Slow AI Adoption – Ep.542

A listener asked what a 30-60-90 day plan looks like from experimentation to production, and who should own AI enablement on a BI team. Getting to an answer means first being honest about where most organizations actually are — which is earlier than the question assumes. This episode is about the blockers that come before the plan.

News & Announcements

• Fabric data agent API is now public — You can now create, configure, update, and publish Fabric data agents programmatically from external tools, pipelines, or backend services. The SDK runs outside Fabric against a public API, which fixes an odd gap: previously the only paths were the Fabric portal UI or the SDK inside a Fabric notebook. Tommy’s remaining wish is an API for how consumers actually use a published agent.

• Power BI reports and Fabric Apps: expanding how you build on your semantic model — Microsoft’s statement that a Power BI report and a Fabric app can build on the same semantic model. Drag-and-drop reporting or a Rayfin app, same core layer. It’s the vision the hosts have been describing for months, now written down by Microsoft.

• Chicagoland Power BI user group — July 16 — In person, with live Rayfin demos. Mike will build a real project from templates, make changes, and work through it with agents, sharing the techniques from his YouTube series.

• Microsoft Scout, Copilot Studio, and Azure AI Foundry — The three references behind Mike’s argument that organizations will increasingly build custom harnesses. Foundry in particular is where you bring a model and wrap your own harness around it.

Main Discussion

Topic: Why organizations are slow to adopt AI, and what the first 90 days should actually be

Mike’s diagnosis is unsentimental: leadership and IT hesitate because they don’t know what agents do, can’t predict the cost, and can’t see the value. On top of that, there’s almost no training available. Both hosts consider themselves AI-ready as individuals and note there is still no shared definition of what that means for an organization.

• AI strategy versus AI theater. Tommy’s distinction, and the most useful frame in the episode. Strategic initiatives target real bottlenecks in real workflows. Theater is generating images and pamphlets because a headline said to.

• We’re earlier than “use it or else.” Both hosts agree the mandate stage is far down the road. Right now IT doesn’t understand what an agent is and the company hasn’t accepted that agents add value — the Power BI adoption curve started in exactly the same place.

• Start with data governance, because the right people are already in the room. Tommy’s answer to “who owns AI enablement”: the governance council already has IT, BI, and business at the table. Without healthy governance you won’t get far, because the agent runs on your data.

• Somebody has to be designated — and security has to weigh in. Give one person explicit responsibility, ideally someone already exploring it. In parallel, a VP or C-level decision is needed on which tools are permitted.

• Look for the mundane, repetitive work. Operations teams filling out the same things repeatedly, account identification, priority lists. That’s where an agent pays for itself, not in content generation.

• Adoption won’t be uniform, and that’s fine. Not everyone adopted Power BI at once either. Proof-of-concept work that lets specific people add value in specific areas is a legitimate path.

• Watch for patterns, then consolidate. If every sales rep independently builds an agent that drafts outreach from their account list, that’s the signal to build one shared agent instead of thirty — otherwise cost climbs with no oversight.

• The agent should teach the user what it can do. Mike calls this an underrated capability: users of anything agentic need to know what it’s for and when to reach for it, and the agent itself can explain that.

The 30-60-90

Mike’s version: 30 days is education — articles, videos, documentation, and thinking honestly about which of your processes could be automated, because most people don’t yet know what agents can do. 60 days is building for one person and one process — the salesperson who needs their day reasoned through, a single workflow aligned to what you’ve learned. 90 days extends that pattern outward, with the agent instructing its own users. Tommy’s emphasis lands on the prerequisite: without a defined process, an agent has no clear reason to exist.

Looking Forward

Name one repetitive workflow on your team and one person who owns exploring it — that pair is more valuable than any org-wide AI announcement.

Episode Transcript

0:02 Tommy and Mike lighting up the sky. Dance to the day, to laughs in the mix. Fabric and AI get your feels. Explicit Measures Drop the beat now. Podcast Kings feel the crown. Explicit Measures Drop it loud. Hello everyone and welcome back to the Explicit Measures podcast with Tommy and Mike. How’s everybody doing? Everybody? All the peoples. Even you in We are doing pretty well.

0:32 We are doing pretty well. Say hello in chat. Let us know where you are where you’re calling in from, what you’re listening to. What’s your What are you doing these days? Let’s get everyone on board here. We should one day do a live call in. I don’t know how we would do that, but okay. okay. no idea either, but we could figure it out. AI can do it for us. I’ve been cooking up some software on the side here Tommy that that would be really relevant around that area. So, I don’t know if it’s actually ready to go, but it would be pretty cool to show the audience here what

1:02 show the audience here what we’ve been cooking behind the scenes here for some potentially a new interaction how we would handle meetings or teams or the podcast moving forward, so. Stuff is is being baked. Software is now easier than ever to create with agents and AI. Speaking of which, that’s going to be our main topic today, so that’s a good lead in. Our main topic today will be talking around agents and agentic experiences. What does that look like for people and users who are in organizations who are maybe clamping down a bit on that or don’t have full access to things or

1:32 don’t have full access to things or organizations that are slowing to adopt AI. AI. So, that being said, that’s going to be our main topic. Tommy, do you have any news items we should discuss before we get going? I do. So, we have one announcement and then two great news articles from the Fabric team. First one is, hey, we got the Chicago user group coming up July 16th in person and we’re really excited because it’s been a while since we’ve been doing it. I’ve know I said it before that we want us to actually start

2:02 before that we want us to actually start doing it on more better cadence. And guess what? We are. July 16th, we have August 24th, and then I believe July or September 20th. We’re going back to monthly, but we’re kicking off with none other than not yours truly, but yours truly, Mike Carlo Mike DeCarlo will be announcing and and actually speaking on Raytheon at our user group. So again, this is July 16th. We’re doing p. m. at the Microsoft building. So this is exciting. p. m. seems to work better for people. more

2:34 seems to work better for people. more people actually asked for that. It’s going to be on the Microsoft Microsoft Technology Center downtown Chicago. Love to see you. Mike, tell us what you’re going to be doing. What are you going to What are you going to provide for for the peoples? So we’re going to go through again around the agentic experiences things as well. We’re going to go directly through everything that’s we’re talking about today on the podcast. We’re going to talk about agentic experiences. we’re going to try and talk through how do we build with agents? What does that look like for us, Tommy and I?

3:05 that look like for us, Tommy and I? We’re going to actually going to do some Raytheon demos. So we’re going to get get into the Raytheon project. So if you heard about it at the Microsoft conference and you want to learn more about it, definitely check out this session. You’re going to want to be able to show up, listen, watch, see a real demo, watch me click the keyboard, build out a real Raytheon project. I’ve been doing a YouTube series around this one directly where we’ve been building from templates and making changes and then talking with our agents. So I’ll just give you what I’ve been learning and techniques and tips or tricks that I’ve been using in that space as well. So So If you’ve never Yeah, and Mike, I think

3:36 If you’ve never Yeah, and Mike, I think if you’ve never been to a user group before in person, who’s like, “Well, why don’t I just watch your YouTube videos?” Like that would I don’t have to actually travel and do things. Yes. Yes. why user groups are still for me something are near and dear because one, the networking. Two, you you get to ask questions. You got the person who built it. You can ask any questions you want. And also you have other people who are working on the same thing. I can’t wait to see this and peg you with questions, but you get a lot of people that are very like-minded there. So bring some friends. There will

4:06 there. So bring some friends. There will be food. be food. If you are first time, maybe we’ll have some swag, but tell it to your friends, share it online. The link is in YouTube and it’s also in the podcast as well. I think that is very fair as well. All right. That being said, let’s Let’s go to some of the other topics you have, Tommy. What else you have going on? on? Next thing we got is Fabric Data Agent, the API is now public. So you can actually build Fabric

4:38 public. So you can actually build Fabric Data Agents into your tools and pipelines. What this simply allows you to do is create, configure, update, and publish data agents programmatically via external tools from pipelines or any back-end services. The SDK runs outside of Fabric authenticated through a public API. So Mike, this is great because the previous only way that you could actually build a data agent was in the Fabric user interface. It was in the portal portal or via the SDK only within a Fabric notebook. So it was strange, but now you can create those artifacts, you

5:09 now you can create those artifacts, you can configure the data sources. You can publish all of this from code. This is great. What are your initial thoughts here because I have a I have a take here that maybe not as positive here because there’s something here missing. But what’s your take? , I I think the data agents are interesting. oh, speaking of which, Tommy, I need to correct ourselves a little bit. yesterday on Wednesday, I talked about how we really need the semantic model to have a place where we put instructions

5:39 have a place where we put instructions or details or other things for the agent. agent. Yeah. Yeah. that exists. In In semantic models today? In the semantic models as we know them today. today. So, inside the PBIP format, specifically inside the prep data prep for AI experience, there is a dialogue box that allows us to be able to Yeah. Yeah. enter in instructions. Yes. Yes. Instructions are saved directly inside

6:10 Instructions are saved directly inside the model. I did not know this. Or or maybe I did, I didn’t put two and two together. but all that to say was we ribbed Microsoft and and Tommy, you even zoomed in to to a mirror a little bit. I I I know yes. You were like or a rune, someone. A rune., rune., we need to have we need to have First First this ability to put instructions or other context inside the model in the model itself. And so, what I what I would note there, Tommy, is it’s there. There is something already there and

6:40 There is something already there and available to you. So, data agent With limitations. It’s instruction character limits and it’s not markdown. There’s a character limit and there’s it’s not it’s I don’t even know what file format it’s in. I thought it was just Timedl. Like a text file. still text. Yeah, that the pop-up is not necessarily Timedl. Which would be strange. I’m not sure. Okay. Okay. I haven’t played with it yet. So, there’s there’s something there that at least lets you see that part of it. So,

7:10 least lets you see that part of it. So, I think I think it’s there enough that be like, “Okay, that’s useful to some degree.” How we leverage it in all the agent experiences not sure. Okay. So, So, Yeah. Yeah. that’s one thing about the semantic model. Let’s step back and then go over to what’s happening inside data agent data agents, which is now, “Okay, do I want to point at a lake house? Do I want to point at something else? Do I want to have multiple things I want to point at in order to be able to have an agent talk across a lake house or semantic

7:41 talk across a lake house or semantic model or mix the two together, right? I model or mix the two together, right?, maybe you want a little bit of a mean, maybe you want a little bit of a couple of tables from the lake house and a table or something from the semantic model. Data agents should be able to adjust the queries and return results or data from multiple sources. So, I think data agents is really like when you are trying to absorb many semantic models or multiple tables from a lake house, like there was there’s no mechanism there to serve that data back out to agents in a in a clean way. The semantic model seems to have a little bit of its own story with the

8:11 little bit of its own story with the Fabric MCP modeling server, we have other features and tools. And by the way, in public preview right now, this is what we did we talked about yesterday on the podcast was there is now the ability for you to use Copilot in modeling for the web. So, now Copilot now exists there and it does use the Power BI MCP modeling server modeling MCP server to help that agent talk to your model. So, all the rich tools and skills that

8:42 So, all the rich tools and skills that we get and that Tommy we love working with in desktop, they’re now available to us directly inside that experience as well. So, anyways, I just want to kind well. So, anyways, I just want to point that out as something that’s of point that out as something that’s that’s relevant here and thought that was useful for us to like unpack. unpack. And I think too, the one thing I feel is missing right now with the API is the consuming view. I would love the data agent to be available via the API. Right now, it’s for the developer to create, which is fine. But again, it’s like how Okay, how is a

9:12 But again, it’s like how Okay, how is a consumer how is a business going to actually utilize the data agent once it is published? That I would love to see the API being used. Because you think about that, that interface or that ability is only in like one of one of three places in the Fabric portal itself, connected to a Copilot agent, or I think you can actually connect it in Microsoft Foundry in in some capacity. But you really don’t have a lot of options in terms of how does someone actually use a data agent? That I would love to see an API for.

9:43 for. I don’t know if you’re going to get directly a data agent, but there is a data agent MCP server. So you can, the recommended way for you to talk to a data agent is use the MCP server to talk to the agent. I the MCP server to talk to the agent. that’s what Microsoft’s guidance mean that’s what Microsoft’s guidance was when I asked them about this. I was like, “Where’s the API for this?” They’re like, “We don’t really give you like directly an API call, but you should be able to use a data agent with an MCP server.” So I’ll I’ll see if I can go dig up the documentation on that one as well. Again, Tommy, to your what you’re describing here is all this area

10:14 you’re describing here is all this area feels a bit fuzzy. It’s like it’s starting to get woven together, but the story is not quite there yet. And so,, when do I use which tool? And am I using Copilot in the service? Are there advantages or disadvantages to this? I think this is actually, Tommy, why organizations are struggling to get their users access to stuff because it does feel a bit like mysterious to some degree. degree. And yeah, that does go into the mailbag today about the org’s slow AI adoption,

10:45 today about the org’s slow AI adoption, which we’ll talk about. But I think yeah, to your point, when I publish the report in a semantic model, there’s a clear story on the placement and and distributing that report to consumers. Yes., there’s still multiple channels that you can do that, but there’s a very clear setup. With this, it’s like, well, what is the best way? What is the way that more the most people, the right people can get access and utilize it properly. So that’s We’re still dealing with that, but I am still happy to see the API there. Mike, I got an article here that’s right up your

11:14 an article here that’s right up your alley. This is also from the Fabric Blog. This is Zoe Douglas, Zoe. and this is now Power BI reports in Fabric apps, expanding how you build on your semantic model. And it’s not so much a new feature update because we already know this is available, but it’s really an explanation of semantic models are really the foundation of business intelligence in Microsoft Fabric. Again, the fact now that semantic models support a wide range of scenarios, not just reporting.

11:45 just reporting. Yep. Yep. And Power BI reports, we know are kind And Power BI reports, we know are that dedicated experience for the of that dedicated experience for the reporting. They simplify the query generation. However, what we’re seeing now with the Fabric apps and especially with Rayfin is you can now office semantic model build enable fully functional managed web apps on top of semantic model data in minutes. Fabric becomes the hosting layer. Developer built Fabric apps sit alongside Power BI reports as well. So, there’s two experiences, but one you there’s two experiences, but one, one foundation because both a know, one foundation because both a Power BI report and a Fabric app can

12:16 Power BI report and a Fabric app can build on the same semantic model. Users can drag and drop build on the report or they can actually do what they need to do in Rayfin. And the broader vision here and I think this is the take here is Microsoft aims to provide really a full range options for bringing the data together, turning that semantic model into meaningful signals, and empowering people with all the ways they can look at AI, build reports, view their data Yep. Yep. by using the semantic model as that core layer here. This is what we’ve been talking about on the podcast for many, many months,

12:46 the podcast for many, many months, Tommy, which is like even before Rayfin apps appeared. The model is key. The semantic model is the the essential part of like where the business logic lives. It’s where our tables are sitting. It’s where the interaction is happening between the different elements on the model. So, for that reason, I really like this idea of making the semantic model like the story, the source of truth on things. and so, I I really feel like that is the right approach on how we are handling the semantic

13:18 on how we are handling the semantic model elements. And so, I think I think this is just an extension of that story, which is hey, really useful., , this is This is This is how we’re going to handle handle or the story of everything on the data engineering side is generating, producing, creating all the data on the front side, and then the outside, the backside of it is now expanding and getting the,, the the models,

13:48 getting the,, the the models, the reports, the paginated things, all the things that we really want it to be doing, it’s happening on the other side. And what’s also great about this, too, is for people who’ve been in business intelligence a long time, they have the experience in their own foundation here, too. When you started in Power BI, Mike, or even years into it, you weren’t thinking, “Well, this is going to be great when AI really takes off.” You weren’t think No one was thinking that in 2019. Like, “I’m glad I’m doing semantic modeling now because that’s the dedicated way that AI is going to work.” We got that very But now we know, and not just as Microsoft

14:20 we know, and not just as Microsoft really,, stamped that, but Claude gave their seal of approval on semantic models. They necessarily They called it a semantic layer, Yeah. Yeah. but it’s a semantic model. And the fact that that holds true for us, the developers building on in a sense, keep doing what you’re doing with some verifications, not overhaul the way you think about data, but more importantly, that not only from a reporting point of view, but from an AI point of view, that job and our task, the purpose we exist does not change

14:51 exist does not change terribly. terribly. I I Yes. I would agree with that one, Tommy. And And I think this other, the other big players in this space are also realizing the same thing as well. Yeah. Yeah. Right? Databricks, All the time., the Snowflake area, they’re also identifying like, yeah, this is This is where the the information should live, right? The semantic model is key to making sure the business runs well and smoothly across these these different areas. So, I really do feel like that’s that is the right approach for things.

15:21 for things. and I I I’m firmly believing like that’s the semantic model is still key. It still needs to happen. Running it, being able to get data out of it is very useful, but then also having the metadata that supports everything that the agents and or computer things need, Ray Fin or, Power BI reports is incredibly valuable here. So, that’s why I think this is such a game changer for organizations to make sure they understand and learn how to do it. Hence, why we’re doing the next user group on this. Exactly. All right, Mike. We have a great Do you have any other news beats from the streets?

15:51 from the streets?, I don’t think so. We have the meet up we talked about, Ray Fin blog post and then we talked about data agents as well. I don’t really have any beats from the streets or anything else that’s extremely relevant. So, with that, maybe we should move on to the next thing. Let’s get to the mailbag. Yeah, so this is we don’t have a name here, but I do appreciate this is a great question and it’s really just a bunch of questions actually that someone wrote in., I think I don’t know if we’re going to be able to answer all of them because there’s I think seven questions here, but the main thing here is here you go.

16:23 Why are organizations moving slowly on AI adoption even with Fabric available? What is the bigger blocker right now? Skills, data quality, governance, and unclear ROI? And what does AI even look like in Fabric and in Power BI environment? What low risk high value AI use cases should teams start with? What are the governance decisions that need to be made before scaling AI across self-service business intelligence? And what does a realistic 30-60-90 day plan look like from experimentation to

16:54 plan look like from experimentation to production? And who should own AI enablement in a BI team? IT, data, COE, business, and why? So, This is a big question. So, there’s a lot here and part of me almost wants to break this down into other episodes, but I think there’s some very some very high-level takes here that we need to get through here and let’s start with and I so I really do appreciate the mailbag here. But I think the big the crux of this is this.

17:24 this. We are seeing organizations as much as they want AI and they are spending money on it, adoption’s different, right? You can say we’re investing in AI and we built tools. Yes. Yes. But just like reports and just like Power BI does not mean you’re going to see adoption. It’s not a one-to-one correlation. With just because we have 200 reports, we’re going to see ex- exponential amount of use in those reports. I think we’re seeing the same thing in AI right now. So, let’s just start with the high-level question here.

17:54 high-level question here. Mike, is does this sound true or ring true in your own current experience with clients and companies when it comes to organizations are moving slowly with AI adoption? Does that seem to be what the case is for you? I think it’s really more of Yeah, yes and no, Tommy. So, some parts of the organization I think are moving slow. I think a lot of people are clamoring to get their hands on it. I also think Microsoft is not giving us the tools that we really need to put this AI stuff in place for us to really build like I saw a blog post from Kim Manis. I

18:25 I saw a blog post from Kim Manis. I think maybe we talked about this earlier, but Kim Manis was talking about how she’s really reworking her workflow, right? She’s still doing the same tasks. Kim was talking about, “Hey, I need to summarize some emails. I need to get some customer goals. There’s some maybe some customer feedback coming in.” There’s lots of sources of information that she’s trying to digest all at the same time to get something in her in her lap so that she can action on and do something with that. And so, I really like this idea of rethinking what your daily workflow looks like.

18:55 workflow looks like. What what does getting work done look like to you? And then the thought here is where do you apply the agents? And where does the tooling get applied? So, So, that’s that’s a rethinking of how I do work. What we work on will greatly shift. It will be different, but we’re still going to be doing work on stuff, but now with agents. I think the new skill to have in the in the 2015-16 2015-16 era, the skill was learn how to Power BI. That was the skill, and I’ve, you

19:26 BI. That was the skill, and I’ve, you BI. That was the skill, and I’ve,, substantially have built a career know, substantially have built a career around this. I have a whole company that helps out with that. I’m now more convinced than ever the new net new skill is not just building Power BI, but it’s building Power BI and Fabric cuz that’s data related things, but doing that also with agents. Anyone moving forward is going to have to understand how to wield agents, how to get them to do things, how to have the agent manipulate files and build stuff and apps and solutions, and all of that encompassing, that’s the new skill set. So, I think I think

19:57 new skill set. So, I think I think that’s where we’re going to go. That’s going to maybe speak more to these 30-60-90 day time periods. Yeah. Yeah. But, I feel like right now the biggest hesitation is from leadership or the IT organization. We don’t know what it’s doing. We’re unclear about how much it’s going to cost us. It’s we’re we’re slow to adopt those things because we’re unable to like really understand one, the value that it brings. And And there’s literally almost no training on this. this. From what I’m aware of.

20:27 From what I’m aware of. yeah. yeah. Well, Tommy, you and I have been studying this for months now, and we’re starting to formulate patterns of like how we can be productive with it. But, there was a period of time we were just doing things. Experimenting. Really being uber productive, and just playing with the tool and figuring out what it could do. what? Many hours that somehow I wish I could take back, but yes, 100% there. And Those were Those were educational steps to get you to where you’re at now. Most of them were. Some of them were like, “Oh, cool. That’s a I could create

20:58 like, “Oh, cool. That’s a I could create a pamphlet.” Don’t need it, but that’s cool. Yeah. There was definitely a lot of rabbit of rabbit a lot of trails that most of them were good. Yes, but yeah, there were some who were like, “Well, that’s cool. Local models. I don’t need that.” Anyways, but you’re speaking of something here that I’ve mentioned before on this podcast, and I will talk to on the top of the mountain continue to yell this. You mentioned the article by Kim, and this is still I don’t I’m not saying you’re missing the point or Kim’s missing the point, but let’s be clear here.

21:28 here. You’re talking about an individual, and even us learning it as an individual how to utilize and apply AI from a micro individual point of view. You and I are what you’d call ourselves AI ready. I can deploy, manage, and utilize AI around my workflow as an individual. However, however, the point being is we’re we still don’t have training on this. We still don’t have what this idea of what AI ready

22:00 have what this idea of what AI ready looks like for an organization, because what does that look like for a team of people, or a group of people, or an organization? And I think this is still the critical gap that we see, because right now, Mike, you and I as an individual can apply AI emphatically, practically, effectively, that’s great. But, how does that actually look like? And I think you and I can also measure success around it for myself or yourself. But, when you think of a larger organization,

22:31 of a larger organization, that’s not the case. It’s not one-to-one in terms of those same skills and also the same agents I’m going to create. So, I think just starting there, AI adoption is a group a culture thing of a group of people coming together to utilize the same tooling and utilize the same process. process. We don’t really have what We don’t really even know what that looks like right now from an agentic point of view. And I think for me, that’s the biggest point to start here.

23:05 I I agree at some level, Tommy. Where would you disagree? I’m not sure if I disagree. I I just I’m trying to draw like a parallel in my mind against something. I believe the parallel in my mind right now is now is if I look at what we’re doing with Power BI reports. So, if we look at what we’re doing at Power BI report level, and we’re looking at what we do for people who create those Power BI reports and people who consume them, right? There’s a level of audience, a larger audience of people who are just like, “Look, just give me my data. Let me go into my report, click the buttons. I’ll get the things that I

23:36 the buttons. I’ll get the things that I need out of the report.” There’s another handful of people that are focusing on the opposite of that, right? I want to create the report. I want to build the analysis. I want to put the analysis pieces together. So, I think there is another area of this that it is is supporting that I’m a consumer versus I’m a builder. I I wrote it down. Again, I look I

24:06 I I wrote it down. Again, I look I literally have that the same thing I wrote. Author, consumer. So, you’re a speak preach, brother. There there’s there’s two parts to this, right? Are you You need to identify are you an AI author, meaning you’re using AI to help you author things, or are you an AI consumer, meaning I build you something with something maybe that maybe that uses AI, or I build you a tool that was created by AI. And I feel like the value add right now is leveraging tools that are created by AI. And then very

24:36 And then very carefully sprinkling in AI to like generate this sentence, or review this, or put some reasoning of the AI into your general process. So, let me go back to Kim’s statement cuz I think that’s actually maybe a bit more relevant here. Inside Kim’s statement, she was talking about about she is she is like she already has this workflow. So I would consider what Kim is doing is she’s becoming a creator of agents, a creator agent. I’m building a process, I

25:07 creator agent. I’m building a process, I know what I want to get done. I’m going to use this agent to tooling to help me build something that like is useful to me and I can leverage that to help me create a workflow, a system, a process that is relevant. So I I think if I look at the definitions we have here, I think Kim is of that creator agent level. Once she figures out this workflow, the next challenge is how do you bundle that up and package it for the consumer

25:38 that up and package it for the consumer side? side? Yeah. Yeah. And I think Tommy, you and I are are much more in the space of we are in the creator agent arena more so than anything else. Right. Right. And so when I look at that perspective, that’s where we understand it and that’s where I think most of our value comes from. from. We’re we’re going to need to learn how to package those things up into manageable publishable bundles. And that’s something that’s

26:08 bundles. And that’s something that’s what Microsoft 365 is trying to do with their agents, that’s what Microsoft Foundry is trying to do. There’s a lot of tools and these are links in the description also as well below here where Foundry, you go into Foundry, you build some you you add the large language model, you build a harness around it. It’s basically like a custom harness. Right. Right. I really think that organizations are going to spend more and more time or more and more resources around building themselves a custom harness around a particular process. And so AI allows you to hyper-customize

26:41 AI allows you to hyper-customize every little thing, experiences, how things interact, all the things. Everything becomes very localized to what you’re doing. And I I like you said, I love that you wrote that down because I’m diagramming here the difference between First, let’s start with the idea of the creator author or the author consumer point of view here. That was very clear roles roles in the Power BI world. Even if you split manage self-service, there’s a clear story, there’s a clear path, there’s

27:11 story, there’s a clear path, there’s clear what I can do, what I’m supposed to do to do responsibilities there. Where things are going wrong right now or ambiguous with the AI world is I think you split that author into two. Are you creating applications reports using AI, which still is something that org is part of adoption? Or are you building agents for people to use? use? And I think you if you asked yourself, if you basically had apps and agents,

27:41 if you basically had apps and agents, both with creator with Okay. Okay. I want to challenge the agent definition that you’re using there. I don’t think I would determine that as the same way. No, and I’m saying they are different an app and an agent. If you’re building apps or agents, those are two different creations. Those are two not necessarily two two different authors, but I think there is going to be in the organization someone who’s responsible or a team that’s responsible for creating the agents, creating the things that people are going to use via AI. Because okay,

28:11 are going to use via AI. Because okay, let’s I’ll back up a Hold on, hold on. Let me Let me Let me step back here for a second. When we say agent, I’m thinking things that are like an agent has memory. memory. An agent has business context. An agent has the ability to make files, create files. That we give agent access to your email, your OneDrive, your SharePoint, right? So, when you say the term agent, I think that is a much like there’s like layers of snowball effect here, right? There’s the large language model,

28:41 There’s the large language model, then there’s the harness, then there’s the agent. [snorts] [snorts] I think these are different things that we’re talking about. And so, what I’m talking about is I’m saying I think a lot of more of our work is going to be around the harness level as opposed to all these other pieces. So, So, I I have a different view on this, I think, because I’m thinking of the final output. The output I’m talking about. That’s why I’m making the distinction here. Because I think I’m I’m I think I’m I think I’m talking about one layer

29:11 I think I’m talking about one layer lower than what you’re talking about. And when you say model creator and agent, agent, Mhm. Mhm. we’ve already moved past what I think what I’m discussing, which is the value comes from these hard these custom harnesses. Right. And That’s you Yeah. Yeah. That’s where That’s where we’re like to me, I think we’re maybe speaking a little different here. I would agree with you, though, Tommy, if I step back up and say, “What’s bigger than the harness?” Well, what’s bigger than the harness? That’s the agent. That’s the memory of the AI. It’s It’s a It’s a you’re giving the

29:42 you’re giving the harness memory, recall, the ability to query things, get other stuff out. So, there’s this other idea of this. And to your point, Tommy, agents, I think, need to be very specific to a particular role. role. Yeah. And I You’re the email review agent. You’re the image generation agent. You’re the funny sticker podcast maker agent, right? You You give You find a specific thing. And where I think people are successful is you identify a very specific task, Mhm. Mhm. and you build an agent for that.

30:13 and you build an agent for that. No, 100%. And I think to pick up what you’re putting down, in the same fashion you would build a model, a semantic model, you may have multiple people working on the report and model. You may have the developer or the data engineer, you may have the report designer, you may have the semantic model builder. Those are individual layers of the final output that I think are going to be Again, part of really things that need to be on a resume. And I think each of those layers you’re talking about are individual things that should be, in a sense, on your resume in terms of what can I do?

30:43 can I do? So, but for me, when I think about an org, to go back to the example of author consumer, let me ask you this. Is the consumer who’s using an app going to be necessarily the same consumer who’s using an agent? And the reason I ask this, and I’ll give you a little more clarity here, when I build an app or fabric app or,, whatever the output is, that is just normal accessible for consumer,

31:15 is just normal accessible for consumer, that that the user is not using AI in that point. They’re just using a user interface. They’re using fabric,, the fabric It was all built by AI. It doesn’t necessarily mean they’re using AI. So, that’s a normal I already know how to use this piece. I know how to use a computer. Sure. Sure. The consumer for an agent, though, I think needs to have some education. This should as going to be a skill just like I can know how to use Outlook and things to my calendar or distribution list. They’re not necessarily creating things, just like I’m not creating anything in Outlook.

31:45 anything in Outlook. Mhm. Mhm. But there is a level of skill that a consumer is going to need to fully utilize the agent. Because it’s, again, prompting, where to find it, how to best utilize it. They may not be creating the agent, but that I think is the other part here. I am assuming if I’m building agent, anything agentic, for my organization or for my department, Yeah. Yeah. I must assume that they have some level of skill of skill consuming and using that, rather than

32:17 consuming and using that, rather than just fully watching and being more passive. They’re going to be a more active player in whatever I build. Would you agree with that? Yeah, I I do agree with that., I think that’s in your shaping what the agent of that of that,, how do you what things are you what inputs are you giving So, let me unpack what I hear you’re saying. The inputs you’re giving to the harness, which the harness is the thing that wraps around the agent, right? The skills, the memory, the markdown files, how efficiently it runs., you could have even multiple large language models that are part of

32:47 large language models that are part of that agent, right? A mixed model agent. All of that stuff is just things that you inject into whatever the agent’s doing. doing. And then, I guess what I’m trying to say here is,, going back to what Kim’s doing, she built the agent, she figured out what the process was, made the agent do this thing, and so now she has a specific agent that does this task., and I I think you want to make sure that agents are like hyper specific to a single piece or task item. You may have an orchestrator agent that’s at the higher level that’s using multiple agents in sequence, but you want to

33:17 agents in sequence, but you want to think about, okay, this is the agent to read and summarize emails. This is the agent to read and summarize customer feedback. And then I want to materialize all that information into a single thought or idea. You touched on something here. Yeah. There’s so much more value in like shared business knowledge in a common place. place. And there’s really no good place to put that stuff yet. 100%. I want to touch I want to highlight something that you said because I think this is very critical from an adoption point of view.

33:45 from an adoption point of view. Mhm. Mhm. Regardless of the solution that you provide, provide, the you the you consumers, and I think there’s got to be a better name here, but the users of anything agentic need to know to know what that agent can do, right? And when to use it. Again, for example, I have a few I have built a ton of custom agents that I utilize heavily in the Harness Notion, right? Yeah. Yeah. And there are certain ones that organize,, all the different files and notes created and,, shows status updates. There’s another

34:16 shows status updates. There’s another one that is has a bunch of fabric skills. Now, how do you interact with them? You either go to the chat or you add a comment, but if I didn’t create them, Mike, right? And someone said, “Hey, I Let’s say I scale someone on my team.” I say, “We’re using Notion and use it.” They’re not going to know if I change the status of a meeting that the agent’s going to that triggers the agent. They’re not going to know necessarily without training that, hey, when we are done with X, Y, and Z,,, mention the our operations

34:47 ,, mention the our operations agent to put everything together. Those are things I know because I created them. And again, it works for the individual and I think that’s a big part of that what you’re saying is saying is Yes. Yes. When is it appropriate to use what agent? So, I think that’s also part that’s going to be part of training. But, when we think about this in fact,, do you have any things here before I move on because I want to see No, no. I think we’re good. touch on anything. Yeah, okay. So, I have a few questions here because I think there’s a few blockers when we think about

35:18 when we think about our adoption of we even though we have this great platform for building agents with fabric and semantic models, I think we have a few questions we need to ask ourselves and some of them are There’s a difference I’m finding when I’m we’re talking to organizations around something called AI strategy and AI theater. How do we actually have strategic initiatives for incorporating AI, incorporating agents into the workflow compared to just headlines that people see, right? Like, oh, well, you

35:48 people see, right? Like, oh, well, you saw what Microsoft did. We’re going to just do that. Yeah. Yeah. And I think there’s a few things that we need to separate here because there’s a ton in between, but let me just start with that when I’m seeing a lot of theater and I’m not seeing a lot of strategy. So, I think we’ll start with just like we have the fabric roadmap. Mike, how would you define having an organization that had a healthy AI strategy or a,, a very solid and productive strategic initiatives for

36:18 and productive strategic initiatives for AI? What does this look like to be healthy? Well, give me your thoughts first, Tommy. What does it look like to be healthy for you? You just asked me a lot of questions. I feel like I’m getting here all the time. You give me your thoughts first and then we’ll I’ll present it afterwards. You got it, my friend. So, well, let’s start with this in a fabric environment, right? I think a lot of companies are looking at and we’ve said this before, they have a they want AI solution and they’re looking for,, a problem. problem. And I think a lot of times, this goes back to really for me data governance data quality. Where what are we actually

36:50 data quality. Where what are we actually trying to achieve with our data? Because again, if you’re going to make an agent work in your company, it’s going to run on your data. So, where are the biggest bottlenecks or pain points? What teams could best utilize, utilize,,, more assistance on what they do? Rather than just creating image generation or pamphlets, right? I think of operations teams that have to look at time and time again all the things that are being filled out or help me identify my accounts. Where are the biggest gaps people have in their current data? Like And again, I would go back to looking at

37:22 And again, I would go back to looking at what are we doing with reporting? Where do people still have report request? If you actually think about the biggest pain points people have with reporting, you could easily make the bridge to how that can be something agentic. And And this starts at the top to me with data governance in terms of what are our priorities around data. What are our priorities around analysis to get people what they need to do their decisions. That begins to actually, I think, highlight highlight where could we then apply agentic

37:53 where could we then apply agentic things? And it needs someone like you and I or someone with the experience both on semantic modeling, and data culture, but also on what AI can actually do. So, AI strategy is much more around, “Hey, there’s a new tool, there’s a new model. Let’s,, let’s just incorporate it. Let’s build a co-pilot agent.” Where are our biggest pain points? What are the goals of the company? And how can we incorporate and begin to pilot and incorporate that in small doses to people? Yeah, but who does that? Who’s the the discovering? I think you have to designate someone in the company. So, I

38:23 designate someone in the company. So, I like your point there, Tommy. I think this is very relevant. I also think that there’s a there’s a the idea or concept around like you have to designate someone who’s in charge of that. Give some responsibility to someone on the team. Who’s doing this already? Is someone already willing to explore it? And then, I think security of your company has to step in and say, “What are we allowing people to use?” use?” Like someone has to make a decision at at a higher level, either it’s a you at at a higher level, either it’s a, C or VP level person that’s know, C or VP level person that’s saying,,

38:54 saying,, “We need to we need to we need to invest in this. Agents are coming. We need to give a proof of concept around a person or a team or a specific group.” group.” Same what we do with Power BI. Same what we do with Power BI, right? it needs to be led at at the organizational level. Leadership has to buy into what’s going on here. So, I feel like we want to have leadership step in and say, “This is valuable. valuable. Here’s a team. Here’s someone that Tommy or Mike is trying to figure out how to

39:24 or Mike is trying to figure out how to use this. Okay, let’s give you access to Copilot Studio. What can we build there? Let’s give you access to Copilot inside Fabric. What does the spend and usage of that look like? Is it is that lemon becoming worth the squeeze for us in that space? What does building with Foundry look like? Right. Right. So, I I I think so, how did I get to where I’m at today? I had to spend a lot of time on the side, outside of work hours, to figure

39:55 side, outside of work hours, to figure out what this stuff is doing, to make sure I could understand and build, and I’m doing a lot of education. And the and the education space right now, it’s like for me, it’s 100% on YouTube. What did someone else build? Right. I’m even trying to share the knowledge that I’m building, right? All all the Rayfin things I’ve been playing with. I’m not seeing a bunch of people build a bunch of public demo demos or demonstrations of Rayfin, coding with things, and building those things publicly out in out in space. Like, they’re just not happening. So, I’m try I’m getting as much as my knowledge from other

40:25 other YouTubers or other educators that in the space that is helping inform me of what we can build. Now that I’ve got some framework around this, I’ve been able to push my team, my internal developers, to all use agents as well. And so, we’re seeing massive gains on the amount of output of work that we can do. We’re taking on more ambitious projects. We’re not scared about doing things that in a language we don’t understand. We have an agent that’s there and we can ask questions of it. But, these are all skills skills that you need to develop. And I don’t

40:55 that you need to develop. And I don’t know if this is a class you can go through. through. I think it’s literally time in seat, and there are some techniques you could learn. learn. Yeah. Yeah. But, this is the same thing like I didn’t have to go to a class to learn how to Google something. Well, Well, Right? Right? You just started using it and found it was somewhat effective or valuable in a certain area. Right? Right? Yes. Yes. Once you did that a couple times, you’re like, “Okay, I’m going to keep doing

41:26 like, “Okay, I’m going to keep doing it.” And now,, let’s fast forward here. My family is now using chat GPT for all kinds of things, and they have coined it as chat. Well, chat said, “Chat said this. Chat said that.” Right? It It’s becoming It’s It’s getting to this point where we’re allowing chat to give us answers for things, or or we can rely on it cuz it’s giving us reasonable answers that sound good and and useful. So, let’s pause there. I think there’s a few holes in the argument here, and and where I’m I hear

41:59 argument here, and and where I’m I hear what you’re saying about from the Google one especially. However, one, one, Google might You’re Excuse me. Companies are not paying for you to Google, right? They’re not paying investment into Google. Google’s available for anyone, right? right? But they’re paying they’re paying for me to like know where to go find information. Yeah, oh sure, 100% but they’re also not saying, “Hey, we invested in Google so we would have really like you to search on Google.” What if you did Bing? What if you did DuckDuckGo? Doesn’t matter. So, So, That’s that’s a skill that’s a skill

42:29 That’s that’s a skill that’s a skill that you had to learn though. I What I’m saying though, Tommy, is Do any company is is in the same way. AI is the same way. It’s a skill you have to learn. Okay. Okay. And this having that skill in your Mhm. Mhm. back pocket is going to be the same requirement for anyone else entering the job market moving forward. I would Here’s where I would find a better parallel at least in from my point of view. I’m not saying you’re wrong but from my point of view

43:00 wrong but from my point of view when I first started my I remember my first company and I, all the college all was done and then I had to use Outlook. And I realized that there was no Outlook course in college and I realized there was a certain skills in Outlook that I needed to get better at because one that is the way the communi- the the company communicated at the time. There was you communicated at the time. There was, Slack was I think relatively new. know, Slack was I think relatively new. There was no Teams. There was what? Skype chat, which was great. sorry, I sha- Shannon but yeah,

43:32 sorry, I sha- Shannon but yeah, Skype was not great. And but really it was Outlook and then there was things like, “Okay, I need to organize my messages. How do you send out an email properly, right?” Those were not things you necessarily learned. And then you realize I took an Excel course in college but I needed if I wanted to do my job better, well one this is what the company used. I was getting sent Excel files. So, that was Now you’re talking about I have to speak the language the company’s using. Now take the example. Let’s say the company was using Google Workspace. The it’s similar, right? Like Gmail and

44:04 The it’s similar, right? Like Gmail and Google Sheets but there are differences. I had to learn very quickly the company language on how they communicated both with their data, which was in Excel at the time, and with messaging, which was Outlook at the time. Mhm. Mhm. Once you invested in Power BI, the distribution, the creation of the person talking and the person listening was also a language that the company had to learn. learn. The Google example, and I think right now, even with ChatGPT, there is no one person talking, one

44:35 there is no one person talking, one person listening that we are going through this flow, and because that’s what you’re talking about adoption. I will not be forced to use ChatGPT even if you using it, you’ll be better at my job, right? There is no correlation to what you’re doing in AI if we even work right next to each other compared to what I’m doing. And I think part of this is, honestly, when it comes to again, going back to

44:59 when it comes to again, going back to the strategy, but very much too when we’re talking about when my when we first started adopting Power BI, I had the buy-in from the sales team, and they told everyone, every sales rep, from now on, if you want to know your quota, you go here. This is the report. Do not look at the Excel files that you created. This is where you go to find your quota, your target to goals, everything’s in here. If you have a problem with it, message Tommy.

45:29 Tommy. Yeah. Yeah. And that was the beginning of adoption. So, right now, we do not have that, I think, part of governance into what we’re trying to do with anything AI. AI. Well, hold on. Yeah. Yeah. That’s already been a decision that was made. So, there’s a couple decisions that got you to that point. Yeah. Yeah. You’re skipping ahead too fast here. Okay. Okay. Because at some point, someone said, “Power BI is valuable. We’re going to We’re going to invest in it.” Ah, yeah, yeah. At some point at some point in time, someone So, like, I like I see what you’re saying. You’re addressing a problem that already

46:00 You’re addressing a problem that already had some buy-in at the leadership level to get to the agent space to get to the Power BI space, right? So, again, I’ll go back to it’s the same stuff we’re talking about right now. It’s It’s exactly the same stuff. It’s It’s going to be , it is I think I know where you’re going with this. this. Yeah, you see what I’m You see what I’m trying to like Well, you mentioned it already, I think.

46:30 Well, you mentioned it already, I think. Who owns the enablement? Yes. Yes. And it’s also like,, not only who owns the enablement, but like, you who owns the enablement, but like,, there’s there’s decisions that are know, there’s there’s decisions that are being made around what’s acceptable tooling to use, what’s what’s not acceptable tooling to be used. That’s already being handled at the organization level. Someone’s making those calls for you, those decisions for you. And so We bought the licensing, in a sense. Correct. We bought the licensing, we’re going to do it. That’s That’s what we’re doing. And so, I I’m believing here that

47:03 I I’m believing here that this this How do I frame this out? Decisions have already been made that this is the direction we’re going with the company. Security has already been reviewed. We’re already happy with it. The conversation when we were talking about Power BI in the very beginning days was, I can’t put my reports in the in the cloud. I don’t want a cloud having my reports. That’s too risky. And And then the argument became, well, you already have all your documents in SharePoint. How much work more risky is it to put it in Power BI? And then some people Everyone else said Yeah. And then

47:34 people Everyone else said Yeah. And then everyone else said, well, that does make sense. It’s a better value. We get more out of it. Let’s move away from SQL Server, and let’s directly only move into Power BI. So, to get to where you’re at, Tommy, where you said, team, you need to use this stuff. This is going to be a tool. Use it or else. Learn it or else. We’re not at that stage yet. That’s too That’s too far down the road to where we’re at. We’re still in the IT doesn’t ex- doesn’t understand what an agent is. The company hasn’t bought into agents add value yet. The company

48:05 into agents add value yet. The company hasn’t bought into it’s secure and these are these are the policies or these are the the applications or programs we’re going to use cuz we deem them as secure and safe to our company. We’re not there yet. Yeah. Yeah. And so I think we’re jumping the gun when we when you describe like this world of like you made a report and now everyone has to use it cuz that’s what that’s the source of truth. We’re trusting this report. And we’re going to have someone who’s who’s should have already been built. Yep. We’re going to certify it. We’re going to put a person on it. Tommy is you’re the guy to make sure this report runs well and

48:35 to make sure this report runs well and smoothly. smoothly. So So So let Yeah. No, I I I I think that’s a great point because you’re right. What my my example already assumed there’s a report created, it was developed, validated, and then shared. And I think this is a good point because also also there’s a clear purpose of the report and there everyone knew when you were going to buy Power BI what it was supposed to do. That’s the ambiguous part right now. So let me propose something to you. And especially as we think about one of the questions here from the mailbag was around what does a 30-60-90 day plan

49:08 was around what does a 30-60-90 day plan look like from experimentation to production and what governance decisions need to be made before scaling AI. I think this is a great place to start and this goes really back to okay, who owns it? So I would highly recommend if you’re going to want to enable AI in an organization, I think it starts with having healthy data governance. To me, I don’t think you’re going to get very far without this being owned by data governance because you already have the right people in the room to me. You have IT, you have business intelligence, you assuming this the council or,

49:39 assuming this the council or, the committee is of leadership who make decisions and have purse strings. So that’s where AI enablement then starts. And then I think it’s defining what the purpose not necessarily of it like each tool we’re going to do, but how do we want to integrate it? Is it a chat? But let me back up here because I’m going to propose this to you. Data governance and that committee should promote and think about how can AI become in a citizen of a team or

50:10 AI become in a citizen of a team or department, a assistant in a different sense of rather than AI assistant. Adoption only works when it becomes a critical function of how people interact as a team. Again, consider the report. Everyone knows where to find the report. If we’re going to validate our data, we know it’s through these sources of truth. We cannot just put a tool out there and say we’re going to build an operations tool for people to handle PDFs without saying

50:41 saying there’s a like in it again taking a step back, we want operations to rely on getting rid from mundane task and to utilize utilize,, agents in their workflow. And then identify those different pain points what an agent can do., like where are the anything again, to your point there’s a lot of skill and experience on what AI can do, those agents are capable of. But it starts with data governance saying, well, well,, again, someone coming in to say what is it capable of?

51:12 to say what is it capable of? Where are the biggest pain points they are having in their process? And how do we treat AI for departments as a first-class citizen? How do we treat it as if we were to hire someone being a critical part of the juncture. And I think that’s a for me, I look at this and I say, you cannot have AI work unless it’s integrated into a team’s normal workflow. I I don’t know how to react to your comments or tell me like I I agree with the things you’re saying, but I think

51:42 the things you’re saying, but I think they’re just so high-level. It’s not like tap it to me it’s not like tangible. What does this look like? I don’t I don’t think this is Okay, you’re saying words and I don’t mean to be I don’t understand the tangible aspects of this. Like so going back going back to the question and like what specifically they’re they’re asking for around, you they’re they’re asking for around,, some of these the comments here. know, some of these the comments here. Right? So if I go back directly to the question and say, okay, let’s read through the the the mailbag here again. Right? Why are Why are some of orgs slow to adopt? Right? What are the biggest blockers

52:13 Right? What are the biggest blockers right now? Biggest blockers right now, I think, are IT skills. Everything you’re talking about is after we’ve gotten through those blockers, I think. So I think there’s some really fundamental challenges organizations need to work through and you need to walk walk through those barriers and I think this is a lot of proof of concept it lets people add value in specific areas. Not everyone on the team has to adopt it all at once. That’s fine and I’m okay with it. This is going to be a longer progression in the same way not everyone adopted Power BI to begin with. Not you

52:44 adopted Power BI to begin with. Not you don’t you wouldn’t give everyone semantic modeling to begin with. You’re going to start in very small pockets. You’re going to start with POCs. You’re going to start with a single person who learns and figures out how to build workflows around agent things. Again, taking the Kims of the world, examples of that, and I think who is this person? It could be someone who’s an analyst. I also think this person could be a director, a VP, a C-level. I think anyone can build with the agents. I think this is one of the advantages of like AI and agents that are coming out

53:15 like AI and agents that are coming out now, which is anyone at any level of the organization is going to find value from these agents. I also think there’s this squelching effect or squishing effect that’s happening to middle management. Gone are the days of really big teams, I think. think. I think large teams with lots of people in them in them is going to the size of teams are going to reduce. The amount of middle management have in organizations because a middle management layer was a lot about

53:46 management layer was a lot about communication and talking to people and getting the people that could write the code to connect the ideas with the people designing the UI and the other aspects of it. So, I think there’s a a a this the AI and skills and AR are are producing some condensing effect. It’s giving really powerful tools to creators of things with agents. And I think Tommy, we’re going to just sit back and watch. It’s not

54:17 sit back and watch. It’s not ready yet. I I it it’s it’s ready in the fact that there are systems and Microsoft is starting to put things in place that are useful. one thing I wanted to test out here is the application called Scout. Scout is Microsoft’s version of Open Claw or Hermes. I want Scout to go read my email and provide a summary and do things on a task or a schedule or something that’s regularly repeating. I think that’s very useful. I think that’s going to be something that we’re going to want to leverage and use over and over again.

54:47 and use over and over again. Agents are here. They’re here to stay. Where we place them, how we use them, I think we’re going to talk more in patterns patterns than we are going to be building a single agent for everyone to use. Cuz I think I’ll build the framework or the starting point or the template of the agent and I’ll give it to you, Tommy, and then you’ll take that agent and customize it to what you need. And so, I think that’s really the what I’m feel like I’m walking into here is we can really look at everyone is now able to build agents

55:19 able to build agents at scale at scale with lots of models. Then we have the conversation around does employing more of these agents add more value to the business or cost us more money? And so, now the value proposition shows up. For me, my team, agents have made us writing code way more effective, way faster, and we’re pumping out more code than ever because of agents helping us build things and shape things. We still need people reasoning about them. We still need people in there making sure that it’s testing and building the right

55:49 it’s testing and building the right tests. That still has to happen. But, we’re going to offload a lot of busy work, task-level work to agents. The trick is how do you customize the harness? How do you customize the agent for everyone’s use or pattern of workflow. The work doesn’t change. Just how we do it does. I’m more I’m thinking through your workflow here.

56:14 workflow here. And then I’m again I’m trying to corporate this to a larger org. So, you mentioned a few things here and let me make sure I got you right here. Something like Scout, people And again, I’m not going to put words in your mouth, but to take a little liberty here. We want people to be able to create their own agents, in a sense not a wild west, but really have this freedom for each individual to contribute, to customize it for their needs. needs. Sure. Yep. Now, we have a lot of agents for a lot of individuals, not really a standard

56:46 of individuals, not really a standard or really a standard process, but each person’s doing their own need. There should be some insight to all these agents and what they’re doing. Hopefully, that, again from the top level down. Then there can be some consolidation, right? On Okay, we’re seeing a lot of patterns here where everyone on sales is looking at who of their accounts reached out to them and drafting email or putting their priority list together., based on their targets. Okay, that’s a great example here where we can create a single agent

57:16 create a single agent Yeah, we agreed. that does that does Again, to me though, I I’m going to take some offense to my previous statement here because part of what I said was relying on reliance of an agent in a workflow, which is to me a part of adoption. I do not think those are just fuzz words there. That’s I think what you’re speaking to. Once we find out programmatically how everybody is using agents, if there are patterns to that, if that is the adoption way we’re going

57:47 if that is the adoption way we’re going here, where we give everyone all access to agents or a lot of people, find the patterns people are using, and then create the skills to really optimize them, which may have some token issues, then we actually then say, “Okay, from here on out, hey, guess what sales team? There is your priority agent. This will identify your top agents. It will look at your email. It will look at your data,, our model for our sales quota.” quota.” That has to be relied That has to then be pushed to the company. And I think

58:18 be pushed to the company. And I think that’s that is an avenue here, but I do take some offense to just the reliance side having being too high-level, because I think when you think about adoption, adoption, it is a common language, which again is the way people do things. It is the common process. Yes. Yes. Now, but I I I see that as a good path. The problem with the play devil’s advocate, you’re still going to come up with cost, right? If I have everyone with the liberty of creating their own agents, and then now looking at six, a large

58:49 looking at six, a large company, company,, each person creates two agents or three agents, and now we have to identify all the patterns. That in itself could be a headache. That seems like a great opportunity. And that also is a a bottom-up approach. approach. Yes. Yes. But But maybe that is a good place to start., the other side of the coin that I I I think you’re recommending you mentioned this too is the pilot side of things. Can we identify in a sense our sales team? It’s probably

59:20 in a sense our sales team? It’s probably a great place to start, but they make the most money or, it’s the most usually the direct way of revenue or the way we create products. Can we actually talk with them and do discovery? Like,, again, we want to create a single agent to assist and have that team rely on. That goes, “Oh, did you talk to the quota agent today?” Yeah, it said X, Y, and Z. It should be part of the lingo as well, and I think that’s part of adoption, but I know we’re getting near time, and to really consolidate this, I’ll ask you because I literally

59:51 I’ll ask you because I literally like this idea of 30-60-90 day plan. I don’t know I know we’re not going to be able to answer this here. Yes. Yes. But, let’s just take a a stab at this in terms of you come in, someone ask you today, “Hey, can you put together a 30-60-90 day plan of just us getting started with AI? What do we need to do?” Yes. Yes. I have an idea here. I don’t want to grill you. do you think, Tommy? What’s your What’s your 30-60-90 look like for you? What are you doing in Give me one action for each. In 30 days, what are you doing? In 60 days, what are you doing? And in 90

60:21 60 days, what are you doing? And in 90 days, what are you doing? we’re doing data quality. We’re verifying your data, both soft data and hard data, and we’re going to give you some findings there to say, “What is the most consistent?” We because we can’t do anything without our data quality. That is the number one thing I would want to want to do. Not day 60, after we’re going to do discovery. It’s going to take 2 months probably to do discovery, understand who has the best data, and who has the biggest needs of process that we can incorporate data agents. Like, “Oh, you incorporate data agents. Like, “Oh, what? Actually,, we found know what? Actually,, we found X, Y, and Z on the operations team. This is a great example. They already have

60:51 is a great example. They already have good data quality. So, now we can say this is a great candidate because we want to identify candidates.” After 3 months, we should be able to have an agent that people are using. I know that’s outline, people don’t want to see 3 months for a data agent, but to me, if you don’t have the right data, data, and if it’s all,, scattered, everyone’s is different SharePoint files in their own OneDrive, and their semantic models are not clean, well, we’re not going to have a great agent. And then if we didn’t identify a good process, we’re not going to have a you

61:21 process, we’re not going to have a you process, we’re not going to have a, we’re not going to have a clear know, we’re not going to have a clear reason for that agent to exist. So, that would be my 30-60-90 day plan. Okay, so I I would say,, 30-60-90 is how I look at this one. 30 days in the 30-day time period, it’s about education. It’s educating yourself on on thinking about what processes do you have, what do you think you can automate, learn more about agents. I don’t think people understand what how agents can be used. So, I think just getting lots of good articles, videos, documentation. I think it’s about

61:51 documentation. I think it’s about research, figuring out where it applies. You need to identify one or two individuals that you’re going to give free reign to help evaluate what’s going on. So, in that 30 days, identify those individuals, do some education training, and figure out where that’s going. So, I think I think 30 days is the training space. space. 60 days, I think is finalizing or really adopting, right? After you’ve understood what agents can do and what tooling Microsoft provides, you got to go build a Copilot Studio,

62:22 go build a Copilot Studio, a a Copilot in the Copilot Studio. What does it do? How does it add value? Can it add value? Does it do things you want it to do? Does it not do the things you want it to do? Right? That’s the 60-day time period of just now, okay, looking at the work that you do and how you build things, what does that look like? Right? Where can I apply agents? So, I I look at the 60-day time period is alignment of agents of agents and building them for individuals. So,

62:53 and building them for individuals. So, think about the one person, the one process, what are they doing? Again, to me, I’m looking at this going, this is a Kim. This is the Kim example. Right? Hey, I’m regular I’m a salesperson or I’m a person who’s talking to customers, and I need to materialize emails and schedules and things. I need to have a a holistic view of like what my day looks like. Help me agent reason through what I should be focusing on so that it’s most important for me in my day. So, I think that’s aligning a workflow with what you now know around agents.

63:24 know around agents. And then I think the 90-day time period is look at how you scale it out. Do the evaluation. Now that you’ve got two months of basically experimenting and building, now you can evaluate did this actually add value? Evaluate and start planning the rollout broader than just the one person. Okay, Tommy, you built the sales agent. Right. Right. Was that useful to you? What training would I need to produce? How would I build this for other people? Could I deploy that agent using Microsoft 365 agents and then tell

63:55 Microsoft 365 agents and then tell people how to use it. One of the things that’s really interesting around these agent things is especially MCP servers, you don’t have to know how to use it. That, yeah. I can ask the agent, what do you do? What do you build? How does it work? And agents are really good at telling me the results. results. Right. Right. So, I really like that effect of, letting the agent instruct the user of it.

64:25 instruct the user of it. So, I think that’s an underrated value add for these agents and the harnesses. Mhm. Mhm. So, I think that’s how I give my 30-60-90. Love that. I actually I love that. This has been a really good conversation. I think there’s a lot of things here. I really do think the biggest barrier if I had to go back to the very beginning of this in the same way we decided and made the decision to say, let’s stick with Power BI. This is a solution or a tool that’s going to be useful to us.

64:55 going to be useful to us. Someone at the organization level has to make the same decision around are we going to use agents and which ones are we going to use? Is it going to be Microsoft? Is it going to be Anthropic? Where do we find value and and make Once that decision is made, now we can really step into step into figuring out how it works. And again, this is now a decision once you buy into it, it, as you do this, evaluate the value from it. And you can always pivot. You can turn it off, you can switch to something else, you can have a mix of models. We

65:25 else, you can have a mix of models. We started down the route of only GitHub Copilot. Copilot. We’re actively looking at adding Anthropic and Cloud Code to our mix because it adds value. It It reduces some of our token usage, but it also gives us some value in other areas. So, I’m substantially reevaluating every month about what we’re doing, how we’re spending, where things are going. Okay. Okay. Anyways, really good topic. Thank you very much for the question. Orgs organizations and being slow on adoption to AI. Yes, it’s a thing. it will

65:55 to AI. Yes, it’s a thing. it will happen, but I think if you push forward with the MVP and POC level and talk to us. Tommy and I are happy to consult through like what is working for AI for us. You have some use cases you think AI would be useful? Great. Talk to us. Let us know if you feel like that’s going to be valuable for your organization. Tommy and I both provide consulting in this space. So, there’s way too many organizations who don’t have AI. have AI. we’re experts right now. we’re in this space. We’ve been playing with this stuff for many, many years now and we

66:25 stuff for many, many years now and we are are committed to it. Like it’s it’s going to be be how companies will do business in who wouldn’t want to sit argue in person? person? Right? Who wouldn’t want to have us talk to you live on an actual call and tell you you’re wrong? Anyways, thank you all very much for listening to the podcast. We appreciate you. Tommy, where else can you find the podcast? podcast? You can find us in Apple, Spotify, wherever you get podcast. Make sure to subscribe and leave a rating. It helps us out a ton. Do you have a question idea or topic that you want us to talk about in a future episode? Head over to

66:56 about in a future episode? Head over to powerbia. tips/podcast, leave your name and a great question, and finally, join us live every Tuesday and Thursday a. m. Central on all of powerbia. tips social media channels. Thank you all so much and we’ll see you next time. Explicit Measures pumping up the heat high. Tommy and Mike lighting up the sky. Dance to the data, laughs in the mix. Fabricating heat, I get your fix. Explicit Measures, drop the beat now. Pop is king, feel the crowd.

67:28 Pop is king, feel the crowd. Explicit Measures

Thank You

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

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

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