View Recording: Best of FabCon 2026: What’s New in Microsoft Fabric

Brian Haydin All right. Well, hey, everybody, welcome in. Glad that you were able to make it. I know that Fabrikon happened, you know, a couple of weeks ago and some of you may or may not have been there. 0:0:18.684 –> 0:0:35.324 Brian Haydin Been there. If you were, you probably heard a lot of these announcements, but if you’re if you weren’t there, that’s better because it’ll kind of be a little bit new to you. So what I tried to do was pull the signal out of the noise because there’s a lot of things that they were talking about. 0:0:57.914 –> 0:1:17.794 Brian Haydin There’s a lot that’s going to be going on. And we’ll pause a couple of times just to make sure that we can answer any of the questions. I’ve got Sunir with me today. So he’s going to be jumping in with some technical depth and maybe a healthy dose of here’s what this actually means for you. And so, Sunir, you want to introduce yourself? 0:1:18.594 –> 0:1:35.634 Suneer Mehmood Hello everyone, Sunir, primarily focusing on data and AI working as an architect here. So I’m really happy to be here. Fair warning, I’ll be the one telling you which features are actually available in your tenant today versus which ones Microsoft announced on stage with the. 0:1:36.794 –> 0:1:41.754 Suneer Mehmood you know, yeah, stage recently, if that makes sense. So happy to be here, yeah. 0:1:40.834 –> 0:2:1.234 Brian Haydin Yeah, yeah, and that’s kind of what the value is that we’re trying to do here today, too, is just to give you the lowdown of things that happened that you need to know about. And we’re going to call out pretty actively what was, what’s real and what’s not real. So there were a lot of announcements. And when you strip away all the stage effects and 0:1:52.594 –> 0:1:52.634 Suneer Mehmood I. 0:2:1.274 –> 0:2:26.114 Brian Haydin you know, the one-liners, you know, underneath it all, unify your data estate 1st and then let the agent help, the agents help. That’s like the one-liner, the big takeaway. So everything else that was announced in FabCon, when you talk about like database hub or Fabric IQ or the new OneLake security features, all the agentic, you know, dev tooling. 0:2:26.434 –> 0:2:38.34 Brian Haydin That’s all downstream of that kind of sequencing. If you can’t unify your data, then your agent, it’s really just going to be kind of flying blind out there. So what else were you hearing out there? 0:2:39.474 –> 0:2:40.114 Brian Haydin Suneer. 0:2:39.634 –> 0:2:57.874 Suneer Mehmood Yeah, like the buzzes, in my opinion, in the halls wasn’t just wow AI, right? It was finally one place to see and govern all my databases. So that’s like very, very encouraging. I’m, you know, I’m looking forward to, you know, take a 0:2:58.74 –> 0:3:7.474 Suneer Mehmood deep dive into a lot of these. Database Hub got more traction than anything else, even though it’s early access. So that’s what my take on that is. 0:3:7.954 –> 0:3:26.274 Brian Haydin Yeah, we’re going to dive deep in there. But before we go feature by feature, you know, we talked a little bit about this. I just want to be up front because I want to make sure that we’re building, like we’re not pretending that everything on stage is brand new invention. Some of what got announced at like FabCon is actually like new stuff, right? But 0:3:26.754 –> 0:3:46.594 Brian Haydin Data like database hub that didn’t exist before this convention. Fabric IQ kind of, but it was basically released, you know, in conjunction with FabCon. You know, the planning in the right back direction is new territory, but some of what got on the state, like what got the energy is important, but it’s kind of like an incremental sort of thing. 0:3:46.954 –> 0:4:10.194 Brian Haydin For those of you that are in the developer world, I see this pattern happen all the time. Microsoft Build gets all the cool announcements. This is stuff that’s coming out, but none of it, it’s all in private preview at Build. But when you get to Ignite, that’s when things are actually starting to be real. Some of that happened here, you know, at FabCon, right? So there’s some private preview things that got rolled out. 0:4:10.914 –> 0:4:35.154 Brian Haydin you know, some things that went from private to just regular preview and some things that went to GA. So what kind of, I just want to make sure that we’re, you know, as we go through this, we’re going to be calling that out and helping you sort of figure that out. So, and we’ll be explicit about that, what’s in your, what’s in the tenant and what’s not available in the tenant. So in our slides, we’re going to have like, you’ll see it, it’ll be GA in. 0:4:35.394 –> 0:4:54.914 Brian Haydin and not GA and preview and all that kind of stuff. So on the platform architecture, let’s take a big picture, a slice across the top, like what does the map look like? You know, it’s not just, this is like taking all those lists of the announcements and like packaging them up into like little areas. 0:4:55.394 –> 0:5:15.474 Brian Haydin So step one, unify your data. What I’m talking about here is getting your data into OneLake, or at a minimum, getting it connected to OneLake via some of the shortcuts or the mirroring that’s out there. One copy, one access point, you know, one set of, you know, one place that you’re going to go get the data. Step 2 though is all about governance. 0:5:15.874 –> 0:5:34.674 Brian Haydin So centralizing your security, centralizing your access policy, making sure that it enforces consistently across all those different engines that are reading the data. Step 3, we’re adding meaning to this. And this is really important. I’ve got some things that are probably going to go off script here and talk a little bit about. 0:5:35.74 –> 0:5:55.714 Brian Haydin But Fabric IQ, like this is ontology, this is planning, this is the graph of your data, the semantic vocabulary that makes the AI outputs like what they are. And, you know, I know we’re going to talk a little bit more about this, but I want to compare and contrast some things with some of the other platforms that are out there, because I’m hearing a lot of signals from 0:5:55.994 –> 0:6:15.314 Brian Haydin from my customers. But step 4, we want like the whole purpose at this point of technology is to enable agents. So we now have MCP tooling, we’ve got skills, we have jump starts. The real agentic developer experience that let you build fast. 0:6:15.474 –> 0:6:25.954 Brian Haydin agents. Now, once we have these Deb 1, 2, and 3 and have the foundation in place, you’ll be able to do some of those things. Suneer, any other thoughts on this? 0:6:26.274 –> 0:6:47.994 Suneer Mehmood Yeah, I just wanted to call out that the database hub that we discussed in the last slide, it’s not necessarily a fifth step, right? It’s a management overlay that spans all those four blocks, right? Think of it as control plane view, excuse me, that gives you access to your entire estate, right, regardless of where those databases live. 0:6:48.274 –> 0:7:1.634 Suneer Mehmood in the sequence, if that makes sense. Pretty much like one leg spans across all your databases. Database hub also gives you access to all of what you, you know, host in fabric. So. 0:6:50.354 –> 0:6:50.914 Brian Haydin Yeah. 0:7:2.354 –> 0:7:23.874 Brian Haydin So as we’re like kind of going through stuff today, think of this is like, you know, everything is going to fall into one of these four categories. So this is kind of like a reference map. So 5 announcements. These are the ones that have actual teeth, right? So they’re either available today or they are going to force some architectural decisions that you have. 0:7:24.434 –> 0:7:44.754 Brian Haydin So as I kind of walk through these, I want to set up a distinction that will make the rest of this session a little bit more useful for you. Two of these are genuinely new architectural bets, things that didn’t exist before FabCon, and three are important, like general availability graduations of capabilities that have been building for months. So we’re flagging a little bit of like, 0:7:45.34 –> 0:8:4.914 Brian Haydin which is which, you know, and both of these matter because you have to incorporate these features of these things that we’re going to be talking about into your roadmap and into some of the architectural decisions that you’re going to make. Now, Suneer has already talked a little bit about the database hub. That is your single. 0:8:5.34 –> 0:8:25.114 Brian Haydin pane of glass across your entire database estate. This is genuinely new now. Like this is something we’ve got to screenshot a little bit later, but you know, it’s genuinely new and it is early access at this point. So you’re not going to see a lot of it, you know, right out of the gate, but we’re going to start to see that. 0:8:25.234 –> 0:8:44.34 Brian Haydin incrementally over time. The direct lake, you know, direct lake on OneLake, the Power BI performance story Microsoft called out, you know, it’s going GA at this point. So important graduation, but not necessarily a brand new idea. The OneLake security model. 0:8:44.394 –> 0:9:5.74 Brian Haydin This is centralized policy. It’s enforced at the query time. Multi-quarter effort that’s kind of coming to a close and it’s still rolling, but I think you’re going to need to verify some of the features and feature parity inside of your tenant as this stuff is kind of rolling out. The Fabric IQ components. 0:9:5.554 –> 0:9:25.314 Brian Haydin This is pretty new workload. I mean, I think like it was officially, officially announced a couple of weeks before FabCon, but that was, you know, basically when it was rolled out. This is huge. I can’t talk enough about this. We did a webinar about it, I think, a couple of weeks ago, or at least we talked about it a couple of weeks ago. 0:9:25.954 –> 0:9:45.794 Brian Haydin And this is a genuinely big bet. I actually think this is a differentiator in how the different data platforms are going to look at agentifying or MCPing, you know, their data. And Microsoft is not the only one doing this, but I think they’re getting in early in the game. It’s going to be a differentiator. 0:9:45.994 –> 0:10:4.674 Brian Haydin And then lastly, like they are really elevating the agentic experience for developers. So enabling MCP servers, allowing you to build skills. Some of the Jumpstart, you know, stuff is new. Jumpstart stuff is coming out. MCP server, it’s not new. 0:10:4.954 –> 0:10:16.674 Brian Haydin Like this, like the protocols been around for a bit, but being able to flip that on in in fabric is a is a new thing, so any other callouts that you have? 0:10:17.394 –> 0:10:42.74 Suneer Mehmood Yeah, I mean, the status badges right on these cards are doing real work. Some were introduced in the keynote with energy, like, you know, but if you look at the learn documents today, the preview is still stamped on more of them than the conference narrative suggested, right? And that’s not, yeah, that’s not unique to this conference. Like, it’s a pattern worth knowing about, so you can set expectations with your stakeholders before they. 0:10:34.34 –> 0:10:35.34 Brian Haydin Always this. 0:10:42.74 –> 0:10:43.554 Suneer Mehmood they set them for you. 0:10:45.154 –> 0:11:4.114 Brian Haydin Yeah, and you know, really knowing that distinction, Suneer, right, ahead of time is kind of what separates like a good roadmap from just like, you know, a compliance incident, right? Something like that. So, hey, you know, I did say if there’s any Q&A, let’s pause, take a look. Anybody have any questions about what we talked about so far? 0:10:50.514 –> 0:10:50.914 Suneer Mehmood Yeah. 0:10:53.234 –> 0:10:53.794 Suneer Mehmood Exactly. 0:10:56.594 –> 0:10:56.994 Suneer Mehmood Yep. 0:11:6.514 –> 0:11:24.34 Brian Haydin Feel free to drop those into the chat or use the Q&A. I’ll try to go back and forth with them. Suneer, call me out if I’m going too fast. So, all right, let’s talk a little bit about the database hub. So genuinely new product surface. This is not a rebranding of anything that’s been out there or already existed. 0:11:13.394 –> 0:11:14.354 Suneer Mehmood Nope, yep. 0:11:24.194 –> 0:11:44.914 Brian Haydin And so it’s a unified management surface inside of FabCon that expand that spans things like your Azure SQL, Cosmos DB, Postgres, MySQL. All these things are kind of being, you know, in one single place. So one place to see the health of the ecosystem, performance recommendations. 0:11:45.234 –> 0:12:3.154 Brian Haydin governance posture, and then also some of the cost, you know, signals are there as well for your for the entire data state. And so Microsoft is going to be layering some of these agent capabilities, you know, on top of this to help assist you in making some of the like the decisions that need to happen as well. 0:12:5.194 –> 0:12:6.994 Brian Haydin What are your thoughts on this? 0:12:7.474 –> 0:12:26.594 Suneer Mehmood Yeah, technically this is a control plane abstraction, right? You’re not re-hosting your databases into fabric, you’re surfacing their management metadata into a single fabric context, right? So the databases live where they live. What changes is your observability and governance surface. 0:12:27.474 –> 0:12:31.394 Suneer Mehmood That’s important to understand when you’re sizing the adoption of work, right? So. 0:12:32.434 –> 0:12:52.674 Brian Haydin Yeah, so I want to be a little bit more specific. I said agent assisted, you know, and it is in early access, but you know, when we heard about this in the keynote, you know, it seemed like the framing was a little bit of, it was a little bit ambitious. What’s shipping today is really a little bit. 0:12:52.914 –> 0:13:14.754 Brian Haydin closer to like Azure Advisor, but now we’re going to bring that inside of FabCon. It is the unified view with AI recommendations, and it’s going to allow you to like act on those things, but think of it as like more of an intelligent dashboard that aggregates signals, maybe suggests some new actions, but not like an autonomous agent that’s managing your database estate. 0:13:16.194 –> 0:13:34.874 Brian Haydin You know, the roadmap here on agents, it’s a pretty real roadmap of what they’re doing. There’s no surprise if anybody that’s following Microsoft on agents. But it’s worth knowing that, you know, what we’re talking about with this agent is, you know, it’s going to be rolling and kind of evolving over time. 0:13:35.554 –> 0:13:55.164 Brian Haydin I think something that’s interesting is the database savings plan. So Microsoft is claiming, I mean, we haven’t seen this in real life yet, well, how’s this going to translate? But you know, bringing these workloads inside of FabCon, inside of the database hub, it could save up to 35%. 0:13:55.234 –> 0:14:14.514 Brian Haydin you know, in the pay as you go. Now, that’s, you know, that’s coming straight off the marketing slicks, you know, that they talked about. It’s not, you know, something that we’ve had a lot of experience, like saying this is what we’re actually seeing. But it’s, they’re starting to have a little bit more clarity on what that pricing 0:14:14.634 –> 0:14:36.834 Brian Haydin pricing and cost structure is going to look like in inside of fabric when you bring some of these operational data loads into it. When I was at Build last year and they started talking about SQL databases and other Cosmos databases coming into it, I even like asked that question, what does this look like? How am I going to pay for this? Like, what does this all inclusive in the fabric capacity or is it, you know, somehow going to be? 0:14:37.114 –> 0:14:56.354 Brian Haydin And they really didn’t have a good picture of that. That’s starting to have a little bit more clarity now. And so if we are going to get these like 35% versus PAYGO, I think that’s great. But I still want to see how this is going to, what the cost structure is going to look like on some of this. 0:14:56.834 –> 0:15:18.114 Brian Haydin So more to come on that. And, you know, I wanted to reflect a little bit because we’re a little late to meeting, it’s been a few weeks since FabCon. And let’s just talk a little bit about what some of the community is actually hearing. And Suneer, at some point, I’m going to ask you like what you’re hearing too. 0:15:18.594 –> 0:15:36.994 Brian Haydin But, you know, one LinkedIn comment kind of captured the sentiment perfectly. So Power BI admins, now they’re the database estate managers. Like, what’s going on here, right? And that’s kind of the real question. Who owns this operational services? Is it going to be the DBAs now? 0:15:37.354 –> 0:15:56.354 Brian Haydin Is it the platform teams? Are fabric admins the one that are going to be the app, like, you know, owning these things? If you don’t start to think about it and you define these things before you onboard some of these production, you know, workloads, I think that there’s going to be more of an accountability gap and more and not really a governance platform. 0:15:57.74 –> 0:16:5.394 Brian Haydin So we got to really, you know, start thinking about this. What are you hearing too? Is there that kind of similar, similar sentiments, Suneer? 0:16:5.394 –> 0:16:25.394 Suneer Mehmood Yeah, I mean, several organizations, different organizations are structured a different way with respect to who manages what, right? So from clients, you know, mostly we have been seeing a DevOps model recently, so I’ve seen data engineers owning the platform and then tweaking the knobs as needed. 0:16:25.634 –> 0:16:26.114 Brian Haydin Yeah. 0:16:26.194 –> 0:16:45.794 Suneer Mehmood I’ve seen dedicated IT professionals who are in charge of the administrative, you know, processes inside Fabric and then the engineering team requesting them. So I’ve seen it both ways, right? So one thing that I’d like to add is this agent recommendation P is, it’s interesting, right? Right now think of it as your. 0:16:46.194 –> 0:17:6.354 Suneer Mehmood database estate plus an AI co-pilot that flags things, right? It’s not taking autonomous actions, but you’ll want guardrails to find before it surfaces in front of stakeholders, right, who might act on those recommendations without validation because many a times there is a gap when you have a dedicated admin versus a data engineer. 0:17:6.754 –> 0:17:14.194 Suneer Mehmood who kind of have a little bit more of deep dive into what’s going on and how this could be tweaked, right? So just calling that out. 0:17:14.914 –> 0:17:34.274 Brian Haydin Yeah, that’s a really good call out. Well, all right, so bottom line, this stuff’s early access, which means it’s a great time to start thinking about what the impacts of this is. And maybe one thing I can recommend is start thinking about and defining what your operating model 0:17:34.674 –> 0:17:53.74 Brian Haydin is going to look like. Start coming up with racing charts, delegation rules. What kind of agentic actions are you going to start to allow as this stuff? That’s where this is going. But have those conversations early now before you have to like flip switches with the agents and turn some of these features on. 0:17:53.794 –> 0:18:12.834 Brian Haydin Then you’ll be ready to scale this out, you know, and start bringing in more of these operational workloads into the system, into the ecosystem. So one like security, I think the vision here is pretty clear. And for me, it seems to be going in the right direction. You’re going to define your access policies in one place. 0:18:13.314 –> 0:18:31.634 Brian Haydin And that enforcement now is going to travel with the data across every engine that reads from OneLake. And this is important, right? So meaning that now that I have an ecosystem that I can have a set of policies that can apply to all of those. Currently with your analytics workload, you’ve got… 0:18:32.114 –> 0:18:51.154 Brian Haydin a disconnect between your policy management in Fabric and all these other operational workloads that are being built by your different platform teams, your different software teams. So this is really going to help keep Power BI, SQL endpoints, all your notebooks, everything. All these things are going to be under that same governance roof. 0:18:51.634 –> 0:19:7.794 Brian Haydin Our back roles, you know, are going to be encoded at the row level and column level constraints. All that stuff’s going to be consistent across the top. Write the policy once and enforce it everywhere at every time. Like, that’s how this is supposed to work. I think it’s a great move. What are your thoughts on this, Suneer? 0:19:8.594 –> 0:19:29.354 Suneer Mehmood Yeah, my thoughts on this is like the real picture can be slightly different than the vision, right? All the vision is in the right path, right? Across workload enforcement is not fully consistent. There are documented gaps, particularly around the column level security in certain Derek Lake scenarios. 0:19:29.474 –> 0:19:49.34 Suneer Mehmood and Spark access paths where enforcements behave differently than it does through the SQL world, right, or SQL analytics endpoint rather. So the policy definition and central authoring story is still solid. The every engine behaves identically part is closed. 0:19:49.954 –> 0:20:8.114 Suneer Mehmood But it has seems that means, like, before you architect your production security posture on the unified enforcement model, you need to test the specific engines your org has, you know, and verify they behave the way you expect, because I’ll tell you, Lena, in a typical medallion architecture, some people… 0:19:52.34 –> 0:19:52.514 Brian Haydin Yeah. 0:20:9.234 –> 0:20:28.274 Suneer Mehmood prefer to have gold in a Delta Lake environment, which is pretty much a lake house, and that act differently than a SQL analytics endpoint, which is a fabric data warehouse, right? And I have seen it both ways, right? And there are reasons why one chooses one versus the other. 0:20:13.954 –> 0:20:14.434 Brian Haydin Yeah. 0:20:28.354 –> 0:20:30.274 Suneer Mehmood So, just leaving at that, so… 0:20:30.514 –> 0:20:49.154 Brian Haydin Yeah, and I would say like it’s also this rolling thing where different workloads are coming in at a different pace. So Azure SQL is not going to come in at the same pace as Cosmos, the same as pace as Postgres. So when you’re looking through the learn docs, I’m seeing things that are. 0:20:49.314 –> 0:21:11.154 Brian Haydin you know, being labeled as, you know, preview still, where it kind of in the FabCon narrative, it was like listed as like GA. So keep track of this. You’re going to see a lot of rolling announcements and, you know, verify what’s active, not just in the learn docs, but also like it might be. 0:21:11.234 –> 0:21:30.634 Brian Haydin be kind of rolling in various tenants as well. So check on those things before you like commit to leadership that, you know, hey, we’re going to turn this feature on because I read about it or I saw this like blog post and Bryan and Suneer were talking about it. It was really cool. So run, you know, at least one design workshop. 0:21:30.754 –> 0:21:49.954 Brian Haydin you know, test it across different workloads, different, you know, deployments and make sure that it’s working in-end before you commit to it would be kind of my thought. So then we’ve got this like new construct direct lake on one lake, which right now Microsoft is calling generally available. 0:21:50.594 –> 0:22:8.994 Brian Haydin And this is, this, this has been around for a little bit, and it’s, you know, it’s kind of one of those things where it’s graduating now, right? And so it’s a little bit of new feature development, but really just kind of like graduating into like a general availability. And it’s been around, it’s been building to this point, you know, for about a year. 0:22:9.394 –> 0:22:27.634 Brian Haydin So here’s why it matters. Right now, your Power BI semantic models read directly from one like delta tables and have like import class performance problems, right? No schedule refresh windows, no duplicate, no data duplication into the semantic models and in-memory cache and all that kind of stuff. 0:22:28.114 –> 0:22:46.754 Brian Haydin The data lives once in one lake, and the engine now reads it there. So the right workloads are going to, you know, for the right workloads, this is going to really change the game. And I know that you’ve been dealing with this a lot, Suneer, in some of the projects that we’ve been working on. 0:22:47.154 –> 0:22:54.514 Brian Haydin How do you think this is going to change, you know, the game? And are there any limitations that you want to kind of call out with what’s being released at this point? 0:22:55.474 –> 0:23:14.434 Suneer Mehmood Yeah, the limitations that I would call out is Direct Lake is not feature parity with import mode, and teams get surprised mid-migration when they find these edges, right? Composite model support is constrained. You cannot freely mix Direct Lake tables with import tables in all configurations. 0:23:7.74 –> 0:23:7.554 Brian Haydin Mhm. 0:23:15.434 –> 0:23:33.994 Suneer Mehmood XMLA endpoint behavior changes in ways that affect some external tools and custom deployments, I have to be honest. So user-defined aggregations work differently than they do in import mode, so that’s another difference. And row-level security has nuances indirectly. 0:23:34.674 –> 0:23:53.754 Suneer Mehmood That can catch some teams off guard, and there are, you know, if there are less patterns were built in with the import mode assumptions, let’s say, but none of these are blockers for most of the workloads. I have, you know, let me clarify that as well, but my recommendation is to build your migration rubric. 0:23:53.874 –> 0:24:3.954 Suneer Mehmood against the specific constraints before you commit scope, because definitely there is advantages in going a direct click pattern for your semantic models. 0:24:5.554 –> 0:24:23.634 Brian Haydin That’s a, you know, again, I would say with a lot of this stuff that’s coming out of FabCon, that’s going to be a generally good advice, right? So, you know, two things that change kind of technically at scale that I think are worth calling out. So first, the one lake layout, how your delta tables are partitioned with VR sword. 0:24:12.354 –> 0:24:12.514 Suneer Mehmood If… 0:24:24.274 –> 0:24:43.634 Brian Haydin structure, directly impacts your query performance. So now that’s a Power BI performance concern and not just a data engineering concern. Second, your security model for one link is your security model for Power BI. And your semantic model is no longer going to 0:24:43.674 –> 0:25:2.194 Brian Haydin just be like BI plumbing anymore. It’s the contract that’s going to exist for most of these AI assisted analytics. I would also say if your model naming and measure definitions are sloppy, like your AI outputs, they’re going to be, let’s say, confidently wrong, right? 0:25:2.994 –> 0:25:22.114 Brian Haydin And I think that’s where a lot of organizations are getting kind of burned right now is by having some of these sloppy definitions and not really paying attention to it and thinking that they’re just going to be able to flip these switches on and everything’s going to be fine. So some SQLCon highlights. I know FabCon. 0:25:22.274 –> 0:25:42.194 Brian Haydin you know, and SQLCon was kind of combined. So for the SQL practitioners, you know, that are sitting here, I’d say 3 things. Disk N, vector indexing, Microsoft shipped some meaningful preview upgrades. Specifically, they removed some of the read-only table restrictions. 0:25:42.594 –> 0:26:2.434 Brian Haydin that were blocking many of the teams from actually using it in practice. If you tried it early and you kind of hit that wall maybe, I would say go back and give it another shot. It’s worth retesting. You might find some improvements. And that’s, you know, this is available in both Azure SQL Database and SQL Databases and Fabric as well. 0:26:2.914 –> 0:26:22.434 Brian Haydin So, still preview, so I wouldn’t bet like the SLA farm on it, but if you need, you know, you know, if you need better vector search, you know, over structured enterprise data, the SQL ecosystem is gonna start supporting you a little bit better. What about what your perspective on that? 0:26:23.154 –> 0:26:24.354 Suneer Mehmood Yeah, I mean, uh… 0:26:23.954 –> 0:26:25.314 Brian Haydin What patterns are working? 0:26:25.474 –> 0:26:43.954 Suneer Mehmood Yeah, the patterns is worth understanding here, right? So this scan is approximate nearest neighbor vector index for high dimensional similarity search, right? So the use case is retrieval augmented generation, right? Like so to make your… 0:26:42.434 –> 0:26:42.994 Brian Haydin Right. 0:26:45.154 –> 0:27:4.194 Suneer Mehmood you know, large language models say the facts, right? So instead of sending an agent to a vector database that lives outside your SQL environment, you keep the semantic search co-located with your transactional data. So that reduces the data movement and latency for certain workloads. 0:27:4.994 –> 0:27:25.714 Suneer Mehmood It’s still in preview, so the prototype, I mean, I would say prototype in like in a non-prod environments and measure the index maintenance overhead, you know, before you scale it, because the moment you know you have to maintain these indexes on your transaction data, you also need to actually balance the. 0:27:25.874 –> 0:27:45.314 Suneer Mehmood with your rest of your transactional systems performance, for example, writes, right? So in one way, it’s good that you’re not duplicating, you do not have to duplicate your data, make the data move from one place, but have a little bit of measurement on how much of indexes needs to be maintained. 0:27:45.794 –> 0:27:47.634 Suneer Mehmood So that’s what my take on that would be. 0:27:48.914 –> 0:28:7.874 Brian Haydin Well, let’s talk a little bit about some of the change event streaming. SQL Server’s play right now for event-driven architecture is this change data without building custom CDC pipelines. So this still preview features, so let’s call that out really quick. But I think the instinct is… 0:28:7.954 –> 0:28:26.674 Brian Haydin you know, kind of strong in the community. And, you know, guys like Kris Wagner and Mike Carlo, some of the guys that I, you know, follow quite regularly, they’re echoing some of this and saying that use, you know, use event streaming where it actually changes your outcomes, not just because streaming sounds like… 0:28:26.914 –> 0:28:45.794 Brian Haydin more modern than batch. So be intentional about it. And if your consumer can tolerate like maybe like a 5 minute latency, you don’t need streaming pipelines. So avoid some of that operational complexity that’s going to come with this. And then. 0:28:45.834 –> 0:28:52.274 Brian Haydin Moving on, I’m just trying to keep an eye on the time here. We got quite a lot of slides left yet, but… 0:28:49.874 –> 0:28:50.674 Suneer Mehmood Thank you. 0:28:51.674 –> 0:28:54.274 Suneer Mehmood I think we are good with respect to, yeah. 0:28:54.114 –> 0:29:13.634 Brian Haydin Yeah, all right, well, all right, so GitHub Copilot is now, I think this, you know, GitHub Copilot now generally available with SQL Management Studio, AI assisted SQL authoring is essentially what you’re getting out of this and it’s going to exist within your tool chain now without a lot of context switching. 0:29:14.74 –> 0:29:36.554 Brian Haydin So that’s kind of a cool, meaningful productivity shift. I don’t think this should surprise anybody. This is, you know, where everything’s kind of going is having these coding tools available to you. But one thing I want to be like kind of transparent about is that this is primarily an Azure SQL and SQL tooling kind of announcement. It showed up in the narrative probably because SQL database and fabric. 0:29:36.634 –> 0:29:55.554 Brian Haydin benefits from this as well. But it’s not like a fabric only story. Your DBAs are going to get this regardless of whether you’re using fabric or not. You know, since it’s a SMS, you know, integration. But that makes it kind of a cool thing too, right? So definitely a good announcement on that one. 0:29:57.394 –> 0:30:16.114 Brian Haydin Quick visual reference. So these are some of the actual UIs that they’re talking about and documentation, you know, like on the left side here is kind of showing that the SSMS co-pilot panel. That’s what it kind of looks like today. 0:30:16.954 –> 0:30:23.874 Brian Haydin Um, uh, I’ll let you kind of go through this, uh, you know, Suneer, why don’t you, uh, why don’t you walk through each of these screenshots? 0:30:24.114 –> 0:30:48.234 Suneer Mehmood Yeah, what I love about this is like, you know, when you actually do a select star, it’s technically correct, right? However, you can see that Copilot is giving advice here, like a senior engineer would give to a mid-level or a junior engineer, right? Like, or an architect would give the engineering team, saying like, okay, are you sure you want 0:30:43.474 –> 0:30:43.874 Brian Haydin Yeah. 0:30:48.234 –> 0:31:8.434 Suneer Mehmood to actually do a select star because it actually does retrieve all the columns from your customer table in this example, right? What of your customers table is super wide and you do not need everything? Do you want to really pull all of that in your columnar databases, right? Instead, you know, it is suggesting a change. Why don’t you just go with customer ID, first name, last name, right? 0:31:8.754 –> 0:31:11.74 Suneer Mehmood So, it’s like it’s not even though the… 0:31:12.274 –> 0:31:21.474 Suneer Mehmood There is no syntactical errors in what is being pulled, right? It is still suggesting a performance improvement. So that’s really inspiring to see. 0:31:21.954 –> 0:31:38.194 Brian Haydin Well, yeah, and it’s doing that based off of context too. It’s not just like, you know, saying, right, like, it’s looking at the entire holistic like procedure that you’re working on and saying, hey, you’re really only using this stuff. Select star is going to be really perform, like not going to perform as well. What about the other two? 0:31:24.794 –> 0:31:25.234 Suneer Mehmood Exactly. 0:31:28.754 –> 0:31:29.514 Suneer Mehmood Exactly. 0:31:32.754 –> 0:31:33.634 Suneer Mehmood Exactly. 0:31:39.954 –> 0:32:3.754 Suneer Mehmood So, yeah, this can, you know, for Azure Postgres, you know, flexible server, so there is vector compression and quantization that would happen, you know, on your top of your vectors, and this time we are saying, like, you know, you do not have to move the data, you can literally place it on top of your, you know, transaction layer, and with indexes, it’ll be able to actually. 0:32:3.754 –> 0:32:23.874 Suneer Mehmood Send your LLM queries that is converted into vectors back into your transaction systems, which is indexed, so that’s inspiring to see. So, the other one is primarily your, you know, SQL tooling management, you know, as I was… 0:32:24.34 –> 0:32:42.594 Suneer Mehmood kind of similar to what we were talking about the database hub, right? Like you can actually have all of your SQL databases managed in one layer. So in a way, I agree with the fact that, you know, Microsoft is betting on like that’ll help you. 0:32:28.34 –> 0:32:28.514 Brian Haydin Yeah. 0:32:43.554 –> 0:33:2.914 Suneer Mehmood reduce 35% of the cost. I think they are indexing on the fact that you have a holistic view of everything under one page and then it helps. More often than not, it helps that rather than having to go to different, you know, cost monitors and different, you know, services to tweak the knobs, right? So. 0:33:3.154 –> 0:33:4.674 Suneer Mehmood you sometimes get disconnected. 0:33:3.394 –> 0:33:12.34 Brian Haydin Yeah, and sometimes just a different team that’s like looking at that stuff, right? So this gives you the visibility right inside of the fabric ecosystem. So yeah, that. 0:33:6.914 –> 0:33:7.234 Suneer Mehmood Yeah. 0:33:10.274 –> 0:33:24.754 Suneer Mehmood Yeah, so this is genuinely useful to track when I’ll add a couple more things. You know, several of the features that were in preview at FabCon have been moving to GA on a rolling basis. That page will tell you exactly what changed and when, so I would recommend keeping a tab of it. 0:33:24.994 –> 0:33:42.914 Brian Haydin Oh yeah, that’s a good call out too. It’s a good place to look instead of like relying on learn docs that you look at every six months, right? Yeah, fantastic. So, all right. So here’s a slide that I think, you know, someone should start paying attention to because Fabric IQ. 0:33:26.594 –> 0:33:26.994 Suneer Mehmood Yeah. 0:33:32.194 –> 0:33:32.914 Suneer Mehmood Yep, yep. 0:33:44.34 –> 0:34:4.194 Brian Haydin It’s not just a feature announcement. I think it’s a platform bet. I said I have a lot of opinions about this and I think we’re going to spend a little bit of time talking about it. So it’s a genuinely new workload, genuinely like new direction. And it’s the one that I think is going to have the longest range impact, especially of the announcements this year. 0:34:4.994 –> 0:34:24.194 Brian Haydin on how you build, architect, think about AI solutions inside of the Microsoft ecosystems. So here’s the problem that what, this is the problem that we’re trying to solve. Most AI deployments on enterprise data fail not because like you’re using the wrong model or the model is stupid, 0:34:25.74 –> 0:34:47.114 Brian Haydin It’s failing mostly because the data that you’re feeding it is ambiguous. So a good example of that might be like revenue. You know, when you talk about revenue, that’s going to mean different things to finance. Who cares about like actual actual revenue, right? But sales has a different definition where they look at revenue as maybe like closed. 0:34:47.234 –> 0:35:5.874 Brian Haydin you know, deals is the revenue that I promised to the business. And you know, that all lives in the side of the data warehouse, but it’s not easy for it to interpret what that means. So an AI agent that says like, hey, I gotta go figure out revenue, doesn’t really, you know, it’s querying your tables and not knowing what tables it’s supposed to be querying. 0:35:6.74 –> 0:35:26.434 Brian Haydin And it’s not, it doesn’t really know what those tables mean. So it’s going to give you a confident answer, but depending on who’s asking the question, it may be the right answer, it might be the wrong answer. So Fabric IQ’s answer to that is, you know, still in preview, but it’s a workload that groups for different capabilities. 0:35:26.794 –> 0:35:45.714 Brian Haydin So ontology, you know, what is the shared business vocabulary, your entities and like relationships? It’s a plan, you know, so the budgets and forecasting layer, it has graph, which is your relationship based, you know, queries across different entities and understanding of that. 0:35:46.234 –> 0:36:8.754 Brian Haydin And then agents that operate within all of that semantic context to sort of manage that information as well. A good example of that would be like, I’ve got this massive, you know, 1000 table schema. I can’t go in and like define every column and everything. So I have agents that can operate on that semantic context and start to understand based not on just 0:36:9.194 –> 0:36:19.794 Brian Haydin some metadata that defines that column or that field or whatever, but also on the data itself. So I know you’ve been working a little bit with this, Suneer. What are some of your thoughts? 0:36:20.354 –> 0:36:39.874 Suneer Mehmood Yeah, from an architectural standpoint, the significant thing is the integration point, right? When a framework agent calls into Fabric with Fabric IQ in place, it’s not just squaring a table, it’s squaring with awareness of your business context, your entity definitions, your security policies, like that, right? So… 0:36:25.794 –> 0:36:26.194 Brian Haydin Yeah. 0:36:40.74 –> 0:36:49.394 Suneer Mehmood You go from like a statistically plausible to semantically grounded, you know, information. That’s a real and a meaningful difference in my opinion. So. 0:36:49.634 –> 0:36:54.354 Brian Haydin That is, that is definitely a big difference, and so… 0:36:52.274 –> 0:36:52.674 Suneer Mehmood Yeah. 0:36:56.34 –> 0:37:15.554 Brian Haydin All of this is really going to hold if your semantic model is well built, right? If your org, you know, has a semantic model that’s like messy and it’s, you know, it’s got poorly named measures, maybe a bunch of different duplicated tables with different information in it, no real governance principles of discipline. 0:37:0.594 –> 0:37:2.114 Suneer Mehmood Yep, so. 0:37:16.514 –> 0:37:34.714 Brian Haydin And you know, looking out on, you know, some of the names of people that are here, you know who you are. You know, the agents, they’re going to reflect that mess in the same confidence as they would in like a real clear world. And we see this with hallucinations all the time. So Fabric IQ isn’t going to fix your semantic model, but. 0:37:34.834 –> 0:37:54.354 Brian Haydin You know, it’s actually going to amplify your semantic model, and whatever quality is there, whether it’s good or bad, is going to be amplified. So, I, you know, just, you know, just think about that a little bit, you know, as you know, as you start to build these things, you’re going to have to pay attention to your… 0:37:54.594 –> 0:38:15.794 Brian Haydin you know, to what your data is telling, telling an AI model and telling the semantic model, which probably I would just say like a little foreshadow here, another reason why the governance baselines that are being here, you know, put in place. And, you know, we’re going to talk about a little 90 day plan. These aren’t optional things if you want to start using some of these tools. 0:38:16.434 –> 0:38:36.234 Brian Haydin So, you know, here’s, you know, here’s how the layers actually connect. So, we’ve got trusted data on the left. You know, we’ve got everything that’s grounded in OneLake. It’s governed by the security model that we talked about a little bit earlier. And that’s going to flow into Fabric IQ in the middle here. The ontology is going to define that. 0:38:36.314 –> 0:38:57.34 Brian Haydin business vocabulary, the plan layer is going to handle budgets and forecasts. You’ve got the graph, which is going to enable all those different relationship queries. And then finally, the agents that are going to operate with all that context. And then on the right side is where you get the business outcomes, the governed Q&A, the planned workflows, and the audit ready. 0:38:57.154 –> 0:39:5.154 Brian Haydin you know, agent actions. You know, this is a lot to digest, you know, Suneer. What’s your take? 0:39:5.874 –> 0:39:25.874 Suneer Mehmood Yeah, the critical shift is like an agent on the right side of the diagram is not asking what values are in this table, right? It’s asking what is the approved revenue number for, for example, quarter two as defined by my finance ontology for the eastern region, for users with my access level, for example, right? That’s A fundamentally more trustworthy answer. And it’s worth knowing that 0:39:22.194 –> 0:39:22.674 Brian Haydin Mhm. 0:39:26.114 –> 0:39:46.354 Suneer Mehmood Databricks is pursuing a similar vision with Genie and Unity Catalog. They have had production deployments of the semantic layer plus agents patterns running longer than Fabric IQ has been in preview. So I mentioned that not to undermine the direction. The direction is right, but because if your customers are asking why not Databricks, 0:39:46.834 –> 0:40:8.354 Suneer Mehmood That’s a fair question. The answer is not fabric is better. The answer is like if you are already Microsoft Stack native, the integration story across Teams, Office, Cloud Platform, and Azure is an operational advantage. Databricks generally cannot match at all. So on pure AI over data maturity right now, it’s closer than. 0:40:8.434 –> 0:40:10.194 Suneer Mehmood The keynote implied, so… 0:40:10.274 –> 0:40:33.874 Brian Haydin Yeah, I think that’s a really good call out. So I have customers, you know, Concurrency has customers that are Snowflake, Databricks, you know, et cetera. And one of the things that like really jumped out at me was one of my customers, a Snowflake customer, turned on MCPs and started like using it over a couple of different data domains that they use on a regular basis, right? 0:40:34.674 –> 0:40:52.994 Brian Haydin And they were super underwhelmed by it. And I started thinking about like, you know, how does, you know, how does that position Microsoft and with Fabric IQ? The reason why Snowflake is failing at this, or not failing, but underwhelming people is because they don’t have Fabric IQ. 0:40:53.74 –> 0:41:12.194 Brian Haydin they don’t have a semantic understanding. It’s just their version of it is just translating, you know, it’s just translating SQL queries, you know, your natural language into SQL queries, which is valuable, but you’re going to get a lot of garbage back from it. I think Data Breaks have been doing this longer, though, and candidly speaking, 0:40:53.554 –> 0:40:54.34 Suneer Mehmood Yeah. 0:40:55.394 –> 0:40:55.714 Suneer Mehmood Yep. 0:41:5.194 –> 0:41:5.674 Suneer Mehmood Mhm. 0:41:13.154 –> 0:41:31.234 Brian Haydin You know, they’re probably a little bit ahead, you know, from where Microsoft is in terms of the semantic modeling, you know, so, but you know, to your point, and I think you called that out really well, if you’re already natively inside of this Microsoft ecosystem, that’s where… 0:41:31.474 –> 0:41:50.994 Brian Haydin where the value proposition is. You get kind of the best of both worlds. So I don’t know. Don’t try to build like a universal ontology to solve this problem. You know, I think, you know, if I was going to give anybody some advice, I’d say start with 0:41:39.234 –> 0:41:39.354 Yeah. 0:41:51.954 –> 0:42:13.954 Brian Haydin one domain and build a small ontology that’s tied to like some of the specific, you know, decisions that you want to make. Thinking back to that customer of mine in, you know, in Snowflake, if they would have like taken the time to look at that data domain and what they’re really trying to expose to it and build ontology across it, think of it in that kind of direction. 0:42:14.314 –> 0:42:33.394 Brian Haydin If you do that, I think you’re going to find some, you know, some quick wins. And, you know, it’s not, I would say like the biggest differentiator isn’t really the platform, it’s the process and the scope of what your ambitions are that are going to drive success in these kind of projects. 0:42:33.874 –> 0:42:53.954 Brian Haydin So, all right, well, let’s talk a little bit about some of the finances and the planning capabilities in Fabric. And this is getting like a disproportionate community attention, and I think deservedly so. So these types of questions around, you know, planning, 0:42:52.114 –> 0:42:52.274 Suneer Mehmood Yeah. 0:42:54.594 –> 0:43:14.34 Brian Haydin you know, and the finance gap have been coming up ever since Fabric, you know, was released and talked about. And the finance teams, they’ve been kind of like just living with problems for decades. Budgets in Excel, actuals, you know, data warehouse, forecasting as a separate tool. 0:43:14.514 –> 0:43:33.34 Brian Haydin reconciling them, you know, we have a bunch of like full-time jobs just taking data from one system to the other. And I think some of the like fabric IQ planning is going to like help with some of that, giving governed context, not with just like. 0:43:33.674 –> 0:43:50.914 Brian Haydin like recognizing different differences or doing some of that analytic workload, but actually having some write back capabilities, right? So there is a little bit of a scope boundary that I want to like maybe put down, you know, on the table before anybody like puts it in the chat. 0:43:52.194 –> 0:44:10.834 Brian Haydin You know, or, or you know, more importantly, even goes back and like pitches any of this as like a EPM replacement. If your organization’s running things like, you know, OneStream, Adaptive Insights, or have any kind of other mature planning tool, Fabric IQ Planning is like in its current preview form. 0:44:11.154 –> 0:44:31.554 Brian Haydin is not like a rip and replace kind of thing. Those platforms have years, you know, of workflow maturity, approval chains, version controls, sandbox environments, GL integrations, all that stuff that auditors like, you know, want and actually trust. Fabric IQ planning is compelling. It’s cool that they’re going to be doing it. 0:44:32.594 –> 0:44:44.754 Brian Haydin But if you have a dedicated EPM tool, I would stick with that for now and just keep an eye on where this is going. You know, any other thoughts on this? 0:44:45.554 –> 0:45:4.274 Suneer Mehmood Yeah, I mean, preview, treat it as a control pilot, right? Auditability and stewardship are non-negotiable for finance workflows. Before you roll this out into your financial planning and analytics, you need clear answers on who can write back what approval chain looks, look, look, I mean, does it what? 0:45:4.514 –> 0:45:23.634 Suneer Mehmood does your approval change look like, excuse me, and how changes are logged, right? And align your fabric IQ semantic definitions with your existing BI semantic model so you’re not creating a new which number is right problem, right? Because that’s what we are trying to solve here, right? And yeah, that’s what my take. 0:45:6.914 –> 0:45:7.394 Brian Haydin Yeah. 0:45:23.954 –> 0:45:24.594 Suneer Mehmood And that is. 0:45:24.434 –> 0:45:42.834 Brian Haydin Yeah. I’m thinking of a couple of customers that we should be talking to about some of this right now as I, you know, as I’m just sort of like thinking this through without the webinar going, you know, through my head fast. All right, here’s the thing. Pick one pilot. Focus on like one simple workflow, right? 0:45:43.394 –> 0:46:3.274 Brian Haydin forecast maybe versus actual, that’s kind of like what I was thinking here. You know what I’m thinking, Suneer. But, you know, pick something clean, right? Pick something that you, well, that you already understand pretty well. And, you know, if it does work, then, you know, this is going to be a story that’s worth telling. 0:45:50.674 –> 0:45:51.74 Suneer Mehmood Yeah. 0:46:4.114 –> 0:46:22.674 Brian Haydin And if it doesn’t work yet, well, you know, it’s you scoped out something that you can you failed fast enough to be able to scope it out of the picture. All right, we got a we got about 14 minutes left, so for the engineers or the real nerds. 0:46:14.594 –> 0:46:15.74 Suneer Mehmood Yep. 0:46:20.634 –> 0:46:20.754 Suneer Mehmood Yeah. 0:46:22.834 –> 0:46:41.474 Brian Haydin you know, on the call. I want to give you maybe a little bit of the honest competitive picture that I think it’s going to be a little bit more useful than maybe just a Microsoft only narrative. So 3 different announcements, Fabric MCP, skills for Fabric, and Jumpstart. I want to start with 0:46:41.874 –> 0:47:0.554 Brian Haydin MCP because there’s a comparison that’s kind of worth making right now. So model context protocol, this is, you know, been around for a bit. And it’s a standard interface that lets AI agents interact with tools. And data can be one of those tools, essentially, but it’s not. 0:47:0.634 –> 0:47:22.834 Brian Haydin to interact with tools. So Microsoft published MCP server documentation and tooling for Fabric, which basically means that you can build agents that directly call Fabric operations. You can run a pipeline, you could query a semantic model, or maybe create like some sort of workspace artifact. And you can do that through a standardized agent of Fabric protocol. 0:47:23.554 –> 0:47:41.594 Brian Haydin So that’s pretty cool. Snowflake has MCP and so does Databricks. They have different versions of them doing different things. So the question I think that’s kind of worth asking right now is like, what actually makes fabrics different? Do you have a, what are your thoughts on what makes it different, Suneer? 0:47:42.194 –> 0:48:3.74 Suneer Mehmood Yeah, my honest answer on that is like Snowflake’s MCP, as people who have tested it are discovering, exposes raw schema context to a model, like table names, column names, data types, whatever descriptions you’ve added to your catalog. So when an agent calls and it’s working with schema, not business meaning, so the result is syntactically valid SQL. 0:48:3.474 –> 0:48:23.234 Suneer Mehmood That is, that could be semantically wrong, right? And the model is confident about it because, yeah, so we have seen hallucinations and stuff, right? So customers who have tried it on anything more complex than a single table can hit a wall fast. So Fabric MCP’s real differentiator is the semantic model layer. 0:48:23.354 –> 0:48:42.514 Suneer Mehmood 20 plus years of analysis services heritage, carrying business definitions, hierarchies, measures with proper DAX logic, right, RLS, and explicit entity relationships. So when an agent queries through a well-built semantic model, it’s getting a curated business view, not a raw schema dump. 0:48:32.594 –> 0:48:33.154 Brian Haydin Yeah. 0:48:42.594 –> 0:48:52.114 Suneer Mehmood Like, you know, which is technically right, but not semantically right, if that makes sense. That’s a substant, you know, a substantive difference, in fact, so… 0:48:53.74 –> 0:49:14.34 Brian Haydin Yeah, it, you know, the conditional I think I want to be a little bit more, you know, be honest about is that the differentiate, this differentiation only really holds if your semantic model is well built. We talked about that already. If you got a messy model, it’s not going to help. It’s going to be just as bad as Snowflake, right? Just with a little bit more infrastructure that’s kind of governing around it. 0:49:14.914 –> 0:49:32.914 Brian Haydin So in its current form, without Fabric IQ ontology in place, Fabric MCP is closer to the Snowflake experience than it is to what the keynote probably implied. So you’re going to have to pay a little bit of attention to that. So skills for Fabric is… 0:49:33.674 –> 0:49:57.194 Brian Haydin You know, is open source on GitHub, and we’re talking about standardized governance-aware skill templates for AI coding agents to author and operate fabric workloads. The value, I think, is really in the defaults security and governance patterns that are already baked into this, you know, into into these skills. You can fork these out, you know, fork them in Git. 0:49:57.314 –> 0:50:19.354 Brian Haydin and build internal golden prompts on top of them. And then jumpstart the reference implementations and accelerators that install directly into your workspace. They’re fast. You’re going to get time to value and be able to like kick out some demos pretty quick. And same thing here. You can fork, harden, and bake your own, you know, into your policies as well. 0:50:19.794 –> 0:50:39.394 Brian Haydin So those are a couple things that, you know, we’ll probably see a lot of a lot of momentum over the second half of this year as people start to experiment with them. And love the fact that, you know, we’ve got, you know, open source, you know, opportunities to contribute as a community as well. I know what Suneer is going to be doing is going to be submitting some pull requests. 0:50:39.954 –> 0:50:44.274 Brian Haydin So, all right, where to go next? What do you think? 0:50:44.754 –> 0:51:4.194 Suneer Mehmood Yeah, I mean, resource call out here, right? These are the actual reports. GitHub.com, fabric skills for skills, the fabric jump start for fabric, you know, for jump starting, obviously, like, you know, MCP server documentation is on Microsoft Loan under the fabric get started section. Pull them up, fork them, start there rather than a blank page. 0:51:4.474 –> 0:51:24.234 Suneer Mehmood The defaults are better than most teams write themselves on day one, right? So I would highly recommend that. So I’ll be real with you, the odd center for MCP is where the time goes, right? Service principles, OneLake API permissions, workspace access, budget two days for that before your first real demo. 0:51:24.914 –> 0:51:28.874 Suneer Mehmood Instead of 2 hours, so that’ll be my take, yeah, yeah. 0:51:26.754 –> 0:51:27.794 Brian Haydin Not 2 hours, huh? 0:51:28.994 –> 0:51:36.114 Brian Haydin Yeah, all right. Well, let’s move on real quick. I know we got, we’re button up on time here. 0:51:36.674 –> 0:51:37.474 Brian Haydin Uh… 0:51:38.794 –> 0:52:0.34 Brian Haydin So what are people actually saying? So I think it’s important to kind of have a context, you know, around what the actual conversations are happening, not just in this conference, but like reading the blogs, you know, what are the people talking about afterwards? Here’s just four quotes that I came across that I thought I would kind of capture. 0:52:0.834 –> 0:52:24.834 Brian Haydin that, you know, and share with you because it’s more practitioner reality and not something that’s coming out of like official, you know, official recap. So on the top left, the database hub RACI question, people are asking this publicly on LinkedIn and it started within hours. Like it just like came right, it was like, 0:52:24.874 –> 0:52:46.514 Brian Haydin probably the first thing that I saw. And it’s a legitimate operational question. The product team really doesn’t have an answer for this yet. And hopefully they’re going to have one, you know, pretty quick as people are clamoring about it. Verify your tenant first, you know, caution, you know, with like a lot of people that are trying to play with this. 0:52:47.554 –> 0:53:7.474 Brian Haydin And, you know, as I’ve said, I’ve been through this like, you know, keynote, GA, you know, reality kind of cycle before. It looks good up on the screen, but then when you go to click the buttons, it’s not there. And even when you look at the learn docs and the tech docs, the button says it’s like you can even see the picture of the button. 0:53:7.634 –> 0:53:30.994 Brian Haydin you know, but it’s not there. So the 403 errors. So there’s definite, definite reality here that there’s errors happening in these APIs in the agentic tooling threads. Not a reason to like, you know, stay away from it, but it’s a reason, I think, like you said it, right, budget a couple hours or a couple of days, not a couple of hours. 0:53:30.354 –> 0:53:30.674 Suneer Mehmood Yeah. 0:53:31.434 –> 0:53:54.754 Brian Haydin So just have kind of the right, you know, right frame of mind when you go into this. And then, you know, the ontology, you know, it is exciting. I would absolutely say that it is exciting. But, you know, don’t try to model everything at one time. Try to like learn some of the muscle memory that’s going to go into it. 0:53:55.634 –> 0:54:12.354 Brian Haydin The people that were at FabCon, they, you know, they’re going to have this natural inclination to just want to jump back in and get everything, you know, in Fabric IQ and you’re going to burn out real quick. What are your thoughts? Do you hear anything else you could call out? 0:54:10.594 –> 0:54:33.714 Suneer Mehmood Yeah, my, yeah, my pattern, I mean, my thoughts on, you know, all this quotes are the same, right? It’s all of the same pattern, I would say. The direction is right, the vision is coherent, the platform is moving fast, but the gap between keynote confidence and the production reality is real and consistent. So treating that gap as expected rather than surprising is what separates the teams that succeed rather than like, you know, coming after six months and saying. 0:54:34.154 –> 0:54:35.474 Suneer Mehmood You know, it didn’t work, so… 0:54:36.34 –> 0:54:54.194 Brian Haydin All right, I’m going to fly through. We got just a couple of minutes. So just backseat, operational reality. I’d say three things that I’m going to call out really quick. First off, there’s a lot of preview sprawl, right? Things that are saying like, you know, GA doesn’t mean it’s going to be available in your 0:54:39.74 –> 0:54:39.554 Suneer Mehmood Sure. 0:54:54.914 –> 0:55:15.554 Brian Haydin necessarily in your tenants. So just make sure you check on that. Operating model, you know, this is going to change how you think about governance and ownership and stewardship. You know, people that have been investing in data domains and data owners are probably going to find themselves ahead of the game a little bit. 0:55:16.674 –> 0:55:39.674 Brian Haydin And then also, like, you know, there’s going to be some learning curves here with new authorization patterns, new CICD patterns, and that’s probably going to cause some friction, you know, with you as you start to adopt this. So what does that mean, you know, kind of moving forward? Verify here before you build stuff. If you’re still on the call, thank you for sticking around. 0:55:39.954 –> 0:55:59.714 Brian Haydin take a quick screenshot of this, or better yet, reach out to us and, you know, we can walk you through some of this. But verify, you know, the maturity, not just the decision is it available, but the maturity of it. Do I want to start building on this product or do I want to wait and wait until this gets into a GA or even like, 0:55:59.994 –> 0:56:19.714 Brian Haydin wait for a little bit longer. And then I’ve got a couple of sides here that I’m not going to be able to get through here and, you know, spend any kind of time, but I put together a quick little like 90 day action plan. And I’ll just quickly talk about each one of those. The first one is, you know, probably the first, you know, week or two. 0:56:20.154 –> 0:56:40.194 Brian Haydin Establish some sort of truth and governance baseline for the organization. Inventory things like the domains and the top semantic models that your organization wants to use. Find out where the pain is the worst, right? And that’s where your focus is, you know, focus should be. Then in the first month, start looking at Direct Lake and OneLake. 0:56:40.234 –> 0:56:43.74 Brian Haydin and build out some sort of an optimization pilot. 0:56:43.994 –> 0:57:3.474 Brian Haydin This probably means like, you know, one or two semantic models that you can use, but the goal here is to really just validate the direct lake feasibility. Check it out, make sure that it’s going to work for your case, test the governance behavior, you know, cross the end-to-end before you proceed. 0:57:3.834 –> 0:57:22.914 Brian Haydin And then, in month two, start playing around with, you know, some of the brittle options, you know, mirroring and ETL, like, you know, kind of workloads that can be that can that can be simplified or enhanced, but this is gonna cause, like, you know, some, you know, some amount of change in… 0:57:23.34 –> 0:57:44.274 Brian Haydin you know, within the organization. So this would be kind of like your month three, you know, kind of activity. And then finally, piloting A semantic grounding and then your agentic workflow. So you see how I kind of piece this all together, starting off, you know, with some of the planning and then the governance. And now eventually, we’re going to start being able to play with these workloads, you know, in agents. 0:57:44.754 –> 0:58:2.434 Brian Haydin So try to follow these, like, you know, phase one, two, three, and four steps, and then I’ll keep you on a good path, you know, within like a 90 day kind of like, you know, framework. So if you want to reach out to us, I’ve got my LinkedIn here and Suneer’s LinkedIn here too. 0:58:3.914 –> 0:58:23.634 Brian Haydin kind of talked a little bit faster to get through this, I guess, but hopefully this gave you a pretty clear picture of what’s changed. But more importantly, it’s not just what’s changed and what’s changing, but what’s actually ready and what you can do next. So we dropped a link in the chat here for you to reach out to us or give us some feedback. 0:58:24.154 –> 0:58:34.594 Brian Haydin If you want us to reach out, just go ahead and fill out a little bit in the survey. Most importantly, I just want to say thanks for joining us today. I know your time is valuable and really appreciate it. 0:58:37.594 –> 0:58:38.434 Suneer Mehmood Thank you, everyone.