia and well you say say proper but I think I think you mean like there needs to be a a threshold there’s A’s bar a bar to jump proper it’s like it’s like we need everyone to have a minimum level of understanding of what’s going on here like that’s what we’re looking at right and then to to that end though to to really beat this point to

15:19 this point to death if we make those things if we make these polies hey we expect this behavior with security securing tables giving access to things for example my earlier statement hey we’d like to have one or two admins on the workspace okay if that’s your policy what are you doing to measure that and again a lot of organizations I talk to they fall off the wagon on this one like they that’s what they we have these great ideas of like policies but no one has the ability

15:50 like policies but no one has the ability to enforce them or to see where they are what like how many people do we actually have in how many workers do we have where there is more than one admin that’s the part that I think more organizations need help with is figuring out okay we’ve made these policies what metrics or what data are we getting out daily to help us meet these goals and objectives because that’s what our policies policies are it’s there you can go get the data but many people don’t actually go get it and actually utilize it as actionable information so that’s another area where I’m like people need to learn more about

16:21 I’m like people need to learn more about how to monitor their solution a lot a lot of organizations fly blind and let people just do things with minimal investment at the administration level that needs to be automated so I think there’s a lot there too as well all right what else are we going to talk about here any other topics Tommy I did also put the link in here for the learn documentation around Access Data roles and I think you’ve got another link here looks like same link I think same link for security stuff yeah

16:51 think same link for security stuff yeah oh yeah same link getting started with one link data access roles okay so those are two main topics all right enough of that let’s transition if you’re good Tommy let’s go over to our main topic today our main topic today is going to be talking about Microsoft Fabric and its connection to the the Azure AI Foundry give us a little introduction here Tommy I gotta be honest I haven’t played a lot with the Azure AI Foundry I’ve done a couple things my general feel is I found a lot of the AI tools that Microsoft has

17:21 lot of the AI tools that Microsoft has built either expensive or there’s much barrier to get so started and playing with things so I’m curious to see where you’re going to take things here and regardless at any AI platform or whenever you’re building AI related things you’re going to need a lot of data behind it so fabric is that place to go get the data from and I think we’ll start small then expand to the bigger topic here but so the AI the Azure AI Foundry if you’re not aware is the evolution of what was called the AI Studio this was simply your ability to take a given AI

17:51 simply your ability to take a given AI model let’s say a chat GPT but chat GPT 3. 5 or Turbo or 01 or whatever the models are are now you can in a sense create your own it’s off the same model off of Open AI you can train it off your own data you can add your own prompts and create your own in a sense chat GPT for your own organization there’s a ton of other things there there’s image generation text generation scanning image recognition but the idea really is simply that in my

18:21 but the idea really is simply that in my organization chat GPT is cool there’s also co-pilot but I want something more customizable I want want something more advanced for my organization or for a solution that I’m working on enter AI Foundry and that’s exactly what it’s there for now why would we be speaking about that on a powerbi Microsoft fabric podcast well one of the major announcements that Microsoft did during ignite was the fact that in the AI

18:52 ignite was the fact that in the AI Foundry in their AI platform well I need data to your point Mike you want good AI you need good data there’s no ifs ands or buts about that a direct connector now in AI Foundry is Microsoft fabric this is for me pretty incredible because our entire lives before Microsoft fabric powerbi was really the last stop of the bus the on the bus route where my data lived it went through the integration it

19:22 lived it went through the integration it went through processing it went through

19:23 went through processing it went through engineering and went through a semantic model and then it ended at powerbi my semantic model yeah I could connect to other reports from there but it did not live anywhere else there was really no other ability for what the data that I’ve already worked so hard in Blood Sweat and Tears on could do well Microsoft fabric is now showing us that I can do the same engineering work as hard on my data but now it becomes elevated in AI Foundry where I can actually use my business data whether

19:53 actually use my business data whether it’s a Lakehouse or or however I want to structure it and use it Direct an AI Foundry to me this is again if you’re not in the AR World it sounds like okay that’s nice but again a lot of organizations the biggest issue with getting their AI Solutions up is their data and now we have this direct connection with Microsoft fabric so I think the small story here for us today is this ability with AI Foundry we’ve talked about AI a lot what does that impact but I think the larger story here

20:23 impact but I think the larger story here Mike is the fact that we’re seeing with Microsoft fabric more and more that is no longer the end goal for the business intelligence team or the business intelligence platform that we’re just reporting on data well I can use it now in solutions that my organization needs it’s a SQL database which we all know what a SQL database can be it’s can be used in power apps it can be used in normal applications or can be used in AI Solutions and I think that’s the bigger story but let’s start small and let’s

20:54 story but let’s start small and let’s on the very subset here of the AI Foundry this is a big deal and Mike I know you said you touched and played with it a little but how much and what you’ve dealt with AI already in terms of the ability to work with the data but what have you seen here yeah let me give you just some context I’ve spent some time in it spaces building out Solutions around you spaces building out Solutions around lots of unstructured data right know lots of unstructured data right and to be honest I’m I play a lot around with our podcast videos our podcast

21:25 with our podcast videos our podcast videos has a lot of potential and I haven’t really been able to find a solution that does a great job so you solution that does a great job so where does where does AI start know where does where does AI start really becoming impactful and again I’m just going to speak to my own experiences where I’ve started using it in other tools and and then maybe where does that start supporting things that we’re going to start seeing in fabric maybe the join here right one thing I think is really interesting is a lot of these services that Microsoft is providing particularly through Azure there’s a whole bunch of azure bayi Azure AI Based Services that

21:58 azure bayi Azure AI Based Services that are already there they’ve been there for years but it’s not like this is a new thing for them so like for example they they’ve rev rebranded a couple things like there’s an Azure AI open AI service right so you can build your own co-pilot have an Azure AI search that’s helping you do Vector searching or hybrid searching inside these different systems you want to build there’s actually a Azure AI content Seft safety as your AI translator translating different languages for you AI to speech AI for language AI Vision so there’s a lot of

22:31 language AI Vision so there’s a lot of machine learning algorithms that have already been bu built and if I think about what I’m trying to do as an organization like let’s imagine I’m in an organization I’m showing up I have a lot of media or images or things I need to go tag one of these major things that we require is a lot of tagging of information right I have a lot of images I wanted to search for something that says like has color blue in it or it has a a picture of a computer or maybe you’re in an industry that’s doing some inspection on building and you want to search for cracked pipes or

23:01 want to search for cracked pipes or electrical defects whatever that may be your business may run on a bunch of images that need to be very quickly searchable and and optimized for like search this is where the AI Services come from and as as me as an organization I’m thinking I I like this idea of build the solution first prove that it works and then optimize later and what by that is Microsoft has already built a lot of services that do this for you was working with a client recently

23:31 was working with a client recently one good area to to apply some AI is if you have feedback coming about your products well can AI yeah can can AI summarize all the feedback coming back to your products like through your website or comments or there’s a database of some where a lot of this information comes from right what’s the average sentiment on each product that I have can we read through the comments and from those comments gather out the key like and dislikes of a particular product I think AI is

24:01 particular product I think AI is actually pretty good at those things the reason I’m bringing all this up is I think Microsoft does a good job of giving you a lot of standard let’s call it building blocks like Legos to get you started you don’t have to build all this stuff from scratch and train everything to beginning like it’s expensive to do that so I think there’s this idea of large companies The Big Three Microsoft Google and AWS are going to provide service layer for just AI AI things this is where I think Fabric and

24:32 things this is where I think Fabric and the AI Foundry is trying to make its position here is okay we’ve got a library of services that are AI related what can we do to make that easier for you to bring your fabric data like for example what if I want to bring my podcast how do I bring that into a a fabric world or an Azure data Studio world where I can utilize that information to be highly responsive right so a lot of your systems I think in AI start out as this

25:03 systems I think in AI start out as this like batch processing right I’m going to process a bunch of comments about our products and provide input back to the users at some point in the future you want to have real-time analytics on that as comments come in I immediately review that comment and say is this something that my sales team needs to take action on immediately because there was a a poor performing or someone wrote something really bad about the product and we don’t someone needs to go look at this right when when do I need human interaction and what of this can be just fully

25:33 and what of this can be just fully automated anyways I’ll just pause right there I’ve said a lot of things Tommy around this one but I I’m on the on the fence here a little bit around like not fence I I’m I’m still trying to unpack like with the scope of what Microsoft can provide in the AI Services space and then you build using their tools and eventually you get to the point where you want to be more cost economical and you stop using their custom tools and you start building custom things that you need but you now understand the value of that AI thing right it’s it’s very

26:05 of that AI thing right it’s it’s very expens I don’t want to spend oh six months developing an AI solution to only find that it doesn’t work or it’s not adding value to our customers or it’s not driving sales like that would be a waste of time and effort so what can we do to go faster to get through that initial stage and then we could always come back to that and say what can we do to optimize it moving forward so two quick things one I’m worried that you have a microphone in my office because that example you gave about the Blue Products I literally was showing my wife the other day no just imagine you wanted

26:35 the other day no just imagine you wanted Blue Products so I’m worried that you’re listening to everything that I say even when we’re not recording that microphone that that podcast microphone is actually wired into my home computer microphone so maybe it has something to do with it but no so but I think that the the implications here and I think the testing side’s huge by the way if you’re going to create something make sure you look at all the settings an AI Foundry there’s one where you can turn the safety completely off and my first thought when I saw that I like why like

27:05 thought when I saw that I like why like you go basically turns the AI safety on how you want the temperament to be and what it could in a sense respond with completely at a zero and I’m like who needs that anyways but the point is you needs that anyways but the point is Mike this is terribly expensive to know Mike this is terribly expensive to get especially with all of the scenarios even in a very direct solution that you may want what people may be asking what the Automation and the intake or input that it’s getting is going to derive a ton of different in a sense results and

27:37 ton of different in a sense results and a lot of unintended results Mike that’s why I’m doing a lot of the testing and I have a lot of local AI models right now because I’m not willing for make one person organization to spend what what could be astronomical money other organizations are the biggest thing is this and especially with organizations who are willing to test and are going to use the cloud an a Foundry is we have a budget that we know is going to be just for testing and that’s the biggest part that they have we already have this influx of data so the testing is going

28:08 influx of data so the testing is going to happen faster the more data you have the more you can refine it the more the output you can see but the problem is not having the data in the right way and not having enough of that data in that structure that you need it you may have the data but again power AI models don’t necessarily require a semantic model model and that’s also been the problem like I have this great data in powerbi I have this great data that I’ve cleaned and I’ve structured and I know it’s only the right stuff but I can’t do anything

28:39 the right stuff but I can’t do anything with it with other solutions that I’m having it would be great to do with the like with AI Solutions because my goodness again if you want your AI solution to work make sure your data works there is no way around that and what we’re seeing now Mike is especially I think why that push especially with one of the three major announcements around the AI solution was Microsoft

29:05 around the AI solution was Microsoft fabric the fact that I can now use any of the data in Microsoft fabric that I deem fit not coming from data bricks not coming from some other AI data engineering solution which are all out there expensive and disconnected to my normal data I can now use Fabric and directly connect that and I’ll feed that into that system and to your the last point that you said too is like in in the sense of what the output or the results are well again all the normal

29:35 results are well again all the normal models that you’re aware of today and again the three major ones here the Google AI Studio Claud or anthropic and then open AI are all available and I can create off of that model and that can be the base of what I want AI to do or what I want my solution to do it’s not all this chat or genitor of AI again a huge one today is image generation or image recognition is blue can I find certain things but the biggest Point here out of all those Solutions no

30:07 here out of all those Solutions no matter what you’re dealing with if you want it to really be the expectation you’re at you’re looking for it needs the right data and it’s probably gonna need your need your data this is interesting because I’m I’m looking there’s a lot of these things around like a rag a rag AI soag yeah retrieval M mented mented graph generative something like that I don’t know what it is but I think the idea really is hey I’ve got a a large language model and I’m adding some

30:37 language model and I’m adding some supplemental data to it that’s going to help it give me better context so this is this is I believe what’s happening when you’re using like co-pilot inside your emails right say hey based on this email what what’s the summary of this information and I’m finding me I’m starting to like I’m trying to utilize utilize it in more efficient ways like I have a lot of times I’ll have customer request will come in I I run a training. tips class for people to learn more about powerbi and fabric and I and sometimes I get requests in there that come through and one thing I’ll use

31:07 that come through and one thing I’ll use is I’ll take the comments out of that or comments or responses from a form and I’ll just summarize it I’ll use co-pilot and say can you just summarize this what are the main bullet points this person’s asking for what services are they asking for right so I sometimes I get requests so I use that as a method to groom some of my sales Channel and so instead of having someone read all of that and then sum Mize it into bullet points I can use the AI to do some of these things as well so me technically I can do more on my own without needing additional people or resources to do these things where

31:39 or resources to do these things where things are going to get interesting think I think here especially in the the AI Foundry is there’s actually even and this maybe part of AI Foundry I’m not exactly sure if it’s all bolted together yet Tommy but you can go to GitHub models I believe as well that’s another one too yeah well that’s not connected to AI Foundry but is it’s but it’s like the same models though it’s like it’s still using like met it’s still using like llama 3. 1 but same organization different division so okay okay yeah so this is there’s

32:09 okay okay yeah so this is there’s another feature that I have played with a little bit is GitHub models and so the idea here is M Microsoft will host these various models for you and again it takes compute to run these things a GPU and it’s expensive potentially to run these things so what they’re doing is they’re saying look here’s a bunch of free models that you can go play with you and see if they add value again I think what the idea here is the barrier to entry on some of these products you need to just understand does this yeah thing you’re thinking of does it even add value and so does the cost of

32:41 add value and so does the cost of running this model outweigh the value of what you’re producing with the output of it right so sorry total side note I’m just randomly thinking about mean I’m just randomly thinking about the AI where this F point though too you have to be conscious of the cost and the computation that it takes and and when I look at fabric I think a lot about fabric fabric is more about like numbers tables data yeah maybe there’s some API things in there but I think at the end of the day most of the things we’re talking about in fabric are cable based information I don’t think fabric is

33:12 information I don’t think fabric is becoming my go- to place for like where I want to store video or where I want to run a bunch of yeah you run a bunch of yeah take this video and turn it into a know take this video and turn it into a transcript there’s the opportunity to do it inside notebooks but I’m not utilizing that yet inside fabric most of what I’m doing is pure data engineering table data that that stuff so yeah I think this is another potential area where like where does the AI fit with where fabric is going because a lot

33:43 with where fabric is going because a lot of the AI feels like it’s fitting on top of like language spe text audio video that seems a lot of like where the AI thing goes because that’s those are difficult problems to solve maybe I love where you’re going with this do this does that make sense like I love where you’re going with it I feel like there’s a break in the sand between like where I feel like AI is fitting right now versus what maybe we want AI to be doing in the future with our data I love where you’re going with this in the direction of fabric too and a side note to your computation point of view I don’t know if the gpus

34:13 view I don’t know if the gpus on your your graphics card but man I went down rabbit holes and I understand everything about my Nadia graphics card and what it takes to run what but to your point and I think this is a really good transition here where we’ve still and I I I don’t know if we still do but I think there’s still the common out the common thought process around fabric is there’s a certain user story I’m going to process data in I’m going to store it I’m going to standardize it and I’m

34:43 I’m going to standardize it and I’m going to report on it that’s the path a b and c for a lot of people on what I’m going to do with my data and fabric you bring up a very very good point on I think the capabilities of fabric that I think a lot lot of people aren’t thinking of let’s say you wanted to do your scenario doesn’t mean it can’t do it doesn’t mean you can’t throw and actually I’ve seriously talked to you Tom me about a lot of this on like behind the scenes here I’m like I would really like to be able to like stop storing our videos on like a SharePoint Drive I’d like to start

35:13 SharePoint Drive I’d like to start sticking them in like Fabric and start leveraging fabric to like do things with the video like how cool how cool would it be to like index all the topics and the the phrases and like what we’re saying like I should be able what I in my vision here is I don’t know where this probably exists someone has a product so if if someone’s already interacted with this or has their product put it in the chat window I’d love to hear about it but I’m pretty sure there there’s a tool out there that would probably like hey here’s 388 episodes of podcast go run

35:43 here’s 388 episodes of podcast go run indexing on all the transcripts find all the main topics and key phrases make make a library of them and then just give me a link to here if I search for something give me all the key topics out of it get me to that area so where does AI fit in this AI fits in the taking the video and the audio and turning it into actual text computer readable text and then another part of the the AI space would be like building like a graph right so yeah exactly Jack right how

36:13 right so yeah exactly Jack right how many times did the the explicit measure podcast say GameChanger and what features were those about right we should be able to ask questions like that against the the topics and content we’re creating here and my opinion is the world is just going to be consumed more and more and more with video and creation content that it’s going to be video format because it’s becoming easier and easier to build that stuff do one of one of biggest did please tell me what if you were to guess what is one of the

36:44 if you were to guess what is one of the largest departments that is utilizing AI today if you look at an organization what do you think which department or one of the Departments that is utilizing Mo heavily AI Homeland Security Department no not government departments organiz take take an take an organization one of the major departments which one is utilizing Ai and the AI tooling the most or most heavily heavily today I’m GNA propose where I think

37:14 today I’m GNA propose where I think they’re using it based on SP I think marketing is finding the most amount of value from AI because it’s so in the past you’d have to again this is this is where I’m thinking like you’d have to pay p a lot of money to the marketing people and say hey I need a image that does this and that would be hours of time of someone sitting in a graphics program and like what okay what styles do you like what colors is your cool like what do you want I don’t think we have yeah but the idea is like we have you have to spend a

37:45 idea is like we have you have to spend a lot of time there to get to that point and now I’m seeing articles coming out that say look companies are saving millions of dollars a year on generating like 80 pie P of content and then letting the AI say look okay here’s all these different pieces of content send them out which one of those are performing and the AI generated content is outperforming the human generated content at this point for like Returns on their money and Investments like this is becoming really really interesting

38:15 is becoming really really interesting yeah in the world of fabric too so when I think when we think data right when I think the whole last nine years of Our Lives when we think data I’m thinking Tabler models rows aggregations something in da that’s data well to to a larger organization data is what’s in SharePoint data is what’s in one drive data is your pamphlets and your media and all the other things that we would go well those are files but that’s still data and yes use an AI movie so to speak

38:46 data and yes use an AI movie so to speak see if you can know which movie I’m talking about I think we need to think about fabric to free your mind remember the is that is that one with the pill the the blue pill well not the very scary to watch especially after chat GPT came out yes true but but no but really the idea here is we have the path a b and c for fabric for since Fabric’s coming out I think for a lot of Us coming from the business intelligence background is fabrics for powerbi fabrics for reporting fabrics for our data that

39:17 reporting fabrics for our data that we’re aggregating counting and in a sense our metrics well the thing is to your point Mike the media side of things and again that’s side of it fabric can do a lot and it is because it is a full-fledged Lakehouse it’s a you is a full-fledged Lakehouse it’s a it’s a it can it can store the know it’s a it can it can store the media we can store the media and again marketing is just a subset here it’s not just that type of media or that that it can be files too and because of things like the AI Foundry now or really I

39:48 like the AI Foundry now or really I think I’m GNA put it the other way the things like fabric that allow us to now connect to other sources where fabric has elevated our data and more importantly the storage of our data into other services other platforms is absolutely huge and it’s time for us to really think about well what else can fabric can do with not just our aggregated data but the data that we you aggregated data but the data that we since live

40:19 know since live by this is going to be interesting I’m not sure where this is actually going to land on the topic of things here but here but are are we going to start see so where does AI fit in fabric right where does this so where’s the lens of this I start becoming relevant to us so we have all the data so pipelines are getting data in you have mirroring of various artifacts you can like mirror SQL database which brings that data in we now have fabric SQL so you can do operational data right inside your fabric environment like an

40:50 inside your fabric environment like an example might be is if you’re collecting feedback forms from someone or you have something on a website where you actually want to get that data in and analy it it might make sense to then start Landing that data initially in Fabric and building the infrastructure directly in Fabric and using that as the landing zone for that operational data because you’re getting comments back or feedback or whatever that is you’re going to want to put it somewhere and you’re going to want to do some analytics on it but I think the notebooks in my opinion here that’s where we start moving away from like the data engineering and the analyst roles

41:21 data engineering and the analyst roles and we move way more into data science Ai and bolting things in there and so there is a lot of Rich features in there and so we’re also getting the ability to add additional notebooks as well so you can run a pi a pure python Notebook on a single cluster inside notebooks now crazy so so crazy and I feel we haven’t given enough do credit well my my one question here though is and I don’t know maybe someone can fact check me in this one are we able to use

41:51 check me in this one are we able to use a GPU graphics processor enabled unit spark cluster inside fabric so my understanding is if you’re going to process video or audio with a spark library or libraries you need to have a GPU enabled cluster to do so and I don’t know if we get those by default what comes out of fabric and my understanding is the GPU clusters are very expensive this is Tommy why you run models on your local machine because you need the GPU available to you to run all

42:23 need the GPU available to you to run all these heavy calculations on things so I still like going out to eat so yeah right you don’t want to spend all your money and again it’s like what works like what is what is what is going to be useful to your organization you need to experiment these things first so you you want those you be a burden two important Concepts I think to to your point you cannot forget how early we are in the game right now when it comes to AI Solutions AI products and just having the AI and the cloud we are I would say

42:55 the AI and the cloud we are I would say we’re like in the 1999 to the internet bubble with AI stuff right now where it’s like internet for everything look AOL we’re going to the number one thing about a AOL back in the day was growth it wasn’t even they were losing money hand over fist but the number one metric in the why their stock went up was growth was just the number of subscribers and I think right now with AI we’re still seeing this we’re still so new into the game and to I think the implications of it that yeah if you were to run this Mike let’s say

43:25 if you were to run this Mike let’s say we’re going to do that test we’re going to run on your marketing media information we’re even though it’s just our podcast well we’re not go you just our podcast well we’re not go we’re going to forgo our vacation know we’re going to forgo our vacation this year too or your taxes I don’t know whichever one thing but it’s going to be a pretty penny and that doesn’t mean you shouldn’t start I think the biggest part here is there’s a lot of testing that needs to be done here too and like if I was to if I were to have someone to come and say we really want to try this in fabric well we’re going to create a subset I would have it

43:55 going to create a subset I would have it own skew I would not include this with our normal fabric skew I would have it obviously it’s its own separate environment in fabric to do so and I there’s a lot of benefits here but I think there’s also the we’re still lot we should you still be very cautious in terms of just getting started I don’t know if that answers your question terribly or if you want to rephrase it no I guess what I’m I’m going with this is like the Azure AI Foundry is

44:25 this is like the Azure AI Foundry is more equipped for handling those graph iCal and media type like the unstructured more of that unstructured data we can do a lot of things with what we have in fabric today but I don’t believe there’s like I’m even looking at like I’m going into the the spark configuration settings right now there is no option to pull down a graphics a GPU enabled cluster that you can use right so let me give you some contrast right I’m familiar working with data bricks and if I want data bricks to go process images video other things or write notebooks on top of that with spark you can do so inside data bricks

44:57 spark you can do so inside data bricks you go into the cluster configuration and you can pick a graphic enabled a GPU enabled item from data bricks then you can utilize that to do the processing of that data today in fabric if you brought a bunch of video in there I don’t think you can process the video today currently because there’s no GPU enabled items so this is where I think a good fit for AI Foundry makes sense is you’ll need to you could store the videos in fabric but you’ll need to go to something else like an AI Foundry process those videos down

45:27 an AI Foundry process those videos down fit out the output of that transcript or topic or analysis and you could write that back into the lake housee and then you could report on it so we’re not it’s not fully integrated yet but I do think that fabric becomes a data source to AI yeah I may not be able to process in fabric like for to your point I may not be able to write the notebook in Fabric and do my analysis there over the media but I can store that data in Fabric and it’s going to get processed on the azure

45:57 it’s going to get processed on the azure Foundry side not the fabric side correct so you have to you going have to have a relationship there between the two parts of those tools right and I do think it makes sense to keep it in fabric because then eventually you’re going to want to get back out into Fabric and utilize it for reporting or other needs as well I think as we get closer unless you had anything right off of that too but I think as we’re yeah I think as we’re getting closer I think one of the biggest takeaways for me here is the fact that our fabric data is elevated now to a point where

46:27 now to a point where I think AI Foundry is one of the first ones that we’re seeing this with but we’re going to see it in a ton of other places where there’s this direct fabric connector with my stored data and I think we’re gonna I’m surprised we haven’t seen it directly in co-pilot like but all the other Microsoft systems but it’s going to be much more than just Microsoft too because let’s not take away the fact or really in a sense undermine the fact that I have a complete Lakehouse whenever I create

46:58 complete Lakehouse whenever I create something in Fabric and the ease of that and what a Lakehouse right now can connect to or what a SQL database can connect to what we’re going to be seeing more and more is not just that I’m going to be reporting with my data and fabric but I’m going to store data in Fabric or I’m going to keep data in fabric to connect any of my other systems any of my other products or platforms that you my other products or platforms that my CRM custom CRM system well guess know my CRM custom CRM system well guess what it’s it’s a Lakehouse that has this direct connection is data this is pretty

47:29 direct connection is data this is pretty incredible for us especially coming from just a powerbi background or from just a business intelligence background it really grows the game I think in terms of what our role is going to be or what your current your current role was it this I agree with you Tommy there and I think we’re we’re at a I I’m worried that we’re in a bit of a hype cycle at this point with AI I still think it’s it’s not quite solid enough to for people to really like implement it we’re finding very specific

47:59 implement it we’re finding very specific small use cases where I think AI makes a lot of sense right summarizing various things I don’t I don’t know if I I I didn’t catch wind of this but I I was hearing some other people like chatting like there’s a bit of like I don’t know overhyping maybe I like I don’t know overhyping maybe so some of the micro Market mean so some of the micro Market Microsoft market price has been dropping down here recently in the last couple days and it it sounded to me like the market is was excited about a lot of

48:26 market is was excited about a lot of AI developments Microsoft night happened but the market was like okay yeah this is this is cool stuff but it’s underwhelming at this point and so I think I’m not sure if we’re at the top of that hype cycle this is the I’m thinking about the Gartner hype cycle right there’s this excitement ramping up to like oh this new technology it could solve all these problems and then as people dig into it they find oh wow this is a lot more difficult than I found when I started it it solves 75% of my problems but it’s not really solving everything I want it’s not quite right so the idea or

48:58 want it’s not quite right so the idea or the concept or integration of these things need to continue to refine themselves one one thing I think is going to be incred increasingly important for us whatever you do consume the internet or however you use the internet but I think it’s going to be harder and harder to discern what things are real what things are actable because AI is going to be everywhere especially in these marketing platforms if you bring Ai and make marketing cheaper to do for companies what is that going to do it’s just going to spawn more marketing advertising

49:28 to spawn more marketing advertising spend or advertising in general from companies right if they can if they can get better results and they can find that it’s cheaper for them to at least create the creatives initially all that’s going to do is just drive more of it so I think we’re in a in a place here where we’re it it’s in early days stages I like learning from people who have already done things before me so me my take away from this one really is research what other companies are doing doing building models where they apply in the

49:59 building models where they apply in the models which programs I have a I have a a list in Microsoft lists guess it’s just a their lists program whatever in that list I just made every single AI that I hear or being advertised I just write down the name of it and like what it does and and just listen to I’m just like taking an input right now and furing out okay what are all these things do I just watched a video yesterday on YouTube of a content creator saying that they are now using generative AI to deep fake their

50:30 generative AI to deep fake their voice and use it as transcripts so like if you’re building content where you’re regularly outputting the same content over and over again for hours on end like creting documentation right it’s mundane you have to just get the details acrossed sometimes there’s mistakes and you have to go back and re-record something or something updates you want to update that portion of the code it’s nice to have generative AI just step in and just fix a one or two minute segment of that using generative AI we talked about this a year ago where we could go

51:01 a year ago where we could go International yeah we could we could do like we can do a whole Italian one explicit measures in different languages like we could we could start expanding but again a lot of this is like these are just the the the initial use cases of this yeah what we need is we need more people to come in and build a complete solution I feel like what the AI is doing is what Microsoft is done with fabric right now is Microsoft is making fabric a data commodity everyone can build data and a

51:32 commodity everyone can build data and a and and reports and analytics right that’s making that easy for everyone is has been the goal and I think fabric is making this easier every day it’s not super seamless yet but it’s getting better every day or every month we we go through the program I think AI is in the same route a lot of core functional building blocks if you’re super technical you can make it work for you but AI right now isn’t super smooth there’s not like a take this data table let me give you some other yeah I’m G to step

52:02 give you some other yeah I’m G to step back here just a minute and and give you some big picture here right imagine I’ve got a semantic model I’ve got my facts and dimensions formed and and flattened out I want to just query that data and send the output of that result row by row over to some AI thing I think we can do that today with notebooks in its current form but the integration needs to be tighter it needs to be easier to do I I’m really do think there needs to be more of this graphical user interface of click this button pick these AIS create this new

52:33 these AIS create this new table of data and then spit it back out and land it in our Lake that’s what we need to get to and I don’t think we’re quite there yet but I see visions of this oh boy this is another episode right this is I’m I’m having another episode one type thing right if I could if if I can put put my thinking cap on and say what does this look like five years from now I think it’s we’re all going to control it by the phone it’s it’s no I don’t think phones are the way to go I think phones are too small to really do stuff but I think I think in general the UI will become simpler and

53:03 general the UI will become simpler and there’s going to be tighter Integrations to models and raw data that’s coming out of systems and you’re going to be able to bolt these things these two worlds together better inside fabric I think that’s where we’re going let me give you and I’ll close with some optimism especially I think for a lot of people too in terms of where we’re going with AI a lot of people the best way to equate where we’re at with AI even the big players and the small players is the internet bubble if you to might if yeah we may be in a hype bubble right now but

53:33 we may be in a hype bubble right now but let’s consider the internet bubble when the internet Bubble Burst did the internet go away did it just disappear it kind it took good St and down so it didn’t disappear fully but we obviously still use the internet pretty heavily today and I think right now to your point yeah it may not be the perfect system yet or the we haven’t found the perfect pathway from a user interface to the process intake and then what the actual product’s going to be but I think for a lot of us too if you’re thinking well it’s just a hype bubble well it’s the hype bubble to say

54:03 bubble well it’s the hype bubble to say the the same way the internet was it’s not going to go away it’s but it’s definitely gonna change forms yeah I think I think it’s gonna there’s going to be some big players in this space that’s going to make this easy for us to get into and that’s that’s what I’m looking for right there’s a lot of a lot of no-name companies right now not creation there’s so many different companies and smaller companies we need like a shake out of the market right we need like exactly there’s a lot of good ideas probably happening all at once but there’s so much of it How can any one person beely be up on all these

54:33 person beely be up on all these different AI models and companies and everyone’s doing their own little spin on this and trying to build programs extremely fast like some point this is going to have to shake out and the the companies that stand or that are left standing at the end of this are the ones that are going to add real value to businesses there’s the ones that people are going to really be able to to invest in so yeah I I there’s going to be a wave of this and I think we’re we’re a place where there’s too much right now we need some of the main players to rise to the occasion and we need the people that are just making noise to

55:03 people that are just making noise to disappear because I think it’s also cluing the the waters a little bit here it’s hard to see what’s the right tool I should be using for some of these AI generated things and while you’re waiting clean your data yeah while you’re waiting when that time comes they’re going to be waiting for the data so that’s exactly what we need to do that’s a good point there Tommy too I think I think it’s get your arms around well you say clean your data I’m also going to Echo another Point here that is just bring it to the same place right I think we’ve had also had a

55:34 place right I think we’ve had also had a challenge of a lot of data exists in many different systems all across our company even just for reporting on what historically has happened we struggle to find it all in one place and I think I think cleaning it making sure we have good layers of data understand what the data is doing is going to be ever more important but we’re also going to need a place to just put it all and this was probably about 10 years ago when I was doing some of the analysis for I was researching data things for another company that I was working for and they were saying

56:04 was working for and they were saying data is growing at an exponential rate the amount of data is being generated is being is growing exponentially and here we are 10 years later and there’s just going to be pedabytes and pedabytes and pedabytes much more data next year than it is this year we’re going to need systems that are going to be able to handle large volumes of data and I think this is why Microsoft’s moving this direction towards fabric pipelines Spark more efficient systems here okay excellent well with that being said we really do appreciate your time I hope this conversation wasn’t too random talking about like where AI fits and

56:35 talking about like where AI fits and where fabric fits I think Tommy and I see a bright future Tommy you’ve been very heavy and pro AI based things in general a lot of playing around with it me not so much I focus a bit more on like the data engineering side for now but I don’t see any reason why these two worlds don’t Collide more in the future and I could definitely see a vision where fabric combined I I feel like there’s got to be a vision here where fabric becomes more AIC Centric I think that’s how Microsoft is going to get more people invested in using fabric connecting it with AI that being

57:07 fabric connecting it with AI that being said thank you all very much for your time today if you like this podcast please give us a like or a comment down below let us know your thoughts on this are you using AI today currently in your organization do you have have you found any major wins with using AI I’d be curious and also let us know are you thinking about using Fabric in in concert with AI be interested to know that as well please share this podcast with somebody else if you found this valuable and with that Tommy where else can you find the podcast you can find us on Apple Spotify or wherever you get your podcast make sure to subscribe and

57:38 your podcast make sure to subscribe and leave a rating it helps us out a ton do you have a question an idea or a topic that you want us to talk about in a future episode well keep going to power. tips podcast because a ton of you have and leave your name and a great question and finally join us live every Tuesday and Thursday a. m. Central and join the conversation on all of power tips social media channels all right we appreciate you all so much and we will see you next time [Music]

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.