View Recording: Best of FabCon Europe: The Fabric Highlights That Matter

Suneer Mehmood 0:05 I can see people trickling in. Welcome, welcome. Right. We’ll get started. Let me just… All right. Welcome, everyone, and thank you for joining. I’m Suneer, Lead Architect for Data at Concurrency. I spent most weeks building on Microsoft Fabric with clients. We have an art together. FabCon Europe in Barcelona produced a flood of announcements. And I’m not going to read you through the list. I group what matters into four parts, following the order Microsoft itself used on stage. And also tell you what I left out and why. One house rule for today, every item I show carries a status label. Generally available versus preview versus coming soon or coming. Here’s my advice. Let’s plan on what is generally available. Pilot what is in preview and wait on what is coming, right? Questions, please drop them in the chat as we go and we’ll take them at the end. All right. If you remember one sentence from today, let’s make it this one. Fabric is no longer the only place where you do analytics, or where your analytics lives, right? It is now what Copilot and agents read to understand your business. And that matters because the semantic models you have already built are now an AI source. Good models give good answers. Messy models give wrong answers, right? You have heard the term many a times, garbage in, garbage out, right? And these non-deterministic models will tell you very confidently wrong answers if the data is not prepared right, right? Microsoft picture on this is the circle on the left. Microsoft’s IQ sits on top. It is made of Work IQ, Foundry IQ, Fabric IQ, and Web IQ. Fabric IQ is the part that knows how your business operates. And the core of Fabric IQ is something most of you already own, your semantic models. You might not be in Fabric yet, but you might be in Power BI. That means you have semantic models, right? And people reach all of this through Copilot. Agent 365 is the layer that governs and observes the agents. One caution for me, this theme is Microsoft framing and a good share of what support it is. It’s still in preview. So for the rest of the hour, we’ll separate what is real today from what is promised. This is the map Microsoft keeps coming back to all morning, and whoever uses Fabric knows this inside out, I’m sure. But this is what Fabric provides in general. It’s that ecosystem or one umbrella that has databases, data factory, analytics, real-time intelligence. Power BI and the new one, Fabric IQ. Underneath all of them across is Copilot, OneLake, and Governance. We’ll use this map to walk you through the rest of the hour. Before fabric took on stage. The first part was SQL Con. Microsoft framed it around 3 pillars. Sovereign environments, mission critical apps, and agentic operation. SQL Server on Azure is locally generally available, connected or fully disconnected. In the demo, they ran a local AI model and called it from SQL with no cloud round trip. That was exciting, right? And in mission critical apps, they are using hyperscale. which can go up to 192 virtual course. Vector searching is available in Azure SQL, so you can use rag similar operations in relational databases. How cool is that, right? And indexes, you know, across your keywords or your, you know, or your, sorry, vectors. That means you can have a combined search of your keywords and your vectorized data. And the indexes stays current on every insert, update, and delete. And then we have agentic operations, right? This is the part that touches Fabric Database Hub. In Fabric, it gives one view across your database fleet. And database agents watch it, recommend fixes. Both are in preview. Agent mode and SSMS with GitHub Copilot is generally available. If you run last SQL estates, we can talk about in Q&A, right? There are a lot of these features will help you. Been on. All right, so let’s talk about how the next 45 minutes are going to be laid out. I’ve kept Microsoft’s own order because it tells a story. First, how you build and operate Fabric, then how you prepare the data, then how you give it business context with Fabric IQ, and finally, how that context shows up in Copilot in reports and in apps. I’ll try to do 3 short demos along the way, hopefully on Ontology, Copilot, and an app built with Rayfin. So hold on to that thought, right? I’ll explain to you what Rayfin is for those of you who have not been a part of our earlier demonstration on Rayfin. At the bottom. There are three or a few labels that I mentioned, right? Like as the labels say, green is GA, so it’s available now. You can plan on it. Orange means it is public preview, so pilot it. Yellow means it’s private preview, so it’s a sneak peek. You can try it out. Gray means it is coming, so don’t plan around it yet. All right, sections in build and operate on fabric. So let’s talk about Copilot, deployment plans, observability, excuse me, and agentic investigation on like catalog and policies and capacity management, F0. I’m very excited about F0 personally. I’ll tell you what that is. copilot in fabric. So… You know, Microsoft has announced 1 copilot for every workload. Today, for those of you who have tried Copilot, how it is available is you have one Copilot for your maybe Fabric data warehouse. You have another Copilot for your Fabric Spark or big data engineering space, right? In your notebooks, in fact. You have co-pilots. Power BI, so in different places. But now there is this Fabric Copilot, which has this entire view of your workspace, all of your objects that is there in the workspace, and such. So this is 1 assistant for all workloads. So you can use it in fabric or your own IDE and command line, and it remembers context, just like any, you know, large language model does do work on, like your co-pilots or not co-pilots, excuse me, ChatGPTs or clouds. It remembers the context. It knows what are the objects. You can ask. Questions about these objects, in fact. So I’m going to try to show you one small. Example about that. So, this is kind of, I have a lot of objects in my workspace. And here I created a demo report earlier, so I am not in the report pretty much, right? I’m asking it a question about the report, so… I’m asking you to prepare a summary for my demo report, right? And I run it. Remember, it’s not in the report. It’s right. It has its entirety, entire view about what all is available in the workspace. So it answers based on that. So it’s right now it’s scanning the report content. It’s generating a summary as large language models do. And then. It, it uh creates. a report for you with the data, maybe with some, you know, there you go, right? It created a summary with the data that is within the report and it also gave you a sneak peek of the chart. How cool is that, right? And some recommendations along with that. So yes, Power BI can give you something. But some recommendations, it starts from a large language model, right? Like, you know, and to those of you who ask that question, this data is not going back, you know, to anywhere. Your data is still in your tenant. It’s secure. It’s that AI agents that work within your tenant. Your data is not being sent anywhere. Switching back to our demo. This is getting expanded. What I showed you was a Power BI version of that. Now, when this gets expanded, you get agentic powers as well. So you can pretty much, you know, Troubleshoot your problems with Fabric Copilot and we’ll get, we’ll talk more about that, right? And such. This is like, yeah, this has one peripheral view of all of your objects. Now, agentic powers can actually look at these reports, look at any errors, look at your pipelines, and then troubleshoot base mean. recommend troubleshooting based on that. Now the one that I want to really talk about is deployment plans. Again, for those of you who have implemented Fabric, we have deployment pipelines. So how Fabric operates is we can actually put your dev system. in a place called a workspace and your prod, entirety of what you have in prod in another workspace. And when you are about to deploy something from your dev to prod, you create deployment pipelines. That gives you a full-fledged understanding of, hey, what is there in your dev versus what is in your prod? What are you missing? What are the changes that you have made in dev? So you have one diff check. You can also integrate that with CICD capabilities, Azure DevOps and such. So what deployment pipelines give you is like you can actually put those in queue for your deployment. Now, what deployment plans additionally give you is some things that we… Lacked earlier was. Um… How do, like people used to ask this question, how do I deploy things in the right order and then hydrate and validate what I deployed? For example, let’s talk about a typical medallion architecture where we have bronze, silver, and gold. We probably want to deploy the bronze first because that’s the most upstream part of it, right? Like you cannot run a silver pipeline unless you have bronze, right? And probably you might want to hydrate your metadata tables before you want to run your dynamic pipeline engine. For example, if you don’t have data in your metadata table, you know, it’s not going to run, right? So deployment plan is a new item that, you know, you can set an order. Right. Around each step, you can add actions before and after deploy. In the demo that they showed, like, you know, they were deploying bronze lakehouse first, and it was followed by running the ingest notebook, which hydrated the data in bronze, and followed by silver and gold chain, right? And then bootstrapping that gold warehouse. running the orchestration pipelines and then validating the environment. Everything automated with a proper plan. These actions can take parameters and the whole plan is stored as a YAML file too, so you can modify this in a script through a script. Right, and it is, it can easily be stored in a source control. This, this can plug into deployment pipelines, right? When you deploy from dev to prod, excuse me, so the deployment plans can now plug into deployment pipelines, which will be the entire, you know, workflow for your pipelines. So it’s very exciting to see that. Now, let’s talk about observability that Fabric is bringing. This is a huge improvement as well. The monitoring hub, they are clearly rebuilding the monitoring hub, and that’s in public preview, by the way. It is one place for the, you know, for every job run, capacities, applications, agent across all your workspaces. So the left picture is what debugging looks like, right? So let’s say, let’s talk about for the data engineers over here or architects over here. When you actually create a pipeline, you would have a parent object or a pipeline invoker, and then underneath you will have subsections where you would, you know, have sequenced. movement of your bronze or silver. Underneath you would have activities, probably your notebooks or your copy jobs or whatnot, right? So earlier, in order to get to a specific error or why the pipeline failed, you had to actually drill into the individual components. What they have claimed in the demo is that like the orchestration view in the monitoring hub, as in the exact place where it actually shows there has been an error, it’ll show you the exact failure that really happened at the. root activity level, right? Because that’s what, as engineers, we want to find out what exactly happened, right? We don’t want to understand, like, you know, a parent failed. Yeah, parent failed because a child failed. A child failed because of even, you know, a… in a child field, right? So you want to go inside each and every level and get to the root cause. So this gives you that error at a very high level self and the right picture. If you click investigate on a failed run, an operation agent reads the telemetry and tells you in plain language what really happened. So here in this example, the silver transform could not find its bronze table. So gold publish and semantic model refresh couldn’t run. And it pointed at the shared helper notebook that builds a table path. So it also created a helper notebook, right? So the agents explain it does not fix. Your team still makes and owns a change, right? But what we get from this is… The time from failure to troubleshooting and understanding that failure drops a lot and fixing it right drops a lot because now we are using the power of agent well within the framework. Let’s move on to OneLake catalog. Two things here. First, the discovery gets deeper. You can browse down to tables and columns with a live preview. So earlier, you could not search for fee levels and stuff, you know, in OneLake catalogs. And search is built for agents as well as people, right, like us. And then policies in fabric. These are also in public preview. So on the left, admin opens govern, then policies, picks a capacity, writes a rule. For example, only the data engineers of certain group can create notebooks, lake houses, and whatever objects, right? On the right, a user outside of that group tries to create a notebook and gets the item creation blocked because now the admin has a new governing policy that prevents them because they are not a part of the group, right, from creating those objects. So… Until now, it has been one or zero. For those of you who have maintained fabric workspaces, know that you can either give somebody a viewer access to a workspace. That means they can pretty much see everything. The other access that you have is a contributor access, right? That means they can pretty much… do anything within that workspace. They can create objects, they can modify objects. So there is no individual, more granular control. So with this policies in fabric that an administrator can create, you can actually provide granular authorization to groups, right? So our permissions. Right? So that’s awesome. Now let’s talk about capacity. The capacity story has two halves to it. The first half is GA, generally available. So you have a unified capacity management. So capacity signals stream live and activator can alert on them. In the demo, the admin got a Teams message when a capacity crossed 95%. They clicked through and found the notebooks driving it. So, and search protection lets you cap a single noisy workspace without hurting the jobs that matter. Then let’s talk about capacity overage, right? Instead of throttling, when you run over, fabrics bills you for the excess and, well, with that at least three times the pay as you go rate. So you set a limit for each capacity over a 24 hour period. Right. So the second-half is what I’m most excited about, which is like the on-demand billing and a new F0 capacity. So again, just take a step back. For those of you who are new to Fabric, Fabric starts at F2 and then goes like F4. That’s a capacity, that’s a compute power of Fabric, right? Let me turn off the caption over here. All right, so… So, um… What we are getting here is like, so that talking about F2, F4, F8, F16s, those decide what kind of storage do you get, what kind of compute power do you get, right? It does not regulate any features that you would enjoy. You get. every feature, but if you want to run simultaneous notebooks with X number of users and such, you need, you know, higher compute powers. But many a times for those of some of us who are established in fabric. We always have the question of like, are we using our capacities in the most optimal way? For example, I have seen clients who run, for example, an F128 capacity and are really wondering if I’m using my capacity to my optimal. So Sometimes they spike, sometimes they don’t. Maybe they are at 50% of their capacity. So they question like, should I go and reserve F64, which is half the capacity for next year, and maybe have the next F64, maybe in a pay-as-you-go fashion so that you only reserve that? from, let’s say, Monday to Friday where I’m like, you know, really spiking or maybe not use that over the weekend and stuff. The problem is, if at all your compute spikes beyond that window, then you have overage. And, you know, more overage means like, you know, either your background jobs will stall after certain minutes. or interactive jobs can have a delay and install. So that’s a little bit risky. So F0 gives you an advantage into figuring out, you can actually provision an F0. That means if you provision an F0 today, you don’t start to pay, right? So you only pay per use. You can also mix it with the provision capacity and run selected workloads on demand with spend limits. The keynote showed this in public preview. Microsoft’s own blog says it rolls out in the coming weeks. So no pricing has been published. Without the rate, you cannot really model for it. So I would recommend, please do not cancel any reservations, but I’m very… You know, hopeful for this. Let’s say I wanna, I wanna, I want a new workspace and I wanna try out a new, completely new thing which I don’t know what the capacity usage will be for. I want only, I want that only for the weekend or next couple of days or something. I can really spin up an F Zero, not worry about, like, you know, I having to spend it if I’m not really using it. So that’s the advantage. Really cool thing. Okay, so for the next about 11 minutes or so, let’s talk about the preparation of your data estate, what’s coming, like a new shortcut and mirroring sources, mirroring of views and security roles, IQ sharing across organizations, and some, you know, modifications or new features in data engineering and integration. GPU accelerated warehouse. I am so gung-ho on that. And then real-time intelligence. Let’s talk about it. Let’s dive into that. All right, so OneLake, integration of OneLake that reaches more of your data. Microsoft keeps removing the reasons to write ingestion pipelines. So a source normally gets a shortcut 1st and then a mirroring followed by that. Google BigQuery is right now generally available. Salesforce Data 360 is in public preview and is bi-directional and 0 copy. Salesforce data becomes, I mean, it’s now available in fabric, excuse me, and fabric data can be, you know, available in Salesforce too without moving or copying it. And for Dynamics 365, that gets data mirroring into OneLake. So earlier write-ups called it coming soon. So check your tenant. Well, I mean the Business Central D360 backend already had it. And Google BigQuery, right? SharePoint, OneDrive, these are GA. And AWS Glue, Aviva Connect, Lakehouse Catalog, these are in general availability as well. So they are expanding the sources, right? So this is very useful for those of you who are multi-cloud. For example, some part of your data is in AWS, maybe Google. You do not really have to move the data into Azure and worry about having to do that. For your exploration needs or maybe one of needs, I wouldn’t recommend ingesting your data. You can literally create shortcuts and then access your data from there. Caveat is that you are actually using the IO, you know, power of your other cloud. So it really depends on how that performs. And sometimes shortcuts work, sometimes shortcuts do not work for continuous operations. And, and, and can be a pain, so in that scenarios for continuous operations and stuff, mirroring is an option, like, and you can get the data mirrored into your fabric ecosystem and then that probably stored, you can you can probably store in a store that in a, you know, optimized fashion, like. delta lakes and whatnot. And then for your continuous operations, you can mirror it. The advantage is you do not have to create a bronze ingestion pipelines, right? You can just mirror it and start using it. So. All right. There are some extensions to mirroring. Earlier we had tables, change data feed. So now we have views as well, which is exciting. So, and that means all the logics that you have created on wherever that view exists today, maybe, you know, in your Azure SQL environment. you can mirror that, you know, without having to replicate the logic or having to replicate the base tables, right? Mirrored security rules is very exciting. So the demo used Snowflake as a source. In Snowflake, a rolled call support group can read only three tables. That was their demo scenario. They created a marin fabric. Now, once it is mirrored. You know, they choose an existing security role from the source, and Fabric creates a matching role and switches on OneLake security. So you can see on the right side, you see the result. Snowflake roles appear as OneLake roles. The Support group role, you know, covers exactly… those 3 tables, right, with the same members. So in short, the security roles in Snowflake are replicated into Fabric and then enforced and preserved there. So an admin did not have to create it again, if that makes sense, right? The keynote again said it is GA, documentation says. you know, security roles as preview, but they are coming. So this is awesome for those of you who have. Different, you know, uh, platforms for running data. One thing that they really indexed on is IQ sharing. This is a new idea. You know, I have not seen that in Fabric before, but I’ve seen this in other platforms. You share not just data, but the business context that goes with it. For example, ISVs can share their IQ, you know, a company called TomTom sharing their geo data. The use cases, one enterprise can share data with another for shared use cases, right? For example, a retailer versus a supplier, right, in order to actually do some research on demands and stuff. The good thing is you can share tables, files, and context, right? And when you share, when you create an IQ share, you can set an expiration date on it. And then Fabric itself generates that link and the recipient uses it to add them into their workspace. and the data is not copied, right? How cool is that? Because my data remains on my side, I can share it, you know, and put an expiration date on it, and they can just start using it without having to move that data, right? The demo that they showed was a genomic company sharing a lake house with a hospital in another tenant. So the left picture you can see it’s a sender, they add recipient emails and expiration date. When the research study ends, the access is revoked automatically, right? The right picture is the hospital accepting the share. Then they join the genomics table with their own patient data in their notebook for their research purposes, right? Um… So one thing is, sharing semantic models and ontologies also is there on the roadmap. So we’ll explain, I’ll walk you through what ontologies are, by the way. for those of you who don’t know. Couple of announcements in the data engineering and integration field is those who are fans of DBT, right? DBT jobs are now generally available in Fabric. They’re fully managed, no command line, no local setup, and they run alongside your notebooks and pipelines. Copy jobs now connect to event streams. So copy jobs. I, it’s one of my favorites in the data engineering section of Fabric because it’s a low code. Many a times when you ingest data into bronze, you do not want to transform it because you want to keep your footprint very light on your source system. So you would rather much, you know, do an EL when you actually, you know, move the data from a source system to your bronze. right, instead of an ETL. So you do an EL. So I use copy jobs most of the time, right? The good thing is, is copy job now connects to event streams in both directions, batch to stream and stream to batch. And that’s in preview. Fabric Runtime 2.0 is generally available. Spark 4.1, Delta 4.2, Python 3.13, all of these are available. Microsoft Cites is up to six times faster than an open source Spark. I personally not have tested it. That’s their benchmark. So I would recommend testing your own workloads. Fabric now has a data engineering agent in preview. It is built on technology from Osmos, a company that Microsoft acquired in January. get to test that. A lot of the cool things happening in the data engineering space, by the way. So, the example on the screen, uh, you know… We built, and I mean, they built an insurance claims lake house from SAP and guide by sources, bronze to gold, and as one delegated project, right? So using these new features. So keep an eye out for that. GPU accelerated warehouse. Oh, this is this is cool. This is in public preview. This GPU built accelerated warehouse is built on NVIDIA hardware, right? So you turn it on with one switch in the workspace setting. The query plan can highlight which parts went to the GPU, right? Um… So only the operations that makes most sense to go to the GPU are being sent. That’s what Microsoft claim is. In the demo that they showed, they didn’t send all of the operations back to the GPU. They showed the query plans, literally. And only the operations which were intensive. And, and really, you know, would make a lot of will benefit from being sent to a GPU workload, where sent, and rest of it remained in the regular workload. So let’s look at the comparison, right? That’s one thing to take a note. They compared it with 250 active users, up to 250 active users. And you can see how long on the left hand side, you can see that like in the execution time with the regular work, regular compute in fabric and the GPU accelerated compute. on the right hand side. So execution time around like 600 seconds, right? And, and, and it was literally 11 seconds when you know the exact same number of active users and exact same number of queries executed. So, look at that performance spike they were using in, you know, different combinations, but this was something to, you know, take, you know. Keep in mind, and this showed the contrast of, you know, how much a GPU assisted workload can actually save you, right? They showed a very complex query with a lot of joins and such, and then they hit it with 250 users and 1000 queries being executed at the same time. Oh, by the way, this was announced in Microsoft Build that happened in San Francisco earlier this year. But this is the first time I’m seeing this, a performance benchmark being compared. So it was very cool to see. Jensen Kong personally came into the… The meeting, well, he joined remotely from Taipei or Tokyo somewhere while Satya was giving the keynote and he announced it. But this is the first time that I got to see the performance benchmark as a part of FabCon. So it was really cool to see that. I personally was there, like, you know, when it was announced in San Francisco. So it was exciting. I was looking forward to seeing the benchmark and how it actually worked. How did we set that up? Okay, moving on, real-time intelligence, keeping a tab of time here. I’m going to move a little bit faster. So business events in the real-time hub, you define a business moment once with a schema, for example, a congested, you know, thumb style at a stadium and activator can actually act on it. So schemas for events, accelerated shortcuts, and event house transforms that reshape event data without storing it twice, right? So Emirates presented a customer story. A cargo planning process went from four to five hours to about like 10 seconds with low latency writes and fast. reads in the event house and one thing that they really called out is like the simplicity to maintain the platform, right? So yeah, so it’s for those of you use real-time intelligence. You know, feel, you know, feel free to tie that out. This is in GA. Let’s talk about Fabric IQ now. Semantic models plus ontologies. I think I referred to ontologies before, but for those of you who have not… tried out ontologies. Many of you, I’m sure most of you have tried out semantic models. We’ll talk a little bit about ontologies too. Microsoft calls it semantic model plus plus, right? So I’ll try to give you a live demo on ontology. And they also showed planning in fabric. So ontologies, this is, as I said, Microsoft calls this semantically a plus plus. Um… So you can have props, let’s say for example, structured, unstructured, real-time, and graph data, right? And then you have semantic models that gives your BI metrics that your Power BI can consume. Ontology is what they give is that gives you a vocabulary about your data. For example, you would have customers, right? That’s your master data, all your customers. And then probably they will, of course, they will place orders, right? You will have orders, transaction data. So. You define vocabularies with two entities now, like you have an entity of customer, you have an entity of order, and then you say customer places orders. So that means like you when you create an agent on top of it, that can actually, well, you can, to an agent, what we do, what do we do, right? Like we actually. Talk to them in natural language, so that the agent can take that natural language understanding, a reasoning model can take that data and understand the vocabulary of your data estate. That’s what an ontology helps with, you know, in its in its very… Basic form, right? They also give you operational intelligence, right? That means like, you know, if there is an activator that needs to actually perform certain operations based on certain events, based on your data, ontologies help you drive it, what needs to be done where, based on. how you have set that up. Remember, key is like setting all of your semantic layers and ontologies in the right fashion. The features are available, but architecture is really in your hand. Um… So. One new thing that they announced about ontologies is ontologies was announced, I think, earlier this year or, you know, end of last year or something. But something that they announced new in ontologies is like, you do not have to build it by hand. A copilot agent can read a document. and understand the existing models and propose an ontology, right? So you can pretty much provide it a PDF and then it can provide you a scaffolding of your ontology, right? It also, ontologies now also support DAX expressions. So for those of you who are heavy on semantic layers. You know, you can actually use DAX expressions to maintain and modify your complex ontologies if you want. You know, and the thing is, from a policy perspective, the entity types can, you know, inherit their policies from their parents, right? And means any data policy that you have from a parent entity to a child entity, it’s like, you know, you do not have to. applied twice. Ontologies also have versioning and the rules of it can be written in plain English so that the agent can convert it. Versioning helps these ontologies so that you know you can actually store them in DevOps and then with version controls and stuff. So ontologies are in public preview. I will highly, highly recommend to you to try ontologies so that you can actually enable natural language question to your data. So you’re using the non-deterministic model, like a reasoning model. You’re asking it questions like natural language questions, like your… like how we would treat your copilot or Cloud or ChatGPT, but it is getting the deterministic data from your back end by understanding the vocabulary of your data estate with the power of ontology. So for those of you who have not tried it, I would highly recommend you to try it if you have a strong modernized data platform already on fabric. This is a question that we get a lot, by the way. All right. I’m going to quickly, I don’t know if we have time for a demo, but… I’ll try to… So this is what I was talking about, right? This is a demo that we typically use. You can see a customer places booking, a representative made a booking, a product appears in a booking, planned, you know, fulfills a booking. So this is how you actually set up these relationships and you bind all of these entities with the data elements on the back end. So you can see that. The plant entity is actually bound to dim plants. You know, and you can ask natural language questions to this. This is a new… Feature, I mean, so this is the new new look of how ontology works, and the, you know, probably the demo will take a little bit longer. So let me see if I can no. So what you set up is data agents. Can pick one of these data agents. Might take a little bit. But the advantage once you set that up is like you literally can query your agents with natural language. I’m going to skip the tour. There you go. O, you can see. This is a demo that I set up for AI Summit, which happened last year. So I’m just asking it a question, what are the historical trends across my data? And literally, I can, that was just, I just reuse the suggested, you know, question from there. But you literally can see what is happening underneath, what all it is analyzing. It is analyzing data from Lake House, it’s analyzing data from Custo. It’s analyzing data from a lot of these places, and then it is giving me an answer, right? So many a times when you are on data, you are like trust but verify. It is not just giving you a textual representation of what your data is, but it is also giving you insight into what all it is doing to actually fetch you the data so that you are confident on the numbers that you’re getting. This is the data that it got. I’m not going to read through the data. This is just spoofed sample data, by the way. It doesn’t. So you can see, like, you know, you get a response. A reasoning model gives you a response in a very, very textual fashion, right? And then it also shows you the steps that it completed. Right, and you can actually ask to fetch the underlying data if you want, like, you know, so, so imagine deploying these agents to your co-pilots, for example, and you being able to just type your questions along the way, and rather than trying to fetch a report, probably a report, like, you know. just like you’re chatting, right? You’re typing and then you get to understand your business. And even like voice, I use voice a lot, right? Like talk to your agents and get the data right in front of you without having to understand the semantics of how to actually navigate a report. This is so cool, right? So I’m gonna… Just go back to this. In the interest of time, I’m going to move to planning in Fabric. So what this gives us, like many of us have actual data, but like, you know, we do the planning outside. I have, you know, clients, customers who do that, and they want to put the data back into Fabric through a pipeline. You do not have to do that. There is a planning feature within Fabric, right? So you can actually look forward based on the actual data that you already have. And then one thing that planning gives is like a tree view of like, you know, hey, like if you change, you know, your planning data somewhere, where all would that be impacted? so you get a tree view as well. And then once you’re planning data, the performance of planning data also was very impressive. You make a change and it does scales of all the dependent changes impressively across your data estate. The thing is, when you actually write the planning data, it writes it back to the semantic model. That makes it available to copilot, to ontologies, and to data agents. The demo then showed plan versus actual in a normal Power BI report to, which is cool, right? That’s what you want, right? Like for planning for next year, because we are all like, we are all at the year end, right? You probably want to plan for the next year. So it happens, right? Every year or quarter, whatever, right? There is also integration with ontologies. An ontology entity can bind. To a planning data, so the ontology now has the past, the presence, and a forward-looking view. That means you can talk like what I showed you, like, you know, through Teams or or through your voice, you know, in different Microsoft platforms to your data agents and get your planning and and your or your forecast data and also the actuals data. Right. So when an agent can see plan versus actual on the same governed model, its answers start to be credible to finance teams, right? So. All right, putting Fabric IQ into action. Through co-pilots and such. through fabric apps and then probably we’ll try to see a demo of Raven that I was talking about. So Fabric IQ is not now available, excuse me, inside Microsoft Copilot chat and Cowork, and is generally available. The Copilot understands the question is a data question. So like, let’s say you’re asking Copilot like something about what happened. last month with respect to your profit or revenue. It understands that it is a data question, routes it to the right semantic model, and the answer shows exactly where the data came from, like I showed you, right? So here I was asking that data agent specifically. But think about having to, you asking Copilot in general, and it understands all this data should come from the semantic model. So let me go fetch that data from the semantic model and give you the data. In the demo, someone asked Copilot to stress test a recovery route for a supply disruption. So Copilot answered from the governed semantic model and cited the Power BI report. The numbers match report side by side. Then it went further. Copilot pulled in Work IQ, who owns each approval step in and in what order. And Cowork built a 14 slide review deck and scheduled the sign up meeting after the, I mean, after asking for approval first of course, right? So that’s where you actually connect your Fabric IQ, your Work IQ, and your Foundry IQ and such, right? So on cost, Microsoft wording is that Copilot customers get this without additional AI token cost. Now the catch, every published semantic model is a candidate source for an answer. So any cryptic column names, duplicate measures, and missing descriptions will produce bad answers that sound confident. So on governance, there’s a control called restrict from Copilot that lets administrators exclude specific models. So sensitivity labels and data loss prevention policies also carry through into Copilot. So there is also a second control on each individual semantic model. Anyone with write permission on a model can decide whether people with read-only access may use AI on it. So that covers Copilot in Power BI and Microsoft 365 Copilot Chat and Cowork, Data Agent and MCP servers. The important part is that this setting is switched on by default. So AI access is something for you to opt out of, not something you opt into. So every model your users can read is available to Copilot until it’s open. owners turn that off. So keep that in mind. All right. So, um… All right, so you probably already have seen this. This is a copilot on a semantic model. Let me see if I can pull that up. I just created a quick demo report. So this is just again, spoof data, some total revenue and some slicer that. you know, changes the data as per your selection of your quarters. Go into… Copilot, let me see if I have some sample questions that I have. All right, in the industry of interest of time, excuse me, I’m going to just ask what was the revenue in Q3 2026? Ann. The total revenue in Q3 2026 was 216803. So if I hit the slicer of Q3, you know, Q3 2026, you get that data. So, well, you might think that, oh, I can already see that data, right? Like, you know, why should I ask a, you know, agent? Well, think about this, right? Think about your cards. having a summary data and you probably want to deep dive into certain questions. Why is it less than last year? Whatever, right? Like, you know, based on the data that you see. So you get to ask questions based on what is shown in this Power BI dashboard, also the underlying data that powers it. So that’s the advantage here. Um… You probably another example of this is like, which region had the highest revenue in Q3 2026 and how did it change from Q2 2026, right? So you can see that by the bar chart, you can see that Q3 2026, it’s Midwest, so it has to tell you that it’s Midwest. Yeah, it is. And it tells us the right number, 75898. Yes, it gave the answer. And it almost, well. Q222, that’s a failure. But like, you know, earlier when I tried this, and you can trust me on this, that, you know, it showed that it doubled. So because Q22026 is around like 366 to 9. So sometimes you get failures, but you know, you get a lot of features. You know, this keeps on improving, right? So. So I stop at that so that we can continue with the rest of our slides. All right, so… Ohh, something that is cool is like modern report builder, and that’s not even in private preview. It’s a sneak peek that they showed. This is why coding of for reports. OK, so there is a new feature that is coming in fabric called Modern Report. That means you do not have to know to, you know, build a Power BI. dashboard. So for those of you who just want to wipe code your dashboards, you just… Explain in text, right? Prompted that, you know, you want to create a sales dashboard and, and you know, you explain about your KPI cards, you know, and the trends that you want to create and such, right? So, and then you save and publish to Fabric, and that’s it’s like any other report, and then you can modify those reports because now it is a full-fledged. report, you can either have a Power BI engineer modify it for you or you can modify it yourself if you really are curious about it. Amy Cousland 53:49 Suneer, are you trying to show your slides right now? They’re not showing up. Suneer Mehmood 53:53 Oops, so sorry. How long has it been out? Amy Cousland 53:55 Okay. Just for just a little bit, since you switched from your demo. Suneer Mehmood 54:01 Oh, I’m so sorry. So I think I stopped sharing instead of going on from. So this is the demo that I was showing you. So sorry about that. So folks, this is an example of Power BI dashboard that is created from a modern report builder. This was not hand coded. Somebody prompted Power BI dashboard, you know, to be created, and then they explained it, and then it created just like a full-fledged dashboard. All right, so fabric apps, this is again an interesting story. So… Fabric apps is, I was telling you, like, you know, so far Fabric has been looked at as something that you run your analytics on, you read your, you know, data through dashboards, through semantic models, through ontologies, you know. Also, yes, you can write back using planning. Of course, now Microsoft is switching the game a little bit. They are introducing something called Fabric Apps, which internally has been called as Rayfin. That’s how they announced it in Microsoft Build, where I was physically present. So what you can do is you can actually create apps from scaffolding, right? Means like you can actually, let’s think about like a, for example, a todos app or something like a, you know, a logistics app. For example, think about if you run a small logistic company and you want your people who actually deliver stuff to put the status in, right, you know, maybe upload a photo of your deliveries, you know, imagine Amazon deliveries, right? But that’s all in large scale, but it’s right now it’s starting in small scale. What I did is like I built a todos app, I’ll show you. Um… Here. There you go, so… So you can see, like I have some to-dos that I have added, right? So I can add a new item that finish the FabCon demo, right? That’s a high priority right now, right? So it’s a high, so it’s color-coded priority items. you know, I can add things and once I do that, I can say that I have, I’m done with it. So this is in order. So I don’t think I can do that unless I check off these kind of things. But what is happening is through these apps is like you get to write data back into fabric. Now, did I create all this app right from the scratch? No, I. Used the Rafin, you know, application that is available within Fabric, and then use the available scaffolding or templates that Fabric provides to actually spin up. What happens is the entire, you know, front end, back end, and even the databases that is required for supporting these applications is spun up within Fabric. automatically. So that’s very cool. You can actually create applications within Fabric within seconds, right? And then modify that using the power of Agent to KI. So something to look out for. Microsoft is actually switching the games. They are going beyond just from an analytic engine, so you can have a lot of. applications built within fabric. Right now in small scale, I would say, but one thing to keep an eye is like you want proper governance when you actually, when your team actually spins a lot of applications and that leads to a lot of, you know, databases in the back end, you really want to have a strong governance around it. So because, yeah. All right, 3 moves for the next 30 days. Make your models AI ready. Right? Fabric IQ is in, I mean, in Copilot is GA and on by default. So sort the models into ready, fix, and then restrict them. And then, you know, stand up a preview in Sandbox, right? One non-production workspace where your team can actually try ontologies, deployment plans, the new monitoring hub. and catalog policies without touching production. Three, own the bill, right? Get a baseline of your capacity metrics, put your overage decisions in writing, and leave your reservations alone until pricing is published. Next thing to watch is Microsoft Ignite in San Francisco. So November 17th to 20th. So I expect pricing news there and some previews moving to general availability. And if you would like to help, sorry, if you would like help with the model audit or a capacity review, CONSULTANCY can run either, you know, one with you. So let us know. So that is. That lives us right at the ballpark, right? Like, you know, so questions. So the next, you know, events, you know, is Microsoft Ignite, as I said, FabCon in Atlanta in March, and then FabCon Europe in Amsterdam on October 25. next year. So this is a sneak peek on everything on the table that we discussed. What is available, take a snapshot if you want, so that you get, you know what is available today, what is in public preview, what needs to be verified, and what is coming soon. So with that, we are right at the top of the hour. Sorry, I could not take any questions, but please forward your questions if there are to the channel and we’ll respond to you if there are.