Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Microsoft Cloud IT Pro Podcast

Microsoft Cloud IT Pro Podcast artwork

Episode 438: Taking your WorkOS and LLM Wiki to the Next Level

Oct 8, 202647 minAudio

Welcome to Episode 438 of the Microsoft Cloud IT Pro Podcast. Back in Episode 429, Ben and Scott introduced LLM wikis. In this episode, they come back a few months later to compare what those wikis turned into after real daily use: personalized work operating systems, or a WorkOS, that tie together tasks, meetings, research, communications, and recurring workflows. Scott walks through the skills that now run his day, from a 9:00 a.m. prioritization pass and meeting prep to meeting capture, end-of-day review, and clipping ingestion, all while keeping governed documents in Microsoft 365 where they belong. They also dig into how a workflow matures from an on-demand skill to a scheduled skill, a loop, and finally a deterministic Python script that any interface can call. Scott shows off his locally served web dashboard pinned in Edge, Ben shares the custom Obsidian plugin Claude built for his command center, and they compare notes on choosing a source of truth (Markdown versus Asana), filtering Teams noise down to the signal, and what these systems actually cost to run each month. Show Notes A pattern for building personal knowledge bases using LLMs LLM Wiki v2: extending the pattern with lessons from production Agent Skills open standard About agent skills for GitHub Copilot Use Agent Skills in Visual Studio Code Building effective agents Conductor: Deterministic orchestration for multi-agent AI workflows Obsidian Web Clipper Prior podcast context Episode 429: Getting started with LLM Wikis Episode 430: Scouting out Microsoft Scout Episode 422: Back to the Terminal: The Rise of AI CLI Interfaces About the sponsors Nasuni is a leading unstructured data platform for enterprises where file data is mission-critical for both people and AI. Nasuni powers the operational file layer where work happens - helping organizations manage, protect, and activate data so teams can work smarter, reduce costs, and operate securely without limits. Intelligink - Would you like to become the irreplaceable Microsoft 365 resource for your organization? Let us know!

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Welcome to Episode 438 of the Microsoft Cloud IT Pro Podcast, recorded live on October 2, 2026. This is a show about Microsoft 365 and Azure from the perspective of IT pros and end users, where we discuss a topic, more recent news, and how it relates to you. Back in Episode 429, Scott and Ben introduced LLM wikis. Today, they're circling back around to see what they've turned into after using them daily for a few months.

Scott walks through the skills that now run his daily life, from a morning prioritization pass and meeting prep to meeting capture and end-to-end daily reviews. We also talk about growing a workflow from an on-demand skill to a scheduled skill, loops, and deterministic scripts. We also discuss different approaches Scott and Ben have to creating custom dashboards and command centers and how they interact with their data. We're live with another episode of the Microsoft Cloud IT Pro Podcast, live from our houses.

Live from New York. It's Saturday night. Oh, no, that's a different one. Different show. We're not quite that popular, Scott. Someday. I know we look that good, and we sound that good, and we could clearly do a primetime, not really primetime, late night, NBC show. Better be careful. We're going to end up talking about copyrighted stuff. We could go see if we could end up on The Onion and the new InfoWars channel or something. We could just spin yarn and conspiracy theories about Azure and Microsoft 365. 365. We could. We could come up with all kinds of conspiracy theories.

My daughter is actually enjoying that right now. Do you know it'll be fine? Yes, my daughter's actually taking a class right now that's all about conspiracy theories, where they spend like a week, because she's homeschooled. So all the kids in the class go home from their co-op, spend a week like looking at different conspiracy theories, and then come back to class the next week and debate the various conspiracy theories. So it's kind of interesting. She's having fun. They did the moon landing today. So the last week, she's been looking at all the different conspiracy theories around the moon landing.

My oldest son is, he's studying history in college. And so he's, so he's in his second year now when he's getting kind of like more into things. And he has had the realization that history repeats itself. You didn't learn this in like middle school or high school, but you're coming to it in college. Okay, well, good realization. I mean, at least he came to that realization. There's some maturity there. I have hope that he'll move out of my house someday. We'll see.

So you know what else happened today, Scott? I called my HVAC company because we have like in Florida, AC is kind of important. So twice a year, we have like regular, we have a plan. We sign up for discounted maintenance and all that. But then they do like twice annual, twice annual maintenance checks. And I got a text message and an email and all of it. Hey, it's time to schedule one. So I called them today because usually that works better than trying to do it online and all of that. And it was like, you're on hold. And then I had someone picked up the phone and it was like, I can't remember the name. Hi, I'm Lisa. This is your AI assistant. Like it was an AI.

Well, he should announce itself that way. Person. I'd rather fall into like a phone tree and interactive voice recording like that, where it's very upfront and clear that yes. Yeah. And that's what it did. Name. But it was a little, it was interesting. I wanted to be annoyed at it the whole time because I knew I was talking to AI, but it actually went really smooth. They superimposed normal office sounds in it. So I could hear like a keyboard clicking in the background and all the stuff. But this AI assistant was, why are you calling? I was like annual maintenance and asked me all the questions it needed.

Like had an actual conversation with this AI agent, just like I would have had with anybody in their office and was able to schedule my appointment, got the text message and the email with the confirmation of the appointment date and time. And as much as I wanted to be annoyed knowing I was talking to AI, it actually was a good experience. And I was pleasantly surprised and impressed with how well and smoothly it went. Good to hear. When you were texting me about it and telling me about it, I thought you were going to tell me like you fired off because there's lots of personal agents out there right now.

So you've got Meadow with Muse, there's some invite stuff like Instinct that's out there. OpenAI announced Dots at their dev day this week. So I thought what you were going to tell me was like, oh, you had your AI personal assistant, because now these things are starting to come with phone numbers, and do inbound outbound calling. You had your AI assistant call another AI assistant, and maybe they got caught in one of those loops of like talking to each other in like binary or like whatever. No, but that would be funny.

Sound and wavelengths. They wanted to have like a waveform representation to go and talk back and forth. I should try that next time is have my bot call their bot. Speaking of bots and Dots, did you also see the somewhat trolling, hilarious news like OpenAI announced their Dots? And immediately after that, Musk went out and bought Dot.com and redirected it to the GrokBot. If you want a domain name about a product, you're going to buy the domain name before you announce it.

Or you just get big enough that you have your own company domain name, right? Like the whole like Cloud, Dot, Microsoft, and things like that. See, that would solve it too. That's the other way to go about it. But yeah, we should talk about AI stuff today. We should. I want to revisit the talk that we had a little while ago about getting started with LL and Wikis and kind of talking about what they can turn into with a little bit of maturity.

So like this will be a little bit more of like a freeform conversation for us. I want to talk about my setup. I want to learn a little bit more about your setup. And I really want to dig into kind of the workflows and maybe like the skills and agents that you're using. Like what are the harnesses that you're using around this that make it helpful for you and empower your day-to-day work, accelerate your day-to-day work. Maybe they make things harder, right?

As you've been experimenting and going through. So what are things that maybe you hit a brick wall with or I hit a brick wall with? And we kind of went in a little bit of a different direction. So yeah, you want me to go first? Sure. What I've been doing? Yes. I'm curious to hear what you've been doing. So I don't want to get to your point earlier about like copyright infringement, things like that. I don't want anything there. But I took what were like various instantiations of LL and Wikis that I've been working with around like various areas of like research for specific projects and initiatives that I work on.

Just like my day-to-day tasks, tracking, things like that. And I consolidated them all into one wiki and it's my WorkOS, right? It's my work operating system. So it's meant to track really everything that I do across work, whether that's working. So I'm a product manager, so that could be working on a new product. It could be working on an update to an existing feature. It could be like building a new feature. Maybe that's not a full product, right?

But it's complimentary to existing product suite, things like that. And I do all the PME things that come along with that of market research, writing PRDs, specs, having customer conversations, wrangling work management systems like ADO. We use Azure DevOps. Maybe somebody's using JIRA, something like that. But really, everything that I do gets logged, gets tagged, and goes into that WorkOS.

So off the cuff, that sounds like, okay, but how do I get everything in there? And what's important and what happens along the way? So I've got a set of skills that are out there that I've put together that work for my way of work and that operate within the harness that I use. So the harness that I use today is Microsoft Scout, the desktop app for Scout. I'm going to be a little bit sad when that goes to autopilot here in the future because I don't know what that's going to do.

Not autopilot because we have Microsoft autopilot. We need to differentiate now, Scott, because we have Microsoft autopilot. Autopilot inside of copilot. Copilot autopilot. And copilot autopilot. That's another topic for another day. We'll get to that one eventually. So I've got this set of skills that help me with inputs and outputs to get information into that wiki and then help me get good concrete outputs out of it. So one would be, for example, one of the ones I use quite a bit actually is an email thread ingester skill.

So my work takes place across teams, email, in-person conversations, teams meetings. If I'm meeting with customers, it could be Zoom meetings, things like that. But particularly for internal conversations, a lot of that stuff still happens in email or there's newsletters and various things that go out. So I have a little skill that's my WorkOS email thread ingester. And because it works in context of WorkIQ and the Microsoft Graph, it respects sensitivity labels, DLP, all those kinds of things.

It's not like I'm taking super secret stuff out of the system and throwing it over in this random text wiki over here that lives in my OneDrive or anything like that. So I use that one quite a bit. And that's super handy and helpful for me for point-in-time kinds of things. I have a skill that runs every single morning for me that goes every single weekday for me. It runs every morning at 9 a.m. local, which is my WorkOS daily prioritization skill.

So what that does is it looks at the to-dos that are in my WorkOS. So I made this very conscientious decision not to use Planner or Microsoft to-do or anything like that. I'm just kind of managing everything in markdown files inside my WorkOS. So your source of truth for your tasks is these markdown files in WorkOS? Yes. Okay. But I find that works because I can tag them. They carry the tags through the wiki. They carry all the grounding and the context.

I can link to-dos very easily to projects or to initiatives and various other contexts that exist in that wiki. So what my kind of daily prioritization skill does for me is it's meant to look at my prior day and or my incomplete tasks that are sitting out there that I've already spec'd out and help me kind of rationalize, here's the number one thing that you should do today out of your existing to-do list based on what happened between yesterday and today through your email, through your team signals, all those kinds of things, documents people shared with you.

Here's another task that you should go and do. And then if it has room for it, the way I structured the skill is I always want to have kind of like three things on my plate in front of me that are like these are like the must-dos for today kind of thing. So that sits out there. I have a question about that one because I have one similar to that. Have you, and this is something I've done with mine with some success, have you had that skill when it looks at your daily to-dos, also look at your meetings for the upcoming day to prioritize if there's anything that you have to have done for a meeting coming up in like the next eight hours based on your tasks and projects as well?

Yes and no. So we should get into that one next. All right. I want to kind of start the conversation here talking about the skills that I've integrated and that I'm using within Scout as a harness. I could be using Copilot, CLI, Claude, Codex, Anti-Gravity. It really doesn't matter, right? A skill is a skill and that markdown is portable and can come over and I can tie my connectors, all that kind of stuff. But there's this concept that I think becomes very powerful of moving from skills to loops and having loops that go out and execute things for me.

So while like my daily prioritization skill is just a skill, I could run it on demand. I could come in and I could say, hey, run this thing. It's both a skill and it's a loop for me. So I put it in the harness and I schedule it and I say, hey, run this every day at 9 a.m. I actually have, I can run the same skill like every 30 minutes basically on a constant loop and have it do constant reevaluation for me like throughout the day and pivot things through. So I have two other skills when it comes to meetings that also function as skills and or more harnessed loops and kind of just on the flywheel of they're going to do things automatically for me.

So one is a meeting prep skill. So I have like WorkOS-MeetingPrep and I have WorkOS-MeetingCapture. So WorkOS-MeetingPrep is my prep skill. So as my daily prioritization skill is running in the morning and it's going through like that harness and that automated scheduling and now it's turning more into a loop, what that's actually doing, one of the subtasks that it has is it goes and runs meeting prep for every single meeting that I have that day.

That's on my calendar, okay? And then as I'm coming out of meetings, many of the meetings we attend are recorded, they have transcripts, they have all of those things. I have a meeting capture skill. And my meeting capture skill goes and evaluates the recording, it evaluates the transcript, it grabs the WorkIQ context, which is inclusive of like the meeting chat. So you can catch like documents that were shared in the meeting, things like that. And that comes through and effectively it turns my meetings into a two-step flow of prep and then capture.

Or if I have a meeting that maybe, let's say like I was an optional attendee or maybe it was a learning forum, it's something that I want to catch up on later. So I wasn't active in it. For like meetings on my calendar that I'm just following or I'm an optional attendee, things like that, my morning prioritization skill doesn't actually run prep for those automatically. It just skips over them, right? I don't need them right now. So for those, if I want to, I can go in and I can basically run, I kind of have this like overarching maybe superset skill, which is generate and capture.

So I can run generate individually, I can run capture post-meeting, or I can run generate and capture, which kind of does both for me. And generating capture is helpful for the meeting that's already happened that I didn't attend because they do different things, right? Prep goes out and looks at my inbox, it looks at conversations with meeting attendees, and it says, hey, what's the context for this thing? Capture is go blend that kind of meeting brief and the meeting prep with the actual outcomes of that meeting and what's next along the way.

And then I have another skill that runs in a loop, again, on an automation, which is my end-of-day wrap-up. And part of my end-of-day wrap-up is that skill evaluates my wiki and it says, what are the meetings I attended today that I forgot to capture? Or I forgot to do a prep and capture on, and it just goes and grabs them automatically for me and kind of brings that context in. And that's good for me because then in that virtuous cycle of tomorrow's a new day, my automation for my daily prep runs again.

And now it's able to catch up on the meetings and the context and things that are there. So I don't have to constantly be like human intervention in the loop to say, like, click a button or go run this skill, go run this skill, go run this skill in my harness. It's just happening kind of automatically and automagically as I go through my day. And then, let's see, because this is an LLM wiki, like, if people are on, like, the Carpathi-style LLM wiki path, they often end up with, like, that raw or clippings or ingest folder for things.

Well, I'm a PM. I do research. I have to keep up on industry news. It could be Gartner, IDC, launch from a competitor, things like that. So I'm just running things in a markdown wiki at Subsidion. So I can use Subsidion Web Clipper. So I send things into a, like, a raw or a clippings directory. So I have a, hey, go ingest my clippings. Like, go rationalize over this. And again, organize it in context of my projects. Tag it. Do all those things that you need to do along the way.

And then I have these kind of, I don't run them in loops. I don't run them automatically. I have a bunch of, I call them, like, my little, like, sub-agents or my helper skills. So I'm responsible for a portion of a business. That business has, like, a monthly business review. We do ongoing monthly product reviews. We do weekly execution reviews, things like that. And those often require, like, very structured inputs of, let's go grab your ADO items, get the current status. What's the dates, what's the update, things like that.

So I have these skills that go out and help me do those things, like WorkOS MBR prep, WorkOS product review prep, things like that. And they all go out and they kind of work and they help me get the rough outlines for those things. I don't let them, like, write everything for me. I have this very visceral reaction to, like, just pure, like, AI-written things. I'll let them bullet it out so that then I can take them and kind of put my own thinking or my own spin on top of it along the way.

And then the last thing that I did, and this is probably, like, super niche, but it's been very effective for me, is I took the things that I was doing in the harness. So I started in Copilot CLI, eventually moved to Scout. Now Scout's kind of like sitting over there. So Scout's really like my automation engine. It's not actually where I go and run these things from. So if I need to go, say, run a meeting prep, so I've got an upcoming meeting and that prep hasn't been run yet,

or I've got to go run a capture, I built myself a WorkOS dashboard. So I have a little local HTML site that's served to me, and that site presents to me my calendar. It presents the meetings that haven't been prepped versus those that have been prepped, those that haven't been captured versus those that haven't been captured. It gets my to-dos there for me. So rather than me having to go into Obsidian, say I'm updating a to-do from not started to in progress,

I actually, I go to this website, and I just click the to-do. It's kind of like to-doist or something for me just sitting there. And I update it right there, and that thing goes and writes the wiki, the markdown for me. So I don't have to go, like, find the markdown file and worry about, like, oh, did I get the tag right, everything. Things are very deterministic for me. So I've got the wiki there when I need it, when I need to go search for it. I do all that through Obsidian. I've got Scout sitting on the side, which is running my loops and running my automations and is kind of the house for those skills in the execution engine.

And then I've got this just kind of dashboard that I live in. And from the dashboard, I can do things, like, I've got, like, my calendar in there. Like, if I went and looked at my agenda view today, it would have this podcast recording as a meeting on there. I can click on podcast recording, and then I've got a little button there that says, open your brief, and it'll just pop me open straight into Obsidian. Like, it takes me right into the vault, right to that thing, and gets me ready to go and where I need to be. So that's been super helpful and super contextual for me.

Because when I come in the morning, I refresh my homepage for my WorkOS, right? I just hit the refresh button. And then all of a sudden, I'm presented with, here's your daily priorities. Here's your top three to-dos for the day. Here's your agenda for the day. Here's how these priorities relate to your agenda for the day. Here's all the things that I generated for you. Here's how many aren't left. Things like that. It's a very freeing thing. Like, in the past, I would have sat there and said, like, oh, crap, I've got a meeting. Let me go to that meeting. And you kind of come in all discombobulated, and you do the context switch, and you come

out to the other side. Then you go, okay, I've got to go to my next meeting. I don't have time to go back and kind of write my notes up or capture them, anything like that. Now it's all just turnkey. It's all right there for me and ready to go. And also, very deterministic. So the last time that we chatted about this, I believe I had also mentioned kind of taking skills that maybe start out with a bunch of overarching shape in them, but then you want to make them

more deterministic over time. So sometimes you have to stop relying on the model and the skill itself. And you need to kind of write scripts and have a script do the generation for you so that things are always formatted consistently, they come in the right way, all that. So for me, my meeting prep skill, it started as a very generic skill. Like, go find the transcript and the recording for this meeting on my calendar and capture it. And then I refined it over time, but it was still a bunch of markdown. I actually took it and worked and iterated on it back and forth a little bit, and I was able

to get it. So now my meeting prep skill is really a script that calls WorkIQ. And it calls WorkIQ in a way that the output is very deterministic for me for every single meeting I go to. It's got the meeting attendees, it's got the transcript overview and the summary, it's got here's the key items from the chat, here's the documents that were shared, things like that. And what that allowed me to do by putting that skill, and now the skill is just a wrapper for calling that script, back to that dashboard thing, right?

That dashboard's there, like I can't click a button on that dashboard and have it go run a skill. But I can have it go run a script, right? And kick that off very easily, like it just goes and runs a Python script in the background. So whether I'm running the skill through the harness, through Scout or through GitHub Copilot CLI, or I'm clicking the button, at the end of the day, the skill and me clicking the button are calling the same Python script, giving me the same deterministic output every single time. It's interesting to hear you talk because I've done some things very similar, but then

I've also done some different things. Like I said, it's a very empowering thing, I think. So this is in no way, shape, or form a badge of honor. I'm actually kind of like scared when I say stuff like this out loud. I have on average 65 meetings a week on my calendar. That's nuts. So that's my average. Some weeks I have 80, some weeks I have 40, but I'm sitting right around 65 on average over the past three months since I started tracking this. I do not attend 65 meetings a week, to be very clear.

Like 20 of those are probably just like learning or like FYI things. But having that skill out there, that prep and capture skill, like that still allows me to ingest the knowledge and have it out there. Like we had a training last week on like GitHub Copilot CLI and like sub-agent usage and things in it. I was like, oh, that's cool. Like I really want to go catch up with that. In the past, I would have completely forgotten about that, right? I would have had it written down on a piece of paper. I would have had like a to-do and to-do to go watch the recording for that. And I never would have gotten to it or come back around.

But because I have this like prep and capture thing, it actually can sit there and contextualize for me and not only get me the information I need, so I can go get the summary and decide like, oh, was that actually worth my time or not? But then it's got all the canonical links back to the full recap and everything that I need out there and that's ready to go. So I'm probably like on some further end of the pendulum swinging. Like I haven't quite found equilibrium there where I'm probably still capturing too much stuff and I'm bringing in too much context.

But I'm still at the point even like a couple months into this where I'd rather have more than less. And I hope I can get to like the less is more kind of thing along the way. But when it's so easy to do, because the loops are there and the harness is there, I'm kind of inclined just to keep it running. And then, like I said, it makes a bunch of this stuff like more powerful for me, right? Because if I'm going and like we're working on maybe like a new like CLI component or we're doing an MCP thing or some kind of like agentic thing and now I've got all these training on

how agents work and how they do stuff, like that all applies, right? That all comes back into what I need to do. So contextually, that stuff could be part of my research for a SPAC or a PRD like that something later. And then I have these aspirations to go build some more loops. And I very much want to do this. So for me, for example, one of the things that I like to do is I like to go on Reddit and LinkedIn to go find feedback from users.

So I've got a couple different places. Azure Storage really doesn't show up on Hacker News for some reason. They must not think we're a startup or something. I don't know. But we get like decent feedback on forums like Microsoft Q&A and Reddit and things like that. So I would love to go and build like a user research loop that can help me go like research my product on like a daily or weekly or monthly basis, things like that might be a little bit harder for me. I saw just today. So today is October 2nd. I think I saw earlier that Reddit is going to shut down all their external search access

and all their RSS feed access. So that's going to make it a lot harder for agents to get at the data that's in Reddit. So maybe it's a good thing. I haven't built that loop and skill yet to go and do that. But the whole thing of starting with skills, automating those skills, and then figuring out how to compound them into these loops that just kind of keep running until you get to the outcome you want, super powerful. If you're managing file storage today, you've probably felt the pressure.

Refresh cycles are harder to plan. Costs and complexity are increasing, especially as file workloads scale across regions. That's why Microsoft customers trust Nasuni. As an Azure certified software, Nasuni provides a global file system designed for enterprise scale environments, keeping file data secure, governed, and accessible across Azure environments without added infrastructure complexity. It's the foundation to keep your unstructured data stable, even as your infrastructure evolves.

To see how much you could save while keeping your initiatives on track, visit nasuni.com I have a few that I run regularly. I have not gotten too far into the loops yet. I have a lot of stuff that I still just run manually. I have a few things that I run as routines or scheduled jobs.

But it's been, yeah, it's a lot of the way I would look at these LLM wikis. And even again, it's probably taking it further than what the LLM wiki did, like with your dashboard and HTML. I actually built my own custom Obsidian plugin that I have as a dashboard. So my dashboard sits right in Obsidian rather than a web page. How did you do that? I'm intrigued. I told Claude, so I actually told Claude to go build me a custom Obsidian plugin. And Claude has built the... Okay, built it as a plugin.

So it renders in the canvas. It built it as a plugin. So it renders as a canvas, like down the left side of Obsidian where you have your different icons. I now have an icon for my dashboard. And my dashboard has a bunch of cards on it. So this is where similar type of concept with mine, only I combined a bunch of stuff. I have multiple LLM wikis in the same vault because I have one for work. I have one for some personal stuff. I have a separate podcast one in my vault. Because sometimes for me, the overlap between a topic we talk about on a podcast and a blog

post, maybe I want to write for my website and a client that I'm working with, there's so much overlap there. I want them to be separate to an extent, but I also want to be able to correlate certain things across them. But then my dashboard has cards that are color-coded and labeled. So I have intelligent cards for my top to-dos, maybe things that are on fire, CRM type information, potential jobs, where jobs are in the pipeline. I have some podcast stuff where I pull in podcast stats so I can see trends with the podcast

downloads and episodes month to month. I have some personal stuff where I actually pull in stock market. I have a markdown file where I put stock tickers I want to track, and then I have a dashboard that every time I go to it, it'll go out and pull the latest stock prices and put it on the dashboard. But then I put little buttons on it that'll call a skill to go do analysis on that stock ticker. So I have my own analysis on stock tickers within my LLM wiki.

And I might have to explore this because one of the things... It's interesting. I am definitely juggling between Scout, which sits out there. It's a desktop app, right? It has to be running. So I've got Scout sitting out there running and... Wow, that hurt. Sorry about that. It was quite a whine by that dog, by that puppy. I've got Scout out there running. I have this website that kind of runs. It's just a pin tab in Edge for me. It's always my first tab sitting at the top. And then I have Obsidian. So having just one less thing, like I don't have to go to the website and I can just live

in Obsidian. Because I do this thing where I've been tempted to... Maybe I was thinking about it wrong. I was actually tempted to see if I could find a way to embed a Markdown viewer in an Obsidian view in my website. But I should just totally go the other way. The Canvas thing is very interesting. So I'm glad you brought that up. I've seen a bunch of people doing similar things recently. So if you're in VS Code and you're using, say, like GitHub Copilot chat in VS Code. So VS Code has plugins, all those kinds of things.

You can absolutely use your LLM to help you develop a plugin that draws on the canvas in VS Code. The other one is GitHub Copilot. So GitHub Copilot has a CLI. And there's a new GitHub Copilot app as well. And that app renders canvases in HTML as well. So I've seen some folks in my organization that are creating custom canvas apps for GitHub Copilot, the desktop app, because that's where they're living. And that's where they're doing their work out of kind of thing. So that's another super powerful concept is like kind of coaxing an LLM and getting it

to build the view that you want and the experience you want in the place that you work. So going back to your skills, I have a lot of similar skills. Like I have a daily review skill that will go back and look at all my meeting notes from the day. It'll look at my calendar. Mine actually goes into, I don't know if you've tried this, it'll go into my team's conversations. And I've told it to look for certain, there's certain teams that I'm in that are super chatty where only like five or 10% of the stuff is relevant to me.

And I'm like, go look through all of these chats and teams and give me, tell me where there's a chat related to a certain topic because that's the only thing I care about. So part of my review will actually analyze all my team's chats for me and tell me if there's anything I should go back and look at my team's chats or if I can just ignore all of them. I run a similar one. That's another automation or loop that I have out there running. So it looks across my inbox and my team's chats. And your team's. And it looks for emails from my manager, my skip, or people on my team.

Okay. Who report to me. And initially that was a, let's get out there and kind of grab everything, right? Like the signal to noise ratio was, it was really bad. It was pretty horrible because like anytime somebody chatted with me on my team, like they could ping me on the side and say, how was the weather today? And I would all of a sudden get this alert that popped up and said, Joe is, Joe is, I don't have Joe on my team, but Joe is asking you how the weather is today. Do you want to respond?

Oh no, that's, that's not important, right? So I definitely had to go and refine that thing in a similar way. I have a little bit of contextual filter, things like that. So it only alerts me now when folks are kind of pinging me with urgency or high priority items, particularly things that align to work streams that were kind of integrated on or that I need to know about what's happening. I've actually been thinking, and one of the things I would love to do is take it to the next level and just have it be able to respond back on teams. So like if somebody says, oh, hey, can you give me the latest update on feature ABC?

If I already sent out the status update for feature ABC this week, I just wanted to send them the doc or send them a link like, or send them like the subject of that email that went out or whatever that was. And that feels like low hanging fruit and something that I could go out there and automate fullness of time. I just haven't been sure if it's worth it yet. Like putting the effort in to have an additional contextual filter, definitely worth it. Like I said, like I was getting all sorts of like stuff, like very much along the lines of like, so-and-so is telling you the weather's nice today.

Like very important. Please respond. Like, no, I don't need to do that. I don't find mine too. It's just certain topics. But then in my dashboard, so I can run all of my skills from my dashboard. I can just click a button in the dashboard and it'll go open a bash prompt, a Claude session in the background and go run the skill in the background and I won't ever see it. So any skills that I don't have to interact with, then I have other skills where I know like I may go back and forth with them on or have more of a conversation with Claude.

Those, the button actually copies a prompt to my clipboard. So I just open up a Claude session and paste the prompt right into a Claude session. Interesting that you chose to go down that path. Everything out of my dashboard. Go open the harness, go open Claude and run the skill versus me where I was like, I'm going to go, I'm going to go have it write the script and have it be super deterministic so I can run that script from a skill or from a button or from here. Some of my skills do still run scripts. It's the same. I still have a lot of scripts where those skills, it's still deterministic.

It's a button, open a Claude session, run a script, run a skill. So one of the things maybe I was going at and something for folks to think about is, and one of the reasons I kind of go down this path quite a bit, is optimizing for token usage and context. So if you're using Claude as a harness, Claude has a fairly large system context already coming with it as part of the harness. And every time you're invoking a skill, you have to load that context no matter what. Like that's there. And so sure, you should have the advantages of cash tokens and things like that.

If Anthropik's doing their job right, if they're not doing their job right on a given day, or stuff does go sideways, you're really in this place where you're burning tokens. And this stuff does cost money, right? Like, I'd be curious if you're comfortable saying like, do you have a sense for how much yours costs to run over the course of a month? If not? I do other things in it. And I mean, I will say I am paying for Claude team now. So I'm paying for the team edition of Claude.

What's the team edition run? Which it's somewhere between, so they have like the Claude max, where it's the 5x or the 20x. I'm actually sitting at like 6.5x. They don't have a 5x or 20x. So I would say I'm equivalent to the 5x, but it's a little bit more. I'm paying $125 a month for Claude. So my token consumption input and output is 100 to 120. I don't have the hard average in front of me, but I would say I probably average. I probably average like $110, $115 a month, something like that, to go through and run

everything that I need to do. But that includes like the things like, hey, going and like bootstrapping the dashboard and things like that. I would expect like continued optimization as things come down. I was trying to find this the other day. Over the last 30 days, I can see things like product stickiness. This is connectors. I can see the skills. Most of these run in Claude code, usage, top skills, tool usage. I was just trying to look super quick if I could see how many gives me value.

I haven't found. I'd have to dig through here like how many tokens I use. It's not free. No, it's not free. But I can fit it all in there with room to spare because I run these skills, but I also use it all the time for writing PowerShell scripts or analyzing data, that type of stuff. Because I do, similar to you, I have a lot of these skills that have turned into Python scripts in the background as well, where the skill calls a script to do a lot of the work

for it. Python seems to be where it all lands. But a lot of what I do is very similar type of stuff that I built to you, except that I have multiple LLM wikis in that vault. And then I did the dashboard in Obsidian. It's been, at least from my end, super empowering because I get to work the way I want to work. I'm always frustrated at the friction of, here's Microsoft to do, and it does to do the way Microsoft wants to do them for that product or things like that. So I would love it if I could say, I could take this thing and package it up and give it

to you and it would work for the way you work. I don't think that's the case, nor do I think it's the point. The whole point is to have the way you work and the set of skills and loops and sub-agents or authentic flows and connectors and everything else that are out there to wire everything up and get it going and have a good set of outcomes for it. I really do feel like it helps accelerate the work that I do. I think it keeps me better grounded. I still have to go through a lot of context switching and I find it super helpful to have

a lot of the grounded notes and things just right there at my fingertips. Go ask a question so I can search in Obsidian or when I do need to use an LLM to iterate over it, great, here you go. I can hop in that directory and invoke my harness of choice and I'm off to the races. I did things a little differently too for my tasks because I have contractors working for me. So I want to make sure tasks are updated in Asana and some of that. So my dashboard, instead of managing tasks here, all of my skills and prompts and everything

ensure that Asana stays the source of truth. So if I see something on my list of my top to do or something that I need to get done, it actually links to Asana so I can click it, open it in Asana and I check it off there and then the dashboard will update. And same thing for on fire. I have two right now where there's an email and a Teams conversation. I have created my dashboard so I can click on those items that I've pulled in and it takes me right to the email or it takes me right to the Teams conversation.

I have one for blog posts that takes me right to the draft blog posts or to the client record in my CRM. So my dashboard, the way I've thought about mine, I've turned my dashboard and my LLM into an integration point so I can pull all the information from all the systems I want and built a central, dashboard command center that I can work from but then link out to all the different things that I need to respond to that pop up on it. I'm on a very similar path.

I have this kind of backlog of things that I want to go kind of integrate next. So one is, I mentioned we use Azure DevOps, ADO, for kind of tracking of epics and features and PBIs and things like that. I'd love to have that integrated view right there for, hey, I'm working on project XYZ, click on project XYZ, here's the ADO items for it, and here's their current state and status, things like that. The other one that I want to build is, at some point, I struggle with this stuff, I don't

know when I'm reinventing the wheel. This one feels a little bit maybe like reinventing SharePoint, but I've got these projects that exist. So when you do the Carpathia-style LLM wikis, you end up with artifacts and entities, you end up with these discrete templates for types of objects in your LLM wiki. So I've got projects and work streams and events and things like that. So for my work streams and my projects, there's these kind of canonical briefs that sit there that have all the detail about them.

And then they have all the wiki links to sublinks and things like that. But sometimes the sublinks they carry are links to documents, so like Word docs, Excels, PowerPoints. So I kind of want to view where I can say, hey, for Workstream ABC, or Workstream A, click on Workstream A, here's all the important documents for it. Like, here's all the contextual things and just like the little snippet and summary of it that's there. And then like the document, to be very clear, like the artifacts, the canonical artifact, not stored in my wiki.

Like that stuff's still IRM, it's got sensitivity labels, all that stuff. But the link to it is in my wiki. So now it's like one-stop shopping for me, where in the past, I would have been going to SharePoint and just going to kind of like the generic intranet search or even like Copilot search. I'm finding it to be kind of annoying because it's an LLM-based search. So you kind of start talking to it like an LLM and it goes sideways pretty quick. So I've got all this stuff sitting. Like the context is there. It's just about, like you said, integrating the views and kind of making it one-stop shopping

across all these various systems. Yeah, my project for my dashboard has a backlog markdown file in it where I just, when I come up with an idea, I throw an idea in there. So when I have some time or I'm sitting on the couch in the evening and I want to build a new feature on my dashboard, I have my whole backlog and I can pick one off the backlog to start building. Anything we miss. I'm sure this is one we're going to come back to because I feel like, to your point, you're always evolving it, changing it, modifying it as everything changes. And as you come up with ideas. The builder's itch comes in along the way.

And then the other thing I like about it is, so you not only get to go build these things, but like I said, you get to adapt them to the way you work. And you also get to adapt them to your taste. Like I said, I don't have the greatest reaction sometimes when I have to go into a Microsoft to-do or a planner, but it doesn't rub me nearly as much the wrong way going into this thing that I built for the way I guided an LM to build it for me, because there's no way

I was writing all that stuff myself to go and put it together. So it actually, it's like, it's a little bit more calming, right? Like while it's bringing in probably, like I said, more stuff. And if anything, like it's allowing me to over-function, but it's at least allowing me to over-function in like my way of over-functioning without like my head constantly on a swivel. That's the biggest advantage I've found of this and frustrations I've had other times that I'm like, how do I integrate Asana with Teams, with Outlook, in a way that works for

how I want to work? This has allowed me to pull that information and put some integrations in place that A, I can never develop or build on my own. And B, I can use products, but then tie them all together in a way that makes sense for the way I want to see information, work with information, manage my day. Et cetera. So that's the part I've enjoyed about this, to your point. It's, I can tweak it to me. So with that, just wrap it up. We went almost an hour today, Scott.

Sometimes we go a little bit longer. So I'm always interested in feedback from other folks. We've mentioned a couple of times and it's good to see people coming and commenting on Reddit or LinkedIn. I would love to hear what other folks are doing there. Like, what's your superpower, your unlock for LLM wikis? Like, are you using them? I think you and I are maybe a little bit more on like one end of the spectrum where we're like in CLI harnesses or a little bit more advanced thing.

Like, I've seen folks on definitely like YouTube using very much like Claude Cowork style kinds of things to go and execute these same workflows. So I'd love to understand where others are experimenting, what they're doing, where they're finding the unlock, and maybe what else we're missing that could be like the take it up another notch. We'll do the final tap thing. Like, does this go to 11? Yes, yes, it does. Yes. Let us know. We'd love to hear it. All right. Well, that's Scott. It's time for the weekend. Almost time for the weekend.

But yeah. Well, as always, thanks, Ben. Not a problem. Thank you. And we'll talk to you again soon. If you enjoyed the podcast, go leave us a five star rating in iTunes. It helps to get the word out so more IT pros can learn about Office 365 and Azure. If you have any questions you want us to address on the show or feedback about the show, feel free to reach out via our website, Twitter, or Facebook. Thanks again for listening and have a great day.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Episode 441: Microsoft Ignite 2026: Autopilot, Workspaces, and Spot Plus

    Nov 19, 202648 minAudio

    Welcome to Episode 441 of the Microsoft Cloud IT Pro Podcast. In this episode, Ben and Scott take a first pass at Microsoft Ignite 2026, starting with a question a lot of IT pros are asking: is the pace of change...

    Credit the people on this episode

  2. Episode 440: The J Factor: Bringing Your Experience to Generative AI

    Nov 5, 202638 minAudio

    Welcome to Episode 440 of the Microsoft Cloud IT Pro Podcast. Recorded live at the North American Cloud and Collaboration Summit in Branson, Missouri, this MS Cloud on the Road conversation has Ben sitting down with Jay...

    Credit the people on this episode

  3. Episode 439: Azure SRE Agent vs. the Azure Copilot Troubleshooting Agent

    Oct 22, 202638 minAudio

    Welcome to Episode 439 of the Microsoft Cloud IT Pro Podcast. In this episode, Ben and Scott dig into two Azure agents with confusingly different names: the Azure Copilot Troubleshooting Agent and the Azure SRE Agent....

    Credit the people on this episode

  4. Episode 437: Nasuni and AI-Ready Unstructured Data with Nick Burling

    Sep 24, 202640 minAudio

    Welcome to Episode 437 of the Microsoft Cloud IT Pro Podcast. In this episode, Nick Burling, Chief Product Officer at Nasuni, joins Ben and Scott to talk about getting unstructured file data ready for AI. Nick digs into...

    Credit the people on this episode

  5. Episode 436: Finding the Entra Tenants You Didn't Know You Had

    Sep 11, 202638 minAudio

    Welcome to Episode 436 of the Microsoft Cloud IT Pro Podcast. In this episode, Ben and Scott dive into Microsoft Entra Tenant Governance. Almost nobody runs one Microsoft Entra tenant. Mergers, divestitures, dev and...

    Credit the people on this episode