Fabric Deployment Plan - Ep.573

Deployment pipelines move item definitions between workspaces, but they have never moved the data those items need. Mike Carlo and Tommy Puglia dig into the new Fabric deployment plan, argue about whether a feature without an agent is now a barrier to entry, and leave with a new word for the shop: deplavity.

News & Announcements

• All About the Power BI / Fabric MCP, Chicago Fabric / Power BI User Group. Tommy hosts the next Chicago user group on Thursday, Oct 22, from 3:00 to 5:00 PM CDT at the Microsoft Technology Center on East Randolph Drive. The session demos both the Power BI semantic model MCP and the Fabric remote MCP, with setup at the start. Mike’s advice is to watch someone turn each one on, because the two MCPs do different things and the setup inside VS Code keeps shifting. You have to register, and your full name needs to be on your user group account to get in.

• Fabric Credit Ratings. Tommy released a site that gives Fabric items a bond-style letter grade, from AAA down to C, based on ratings from him and other MVPs. Each item is scored on feature capability, user experience, CU or token cost, and outlook, with a note on why it earned its grade and a rating history as new rounds are released. Pipelines landed at AA+, Dataflow Gen1 rated above Dataflow Gen2, and dashboards sit at B minus. Mike’s ratings lean toward data engineering and Power BI, since real-time work needs real-time sources and more capacity than he keeps running. Tommy is asking MVPs who work in real-time or data science to reach out on LinkedIn so those items get rated too.

• Advanced NL2SQL and advanced DAX generation in Fabric data agents (GA). Instead of writing a query in one pass, the data agent now explores the schema, tests joins, and looks up real filter values before it generates SQL or DAX. Microsoft’s internal benchmarks show smart-answer accuracy rising from 78.8% to 90.9% for SQL and from 89.15% to 91.86% for DAX, and the improvements roll out to the standard runtime starting November 2, with completion expected by November 10. Mike doesn’t chat with data agents much. He builds with agents directly, and right now he wants to know whether an agent can build a report.

• Fabric Policies: granular governance at scale (preview). Policies let admins set rules at tenant or capacity scope instead of relying only on tenant-wide switches. The preview covers item creation by capacity, selected security-sensitive workspace settings, and, coming soon, external data sharing. Mike’s use case is giving a team a slice of Fabric, such as notebooks, lakehouses, semantic models, and reports, without also handing them real-time analytics. He hasn’t tried it yet, but he sees it filling a gap the community has asked about for a long time.

Main Discussion

Topic: Filling a gap with the Fabric deployment plan

The deployment plan was announced at the Fabric conference in Barcelona. Per the Learn article, it’s a workspace item that puts each item in its own deployment group, orders the groups by dependency, and can run actions before and after each item deploys. You attach it to a deployment in Git integration, in a deployment pipeline, or through the REST API. It’s in preview, and a Fabric admin has to turn on a tenant setting before anyone can create one. Tommy calls it a process changer: if you adopt it, your team’s process has to change with it.

• Lineage gets the order right and still fails. Tommy walks through Microsoft’s sales example. A warehouse view reads a lakehouse table that two notebooks produce. Lineage deploys the lakehouse first, but the table doesn’t arrive with the item definition, so the view is created against nothing and the deployment fails. The plan runs both notebooks as post-deploy actions on the lakehouse group and makes the warehouse depend on it. To Tommy, this is Microsoft admitting that lineage covers metadata, not when data actually shows up.

• Infrastructure, code, and data. Mike breaks deployments into three layers: the item itself, the code inside it, and the data it needs. Deployment pipelines handle the first two. “There is no data deployment.” A notebook that hydrates the data as part of the plan is the piece that has been missing since deployment pipelines came out. He also wants every environment parameterized, so the dev lakehouse GUID swaps to the next environment’s lakehouse when the code moves.

• The edge cases are real. Mike has hit a parent pipeline that wouldn’t deploy until its child pipelines existed. He has also seen a lakehouse shortcut fail in test because the upstream table didn’t exist there, then revert to pointing at the development data instead of just breaking. A pre-deploy check that confirms every table exists before the shortcuts move is exactly the kind of step a plan can run.

• A middle ground, not a replacement for Git. If your team already lives in Git and GitHub Actions, Mike doesn’t think you need a plan, because you can sequence API calls and notebook runs in actions yourself. The plan is for teams that need more control than deployment pipelines but aren’t ready to do everything in GitHub or Azure DevOps actions. Tommy points out that the plan is YAML underneath, but you build it on a canvas, so you don’t need GitHub Actions experience to use it.

• Is “no agent” a barrier to entry now? Mike admits he relies heavily on an agent whenever he builds deployment actions, and many companies still don’t give their people one. For those teams, the canvas is the way in. Tommy called a timeout on the idea that anything without an agent is a barrier, so Mike offered him a raise to work without AI. Tommy said no, and when the offer flipped to 20% less pay to keep AI, he said yes. His Fabric Credit Ratings site took about a week with agents. It uses Fabric as the backend for raters, credentials, and submissions, and it publishes each released round to a static GitHub Pages front end.

• Wait for GA, then match the plan to the team. Mike won’t recommend a preview feature for production workloads. Even after GA, he’d look at team culture first: a team that has never used Git and has only dabbled in deployment pipelines needs some pipeline experience before stepping into plans. He frames it around a release engineer: developers only touch dev, and someone else moves items between environments and makes sure the data refreshes. Tommy’s summary is that deployment plans are to Fabric what deployment pipelines were to Power BI. He also flags the preview limits on the deployment plan actions page, including one plan per deployment operation and no deploying into more than one workspace in a single run.

• Deplavity was born. Mid-sentence, Mike tripped over his words and coined deplavity, and Tommy told him to trademark it before Microsoft takes it. The episode closed with a mock prescription drug ad that Mike credited to John: “Ask your capacity admin about deplavity.” Today’s sponsor, for the first time, was deplavity.

Here’s the deplavity moment, in case you missed it live:

Looking Forward

If deployment pipelines keep tripping over data that isn’t there yet, rebuild Microsoft’s sales example in a dev workspace and see whether a post-deploy notebook fixes it, then wait for GA before anything reaches production.

Episode Transcript

0:09 measures, explicit measures drop it loud hello everyone and welcome back to the Explicit Measures podcast with Tommy and Mike. How are you doing? Tommy? Good morning. Good morning, Mike. Doing excellent. Good to hear it. We are jumping in today. Our main topic for today, we’ll be talking about this new release from Fabric Conference over in Barcelona, talking about the deployment plan, creating a deployment plan in Microsoft Fabric.

0:40 So this is different than a deployment pipeline. This is a plan that you can execute to help you deploy different items in certain orders. So that’s we’re going to talk about that one next. All right, that being said, we’ll come back to that topic. Tommy, give us some, some news. What are we finding around the Internets today? So one of the first things around the Internet is the next Chicago Fabric User Group is happening this week, this Thursday. You have two days to sign up. Really by noon tomorrow, we’re meeting downtown and we’re going to install and see what the Power BI and Fabric

1:15 MCP can do and what it’s capable of. So as always, we’d love for you to show up. We’d love if you want to show a win or something cool that you’re working on. That’s how we’re going to start off the user group. But again, then we’re just going to go through a tutorial of what the MCP is, the Model Context Protocol, and just showcase it. We’re really excited about that. We’re going to see installing both the the remote for Fabric as well as the Power BI local one.

1:45 We’re really excited about those two. I’ll put the meetup invite down below. This will be in Chicago at the Microsoft office there. This is the Chicago User Group. We’ll put it here. It’s good to see these things happening. I have noticed recently though that there has been a bit of shifting around how you use the McP servers inside VS code. I’m assuming that’s what you’re going to do it in, Tommy. It’ll be good to really watch a demo of someone like turning it on, getting it set up. How do you talk to different things? And Power BI and Fabric MCP do two different things, so you need to understand the difference between the two

2:19 items. So anyways, check that out. Really excited. Coming up soon. Make sure you register. You’ve got to register to go to the event. Yes. And your full name. Name. Your full name needs to be in the user group account, so then you can’t come in. All right, so that’s the first one, Mike. Next one is something I’m. I’m personally proud that I’m releasing today. So, Mike, there are 75 plus artifacts in Microsoft fabric. And you and I always talk about what’s working well, what’s not.

2:50 I think for a lot of people it’s hard to gauge whether or not things are worthwhile. Right. And we talk about this all the time. We talk all the time. Yeah. Like, okay, notebooks. When should you use notebooks? And I was inspired by a few of the books I was reading to make it easier for people just to get an idea of what’s worthwhile. So I am releasing something called the fabric Credit ratings. Again, it’s off something based on what you call investment grade. This is how we rate bonds for in, in the company or in the country.

3:23 And it’s just a simple letter grade based on submissions from myself and other MVPs on Just Raining out the different artifacts and fabric. And it gives bas every feature a grade, a letter grade and the letter grade we use the same credit rating as Moody’s. You don’t have to worry what that is. You just have to know that things that are aaa, that’s the best of the best. And what is a,, a good investment in terms of investing your time and for your company and goes down. Explains all the, the ratings there.

3:54 But Mike, I’m excited about this because is it the definite guide? Not necessarily, but is it a helpful place just to say what’s worthwhile, what are other MVPs? See, if you are an MVP and you want to help contribute to it, just message me on LinkedIn as well. Again, the. The whole idea is to give a simple, convenient way of understanding where should you invest your time and resources. This is available@fabriccreditratings. com and this is something that I’ll also be posting on the socials today as

4:25 well, actually showing that it’s out in the wild. But again, just Mike, for example, some of the things that you and I showcase or some of the submissions here that we’re looking at is everything’s broken up into data engineering, storage and databases mirroring real Time Intelligence, Power, BI Fabric, IQ, OneLake, Data Science and AI Developer Controls, MCPS Developer and Platform governance, security, and all the different artifacts in there, like no surprise, pipelines Got a double A plus rating which is the second highest

4:56 rating. Our notebooks are double A as well. So just. It falls along the same lines that we, I think we’ve talked about but again just to make it easier for people people, this is also going to show historical ratings as we do this either once a month or once a quarter. And there’s a description on why that rating was given. So Mike, for me I really like this. I’m really excited to do this because I think a lot of people just need a convenient way of is something worth the investment is what are the things that you may be excited to try but they’re speculative.

5:26 They’re, you’re. You’re at your own risk. It’s usable but there’s some gaps. I think I’m confused. I don’t even understand normal, normal ratings as well. But you don’t have to worry about what a credit rating is. You just need to know there’s a letter grade. The higher the letter grade the better that product is. True. I do think the, the app looks really good. Tommy, your scales the top. I never understood bonds to begin with but having a nice little designation at the top of the screen, super

5:59 helpful. Very easy to understand like what the gradings mean. You also color coded it very well so as the grades get better the scores change in color. Which also makes me like okay, I know what I’m doing. So yeah, take this for what you will. This is just some opinions of some random people on the Internet. , you’ll. You’ll find your best places where you like to work with fabric and pieces there and hopefully as things improve over time, features get built as items get created in fabric or new items exist or show

6:29 up. You’ll see hopefully some of these creates move up, down slightly. I don’t anticipate a lot of these things changing rapidly unless something major happens or something really gets improved or a lot of things that add there. Anyways just a a fun site to give an opinion about, comparatively looking at everything at a landscape level where, where do you spend your time? What’s adding value to Tommy and I as we build? One thing I will observe though more of my opinion. There’s a couple. You can actually leave a couple items blank.

7:00 More of my opinion is definitely centered around data engineering and power bi and report visuals. I did not have as much of an opinion on the data science stuff and I didn’t have as much of an opinion on real time as those weren’t really applicable workflows for what I like to use. And I think for me, Tommy, the observation here is in the real time you actually need real time data sources and then you also have to be willing to spend the money or the capacity units to make real time work.

7:30 Yeah. And most of what I work on is. Or at least when I’m demoing or testing things, I’m on a lower end sku. So I’m not just going to leave on a real time something for long periods of time because it eats up a lot of my smaller capacities. So I think again also part of what I do is I’m on the lower end of the spectrum of cost and efficiency of things and things that are going to cost a lot more to regularly run. Just not as interested in. Yeah. Funny enough though, Dataflows Gen1 got a higher rating than Dataflows Gen2, which I’m not surprised about.

8:03 price point, everyone gets access to Gen1. I know Gen2 is only a fabric thing like it’s better but it doesn’t seem like it runs more efficiently than it. The UI I would say is in gen 2. 2 is much better. Right? It’s way, way easier to use. It is, yeah. 100%. If you are in real time and or data science and you’re an MVP again hit me up on LinkedIn because we want to fill out the rest. There’s a lot of artifacts that if we didn’t feel. If I didn’t feel comfortable submitting, they’re not ratings not there.

8:34 So we love to see that too. And one of my favorite ones, Mike Dashboards got B minus which is one of the lower ones. It’s basically just above,, distressed. So it’s. And that makes sense because are people actually using Dashboards? A few things if you want to know about the score. I’ll just run through this. There’s three things, four things. Feature capability. So does it have all the features that you need? The user experience, the how easy it is actually working, see your token cost if it’s applicable for that particular feature and what the outlook is.

9:07 So yeah, pretty cool. Again, check it out. Fabriccreditratings. com awesome. Fun little example there. Neat little website built there and then let’s move on. Next, next. All right, our next one, Mike Data agents. We got Data agents and governance. First one is around major, somewhat major updates to Fabric Data Agents, Advanced NL to SQL and advanced DAX generation are available in Data Agents. And what this simply means instead of generating a query in a single Pass the fabric data agent now explores

9:40 schemas, validates assumptions and checks filter values and refines its approach before it produces the SQL or dax. Because the problem is, especially with enterprise data Mike, it’s one thing for you to run an agent across a single CSV file, but when you have schemas that contain hundreds of tables and columns and business definitions exist outside the database where the terminology may not be stored, the NL2 SQL uses an agentic workflow to basically find a

10:12 relevant example query and adapts. It clarifies any questions and does data exploration so that two new SQL source capabilities for schema descriptions and also topics that you can add to it are available in the preview. So you can have markdown based guidance documents and business meetings, tables and columns. And then the advanced DAX generation, they said the accuracy improved from 89% to 92%. For SQL, the smart answer accuracy improved from 78 to 90%.

10:47 And there’s just a better consistency because the advanced DAX is doing tool calling and iterative reasoning in instant value indexing. More consistent answers. They’re going to be rolling this out in the standard runtime on November 2nd with completion expected by November 10th. Excellent. This is, I think another one of these really good ones. Again, slowly slow improvements here. They’re getting better at letting you use natural language to go build things.

11:17 I don’t use a ton of this data agent in this way. I don’t typically talk to the data agent. Usually I know what I want. I’m spending more time instead of Tommy, Tommy talking to an agent in line like in a, in a chat thread. It’s just what this feels like.

11:30 I’m doing a lot more building agents directly with the tools I have at my hand my disposals. I’m really interested right now in like can I build reports with agents? How can I get a report built with an agent as I’m working through that. So this is a neat workflow. I think they’re. I’m glad they’re improving it. I’m getting. It’s better. NL2DAX stands for natural language to DAX. So natural language is you speak it and then it works and then the DAX is then run. So happy they’re improving this. This is going ga. So it’s coming whether you like it or not.

12:02 So yeah, probably a good announcement. Play with it, see what it turns into if it’s good or not for you. Love to see what your opinions are on the community. All right, the last one. Mike, I think you’re gonna like this. So Microsoft is inducing fabric policies and they’re in preview to provide fine grained governance controls across Microsoft fabric. And what this is simply allowing is instead of relying on tenant wide settings, organizations can apply policies at both the tenant and capacity level.

12:33 What this matters is you can apply governance rules with organizational boundaries, restrict high risk actions and manage governance consistently. So what does a policy do? It allows. Right now in preview, governance supports the following. You can do item creation or allow certain items to be created, selected security sensitive workspace settings and external data sharing which is coming soon. The big one here, the main meat of the the dinner, the a la carte or is the

13:04 item creation controls capacity administrators can control who can create specific fabric item types with workspaces assigned to the capacity. The examples allowing certain security groups to create dataflows, Gen 2 items or data agents only. Only in approved workspaces, requests that do not match the approved rules are blocked. A few key points, the relevant fabric feature, the setting that allows and a matching active policy rule. So the policy rule would be based on workspaces, the users and what item it is.

13:34 You create multiple policy rules here. So I think this policy thing is something that the community has been asking for for a long time. , I’m not willing to give people fabric because I can’t Turn off Dataflows Gen 2 or Genfo Dataflows Gen 1 or whatever the, whatever the setting is inside the workspace. So I think this is going to be able to give people much more control. I think this is also going to make sure that the central governing body, the BI center of Excellence, they’re

14:08 going to have more things to think about. Right. This is going to just give you more control to really decide has your team really been chosen to build all the items? Right. For example, we just talked about earlier about real time intelligence. I may want to give you an access to a workspace that’s using notebooks and lakehouses and semantic models and reports. That might be like the scope of what I think your team should be using or that’s what you need to build your pipelines or the data things. If I give you fabric today, right now, you can do anything. If it’s in fabric, you can run it.

14:39 Okay, great. It’s nice and easy that way. But I don’t necessarily want everyone to have real time analytics. One of our major complaints we’ve had. Right. So, I may want to give you access to a portion of what fabric is doing, but not all of fabric. I think this is the, the Plan here to give more fine grain controlled access here you set the policy and then the resulting workspaces or if things are enforced and, and get set to an audit log so you can see what’s going on there. So I think this is very useful.

15:10 I like this idea of a policy set. So again, I think this is a filling a gap. I haven’t played with it yet, so can’t be fully opinionated about everything. But directionally I think we’re in the right place here. All right, do you got anything else for me? That’s all the news I got. That’s basically all I’ve got as well. We’ll go with that one for now. And let’s jump into the main topic today. So our main topic today is coming back over to a Learn document. So this is the announcement that was made from the Microsoft Fabric Conference in Barcelona.

15:41 This is how to create a deployment plan. And let’s jump into this one. Here’s the documentation. It’s also on the chat window as well. This is another one that I think is very useful. So it may go different a number of ways here, Tommy, but if you think about deploying items, I think the BI teams that I’ve worked with or when you think about reporting reporting teams just have maybe a development environment and a production environment, dev prod, that’s basically it.

16:14 I build some models, I build some tables and I push out a report and then that’s basically it. As we get a bit more IT centric, there’s a lot more software development capabilities, patterns, best practices that we’re bringing along here for the ride. Right. Microsoft has made the pbir, we now have Tymdal. So things that can be tracked at a much smaller level and we can roll out specific changes in our

16:44 pipelines. Deploying the physical items are one thing. Getting the data into them is, yes, entirely a new problem. And so I think about, when I think about deploying things, I think I’m thinking about dev environment, test environment, production environment. Usually those are broken down into different workspaces. You could have potentially multiple workspaces happening, right? So you could have a row of workspaces that are just semantic models and reports and you could have another

17:15 row of workspaces that are just data engineering lake houses and loading data. So I think the challenge here becomes, well, what happens when you have three environments and then you also have three batches of data, Dev data, QA data and production data. And what does that look like? So the thing, this is what this is trying to solve is this ability to like, okay, we need to run a pre action on top of this lakehouse or I’ve run into problems, Tommy, where I’ve tried to deploy

17:50 a pipeline, right? If you have a parent pipeline and then you have children pipelines beneath it. Well, the children pipelines need to exist before the parent pipeline deploys because I can’t deploy them all the set at the same time. So there actually is an order, a sequence. And that has just been happening in deployment plans just by itself. We haven’t really had control about what’s deploying, when and where. But I’ve gotten into situations where I’ve tried to deploy a pipeline and says, you can’t deploy this or it

18:20 fails because I’m trying to deploy the parent and I haven’t deployed the child first. It doesn’t know what to attach to and it’s missing things. So I’m like, this is a mess. So what do you think, Tommy? What’s this feature like to you? There’s two things here. I think first off, this is I always love when Microsoft releases something that I call a process changer. If you use it, you got to change your team’s process immediately because it doesn’t work though, without having your team do a different set. , intake what the actual actions are and what the expectation is.

18:52 So this is a process changer. It’s funny though too, because deployment pipelines fell off once Git really became the standard in fabric and power BI workspaces, right? Like deployment pipelines made sense, but then we had all the source control and BR branching and logical branching and deployment pipelines almost felt out of place. They still made sense, but it didn’t feel like part of the whole flow of things because your main central repository, for lack of a better term, is the repository is your workspace.

19:28 And deployment pipelines were like, okay, well can I use both? Because they felt like such different flows. Deployment plans, I think really help allow. Think of them as I think of them, Mike, as if you ever use GitHub Actions or repo actions where whenever there’s a commit, there are certain actions that may occur on that repository. And deployment plans allow a workspace to simply define explicit order of deployment or deployment in order of time.

20:00 So to your point, we’re going to create the table, then we’re going to load the data in. We can run automated actions before or after any deployment. And the biggest part is dependencies. To your point, the pipelines did not have that where you can add dependencies that are not normally captured by fabric lineage. Fabric lineage is great, but it does not encompass everything. And if you’ve dealt with deployment pipelines, that and you can use that same plan with your Git integration, deployment pipe lines or REST API.

20:33 So to your point, these are really useful when we can deploy a items in a specific sequence that lineage would not recognize. You need to execute notebooks or pipelines or other supported items before or after deployment. That to me is the the,, the star of the show here. And you want the both lineage dependency and handling custom deployments. But yeah, oh, you’re on mute opponent to pipeline doesn’t let you do like in the pipeline some action and sometimes you need some

21:08 of those things. So the second article here, which is,, actually learning a bit deeper about what our deployment plans. Right. Let me put this here as well. it talks about some examples here, right? One of them, one of the examples is, well, you can use it to refresh content after it deploys, right. So there are actions right now. Okay, I want to deploy and Tommy, I want to go back to your comment as well here. This will be a double comment here.

21:39 Double comment. Refreshing content after deploys. If we were just using a peer deployment pipeline, we would deploy potentially the, the lakehouse, maybe a notebook and maybe a semantic model depending on where your data comes from. A lot of times companies will have a development server with development data on it, a test server with test data on it, and then a production server with production data on it. So as you’re deploying these things, you may want to rehydrate or reload data after the deployment is complete. Right. This example they give here, you have,, the sales warehouse exists after the item deploys.

22:16 There’s nothing, there’s no actions. But then you have the sales semantic model. Well, once the sales semantic model deploys, you need the sales semantic model to refresh itself, get the latest bit of data, right. And then it has the refresh information. So typically in deployment pipeline you just do that by yourself and there is no refresh action. There’s no like actions after the item has been built. So I think that makes a lot of sense. They have another one here for validating anything before something deploys, right. , having checks in place.

22:47 So in this example here they have like some pre flight checks. Hey, this notebook needs to run and have a full successful run. Check the server exists, check. This is on check. We have data check. We have this check, check, check, check. Right. You can just define certain things.

23:02 , you need to be in place place before the next step of the deployment occurs. Again, makes total sense. Then you go through and after the checks approve, it moves on. So the nice thing here is the notebook can run and if you write the notebook in a way that says okay, notebook fail, that failure is then passed back to the pipeline. And the point pipeline says, oops, stop. I’m going to stop at this step and not proceed forward. So I think this is another really big moment here of like not only are we running these pre and

23:35 post actions, but the information that’s running inside the notebook is actually being passed back to the deployment plan to say if it’s a go or no go to the next stage. Again, I think this is really well thought out. This makes sense to me. And the big part here too is this is filling such a gap for us where to be honest, Mike, I, because of source control and the git integration, I was really stepping away from pipelines because it’s like they

24:05 just again they both felt detached from one another where all the actions you’d want to do, again you didn’t have the full experience of what you would really need in a deployment pipeline to do. So I think right off the bat, if you’re wondering, can I reuse deployment pipelines now? Yeah, but you got to use them with a deployment plan, Mike. And there’s three main items in a deployment plan I think it would be good to go into. And I really am happy of how Microsoft thought this out because I think this is all really well thought

24:36 out. Any deployment plan that you have is going to have three things. Deployment groups, the steps and actions. Mike is again, it’s actually written in YAML too, which I didn’t even recognize. So it really is like GitHub Actions. that’s what we’re dealing with is awesome. Which is awesome. Those applying the groups, you can actually add items to specific groups, say okay, these items, this group of items is going to be deployed. Then we can in a sense run either a pre deployment action if I want.

25:07 Before we even deploy these things, we’re going to run actions like a notebook or I can say pre deploy and then run actions after they’re done. And then you can run those post deployment actions. What is really important is the deployment order alone isn’t enough. This is in one of the articles about deployment plan examples because you can see the lineage and fabric and I really want to Dive into this, Mike, with you. You can detect the warehouse view reference to the lakehouse table. So it deploys the lakehouse before the warehouse.

25:38 That might be an example where you have a lakehouse and you want to give it to the warehouse. The problem is lineage alone. If you were just looking at the lineage view, view doesn’t show that the table must be populated before that view is actually valid. Right. It’s not just you have a blank view. The table doesn’t arrive with the item deferent definition. So two notebooks produce it at the run time. So if you were to deploy a pipeline in that order without the notebook actions, it’s going to fail every single time because every item is copied, is creating the views, creating the table.

26:12 But the warehouse view is created against something that doesn’t exist because your notebook hasn’t run with deployment pipeline. With deployment plans, you can say, okay, we’re going to have the lakehouse, we’re going to run this notebook to populate it so this thing actually exists. And then we can deploy the warehouse view. This whole order, I think opens up so much for us. Because I’ve been having a lot of questions lately, Mike, with some of my clients around git and deployment pipelines and where does this stand? What’s the. How essential is this to an organization’s workflow?

26:46 I have been more personally recommending like, you got to use source control, you got to use git. I’ve always been, I don’t want to say I’ve always been averse to pipelines. That’s not the case. It’s just, it didn’t have everything and it was always like, well, you can use deployment pipelines, but I didn’t see them as an essential,, piece of the puzzle. Source control is for me now, by the way, regardless if it’s in ga, it is high on the list. Yes. But now that we have this gap that fills in what lineage couldn’t these actions alone.

27:22 Let me just bring it to you in your take. In terms of do we look at this and say, team, we need to invest in this today. I think I’m going to seriously look at what our current deployment pipeline looks like. Our current deployments look like. So I’m gonna really. It’s gonna depend on how your team likes to deploy. Right? I think we’re. I think we’re at a place where if you are really into deployment,

27:59 if you’re, if your team really understands git and get an integration, I don’t think you need deployment plans. So it feels like to me this is another one of like,, in a GitHub action we can do anything, anything we want. Yeah, yeah, right. We can sequence everything’s API calls. You can run notebooks from API calls, you can have information coming back from this API call. So you could describe all of this, I think inside a series of GitHub actions that run not only deploy

28:30 an item, but then run said notebook during the deployment. I think it’s possible. It now requires your team to know how to run GitHub Actions and now requires your team to then do a lot of testing on this to make it all work correctly. Sequence things the appropriate way. If that is your workflow. I don’t think you need to go into deployment actions or deployment plans. I think it’s going to be a major help. However, if you are using deployment pipelines and you are running into some issues with data loading data.

29:02 Another one I think of is like there’s a weird bug right now. Maybe it’s not a bug, maybe it’s by design. I don’t really know. , if you go into a lake house and you have a source system, let’s say your source system is something like Snowflake or Databricks and you’re trying to shortcut into those tables outside of fabric, right? If, if a table does not exist outside of fabric, you should not create the shortcut. And if you use a deployment pipeline, if you have a lakehouse with shortcuts in it, you can still parameterize

29:37 the lakehouse to point to a different environment Dev test prod. But if your dev environment has Tommy’s baseball data as a fact table, and I move that same shortcut to the test environment and Tommy’s baseball data table does not exist upstream in the test catalog or whatever the system is, that shortcut will fail to deploy. And instead of just saying I’m broken, it will go revert back to the development data.

30:09 What it was working. So basically it fails and it reverts back to its default state, which is very much not ideal. Do not want that. And, and so, so there’s like some really weird things going on with shortcuts as you move shortcuts. So it’s all these like little edge Casey things that are like happening that are like very specific to like why you need to test things, right? For example, this is like the pre test, right run. Before I make all the shortcuts, let’s test that all these tables exist.

30:39 Check, check, check, check, check, check. Okay, pass. Now do the next step. Right. So I think this is a very powerful tool for teams that need more control, a bit more than just deployment pipelines, but they’re not willing to go full bore GitHub actions and just do everything or Azure DevOps actions and do everything in code. Right. I feel like this is a good happy medium between the both. Well, I’m gonna, I’m gonna agree and then push back on two of your comments here.

31:10 So the first one is you said, is it a bug? And this is. I don’t know, this is probably not pushing back. But the whole bug that you’re talking about, it wasn’t a bug in deployment pipelines, is a bug in Lineage. Because the problem is you can look at the lineage and go, oh, I guess this is how I set up my pipeline. Microsoft is more or less admitting at this point that lineage doesn’t cover everything. Just looking at lineage is not the whole story. Which is exactly where deployment plans come into. Because they’re even saying lineage doesn’t show the table must be populated before the view is valid.

31:43 Even if you look at the lineage view, it doesn’t necessarily what actions need to occur. It’s only the metadata. Right. Which most people look at a lineage view. Right. If you’re starting a deployment pipeline for the first time and go, that’s the way I’m going to, in a sense, create my pipeline. That’s the way things need to go in this order. But again, the problem with that is it doesn’t. That data doesn’t arrive or the table doesn’t arrive with the item definition. So even if you were to run that pipeline based exactly on what you saw in Lineage, you normally would

32:15 run into errors. It’s not necessarily a bug in the pipeline, it’s a bug in the lineage in so many words, because it doesn’t find when things actually get populated. So to me, it’s like Microsoft going, yeah, lineage is good for metadata, but not good for the developer in some ways. And that’s where that deployment plan comes in. Yeah. Can I just jump in here? Sure. What I think you’re talking about is there are three items we care about when we’re doing deployments. We care about the infrastructure, which is the lakehouse.

32:48 Can you just make the item of the lakehouse? Right. Can I deploy a semantic model? That’s another. Can I deploy the item itself? That’s the infrastructure deployment deployment pipelines handles, in most cases, infrastructure deployment. The second thing is, well, there’s code that goes inside the item. Now can you deploy notebook, yes or no. But then there’s code that runs inside the notebook. You need to deploy the code in the notebook.

33:19 Also, when we’re talking about deployment solutions, if I deploy a notebook named,, Tommy’s Amazing Notebook, tomorrow when I make a change in that notebook in dev, I need to redeploy the code into the same item. So the infrastructure is not changing, but I’m now changing the code inside the notebook. So that’s another. The code of the item now needs to be deployed. And the thing that the deployment pipelines does not cover is the data.

33:50 There is no data deployment. And so what I think this is solving is, yes, it’s not as probably clean as like an automatic tool, but you can run a notebook that hydrates the data that then deploys the data that you need. Right. So I think this is another one of these situations where that’s another really good example here. Another one I also maybe point out to Tommy. I think deployment plans can also use variable libraries. Yeah, I think so too. So I’m going to have to double check that comment.

34:20 But I really like the idea of variable libraries because when you build infrastructure and code and deployment of data,

34:31 those are the three things I think we’re discussing here. When you’re deploying the code, you need the code to be parameterized. This is the GUID of my lakehouse. When all my notebooks run, they should run on top of the this lakehouse, which is the dev lakehouse. When I move to the next environment, that lakehouse need to switch physically to a different guid, a different item that is okay. Now everything needs to be parameterized to point at this new location. I think this is something that the variable library does a good job of explaining.

35:07 Defaults dev test, prod parameters. And then you can use or inject them during deployment in the environment that you need them in. So these are all like. There’s a lot of concepts from software development that’s showing up here that is very useful for us to understand. Yeah, and let me go through an example here of the different ways you would take your normal pipeline and put this into a plan. And I think this is probably the best example that Microsoft has right now. I want to give you a scenario here.

35:38 Let’s say you looked in your lineage and you had a deployment pipeline that had first a notebook that writes to a table in sales lakehouse that’s read by a notebook that publishes the top customers that publishes a table in the lakehouse. And that’s read by a view in the warehouse. So notebook writes to lakehouse, read by a notebook publishes to a table and read by a warehouse in that order. Even if the lineage looks right, in a normal pipeline, that fails because every item is copied.

36:12 You’d see all the items there. But the warehouse view has no data because in that or everything’s deployed at the same time. Yep, correct. So that’s the problem. It’s not technically, yeah, things are deploying like simultaneously, but like. Right, right to your point. Yeah. You can wire them all together and you can move them between environments, but you’re only getting two out of three of the things. Exactly, exactly. So with the plan, what actually happens is you have two deployment groups. You have a group called sales Lakehouse, Sales Warehouse, the item deployed. The first one is the sales lakehouse item.

36:43 But the difference is you have two post deploy actions. You run the first notebook and then you run the second notebooks. The second deployment plan is called Sales warehouse, which is that item that you already had and it has a dependency, which is a huge part saying this depends on the sales lake house being complete. Done,, valid. Exactly. That allows that lineage happen. And Mike, for me, I want to make a note here because you mentioned that if you are already using GitHub Actions or in DevOps YAML Actions, that this makes sense.

37:18 I would venture to say because of the canvas setting, there’s YAML in the back end, but you can build this all in a ui. You a nice pretty canvas user interface. This is the difference. Right, right. And I think for a lot of people, most people, this allows people to get introduced to that. You don’t have to have experience with GitHub Actions to use this. Let me, let me also be really real here, Tommy too. Be real, be a hundred, be a hundred, go 100. How, how many people are able to build a deployment pipeline without an agent by themselves right now?

37:59 I’m just saying, like whenever I’m building deployment things, I’m relying heavily on an AI agent to help build the right or actions. Okay, the whole thing, the whole like, not the pipelines but like,, there’s this idea of like what is a user interface and what does the user do versus, so that to me that blocks a lot of companies and organizations because now you’re saying, okay, one, Tommy, go make a deployment thing. But here you don’t get a large language model or an agent to help you go do this. We don’t think that’s secure. So therefore, Tommy, you’re On your own, figuring out how to write this action from scratch,, I really,

38:35 the, the amount of code that’s written is not complicated, but the fact that the agent knows every single thing, man, Tommy, I know this too, because we’ve been doing. Tommy and I have been talking about this for a while. We’ve been using fabric as a back end for a long period of time. And when fabric becomes a backend to your systems, you need agents to help you build all the deployment pipeline stuff just for an app, just a website. Right. I have code, I need to get it into GitHub pages. I have code, I need to get it.

39:06 To be honest, Tommy, for the longest time I was extremely intimidated by GitHub pages because like, how do you set it up? What code needs to be there, what needs to be in the action? I didn’t know what was going on and I had seen other people using like examples from GitHub, but every time I ran them, they didn’t quite run correctly. Now with agents at my fingertips, it’s a one liner for me. It’s literally show up and say, hey, agent, I’m going to build this website. Here’s the tech stack I’m using. This deploys as GitHub Actions.

39:37 It knows exactly what we’re doing, does all the CI CD things and then all I have to do is go into the repo and just say, okay, let’s just make sure we turn on the deploy from actions. And then boom, it’s up and running. And if it errors, I just go back to the agent and say, failed. Here’s the error code. Oh, I know what that means. Fixes it and moves on. So for me, I really like this idea of like getting the agent to help me set up these deployment patterns. It’s so rinse, wash and repeat.

40:07 I guess it’s so common. Yeah, so, so okay, align that. Then to me was what’s happening fabric. Because I love that example. But tell me how that works when we’re talking about the deployment. Yeah, so maybe let me bring it back home here a little bit. So there is a risk to this, right? The risk is if I’m in an organization that doesn’t let me have an AI agent, I now need to set this stuff up more on my own. Right. I need to be able to have a ui, something that’s like, if I’m not used to writing on

40:38 these, these actions and things, to me, I look at this as a barrier to entry. The barrier to entry is if I don’t have access to AGEN agents, if I can’t show up and just tell them what I want to build, I now need like the human experience of like, what does people do to build something? So the deployment plan is the UI that I would use. I don’t need an agent to do this. I can think through the sequence of items that need to be deployed. And so while I could build all this with actions and I don’t have to, it’s now part of the

41:08 product. And I think there’s a lot of people in that camp where we’re coming from a power BI world. We’re starting to get more fabric, we’re starting to get more deployment savvy. Yeah, Deplavity. Deployment savvy was what I meant to say. Like two words instead of just one. So because we’re getting the business teams that were coming from BI are now getting more of this IT centric space, having a UI that shows the three patterns, infrastructure, the code and the data, and sequencing those things.

41:40 I think we now open the door to making this much easier for users who don’t necessarily need to have or don’t want the complexity of going to go talk to an agent to build everything with deployment actions. Okay, first off, you have to trademark now to Plavi or Microsoft’s going to take it from you. They love nothing more than taking two giant words and just compound making a really bad name for the new product. Yeah, get to play. That’s a T shirt, so go get the flavor. Okay. And then the other thing is timeout. I deplavity to prod.

42:10 Right. I. I only. I deplavity right into prod. I don’t think that would be very deplavy of you, really. So. But I gotta, I just gotta put a timeout here. You said, and I just want to make sure I heard you right, that this day on October 20, 2026, that if something doesn’t have an agent, it’s a barrier to entry. That’s where we live now. So even though, because you said that about the pipelines, even though deployment plans have a really great canvas, I. I’ve got to admit, looks pretty.

42:41 It’s intuitive. Okay. So I don’t have to have any agent experience to do this, but if I don’t have an agent, like, are we already at that point now in our life where if something doesn’t have an agent, we’re not going to touch it? I’m gonna. The hours I spent on Dax. Right. Tommy, I’m gonna respond your question with yet another question. Tommy, I’m gonna give you. Tommy, I’m gonna. Yeah, it’s probably the answer. Tommy, I’m gonna give you a promotion.

43:12 Okay. You’re gonna make. Let’s say it’s 10 more. But you’re not allowed to use AI. No. Would you take it? Okay. All right. That’s so easy. I really need you. Can I have. I’m gonna give you. I’m gonna give you 20. I’m gonna give you 20 more. But you can’t use any of your AI you have to do it all by hand. I’m going to answer your question with a question. How about that? So how. How. How high do I need to go, Tommy? No, no. How much? The other way. I’m going the other way.

43:42 Let me answer this. You’re going to get to pay 20 less, but that’s the only way you can use AI are you going to do that? What you make in a year, you can keep making, but no AI if you want to use AI, you got to get paid 20% less. Yeah, because I’m probably going to work, like, three more jobs. Okay. Yeah. , I think. I think it’s at some point. So. To your point there, Tommy, like, until things catch up and the consulting firm’s going to get their head

44:13 around themselves. Right? The. The. The AI space is going to start compressing consulting salaries or wages down lower level because it’s. It’s providing much more access to many more people to do more technical things. Right. That. That’s going to compress at some point. But there is. So I was actually watching. I love watching these people on, like, YouTube. They have all these, like, really funny or comical things. So there’s a gentleman, like, doing a little bit of a skit with himself. He goes,, the company responds with, hey, Mr.

44:46 Consultant. Hey, we’re not going to use you anymore. We’re just going to use AI and just build what we want to build with AI and the gentleman says, okay, well, that’s fine. You’re. You’re more than welcome to use AI but he goes,, I was going to use AI Anyways to build your solution. Do you think a engineer with 15 years of engineering experience with AI would build a better solution than you with no experience? And AI would build. And the. The business. The. The business comes back and says, h. You’re right. You’re probably going to build a better solution with the AI at your disposal than we would without

45:18 the experience with AI as well. And there might be more hiccups along the way. So,, I think there’s this idea of like we, we think that AI is coming in to replace things and actually reality, I don’t think that’s doing that. I think what AI does is it actually adds more capability to the experts that are already there. And if you already know how to narrate and direct the AI to certain things, it will make you, you know, that much sharper, that much more productive, that much more. Yeah. Capable in that space.

45:48 So I guess where I was going to go with this one, Tommy, in the whole cheeky,, segment here around like, how much are you willing. How much do I need to pay you, Tommy, to do your job without AI? I don’t know what the number is. I, I don’t know.

46:01 I think we’ve, I think we’ve already hit the tipping point. So my, my whole point here is I think we’ve already hit the tipping point of, for companies who have leveraged AI, there is no going back. You’re not going to go from this. I can build full applications in a weekend. Tommy, how long did it take you to build Fabric Ratings to get everything right? I went through a lot of iterations. There were a lot of. But it was probably like from a week of time, which is incredible. Thinking about all the components.

46:31 Yep. Describe for me a bit more of your architecture. Right. I know it’s not just a simple like website. You didn’t just say make a website. Right, Right. What did, what did you tell me about like,, how. So I, I always want to use fabric as a back end and I, I’m trying to do that more and more. The problem though is if I want a public facing site, if you get a lot of access, it would have to read the database every single time, which is going to be a lot of cost. And you don’t want, you don’t want people running every time you’re the same data. Right. That’s going to be a problem. So the front end behaves a little differently.

47:03 But I also want, I wanted to create something where I can invite other users and have credentials stored. So like I’m saying, hey Mike, you want to submit ratings? Great, I’ll create a user for you. It’s not using Microsoft Sign, it’s actually can create and generate a user on the back end using Fabric. There’s like an admin portal to it and that will spin up the database and everything loads. It takes a little long the first time if it’s been like 24 hours and. But it will actually go through and allow that user on the back end using a fabric app with anonymous

47:34 access to be able to go through and submit ratings. If they’re a rater, admin has these credentials stored or the permissions that person has and they all get submitted. And I basically the. As the admin release, let’s say we had 10 people submitting ratings, right? Or 100 people submitting ratings. All those ratings would be there, they’d be average, but they wouldn’t be on the site yet because I didn’t want like the score to change every single time that someone submitted ratings. So all that data is stored in fabric in the back end and when I finally release it, saying, okay,

48:05 round quarter four round is going to be released, then that will update and push to the site. So once you go to the site, it’s super fast. It’s not reading fabric at that point, if you log in it is. And outside of that it’s GitHub pages. So I can do this work control too. So the front end is GitHub pages. The back end is a mix of a UI of GitHub pages. But really fabric has a backend completely. And again there’s a lot of iterations because I wanted again the ability to create users without having them be

48:35 part of my Microsoft tenant. What if I wanted, what if Kurt said, what, I want to submit ratings or someone else came up to like, I want to submit ratings. Okay, well I’ll create a user for you. I don’t have to add you as another user. In my tenant, you will have your own login information and then they submit any time I can re, you know, review when everyone’s submitted and release it and that becomes the front end. So. Yes, well, I was, I was really just focusing on like you built a very complex infrastructure, right? You have this half of the app is in transactional data and then the other half of

49:12 the app is like static data that when the transactional data is done, you land the transactional data into a static something that does not change. So you have this hyper efficient side of the. This part of our data does not change frequently. I’m going to just serve that as a very static web page over and over again. The other side of it is, okay, well the same app now has another part of it. There’s a series of screens that are dynamic, they’re always running, they’re updating, there’s constantly Changing. So when you’re doing the SQL things, when you do the SQL side, you have a login search for it

49:47 turns on, store the data server turns on when you’re done, process said data. Static pages, cheap. You’re able to stop thinking about, I’m just going to throw a SQL database at this and run everything through SQL. Now, Tommy, you’re designing for optimization of websites. So they’re fast, they’re good, they have all the things you need in them. And again, this is where I’m saying, yes, you spent some cycles on this with an agent, but imagine building that on your own without any agent.

50:18 It would not be built. Yes. So this weekend I built two more games that I’ve been playing with. I sent them to Tommy. They’re fun and, and so like kids, I’m like, I, I’m, I’m using, I’m using one of your games right now as a reward for them. Okay. Yes. Yeah, that’s amazing. Okay, good. But this is the point though. The point is without agents, you could still do the same stuff. It’s just going to be much slower. And so I think, yeah, this is, this is why I’m thinking like,, again, not every.

50:52 Again, I’m going to go back to my point earlier, which was not every organization has given everyone free rein to agents the way that Tommy and I have in our, in our companies. Right. And so because of that there will be a need to do these sequence of deployments of things that I’m not always doing anything through cicd. There is some challenges and some friction there. This deployment pipeline plan solves a lot of that friction. Right. Loading of data, running of notebooks, making sure things are like notebooks are executing and sending messages back to

51:22 the plan. I think this, this whole problem they’re trying to solve has been something I’ve been complaining about since deployment pipelines came out. There is no action around the data how to deploy it. This solves that problem. Yeah. So let me tie back in as we get near the end here. Right now, I didn’t see a lot with the what is appointment plan about using AI with it, which I’m sure there’s going to be skills here. There’s going to be an mcp, hopefully there’s an MC server. Yes. But like why, again, this is like you don’t really need it.

51:55 Okay, so let’s get, let’s dive into that. So you’re saying I see this, my statement is I see deployment plans with the nice Canvas with someone who’s a data engineer to be able to use all this without AI. Right. Because if you’re using YAML, if the only thing, hey, we’re going to do deployment plans, it’s GitHub Actions I’m probably going to expect at this point. Agentic experience. Well,, before I get into that, because there is YAML, but the fact that anyone can

52:26 go into fabric, use the UI and honestly, very logically, intuitively build it out. Yes. Is such a big win here because I think to your point, you and I talk about AI and people know we do. We’ve gotten comments that we talk too much about. , it’s the world we live in. A lot of organizations don’t have that ability. They are, they, they may want to. The person listening right now may go, dude, I love AI. This is all these agent things I’ve been clamoring for, but my company’s not releasing it, so there’s always that barrier. These deployment pipelines, I think were just, they were very well thought out.

53:00 They were very well thought out to be solving a very specific problem too. Yeah, this is, that’s another good point. It has a very clear plan on what it’s supposed to do and also how to build it. And I think hopefully testing is a big part of this as well. Now, what’s interesting is the GIT integration with this is you, their YAML action can be part of your source control. So honestly, Mike, if you wanted to use an agent with this, it’s YAML and it looks, it smells

53:30 like, it smells like GitHub actions and it looks like GitHub actions in terms of what the YAML is, the schema, the version, there are the groups. So all that looks very similar. The problem is it’s, it is not meant for a user to create. Like this is going to come straight from the canvas because it has like the logical IDs and what the action names are. And there’s a lot of different things you can do with that plan YM YML file.

54:01 But if you want, you could have an agent run across that. But as we get near the end here, this is still preview, so let’s still make a note of that. I’ll put you as the data czar or the bis are of your organization who’s already. The scenario is they are already using deployment pipelines. They have the frustrations, they have source control too. Are you expecting them? What, what are you telling them about deployment plans? Are you going to tell them, let’s look into this and build a sandbox with this.

54:32 Are we going to say we’re not touching this till this is ga, or are you going to full bore like we want to adopt this? I want to adopt this by the end of the year or, by the end of November. Let’s make a plan for that. Where would you stand? As of today, I will not recommend anything that is directly coming out of,, preview or public preview until it goes to GA to go using any production workloads.

55:02 I do think we want to understand how well this fits our business needs, right This. I look at this and I go, I really need to evaluate what company culture we have. Do we have a company culture where we already have a deployment team doing things? Larger organizations seem to have team members they’ve hired specifically to help with deployments. In those situations, I might err on the side of going more down the git route and just actions. Another thing, I’ll just be mindful here of if we’re going down straight git or GitHub.

55:36 This also assumes everything else you need is already defined and in git already. So you’re having git on workspaces, you understand what git is, you understand branching. Not that that’s impossible for people to learn, but there are teams that just, it’s new to them, they’re not as familiar with that. So I think I’m going to be really looking, being critical about like the culture of the company and what they have access to and what they can do inside their organization, what, what level of comfort they have. My argument is many organizations haven’t figured out yet, don’t know what’s going on yet, still trying to work on

56:07 it. Even though that is all ga, they’re still trying to figure out, well, how can we just do it in the ui? What’s easy. So I’m gonna, I’m really gonna evaluate the, the culture of that team and what we’re deploying things with in that team to really figure out do they have this,, the, the chops, the skills. Am I, am I going to have to teach them a lot of other things in addition to get in integrations and get actions? Or is this a team that’s already, okay, we’re already comfortable with deployment pipelines.

56:37 We just need a little bit more assistance here. Now I can step into a deployment plan and get a bit more functionality out of it. And I think the deployment pan, the deployment plan helps get you closer to that end. Game of I want to deploy things, I want data to refresh. It should just work. Okay, I like that answer. I’m going to tweak some of the variables and I want to see if your answer changes. Let’s say deployment plan. Deploy deployment plans. Rga.

57:07 Let’s say they are officially GA today. We just didn’t realize it, but the team that you’re working with does not use Git and they’ve only dabbled in deployment pipelines. But this is a GA feature and there’s a clear need for it. Is this something you’re going to push for them in the next two months to start utilizing and relying on? If this were to be ga? No. Get experience, a little pipeline experience.

57:38 So use this. If I have a. I would say a little deployment pipeline experience is required to step into this new deployment. I think, I think you need to understand. I think you need to. Again, the whole concept of like this is. It’s the same thing as a deployment pipeline. It just gives you a bit more control around sequencing and pre and post actions. So I think this is like a step that builds on top of a deployment plan or if I’m just deploying a semantic model through. Sure. An environment. So I feel like deployment pipelines were very much for like, I need a report and I need

58:09 a semantic model. Right. I deploy the semantic model through. Yeah. I have users going into a workspace and they’re refreshing the semantic model. Right. All those things are fine. Work just really well. Right. As soon as I need to start,, having more hands off to these things. And again, we, in the past, Tommy, we’ve always communicated. I’ve communicated to our clients and stuff like that is there’s this concept of a release engineer. The release engineer is involved with. They are able to touch test and production.

58:41 Right. When you’re using a deployment plan, all of your engineers building the things can only, could only touch development. That’s the goal. Your developers only touch a single environment. Someone else is in charge of moving those items between the environments and making sure the data is refreshing and things are good there. Deployment pipeline plans makes that even easier. So I, I think that’s where I’m going. I think some experience or some knowledge around what a deployment pipeline is is useful.

59:12 Yeah. And then I’m extending that idea further with that team and saying, well, let’s now talk about deployment plans. This also might be very helpful for you and make it even easier for you to switch between different variables and different environments and other things that you need to be doing. And then you again, you’re starting to take that team from. I’m just a. I need to get these things between the environments to. I need to get these things between the environments and I need to hydrate data and I need to test things. It’s adding more robust process to that thing. I think you just hit the nail on the head.

59:42 I think you. You just said what I wanted to say at the beginning of this call of our episode where deployment pipelines were power BI and deployment plans are the fabric with deployment pipelines are to power bi because once we enter the world of data engineering, pipelines missed the mark. It just didn’t have everything where honestly, if you are focused on deployment on fabric and lakehouses and notebooks, this is easier than a pipeline.

60:14 Honestly, it’s easier to get your grasp on and easier to create. Conceptually, this makes sense if you’re just doing semantic models. Well, all you needed was a pipeline. Really, this doesn’t make as much sense. So if you are a fabric shop, yeah, the deployment plan is easier to think about and grasp and create. I. I hate to end on this, Mike, because I know we’re getting at the end, but one thing that we mentioned here is this is in preview. So again, I mentioned to Mike about what would you do if this was in ga?

60:46 And the reason is please make sure take a look on deployment plan actions and creating a deployment plan. There’s a ton of limitations here. So I hate to say this, I,, because there’s so many cool things here. But the thing is there are limitations here. And for example, you can only. You can’t use one plan to deploy into more than one workspace in a single run. You can only attach one deployment plan to a deployment operation. There’s a few things in terms of size of files that you return or the length of how many groups

61:20 you can have. It’s pretty large. The limitations for plan actions and limitations during deployment. So make sure to look on this. This is on the deployment plan actions page of the docs. So this is why we say especially when you’re in preview. Let me put it this way. I agree with you. This is something I’m going to wait for ga, but I’m going to be very antsy about because I can’t wait for this to be GA to bring this in. So I totally agree with you on that one. So, Tommy, in the era of,, your funny joke here, and I think John Kirsky.

61:50 John, thanks very much. Deplavity sounds like a prescription drug to solve your deployment fabric artifacts so actually Tommy, I’m going to pitch it to you. Let me know how you think, all right? Millions of fabric users struggle with fabric drift Deployment drift every day. Ask your capacity admin about deplavity. Deplavity is the once a day treatment that rebinds your semantic models, remaps your connections, and convinces your deployment rules to actually apply. Clinically shown to reduce why is it pointing to dev moments?

62:22 Up to 94% do not take deplavity if you are allergic to YAML have currently been in a git conflict. Side effects may include sudden confidence in git integration, an urge to parameterize everything, and telling strangers about workspace identities. In rare cases, patients have reported their capacity throttling while they were simply looking at it. Do not operate heavy data flows while taking deplavity. If your refresh lasts longer than four hours, contact a Microsoft rep right away. Deplavity may interact with tenant settings you did not know existed.

62:54 Not approved for Premium per user Consult your fabric admin before stopping, starting, or thinking about deplavity. Deplavity because artifacts deserve more to arrive than when you sent them, may cause headaches and data blockage. That’s good. I like that. So granted, I. I hate. I loathe those commercials on tv and I think all of them should get. I don’t know. I don’t want to say I don’t like those commercials, but I love that thing

63:28 anyways. That is hilarious. Thank you for listening today. Today’s sponsorship, brought to you by Deplavity. That’s good. All right, we have a sponsor for the first time. For the first time, Deplavities. Tommy, where else can you find the podcast? You can find us on Apple, Spotify, wherever your podcast, make sure to subscribe and leave a rating. It helps us out a ton. You have a question, idea or topic that you want us to talk about in a future episode? Well, put your money where your mouth is. If you don’t want us to talk about AI, send us a mailbag at powerbi Tip Slash Podcast.

63:58 Leave your name or a great question and finally join us live every Tuesday and Thursday, 7: 30am Central on all Power D Tips Social media channels. Thank you all so much. And here’s a little earworm for the end of the episode and then we’ll see the better be.

64:32 Jack from Finance builds his first report at night. Experts smile whispered his team sees the light. Five minutes to wow. And the room starts to leave Power be arising from the ground unseen.

65:16 Joy in two cups. Bottom of belief it’s right in someone’s head A colleague shares a fire the spark goes wild Adoption isn’t for

66:00 it’s a rising time I’m on the floor before the tower woke up I know, I know Jack from finest pouring joy to come.

66:36 They said Enterprise would choose another name but the people on the desk still played the game because the two felt close because the wind fell near bottom of the this how the truth appears don’t wait for the man

67:08 from a distant FL Start with the person who needed one door Give them five minutes and a reason to believe the rest of the company loves to breathe There’s a night festival the base on the marches Think of Jack and how the quiet ones progress how of me I didn’t shout from above all because we built it out of love. Sa.

69:01 Feel the crowd Explicit measures explicit measures Drop it loud.

Thank You

Want to catch us live? Join every Tuesday and Thursday at 7:30 AM Central on YouTube and LinkedIn.

Got a question? Head to powerbi.tips/empodcast and submit your topic ideas.

Listen on Spotify, Apple Podcasts, or wherever you get your podcasts.