Using Harnesses for Fabric Projects – Ep.544

Somebody asked how to track report requests through an agentic workflow. Answering it properly means describing the whole apparatus — the second brain, the skills, the triggers, and the tooling that connects them. Mike and Tommy compare very different setups (Notion versus a local open-source stack) and arrive at the same four components.

News & Announcements

Beat from the street — Rayfin limitations. No links this week; Mike brought findings from actually building with it. He set out to replace a stack of Microsoft Forms with a Rayfin input form app and ran into the edges: Rayfin blocks anonymous SQL, which broke the form outright, and getting data into Fabric in the first place remains the awkward step. His observation that Fabric can’t take a webhook drew the sharpest reaction — it’s a basic integration primitive to be missing. He eventually one-shotted a combined admin and form tool with Fable instead. His summary: once the data is inside Fabric, he’s confident; it’s the first hop that’s still hard.

Main Discussion

Topic: What a harness actually consists of, and how to bolt agents onto the tools your team already uses

Both hosts have built elaborate personal systems, and neither is selling theirs. The point of comparing them is what’s common underneath.

• The four wheels of a harness. Mike’s answer when pressed for a ten-second version: skills, agents, properly organized context, and the ability to talk to other harnesses. Everything else is preference.

• Stack small agents rather than building one big one. Mike’s effectiveness came from chaining many small agent-based steps and using agents only where they earn their place — not from one monolithic assistant.

• Tommy runs Notion as a second brain. Agents there can be triggered by a status change, invoked from a page comment, or chatted with directly. He runs separate agents against different models and consolidates their output, and skills mean his agents already know his workflow before Claude Code ever opens.

• Mike runs Logseq for the same job, for free. Markdown files, a GitHub extension syncing everything to a repo, MCP servers so his agents talk to it directly, and no vendor holding the data. Same second brain, different tradeoffs.

• Loop should have had a git repo. Mike’s specific complaint: Loop is markdown, does much of what Notion does, and then buries everything in SharePoint. Sync it to a repo and it would be immediately useful to agents.

• Don’t ask Fabric to be Notion. Tommy’s position, and Mike mostly concedes it: Fabric is for data and it doesn’t need to become a knowledge platform — as long as it lets external harnesses interact fully with it.

• Use the wiki you already have. For teams on Azure DevOps, Mike’s advice is to skip adding another tool. The ADO wiki is markdown, searchable, and links to issues and tickets — one client has agents writing all of it, producing a genuine documentation base.

• Triggers and actions already exist in your repo. Both GitHub and Azure DevOps can fire actions on check-in, which is where you hang agent work in an existing process rather than inventing a parallel one.

Request tracking, stripped back

Mike reframes the original question by removing the tools entirely: someone looks at a report, sends you a message somewhere, you receive it, you act. The agentic version doesn’t change the process — it changes who does which step. Which is why his answer starts with understanding the process you already have, not with picking an MCP server.

Buy less software, build more tooling

Mike is deliberately removing paid programs from his business, replacing them with a large language model, some API calls, and custom tooling he controls — tooling that can talk to his other tooling. Tommy doesn’t fully disagree but pushes on where that stops making sense. The barrier is genuinely low now: point Fable at a feature list and back a Notion-alike with lakehouse storage, and you have something your team owns.

Looking Forward

Take the tool your team already lives in — the ADO wiki, the repo, the tracker — and attach one agent to one step of a process you already run.

Episode Transcript

0:02 up the sky. Dance to the day to laugh in the mix. Fabric and A. I get your feels. Explicit measures. Drop the beat now. Pumpkins feel the crowd. Explicit measures. Hello and welcome back. Today is a Thursday. We got another episode ready to go for you today. Tommy, hello. Welcome back to the show. Oh, it’s another beautiful day, Mike. It is. Well, I gotta before we get

0:35 It is. Well, I gotta before we get into our little intros and some news items or things that are happening, let’s quickly talk about what is our main topic today, Tommy? Our main topic today is a question and we’re talking about harnesses for fabric projects. This is a mailbag, right, Tommy? This is actually a YouTube comment. A YouTube comment. Even better. We’re getting more comments on YouTubes. So,, we’re trying to be responsive on those things and and watching and monitoring that. So, if you have questions or have topics or ideas, let us know. Let us know in the comments section or in the, u discussion of this

1:07 section or in the, u discussion of this video. We’d love to talk with you live. All right, that being said, Tommy, let’s do a little bit of a introduction here. I I’ve been un unsurprisingly, Tommy, I’ve been very excited to use Rayen. I’ve been thinking this is the the next level of stuff. I’m so into this. You’ve been preaching it from the rooftops. Yes. Oh, man. I done I did a whole week of every video, a video once per day for a whole week to get people started on it and like getting it going.

1:38 and like getting it going. I tried building a project the other day, Tommy, and I thought this would be a pretty simple project. My project was I get frustrated when I can’t get data into fabric or easily get data into fabric. It’s a lot of work. And I have I have this,, mentality of I want it to be as simple as possible. So if you imagine, I have a website and some apps and software that we sell. We have a number of contact forms across different social media platforms. And I was looking at them the other day and I’m like, dog on it, I’m not doing a good job of like responding to these

2:09 good job of like responding to these messages. I I don’t know they’re there. I have to go to a specific location to go check for them. I need them to be like that would be a great place to put an agent. like have a an agent in the front watching, monitoring, making sure I know what’s going on, summarizing things and and bringing my attention to things when I need to., we talked recently around Kim Manis and how she’s changing her workflow with agents and using that same way. I’m like, I should do the same thing. So, I went over to Rayfin and said, “Okay, I’m going to build this input form app. It’s going to

2:40 build this input form app. It’s going to have a SQL database in the back end and I’ll just host the page on Rayfin and boom,, I’ll embed that in various websites and good to go. Well, Tommy, I learned there’s a limitation. Oh, no. Oh, when you when you use SQL from Fabric right now, I don’t know if this is going to change, but right now the current pattern is if you’re trying to use SQL from Rayfin, you need some user to be signed in. So my fully anonymous form that I was going to fill out

3:11 form that I was going to fill out having,, just embedding it in an iframe on my contact page won’t work. It it would not work. Pushed to a fabric there. It’s requiring some authentication of a user to have the SQL database. what you really want to your to our point here Tommy it’s making a secure connection between the report side and the SQL database side but the report the the report we call it the the front end side is somewhat there’s something that was requiring it to go

3:42 something that was requiring it to go talk directly to the SQL database they can’t store the credentials it doesn’t have any way of anonymously connecting to the database and making changes or updates there so that was disappointing I even went back to the product team I was like hey dudes can I do this or like no doesn’t exist. So So yeah go ahead. Yeah question. isn’t there like a workaround here where you could if someone has a contact form and again I’m being very crude here but for for the lay people usually can go to like a Google

4:12 people usually can go to like a Google Drive or an Excel file when would that be able to be read by the the fabric too because then there’s some authentication or permissions on behalf of the user. Yeah, you could do that. Like that’s another that’s a great other option, Tommy. Like I could have just sent it to a Google form. I have Microsoft forms running right now because you’re thinking but it should work this way. But it should work this way. Like it should just work. why again Tommy? I So what I was previously doing was I had a series of Microsoft

4:42 doing was I had a series of Microsoft forms showing up and I was also paying $20 a month for N to to host the form, right? So I basically had a web hook at N. You would send a message, the message would be packaged up, it would send the API call and boom, I had the message. One of the things that I really disappointed by, Tommy, around fabric, fabric has no ability to have a web hook. You just can’t randomly send fabric some data from any system anywhere. Most other web applications

5:12 anywhere. Most other web applications have some web hook capability. Fabric does not have this, which is a news a useful use. It’s weird actually that it does like most things do. Yeah. Almost every other application, even like Power Automate has a web hook type thing. You don’t have to authenticate to send it a data or send it some information. So, okay, Tommy, so what did I do? How do I solve this problem? So, I was disappointed. I was like, a man because I want the data in fabric. If my data to run my applications is in fabric, if my data to

5:43 applications is in fabric, if my data to run my intake form, my customer data, like if I if I’m going to double down and make Fabric the place to do things, I want it all inside Fabric. and I don’t again, Tommy, I’m I’m paying a lot for Fabric. I pay money to have it run. Oh, yeah. Oh, yeah. So, I don’t want to pay for more apps. I don’t want to pay $20 here and, you don’t want to pay $20 here and,, $85 there for Dynamics licensing know, $85 there for Dynamics licensing and this licensing that. I don’t want six different licenses to get stuff to work. I want it to be in all in fabric and build what I want. So, what did I do? So, I said, “Look, I need to step up

6:14 do? So, I said, “Look, I need to step up and step back and rethink this.” So, I actually rebuilt my little forms app. And this one is a little bit fun. I went over to Fable. So, Fable is currently available for clawed users., we’re just talking about it. Yeah. You don’t have many days left. Nope. and I oneshotted an admin and form tool all in one go, a full application. So I said, “Hey, Fable, I need to be able to have multiple forms on

6:44 able to have multiple forms on different websites. I need them all to send data directly to this SQL database that’s here. Here’s how I want you to use this GraphQL that the SQL database has. I want you to grill me about all the features and requirements. I want an admin page. I want to style things. Everything I gave it all the requirements.” And it was probably like two or three paragraphs of information of like as I was thinking about requirements and then I asked it to grill me or ask me more questions, ask me more questions until you feel like we’ve made it to a mutual understanding. That grill me feature. It went through. We probably answered

7:14 It went through. We probably answered about four or five rounds of additional questions. And then it said, “Okay, I think I understand what you want.” Boom. It built the entire thing all in one shot, Tommy. It was pretty impressive. And then with a little bit of trickery, I was able to get it to host on GitHub and now building. And so now I’ve got an agently built app exactly what I want. I’m making slight tweaks with the app right now. And now all the data is coming directly to fabric right to my SQL database. So now that the SQL

7:44 to my SQL database. So now that the SQL database has it, now I can start automating and making notebooks and have any like now I can pull in agents to to do. So the the trick of this is Tommy and maybe this is the point I’m trying to make here is once I get data into fabric fabric everything else is much easier to to handle handle because then I can write user data functions that that run on a schedule. I can write data engineering things. I can go invoke things from Foundry. I can put an agent on top. Like everything gets easier once I’m inside Fabric. And

8:15 gets easier once I’m inside Fabric. And so it’s that first step of making sure we get that data into fabric that is a little bit tricky right now. But once that’s in, I feel like I I have much I’m much more comfortable and feel confident about what’s going on in there. Dude, I’m telling you, the world that we live in now, I literally right before we started this call, I’ve been doing a lot more with Alex P’s fabric task flows. Isn’t that amazing? It is amazing. But why is the rest of Microsoft not embracing that as much as like I think it’s really powerful. Well, I’m gonna I’m gonna bring it to the next level

8:46 I’m gonna bring it to the next level because I’m like I so we’ll we’re gonna get into this more because I’m actually have a harness that in a skill to help build my prompt for the task flow based on context. But I’m like I like using I love cloud code. Yeah. Yeah. But there’s sometimes especially when there’s a lot of steps. So I’m like you there’s a lot of steps. So I’m like what probably 30 minutes before know what probably 30 minutes before we jumped on I’m like build me a UI build me an app that basically I can provide the requirements because I can manage the different projects in the task flow because the tassel is a lot of steps because he has to use the

9:16 steps because he has to use the co-pilot agent and to go through the discovery and it’s like well I would love to see this but the and one thing I want to just point out here to Alex then what he built there he did two videos with us on one on the podcast we talked about it with us task flow assistant and then he did a a video on agenda IC thinking which is our other podcast around agent stuff agent and AI things right right I’m really impressed Tommy with what that tool does for one and then for two when I look at it I go there’s a lot of scripting in there

9:47 there’s a lot of scripting in there there’s some AI reasoning but Alex did a really good job of like giving the agent really good context and the secret I believe of that tool is the tool has been scraping the learn documents and has a lot of good definition of like common terms, common languages and it it basically took the data from the learn docs and turned it into markdown that the agent can understand and read and I feel like that’s the trick here

10:18 and I feel like that’s the trick here and yes there’s agents involved but a lot of that there’s a lot of scripting in there it there’s a lot of known things that it built to understand how to build all the items in fabric which is really cool dude it it really We’re getting to a point now and I think what we’re understanding more and more is how much it’s not just your prompt. It’s not just the harness or so to speak,, like how the ability of the model, but the really the harness behind it. So, I

10:48 the really the harness behind it. So, I don’t want to get ahead of ourselves, but Mike, did you have something else? That was my my main like news or intro item. Like I found a limitation with with Rayfen. Rayfen doesn’t allow anonymous access to like a SQL database with the Rayen front end. it’s missing., I’ve been told that we’re going to get fabric user data functions which would help out with some of that and being able to authenticate more. Again, this is just one of my personal beasts with fabric. There’s no web hooking. I can’t just randomly send fabric some data like I can at Power

11:19 fabric some data like I can at Power Automate and get it into the system. That just seems like a miss in my point. I know that is really strange. I’m surprised that they haven’t even built Power Automate into fabric yet to people listen Power Automate is a outstanding tool on its own. So Mike here’s some flowers here. What do you mean what does it not do? Power Automate is a good tool. It just looks like it’s built in the ‘9s and it’s workflowing. I don’t So Power

11:49 and it’s workflowing. I don’t So Power Automate is like a tool I think Tommy and you can tell me if I’m wrong here. Power Automate is taking like a business a business process flow and actually I’m going to I’m going to pick you here. Only say it first and then I’ll come back to it. Power automate is like a business process flow that’s just it’s it worked for what it does but it’s just really old and clunky and it’s not actually I don’t think it’s very easy to use compared to other tools. Well,

12:12 Well, so what did you think I was going to say, Tommy? Because I thought you were going to say Power Automate was made for before agents. Oh, [snorts] okay. No, no, I was going to say that because anytime you could fill ages into your sentence. Well, what I will say though is I really like the program N8. It’s really good. Like Like that’s even more difficult., yes. I think the learning curve on NAN is a bit higher than Power Automate, Tommy. I’ll give you that one 100%. The learning curve is a bit steeper there. But when it runs and once you understand

12:45 But when it runs and once you understand how inputs and outputs work on every single node inside the system that it builds, honestly, Tommy, it’s really nice. I really like it. It’s Microsoft continues to build these like workflow things where you have like,, it’s horizontal or it’s vertical. Stop it. Stop doing that. That’s not how our brains think, Microsoft. We don’t just build everything horizontally or vertically. And I don’t want to scroll forever and ever if I have a really big workflow. I want to be able to zoom out

13:15 workflow. I want to be able to zoom out and zoom in to the workflow in the areas that I see. That’s how my brain works. I think they updated first off everything you just said about the learning curve and understanding inputs and outputs is you can literally copy and paste that with Power Automate with what you just said to me at least. Now granted part of my previous job it does everything. Yeah, there. But I don’t think I don’t think Power Automate has as good of a hook as NAN with the AI and large language model experiences. True. That part’s that part they they’ve got that dialed in and they can

13:46 they’ve got that dialed in and they can customize it. And they’ve even got like their agent node I guess is what it’s called is what they’ve got or their AI node. I guess it’s more like an AI node. You can have any model you want. You can give it custom instructions. It all bolts on to that item. And then you can even have an output that says it it formats the output exactly how you want it. So the agent can give you some junk and it and it auto formats it to something that dude it’s way better. Way better. better. Okay. First off, you’re bringing up something you you’re hitting a trigger for me because this something that I

14:16 for me because this something that I relate with is is I now have gotten to the point where any AI assistant in any application that after I give it a prompt or like a problem and it tells me to go search the web web or rec rephrase it, I lose my mind. So I was I was using QuickBooks the other day and I’m like, “Hey, how do I avoid this? It says just charge. I want to send this.” And it said, “Yeah, D.” I’m like, I can’t see your invoice. I’m like, it’s on the screen. And then he’s like, well, would you here’s the here’s a Google search you can do. I’m

14:46 here’s a Google search you can do. I’m like, you should be better in Power Automate with C-Pilot. You’ll say, I don’t know what this problem is. They go, have you basically says, have you turned it on and off again? I’m like, come on, dude. If you’re going to have that, it should be able to answer those questions or look at the back end. So, 100%. And like it turns you on the other other side. We’re getting way off topic here. here. No, I don’t think so. I think this is relevant, Tommy. I think agents and things and how we like but what I think we’re what we’re Tommy what I think we are starting to talk around the

15:16 think we are starting to talk around the periphery of is the harnesses that we use around the large language model really influences our feeling about does this model even work what I this model even work what like mean like oh yep yep yep wait to bring it back I’m I’m gonna rain it in here a bit and pull it back a bit so if we think about yeah we’re land this topic right So when I have a good it well integrated system system working with agents seems like magic it

15:47 working with agents seems like magic it seems it seems easy to use I’m able to get the ideas out of my head into the agent and the agent is building what I want right now I’ve been using over the last since January December January right time frame I’ve been using exclusively GitHub copilot I’ve just now more opened up I I’ve known about claude I’ve played with it a little here and there, but I’ve now buying subscriptions to it. I’m now heavily using it. I’m giving it to a couple strategic team members to test it out, see where Claude goes. goes. Tommy, the harness they’ve got is just

16:20 Tommy, the harness they’ve got is just second to none. Whether it’s the CLI, whether it’s their Claude code, whether and I can I can already see VS Code trying to mimic and copy a lot of the features that come out of cloud code. Really, really nice. I I love working on cloud code on my mobile app. I can set it up on my desktop. I give it access to like my repo. I can give it some instructions. I have it chew away some things., one other system here, Tommy, that’s that we’ll we’ll kind Tommy, that’s that we’ll we’ll maybe discuss here is we Tommy and I of maybe discuss here is we Tommy and I do a lot of long form videos. Every week

16:51 do a lot of long form videos. Every week we’re doing two hours of long form content. content. So, with a buddy of mine Armando, he and I sat down and said, “We have a lot of long form content. We got to figure out how to build something.” So, we made a joint repo between the two of us and we have now created the ability for an scripting plus AI agents to go through, take our long form video, transcribe it into speakers. From the transcription, we’re then able to come back and have it pick out the most relevant topics that

17:22 pick out the most relevant topics that are going to be used for hosting to to social media. And there we go. We now have an automated system. It’s still I’m still refining it. It’s not quite clean all the time. It still has issues. So sometimes Claude finds halfway through there’s a bug, something doesn’t work, whatever., Claude’s still refining the idea and the schema, but I’m continually prompting my my system, okay, Claude, run this pipeline, pipeline, right? right? Cut the cut the clips, upload the clips. And now I’ve got it down to a a system

17:52 And now I’ve got it down to a a system where I can give it a oneline command and it knows where to get the video. It knows how to cut the cut the clips. It knows how to render the clips. It knows how to put graphics in the clips. Everything happens. It even uploads the clips to a place where I can view them. So, I can be on my mobile device and say, “Just share me the URL links of these clips that I can go test.” And it does it all. Everything. Just boom, done. I’m like, “This is really slick.” So, now I’m building a lot of,, very small things, but I’m now stacking

18:22 very small things, but I’m now stacking a lot of these small agent-based processes together, and it’s now becoming more effective. and I’m only using agents where it makes sense. This is a great segue, I think, into what we’re talking about here because to me, you talked about these smaller agents agents making up the hole here because one thing one thing if you were to say, give me one success, one thing you have to do to make AI work for you. I and I would say it’s not

18:52 for you. I and I would say it’s not about the prompt. It’s all about the harness at this point and how you plan it out yourself. And let’s actually get into the topic here. This is, I think, a great way to talk about what the her hip h rp. If you’re listening on YouTube, we got you. So, this is a great one. So, this is from one of our previous episodes, and they said, “Great episode. Could Tommy elaborate on his workflow using notion for request tracking and then give that to an agent

19:22 tracking and then give that to an agent as context via MCP when Mtheas and Mike talk about GitHub issues for this as well. I would love a full episode about this outer business context to agentic building and how to make it available. Also think of people in corporate environments with tools like GitHub are not available. We use ADO neither notion. We have loop or oneote. Thanks. Thanks. Oh boy, this is going to open up a big can of worms. So, so first off, thank you for listening because this is pulling this is a question across two

19:53 pulling this is a question across two different podcasts coming together. So, it’s talking about the explicit measures podcast podcast listener. Yeah. And it’s talking about the Agentic Thinking podcast, which if you’re not listening to Aentic Thinking, if you want to learn more about agents and how to build them, on Tuesdays, Matias and I do like news and talking about what’s happening in the space. And then on Fridays, we do like a demo. So it’s more like a show slpodcast, but on Fridays we try and demonstrate. Let’s turn on our computer. Let’s show you how it works. So that’s really our mentality there.

20:23 So that’s really our mentality there. Okay, that being said, tell me where do you want to take this question? question? So I think the first part is let’s really talk about the idea of building out that agentic or request tracking the whole workflow. Some of the brief ideas here, right, in terms of what he’s talking about. So, I have mentioned many times on here that I’m in a huge notion fan. I’m a notion nerd at this point in terms of I rely on for my work. I rely on for helping out

20:53 for my work. I rely on for helping out with youth mystery. I rely on it for every single part of my job. Not just for writing notes, but for meetings. And what I’ve expanded that to is because of what they’ve made available is really having these assistants that can now speak to other locations. It is in the tech. Yeah. Let let me just level set real just real quick. I I think you’re going to get into hang on to your thought. I think it’s really going to keep going how you how you’re using it. I have used notion a little bit with Tommy.

21:23 Tommy. Mhm. Mhm. most of our audience probably has not used Notion. You may have heard another program called Obsidian is another one that’s out there as well. Very similar. So my understanding and again cor Tommy I’m going to just throw out my understanding of what Notion is and then you can correct me if I’m wrong. Notion was around before agents were available. Notion is a program that is like a knowledge graph to some degree. So it lets you have like lots of markdown files or folders and you can put in notes, information, you can there’s

21:54 notes, information, you can there’s like different structures. You can have like a table, you can have a list, you can have all these different like formatted options in one system. Now, when agents came out, notion seemed to embrace AI and agents really well. And so, you pay a bit more, but you get like notion with an agentic experience on top of it now. And so, now you can then trigger an agent to go do something. And it may take information from notebook A or or page A, reason about it, and then

22:24 or or page A, reason about it, and then go put it somewhere else. I know on the the podcast, Tommy uses it heavily. He actually has three different agents. He has an agent for chat GPT, one for Google, and one for Claude. And so he says, “Look, here’s the episode. Here’s the main topic. Give me some questions from each of these three agents, and then consolidate them down to like an episode where we would actually have questions or some topics you want to think about.” So, you’re physically using multiple AI agentic experiences integrated into the system and it has it’s able to have context and

22:54 and it has it’s able to have context and awareness. Is that a good summary there? That’s pretty Yeah, that’s pretty good. So the idea is a second brain in terms of how things relate to one another. Every page is based as blocks. I can have code. I can have files. I can have section headers etc. So there’s all this content. The real power came years ago when they came up with this idea of databases. Obviously we know what a database is but it’s different where each record is its own page. You can have pages within that. So good example we have a database for the podcast and every episode is a new page.

23:26 podcast and every episode is a new page. It can go as far as you want here. So for example, all my projects are in a project database, but I can have multiple pages in there referenced by each other. For example, every project I have has a page called resources index with a list of other pages underneath it. So now notion understands how these all pages are related to other. What’s the parent page and all what are the pages underneath it? Well, obviously all the pages under resource index make you the pages under resource index make are have some point that they make know are have some point that they make sense. sense. Yeah. When agents came out, they they

23:57 Yeah. When agents came out, they they were starting with AI years ago, like a few years ago, and this ability to edit on a page, but agents came out and it really completely took,, what notion was, and it just,, amplified really everything. I used notion a lot before. I don’t go a day without looking or using it. So, for example, example, I’ll explain my whole project workflow or briefly or do you want to go somewhere else first? No, no. I think and I think you’re

24:26 No, no. I think and I think you’re you’re on point here, Tommy. The only thing I would maybe add here is there’s a couple little interesting things that are unique, I think, to notion. There’s this idea of like an automation or a trigger. So, there’s some there’s a there’s something that you can actually have like I’m working on a page in this notion thing and then I can trigger something else to happen, right? when when I add a page, do do this. When I [snorts] [snorts] some agent status of a page, change the status of something or whatever that is, it knows, hey, that status is linked to some action

24:57 status is linked to some action and then I need to go do something. That action could be send emails. It could just be call an agent. It could be make a new page. It could be do some analysis. So, you can actually ask it to do additional things, which I think is really useful. Yeah. So I think and again I I think I strained the architecture of a project. Every project that I work on has like milestones in it with this is another database related. There’s meetings related to it. So I have this project page. So let’s add a project called Mike and Tommy. Sure. Or EMP like this podcast. We have

25:29 Sure. Or EMP like this podcast. We have this right now. We have milestones and things that we wanted to do. That would be a database that’s related to projects and obviously they would have relation to EMP. I also have all the meetings that we did. So these are all these different databases that are all like single databases but that has basically like a relation to it. So notion understands all these different context of all the meetings for EMP all the milestones for EMP all these other resources I may have given it. it. Yep. Yep. Now these are the part for me when it comes to a harness to provide

25:59 when it comes to a harness to provide business context. Well, I I white I whiteboarded this initially to really plan out how I wanted agents to work. And I realized that a project is the core,, where it’s the hub of whatever I’m working on. It doesn’t necessarily have to be a project for work. It can be like I want to build this app. Well, I need to know the context, all the different resources that I may want to provide for it. Maybe I’m building a semantic model. I want to provide any data diagrams that were given to me or PDFs or reference of the metrics that are needed.

26:30 are needed. The other part then is the skill and agents Mike because agents in notion or a lot of these tools do not limit what they are possible to do because an agent can run through I can use it in four or five different places in notion. I can comment on a page for it to do something. I can chat with it to update a page or project or I can change a status to your point of trigger. Yes. The goal here and the reason why this all works is because the agents

27:00 this all works is because the agents I’ve created, Mike. So, this is a really important part and I’ll I’ll briefly explain this and then I’ll let you see what you think and your take. But the agent part of this is critical because rather than just having a single I’m going to chat with the notion AI and have do anything. I was very conscious and very intentional about what I wanted agents to do. So I have three main agents, Mike, that are around my projects. And you actually inspired me with one of the names. It’s called PET. It’s made a meatball meatball and I call that it’s really

27:30 and I call that it’s really my my That’s good. It’s my consultant’s consultant. It has all of these resources on fabric, all these resources on my consulting firm and the previous projects. Its goal is to look at any status update to look at what I’m doing with a semantic model to look at my meeting tie back to the project and help say okay what are the next steps help me plan things out but also as an experience of what fabric is in PowerBI is so anytime I’m like hey what’s what are we looking at you

28:00 what’s what are we looking at you what’s what are we looking at are there any blockers when it know are there any blockers when it comes to X Y and Z it run through okay well let’s make sure that we update that prep for the call the second one I have is called operion which is operations organize the project organize pages together make sure status updates are there if I had a meeting and someone said we actually don’t have this column well operations will say that’s a blocker to build your semantic model and it provides all these things and the last one is the status update all these

28:31 last one is the status update all these can also talk to each other which is how I’ve instructed them so Mike you have all these things going on I’ve given my agent a set of skills that knows fabric and helps build the idea for a model. I have the context I’ve provided for the project, my statement of work, what we’re trying to do and my meetings. All these things now flow back and forth to one another. And this is before we get to cloud code or we before we get to building. This is really building the second brain where it’s not only going

29:01 second brain where it’s not only going to just say here’s what you did in the meeting, but I have the conversations like look at the semantic model page. Anything did we talk about today? Would that block something? Also, they provided this email of the a CSV file of what the API looks like. Sure. Sure. Does that block anything? And its ability to do that has been amazing. So, let me pause there. There’s a lot of things I know I talked about there. I just wanted to provide the proper lay of the land. No, I think I think that’s a really good like outline of like what you’re doing and how you’re using Notion., I also

29:33 and how you’re using Notion., I also want to maybe address some of the topics here around,, this,, our our question is really around, well, how does that differ from like Loop? Like, and this is one of the areas that I think is a bit disappointing for me from the Microsoft side is it would be nice if Loop did this., honestly, Tommy, the things that you’re describing now and how you’re leveraging agents and pulling them into your notebook, Loop does similar things. There is an agent inside it, but it’s not as powerful. They didn’t they haven’t really really Oh, no. the story isn’t really like

30:03 Oh, no. the story isn’t really like whipped together in in a more consistent useful way., another one that I’ve heard a lot on the from other PowerBI users and and their MVPs is Obsidian. Obsidian is more similar to Notion and less,, less like or not as close to what Loop is doing. For for years, Tommy, I when Loop came out, I was like, great, this is a better way of using One Note. And I jumped really into Loop and I have been using it. I have a lot of notes in Loop. But then I got I I basically was like okay Microsoft built Loop but hasn’t really

30:33 Microsoft built Loop but hasn’t really been innovating on it. Hasn’t really been making it more notion like I want it I want Loop to operate like a notion. I want Loop to act like my second brain. So I personally Tommy have have moved away from that and I’m right now I’m in in the mode of I don’t want to pay for more things, right? right? I want apps and software that I can use that are open source that are not paid software. So, this is not the exact same thing, Tommy, that you’re speaking of, but I use another one called Log Sequence. Sequence. Logse., Logse Seek is a program

31:03 Logse., Logse Seek is a program that’s similar to Obsidian. Sounds like indie band. Yeah, it does sound like indie band., but that that’s a program I’m using and it’s the same it’s a similar concept, right? I can I have a notebook. Every day has a journal. I can come in and type what I want. I can use hashtags and and,, tag things across different pieces of of information. It makes pages for me. So if I tag EMP or I tag a customer name in any note anywhere I make it. It’s it’s like I can have a consistent

31:34 consistent they call it a journal but it’s more like I just write down everything I’m doing for that day and then that day’s information gets logged into year month day context and then it’s all searchable and then it’s actually and it even builds like a little mini like local network graph right so I can see which things are related like if I go to the month of May here’s all the notes that were from the month of May I can like hone in or or drill into a graph database that shows me my related ideas and how

32:05 that shows me my related ideas and how they stitch together. This is where the agents come into play though because those types of tools, they’re great and I don’t want to diss Logse in any way and I can’t wait for them to come to downtown Chicago to play their newest hit. I thought you called me last Thursday because it sounds like an emo band a bit. But the problem though is you all the input and then where does it go? So it’s great that you can have those relationships and those diagrams, but Mike, I’ll argue with you or let me let me sell you on the ability to pay for something that provides that value. So

32:37 something that provides that value. So you have all these relations, all these diagrams, all these dates and years in log seek. Now the difference when I’m paying for something like notion is I just go to the chat for my agent say I need to take a look what of my particular users if it’s based on a database or have this reference here and where like what do I need to work on what do I need to do what are all the statuses in my deliverables and not only is it just going to chatbot me it’s like hey the client just sent

33:08 me it’s like hey the client just sent over this document what does this change goes through everything and say let’s change the status and now we get to the other part here where not just all the agents interacting and in notion but now let’s talk about the output let’s talk about where why is this important when it comes to the fabric projects right this is all well and good if notion just did this right just ability to update itself itself but notion has done a very good job a

33:40 but notion has done a very good job a great job actually on being available in a lot of different tooling. It’s available I Yeah, let me pause. I’ll I’ll I’ll I’ll let me just react to your comment there because I I think you may not understand exactly what I’m doing with Log. So, you’re right, Tommy. Like, it’s nice to have a paid service that I can go. Let me let me I’m going to feel free to not answer this question, Tom, if you don’t want. Okay. Okay. If you’re paying for Notion, what’s the plans for Notion start at? What What do they cost? Are they like 20 bucks a

34:10 they cost? Are they like 20 bucks a month? Is it like 100 bucks a month? Like what’s the price? They’ve gone on credits now. Okay. So for for agents, makes sense. Everyone’s going pay as you go models, which makes sense. Yeah. There’s about $20 a month for the base in terms of using the pro, but then using the agents, you have to basically purchase set of credits that you get rid. Yeah. And then you burn those down. Sure. Okay. Fair enough. So, the reason I’m asking that question, Tommy, is like I don’t pay anything for my program.

34:40 I don’t pay anything for my program. Mhm. Mhm. And my agents have MCP servers and can talk to it directly. And also all the data doesn’t go into wherever notion puts it. I don’t know if Notion has this as well, but it might also have the feature, but in my in Logique, I have there’s like an extension and a marketplace, and I’ve added in the GitHub extension. So, all the files are automatically added to GitHub. It’s a folder structure. It’s very easy to use. So in that regard, I have it on my mobile device. I have it on my computer and as long as my computer or wherever

35:10 and as long as my computer or wherever I’m using my agent has log installed, there’s a whole bunch of local APIs it knows how to use and there’s an MCP server so it can just talk to it and make notes for me or go find things through my notes. Right. To your point, Tommy, the second brain is useless if you don’t have a place to like put all the information that both you and the agent can work together on the same information. Right. Th this is an area of this idea of the second brain being able to have agents look

35:40 being able to have agents look through your notes efficiently. That’s I think really important regardless of what tool you use. I think that’s and and right now I there’s this idea and I don’t know if this is a real concept or not Tommy but I want apps that are both human related and agent related. Uhhuh. So everything I do now is like this application has data in it. I want to make sure the data is accessible to me. Go in look at it. And this is one of the things I really like about fabric right now is fabric is

36:11 about fabric right now is fabric is getting it is the the encompassing environment for all of my data. But it’s got a pretty decent hooks into how do I use agents to go build a pipeline. you just talked about task flow assistant and how to use that. That’s a that’s a full agentic solution that helps you build direct items and artifacts inside fabric. That’s what we want. We want the ability to go

36:38 want. We want the ability to go seamlessly between I need to be able to do the things directly inside fabric, but I also need my agent to do the same things or I may have the agent set it up and I may tweak it and let’s go to that part. Yeah. Yeah. 100%. So what good is a brain if you can’t use it, right? So and here’s the thing. It’s all well and great here. Now one of the big thing I we talked about agents but notion also like a lot of tools has this idea you can add skills. Mike this was an integral part for me because I didn’t want to just a random

37:10 because I didn’t want to just a random agent model to create things for semantic models in fabric. It didn’t know what it was. It’s not going to do it right. it right. Yeah. But Microsoft already created some skills for fabric. So I was able to convert them to notion skills. Have that in and I’ve every project I work on, I always have claude instructions pages. So I say, hey, here’s CL by the way you’re I need you to build we’re going to use the fabric MCP. Create a cloud instructions page for the semantic model. Mhm.

37:40 model. Mhm. Like notion has done a great job, a lot of tools have done a great job of being able to talk to other tools, other applications and systems. Yeah. Yeah. So this workflow what it creates in the cloud instructions is not a simple page. It scrolls man in notion because it provides what blockers there are, what to check. It understands what semantic modeling is. It knows what measures to create. and understand the business context from every meeting that I had and we know what to look out for

38:10 I had and we know what to look out for and knows that give a recap to update notion at the end. So now you have this whole brain of notion where I go to cloud desktop or an IDE such as cursor or VS code say read the cloud instructions for X Y and Z you cloud instructions for X Y and Z semantic modeling let’s start here know semantic modeling let’s start here and that notion p it does a perfect read says oh I noticed that we have worked on this we left off here because it sees all the things in instructions

38:40 it sees all the things in instructions it’s telling me to use the MCP I don’t have to provide things of instructions in cloud code or cloud desktop because it’s already done in notion. It understands my workflow and it just picks up picks up to be able to do the edits, be able to do the modifications and at the end it’s like let’s update notion tell it what you did which again flows back triggers notion to go oh let’s change the status because we’ve completed that deliverable we’ve created those measures. So you got

39:10 we’ve created those measures. So you got to understand here it’s not this whole harness idea is not just about a harness sitting by itself. And I think that’s what I’m really trying to push here. It’s not just I am building something in notion and then going to the PowerBI desktop and editing. Yeah. Yeah. Harnesses work best. What I’m finding is when they have the ability to talk to one another effectively. Yes, I would agree with this one. And maybe if I step back just one one step above like high level here, Tommy, what I think you’re doing is, and I think

39:40 I think you’re doing is, and I think we’ve talked about this in the podcast a little bit. This is another idea or concept I’m trying to introduce here to the community, which is you’re building custom harnesses. You’re building a custom harness around your process. What you’re doing,, Tommy, how to get your work done. the medium of how you get that harness or collection of harnesses working together is just happens to be what you’re doing with notion and you’re using fabric. So those are like two combinations of things that you’re you’re leverage two

40:11 things that you’re you’re leverage two tools that you’re pulling together to to to make that useful to you. Is that a fair statement? I it’s 100% correct. Yeah. Okay. And then that’s where my mental model goes., one other just random side thought here about this one too as well. One of the other downsides of Loop being a sharing tool or Loop being this Microsoft missed a bucket on this one. They should have made Loop they they shared everything via SharePoint pages, but it goes all like all the pages in Loop go to like your

40:41 all the pages in Loop go to like your personal space. It is markdown. It is doing a lot of things that notion and stuff does, but instead of like localizing it to like a single folder, they just stuck it all in SharePoint. fine, not a big deal. But Microsoft should have let loop have like a git repo repo and everything you do syncs with and without that git repo, which would have been a lot better because then because then you could just say, hey, link up,, I’m going to share access to this git repo and boom, it just works. Like, so to me, that would be like a much better use of the tool. And that’s why I like this one

41:11 the tool. And that’s why I like this one that I have right now, which is called Ludsek. It’s all private data. It doesn’t go to any server. It stays on your local machine. If you want to put in a git repo, you can. and and then I synchronize my phone and my my desktop using the git repo and I use pull requests and I push requests and so I have things staying in sync. It’s a little bit more effort on my end but I have access to all the data. It’s all text and format and I have I control everything. Why I can’t build this as a a feature

41:41 Why I can’t build this as a a feature inside fabric yet? I don’t know. But that would be something I’m going to eventually maybe get to is like have this all again I want to put all the data in the same place. I think Tommy what you’re describing here too is whether you’re building a second brain whether you’re building data whether you’re using notion the the ability for you to pass information around between multiple agents your cla notion going in and out of notion taking an agent going to notion and having that agent use notion like that’s the

42:11 agent use notion like that’s the seamlessness of what we want. We want the agent to act like another team member on the Tommy team. Yes, that’s the perfect way to think about this, right? So, yeah. And I it’s almost like I have a department and teams in it is probably the best way I could think about it. But notion is a department and that team is working around the but I’m having conversations with them. I’m not just telling them what to do. So, yes, Mike and I think this is such a huge part here where I’ll give you an example. Well, I don’t know for those who have seen the show Billions,

42:42 who have seen the show Billions, there’s it’s a great show on I think it was Showtime and there’s this guy named Bobby Axelrod who’s this hedge fund manager and he’s not he’s one of those like anti-hero people like he’s someone you root for even though you shouldn’t and he says he’s like obviously I liked his tone and some of his delivery so I’ve created like skills and I have not and actually an agent now that’s like hey I’m working on this project and they’re not sure about this. What would Axel Rod say? Basically, so

43:13 What would Axel Rod say? Basically, so you have these assistants where it’s not just do things for me. It’s let’s work something out. Let’s that the interactive side. But Mike, the whole harness idea too is not just about one thing outputting to another. It’s not just notion saying, hey, I’m just going to have,, you’re going to tell something else what to do. Let me introduce another part here where I think your harness needs to be a central repository whether it is GitHub or whether it is notion I don’t think

43:43 or whether it is notion I don’t think that’s necessarily it’s critical that you have a central repository where context is being stored and how it’s organized. Yes, Yes, because because yeah, let me actually I’ll have an example, but I just want to really har,, harness or focus on this idea of you have to have a central location where you’re going to keep your context for your harness. Yes.

44:14 I need you to expand a bit more on that one, Tommy. Like, let me give you give me some more like stuff to react to on this one. this one. Let me give you a perfect example. Do some examples. Yeah. Yeah. Yeah. So, this is a good one. So had a discovery. Here’s all these API calls that we have to do because we’re going to be pushing this to the fabric. I don’t know what the data is. It’s the first time with probably 25 27 calls with different equipment,,, different pieces of information. Yeah. Yeah. Well, I thought of it. I’m like, well, Postman’s great, so I’m just going to do this all in Postman. Then I had this

44:44 this all in Postman. Then I had this great idea, Mike. I’m like,, you great idea, Mike. I’m like,,, Claude does have this idea called know, Claude does have this idea called the Chrome extension. So I went, “All right, Pulpete or my consultants consultant and I have another agent called Fabric Advisor work together. Let’s create some chrome Claude instructions for the Chrome extension to go to Postman, do all the API calls we got from the meeting and the emails.” Sure enough, Mike, I had Claude running Chrome and going through and basically doing all the calls to say, “How does that relate to

45:15 calls to say, “How does that relate to my existing model?” you getting what the table information was creating everything in Postman for me my I was having coffee and then I played catch with my son while it’s doing this discovery in Postman so that’s okay again that’s still output in action its goal though the purpose of this exercise was it to report back to notion and report back to my agents its findings and this is what I’m talking about it’s going to get stored somewhere it’s not just doing this output in action the the

45:45 just doing this output in action the the purpose of using the other out external tooling was to report back and update the organization we already have in notion in the notion project in that harness and that’s what I I think is the best one of the better examples I can give you is it’s doing Chrome extensions it’s going to postman it’s discovering its findings it’s going through schemas of other applications I need to pull in and it knows exactly what to do but it knows what has to report. What What are

46:16 knows what has to report. What What are the results that you’re trying to do?, give me a status update. And this is the critical part where I’m talking about your harness needs to have a central location of information. Yes. Okay. Now, I think I’m with you., as you were describing this, Tommy, like I haven’t adopted the same exact process that you have with Notion, but I’m doing very similar things where, and I think more of my knowledge right now resides in repos. So how I think about my business or how

46:46 So how I think about my business or how I think about what I need to work on it’s it’s can I give agents access to my repository and so on my local machine I’ve got like numerous projects there that are all together and so I think this is similar Tommy to what you’re doing with notion as well like so I can give my agent hey here’s here’s the access to the folder that I want you to go handle we’re going to start a project here’s what I want you to go And the interesting thing here that I’m

47:16 And the interesting thing here that I’m using and maybe this is slightly different than what you’re talking about Tommy is the fact that I’m using repositories. Each repository can have its own set of skills. It can have some uniqueness to like what that project’s doing. So I one of the things that I’m I’m not sure Tommy this is the right way of thinking about it. we’ll see we’ll see., yeah. yeah. The large language model is just text in, text out, right? right? The harness is anything else that’s not the large language model. That’s that’s

47:46 the large language model. That’s that’s all harness related stuff. The harness is a a tool that allows you to have like let’s customizations. I I want to I don’t I don’t want to call them that. I want to call them more like plugins, skills, hooks, things like that. Those are all customizations you can put on top of the harness to make it do get better output of what you want. My workflow right now is a lot around this log seek thing which is a repo. All of my statement of works is a repo. All my statement of works are a markdown

48:17 All my statement of works are a markdown file. So I can actually ask the agent to go comb through this thing and figure out what’s there. Now what I haven’t done Tommy which is something I probably should do eventually would be can I move all of my statement of work writing into the same layer as what log SQL is doing right so it there’s no I don’t have a knowledge graph on top of all my statement of work writing it’s just in a repo that I know where it is so that’s maybe the one missing part of my process that I think maybe you have or you get an advantage with out of

48:47 or you get an advantage with out of notion

48:49 notion because you’re able to like you’re you’re landing ing all your data in one place and it knows how to relate all the data together as needed. And so that might be something that would be useful here. But I want to also dovetail this to Tommy where my mind goes is how does this harness experience work in conjunction with what we do in fabric and what we do inside Microsoft 365. 365. And so this is another area that I’m like Microsoft hasn’t really dialed this in. What I do believe though is the the

49:19 in. What I do believe though is the the work IQ work IQ is a big win for me. So when I’m building things in my second brain, right? Well, I’m doing Teams meetings like Teams is how I operate. Microsoft products is what I use. SharePoint is where I store things, right? So if you think about SharePoint and then Teams and then video and then transcripts, like those are all really,, email, those are the main parts of my tooling that I use to get work done. Microsoft hasn’t really made it easy for agents to get access to

49:50 made it easy for agents to get access to that stuff. Like it’s not like all that’s bolted easily into Notion. I think Notion does a good job of it, but Notion’s outside of the Microsoft ecosystem. It’s not like what I would want to see is a notion-l like thing appearing inside Microsoft. Yeah. Unfortunately, we’re not going to get there. And I think the the critical bridge there is MCPs right now. I because that’s how you get there to now. Yeah. Yeah. And we’re not I don’t know if we’re going to ever have that like notion like or second brain like experience in

50:21 or second brain like experience in fabric. fabric. You’ve mentioned this before. Why not? Why not? But so so like that’s I understand Tommy. Tommy. I understand. I understand what you’re saying. saying. We’re not getting ice cream today. Why not? But I want it. I want ice cream. Give me ice cream Tommy. So, so I understand your comment like there’s is there not a but why shouldn’t there be a notion like thing inside fabric Tommy and is this if we’re if we’re saying we see so much value from what

50:53 saying we see so much value from what we’re doing inside notion right now the way this is working can can we just build the version of notion that we would want inside fabric and make it do what we want to do there maybe that’s what we should be building Tommy if if we’re finding so much value out of it right now as a standalone product that’s outside somewhere else, right? What if we shifted that? What if we shifted an item that’s in a workspace that is,, your notion landing page or, that would make a lot of sense because then all your business context can live there. You can

51:24 business context can live there. You can have all your models and where they go and links to pages and, as you’re working with the business, it could capture notes and pull them in and,, digest them down. And then that would make a lot more sense because then I’d have agents that are centered around specific knowledge areas that have access to my model like this. This makes sense to me. I would argue with you though the fabric UI is not going to beat that. But I think that’s why they’re building MCPs in the APIs because fabric is becoming more I feel like you’ve said

51:55 becoming more I feel like you’ve said this before like a backend almost in terms of the services it provides. I don’t I almost don’t want in the UI though to have all of these other things in my notes and I have to pull from there. It’s not going to be a great experience. It really is not because Mike, what you’re trying to think of with the relation organization. The beautiful thing about these different harnesses we’re talking about these second brain harnesses is you can customize them to work for each person or team. If you try to do that in

52:25 or team. If you try to do that in fabric, I think and I’m getting a little technical here. I think you’re going to get this issue of well there it has to work in the way that fabric workflow or and and that UI that experience that organization. So for me I’m okay with fabric not having that as long as they give me the ability for my second brains and my harnesses to interact fully right. Oh man, Tommy, you’re saying good stuff. But this is this is this is what drives me bonkers right now. Tommy, I’ve got

52:55 me bonkers right now. Tommy, I’ve got GitHub with files and things for my statement of works in it. I’ve got notion with notes and things I can interact with agents. This goes back to my very early comment when I started this whole the podcast today, which was I get annoyed when the data doesn’t all live together. Like I should be able to throw an agent at all of this. Like the agent should be context aware of like my emails, my notion, my git repos, my log seek, like every all the that’s the

53:25 seek, like every all the that’s the knowledge of our business as we work together. That’s what makes agents really extremely effective is giving them that rich context of what’s going on. on. And it’s it’s absolutely true and I think that’s it’s going to be more and more important. I think when we talk about AI skills for users and teams, this is the part this is the part that I think we’re really talking about where here’s the thing Mike even if you did have it in fabric like let’s say again going more in that example here well is that living in fabric is it only living

53:56 that living in fabric is it only living in a GitHub repo because the the success that I am finding is I found a harness that can talk to outlook it can talk to my meetings it can talk to MCPs and the fact that all the critical information that I need context is more than just the prompt and again we’ve talked about the difference between context engineering and prompt engineering I need a place that becomes the central hub of my brain and fabric’s not going to be that fabric is meant for

54:26 not going to be that fabric is meant for data is meant for those tasks and those projects it’s not going to do that part and again if I think you’re limiting it fabric is not a do I love fabric I tell people this now okay fabric is. But I understand like it’s not meant for that. Don’t go the Apple route, my friend. Apple should have stayed on freaking phones and software and then they had to build watches and stupid glasses and now I like my Apple Watch. It’s good. No, the Apple Watch is fine, but it dies

54:58 No, the Apple Watch is fine, but it dies every six hours even though it’s a freaking watch. But that’s my own problem. So, but all the Okay, but what I’m saying is pick the core lanes. Fabric should not be a second brain. I know what you’re saying, but I love that Microsoft providing the ability for harnesses to talk to it. Should it be easier to do? Yeah, Yeah, but think about it. Like you can use fabric task flows to deploy. I can create an MCP and have Claude desktop freaking build me my notebooks and validate them while I’m drinking coffee.

55:30 validate them while I’m drinking coffee. I don’t even have to touch fabric. com. But that’s I think that’s my point, Tommy, is you’re you’re describing to me a world that’s like you’re doing things on the peripheral and I’m just trying to be a bit more organized with like where the data comes from, where do I put it all. It doesn’t fabric doesn’t have to do all the things, but fabric should serve the storage system and the and the integration system for all the things we’re talking about, right? cuz it will be your your corporate enterprise solutions

56:01 enterprise solutions they’re going to have data inside fabric that’s where your enterprise information will live so I just see that being the gravity that’s pulling other data systems into the single place and so why not allow notion to point at a folder in a lakehouse storage account that is the notion folder why not pull all of your team members for a specific specific team into a similar second brain like application to when I look at this Tommy the barrier to to create

56:34 this Tommy the barrier to to create something like this is so low like so if I just gave Fable and said hey Fable here’s a list of features that are all notion related go build me an application does the exact same thing but then use lakehouse storage as the the baseline it could do it there I’m not most of these applications can be replaced with something that you build that works for your team. And so this is maybe maybe where I’m going here. The concept is,, I’m of the opinion right now that the agents and the AI world are going to get

57:04 agents and the AI world are going to get so good at being able to create tools and systems that we’re going to just be able to build knowledge and custom tooling tooling like hyper customized for exactly what you need. We’re going to walk into a cloud code session, Tommy. and say, “I need something that does XYZ things and I like how this app works. I like how that app works. I want it all stored here.” And it’ll say, “Yeah, let me build out a plan for you.” And it will rebuild the majority of Notion without

57:34 rebuild the majority of Notion without having notion involved. It’ll just work. And so I I I’m of the opinion like that’s where the next wave of AI stuff is just going to build really hyper customized workflows. Mike, so let’s actually talk about that. If we were to strip everything away, forget notion, forget all those things, tell me what are the minimal viable like elements that you need in a harness to be successful with fabric apps or fabric

58:04 be successful with fabric apps or fabric projects. I I want to I want to I don’t like the answer. I don’t like that question. I I I disagree with your question, Tommy. I want to rephrase it entirely. Yeah. Well, I want to rephrase it in in differently in general. Okay. I want to I want to take it from how do I want to interact with agents along my workflow? That’s I think the question, question, right? right? Okay. Okay, I’m listening. Yeah. So the reason So I Tommy, let’s talk

58:36 So the reason So I Tommy, let’s talk about workflows of things and this is maybe where this question comes from as well like MCP and request tracking and how does all that handle, right? What is your workflow? Let’s talk specifically around the question here which is request tracking. I have a whole bunch of reports that I’ve built. Someone needs something or has a question about the data. How do I track that request in a process? If we just step back, no agents, no tools, someone’s looking at a report, they’re going to send a message to me somewhere. I will get the message. I will then

59:06 I will get the message. I will then acknowledge that I received it. I will then think through what they want. And then I’ll go research and go look at what’s inside the data or try to answer the question. If I don’t have enough information, I’ll go back to them and ask for more information. Right? That whole interchange, that process is custom to me maintaining and running my center of excellence around PowerBI reporting, whatever that may be. Could be anything, an app, call center, doesn’t matter, my business. So if we look at that process and that

59:37 So if we look at that process and that process is repeated hundreds thousands of times per year of just this going back and forth and communicating and doing things I want to look at that process and say in this process flow where’s the right place to put people and where’s the right place to put AI and agents to think about and reason and create.,, Tommy, another one, another good example I have here is I’m giving presentations. I’m seeing presentations from people built entirely in HTML in HTML and they’re not even doing

60:07 and they’re not even doing presentations. They’re not writing them. They’re just saying, “Here’s the outline agent. Go build the presentation. Here’s what I want you to to talk about.” And it it generates all the other stuff. And so, so, if I if I look about my my workflow, one of my workflows may be building presentations and presenting information, right? what how do I integrate AI agents into that process? Can I talk to my microphone? And I was on a call yesterday with a lot of MVPs on Khaled put it together on LinkedIn and it’s on the YouTube channel as well if you want if watch it. It was

60:38 as well if you want if watch it. It was about an hour and a half of just talking about where fabric and ray and agents sit. In that conversation we’re all talking about how agents are changing what we do but not stopping the work that we’re working on. And I think that’s the concept here. So if I if I step back from that and say our questions really should be around, do we understand our processes? Do we

61:03 do we understand our processes? Do we understand our workflow? Once we can clearly articulate what we’re doing, then we step back and then we can answer your question, Tommy. Where do you apply the agents in that step? Right? Agents may be giving me a first pass on descriptions in a in a in a semantic model build. agents may be doing some research for me to understand where the APIs are coming from and how the calls are working and what data comes back from the API calls. Those are all great agent tasks, but to me those are small tasks on top of a broader process flow.

61:34 process flow. Does that make sense what I’m describing there to a point? No, I agree to an extent, but I I’m still going to argue how much diagramming the technology involved to do that. And I think that’s taking that other step here. here. I think techn is remedial at this point. No, absolutely not. I think the agent can build whatever No, I think the agent can build whatever technology you want. Yeah. Okay. This is my fundamental difference. I think Tommy, though, is like you’re saying you’re picking some pieces of technology to like make it work for you. I’m saying I think the technology is

62:05 I’m saying I think the technology is irrelevant at this point. I think the agents agents can build whatever technology we need. Now, Now, I don’t want to have an agent go out and rebuild a semantic model, the the infrastructure, the hardware, all that stuff. Yeah. that that’s a technology piece that I Yes, we had to continue. But if but I but I’m trying what I’m saying is I’m stepping back from that and saying anything I need I’m building more and more custom software technology pieces that fit what I think the workflow should be and how I go to business and how what makes me my

62:36 business and how what makes me my business special. Right. That’s that’s what I’m I’m deleting many many paid programs from my my business. Sure. to remove them and then instead I’m building I’m going to pay for this large language model this series of API calls to use AI but then I will build my process flow and I’ll build custom tooling around what that looks like and now my custom tooling can talk to each other. other. I’m curious about that because I I do see what you’re saying. So I’m not going to say I’m completely vehemently disagree. The the one thing my brain

63:06 disagree. The the one thing my brain goes to though is from an enterprise or team point of view, right? So that you need something that’s going to work for everyone at that point. And also you need that tooling to be able again for me that critical part here is that that harness or that technology has to have good relationships with the other technologies I’m going to use has to be it’s a data contract. It’s a data contract, right? So it’s a it’s a but I think we would argue though

63:36 it’s a but I think we would argue though Tommy if we just decide about the data contract and say here’s my list of markdown files here’s my graph database and how I’ll add things those are the only requirements we’re talking about. Mhm. Mhm. Whether you use notion or other tools or even build a custom tool. Maybe there’s a custom tool you want to build for the sales team. Maybe the sales team needs specific information. Okay, not a problem. But if that happens, right, if we’re building a custom tool that does that, I’m looking at it going, “Oh my

64:07 that, I’m looking at it going, “Oh my gosh, we can abstract any tool away from their system and we can focus on where do we store our knowledge and information? Can the humans see it and can the agent see it?” And maybe I’m oversimplifying the problem. Yeah, I I there’s a bit of o I don’t want to say oversimplifying. I think we’re going to get closer and closer to really every organization really having these custom applications that way. I would say in the meantime because I know we’re getting near time here. I just for me I want to touch on the here and now at least for myself where if someone

64:38 at least for myself where if someone asked me point blank I only got 10 seconds because I’m grabbing ice cream. What do I need to have in a harness to make it work? I would say there would be three elements. You need skills, you need agents and you need the proper organization organized context. Those to me are those three three critical junction or four. The fourth one is again the ability to talk to other harnesses. Those four wheels make up the car that runs my business and that makes up the car that runs fabric projects for me. Again, skills,

65:09 fabric projects for me. Again, skills, the proper skills, the proper agents, organized context, and ability to talk to other tooling. I don’t care if that’s built on its own or on buying. That sounds I liked all of those things. Yeah, that’s that’s where I want to end. But Mike, I I I just want to say love the conversation. I think we’re what you said at the end and really the ability to build the own tooling, the workflow cannot be understated as well. So, all right, that’s all I got. All right, I’m going to quickly wrap up here and just do a couple items here specifically to what this user asked

65:39 specifically to what this user asked about. So, this user asked about a couple things like they said, look, we don’t have cloud code. We don’t have GitHub. We don’t. So the demo Matias and I showed was we were using GitHub and we were having an agent build issues and put issues back out on GitHub to go fix, right? An agentic loop around development, right? go review this semantic model, come up with all the best practices for it. Okay, for each best practice, make a task an issue to go fix it. Right? Okay, new agent, go pick up this issue

66:09 Okay, new agent, go pick up this issue and go fix it. Right? So that’s what we were doing. You can use Azure DevOps in the same way. There is an Azure DevOps MCP server. So just I want to answer that question directly. That is just because you’re using Azure DevOps as your company’s tooling. You can use that as well with an agent as well. So I’m going to put that out here. There’s a Microsoft repo called Azure DevOps MCP. It has multiple releases. The last release was two two weeks ago. So it’s fresh. and you can use an MCP server which allows you to list your

66:41 server which allows you to list your ADO objects, list teams, create a wiki page, update the wiki page, get the content from the wiki page. So all of these things are again what we’re talking about a little bit Tommy is we have a team we have chosen a tool called Azure DevOps. The the tr the trick of this is how do we bolt on the agendic thing onto that tooling? Okay, so that’s one. The other question you had here, they’re saying, “Hey, that they’re using loop or onenote.” I totally agree with you. I understand.

67:12 totally agree with you. I understand. I have been using Loop and OneNote for many, many years. I don’t feel like either of those tools do a great job. Some people have made some MCP servers for Loop. there is an this is another OneNote MCP server that exists. Again, this is made by someone else. It’s not made by Microsoft, so I can’t really vouch for this one. So, this one’s there as well., so that’s there. And,, again, again, I’ll be honest, Tommy, I

67:42 again, again, I’ll be honest, Tommy, I don’t really love Loop or OneNote because I feel like I don’t know where the files go. I don’t feel like I get really good access to like the underlining data of those tools with agents. It doesn’t it doesn’t give me a good I doesn’t do a good job there. So, I don’t love it. I would encourage you to use or check out log seek only because it is re you can use it in a in a repository which means it would work with dzure devops and log seek

68:12 work with dzure devops and log seek would work with agents as long as you install it on your computer. It’s an open source product. it there’s no paying or subscription. you can use it and I if one if one note and loop would let me have a filebased storage pattern I think they could be much more effective. the other note here I maybe would add is if you are using Azure DevOps you may want to double down on the wiki page that might be a better knowledge space for you to put your

68:42 knowledge space for you to put your data. So, think of it as a team, right? I’m in Azure DevOps and we have a whole team people trying to build a knowledge base around what we’re doing. Don’t use Notion. Just use the wiki that’s inside there. It has markdown files. There’s a lot of information there. You can go scan it. You can search through things. So, I have a client right now that we’re working through. They heavily use the wiki page, but they use it. They don’t write anything. They make their agents write everything in there. So, there’s this wealth of documentation and knowledge there. And in that wiki page, you can link back to issues and tickets and

69:12 link back to issues and tickets and tracking. So if you pick if you are bought into the Azure DevOps ecosystem, there’s there is some opportunity I think to expand more of your usage there. All right, last question here from the chat. couldn’t the harness live in a workload? The short answer is yes, and I’ve already built a number of them. So if you go check out power Yeah. Yeah, if you go pick check out the custom workload that is made by PowerBI tips,

69:42 by PowerBI tips, it’s called Power Designer. Power Designer now has seven different tools from tracking your lineage across multiple semantic models. We have an accessibility checker. We have a layout generator. We have a PowerB report theme generator. We have a theme manipulator. All of these tools can be used with agents. And so I’m actually heavily invested in making aic tooling with all these existing ecosystem pieces of fabric. So back to Tommy’s point

70:13 of fabric. So back to Tommy’s point earlier,, why doesn’t notion just live in fabric? I think it’s actually possible for you to get a very notion-like experience directly inside fabric. And I think that would be extremely relevant for users. I I really I want to mention this because you said some awesome things. If you use fabric if you use the get and source control well the GitHub has agents which you can trigger whenever there’s a commit. So if you wanted to update the context in that workspace there’s there’s so many other places here

70:43 here and and that same thing happens in Azure DevOps. So if you’re doing GitHub or Azure DevOps they both have these actions and triggers and when changes are coming in then you can trigger an action to happen and go do specific things and you can call off different agents to go do those actions after you’ve checked it in. So, anyways, really cool stuff. Very neat things here. So, great question today. I really love this one., using harnesses for fabric projects., it’s here. We’re doing it. We’re building it., stay tuned. I think you’re going to see more and more things come out from the

71:13 and more things come out from the community, from Microsoft, and from developers on top of fabric that’s going to continue to enrich that fabric experience. And so,, I would highly recommend go check out Power Designer. I think you might find some usefulness there. That being said, Tommy, thanks so much for the episode. We’re way over on time here., we appreciate you listening. Good conversation here talking about agents and how this works with your data and workflow and how do you customize agents inside your business. That being said, Tommy, where else can you find the podcast? You can find on Apple, Spotify, wherever get your podcast. Make sure to subscribe

71:43 get your 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? Well, try to do it on the mailbag. Head over to powerbiotips/mailbag. Leave your name and a great question. And finally, join us live every Tuesday and Thursday, a. m. Central and join the conversation all PowerBI tips social media channels. So, last one quick question, one note there of [snorts] change, Tommy. If you want to go make notes in the mailbag, it is powerbi tipsodcast and then at the end there, you’ll see the you’ll see the list down at the

72:14 the you’ll see the list down at the bottom there. There’s a mailbag there that’s explicitly there. Speaking of which, I did not add the mailbag into my forms tool. Do that now. Now I now I got to go I got to go do that so I can embed my life easier. Please make Tommy’s life give it a give it pull all the data into the same place. So I will have to make sure I update that as well and make instead of a Microsoft forms I’ll just push it I can trigger it. I can have it just go right to notion from there. So that that would be now we can do some so I didn’t even think of that. So one more form that I can

72:44 that. So one more form that I can replace with my brand new tooling that I just made. So super exciting, very good stuff. That being said, thank you all so much for participating. We really appreciate everyone jumping in and hanging out with us today. We appreciate you all and we’ll see you next time. Measures pump it up high. Tommy and Mike lighting up the sky. Dance to the day to laugh in the mix. Fabric and I get your feel. feel. Explicit measures. Drop the beat now. H 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.