Latest / Microsoft Cloud IT Pro Podcast / Episode 429: Getting started with LLM Wikis
Transcript
- 0:03- Welcome to episode 429
- 0:06- of the Microsoft Cloud IT Pro podcast recorded
- 0:09- live on 05/29/2026.
- 0:12- This is a show about Microsoft three sixty
- 0:14- five and Azure from the perspective of IT
- 0:16- pros and end users, where we discuss a
- 0:18- topic or recent news and how it relates
- 0:21- to you.
- 0:22- In this episode, Ben and Scott dig into
- 0:24- the concept of LLM wikis, specifically
- 0:27- building
- 0:28- personal knowledge management vault using Obsidian,
- 0:31- markdown, and AI tooling like Cloud Code, GitHub
- 0:34- Copilot CLI, and Copilot Cowork.
- 0:37- Scott walks through how he wired up Obsidian
- 0:40- Web Clipper and an RSS dashboard plugin to
- 0:43- feed articles into his vault automatically,
- 0:45- then had the LLM help build a Python
- 0:48- script to automate the ingestion workflow and cut
- 0:50- down on token usage.
- 0:52- Then he expands into how Copilot Cowork fits
- 0:55- into this workflow as a scheduling harness with
- 0:58- practical examples of using it to pull email
- 1:00- from an inbox daily, convert messages to markdown,
- 1:03- and generate a prioritized to do list. Let's
- 1:06- dive into the world of LLM wikis.
- 1:11- I have a theory. Windows? No. I mean,
- 1:13- yes. I run too much stuff. Yes. But
- 1:15- no. Well, I think we all run too
- 1:17- much stuff. I've noticed lately that and I
- 1:20- think I talked about this a couple weeks
- 1:21- ago, the whole I've got cotypist on my
- 1:25- Mac doing type ahead prediction
- 1:27- with with a locally downloaded model.
- 1:30- I've got
- 1:31- Type Whisper
- 1:32- running in the background with a model loaded
- 1:34- in case I just wanna do text dictation.
- 1:36- Like, I I I I was thinking about
- 1:38- it the other day. At any given time,
- 1:41- local model running, I probably have at least
- 1:45- four instances of the same model or a
- 1:47- similar model all loaded up because these apps
- 1:50- don't have, like, a common cache interface. Like,
- 1:52- they don't use, like, hugging face cache or
- 1:54- things like that. They all download, like, locally
- 1:56- into their packages and their directories.
- 1:58- So at least
- 1:59- for me, I think a lot of the
- 2:00- things that I'm running into lately are actually
- 2:03- resource contention. Like, I haven't materially changed my
- 2:06- workloads
- 2:07- or my workflows, but I have a bunch
- 2:09- of the bunch more of these apps running
- 2:10- in the background that I didn't have before.
- 2:12- I was trying to do a little bit
- 2:13- of an audit, and it's not driving, like,
- 2:15- crazy memory pressure, crazy CPU pressure.
- 2:18- But, like, when some of these models load
- 2:21- and unload, like, you've got, like, I don't
- 2:22- know, like, Gemma four sitting there ready to
- 2:25- go. It's like, oh, hey. I had four
- 2:27- gigs of RAM that just got evicted for
- 2:29- this thing over here, and then it's gotta
- 2:31- reload rapidly for this like, it I don't
- 2:33- know. That that would be my guess is
- 2:35- we're all running, like, so much, like, local
- 2:38- AI slops that we don't even know it.
- 2:40- So little of it we can control. Like,
- 2:42- I don't remember if we talked about this
- 2:43- or when this came out, but Google Chrome
- 2:45- now downloading, what, a four gig
- 2:48- LLM model in the background as part of
- 2:51- the browser? The Gemma model in the background.
- 2:53- Yep. Yep. So that same model is the
- 2:55- same model that I'm using in Cotypeist, which
- 2:57- is
- 2:58- similar to what I'm so so, like, I've
- 3:00- got this app called Type Whisper.
- 3:02- Anyhoo. So Type Whisper is one of the
- 3:05- dictation apps. But rather than using, like, Whisperflow
- 3:09- or Super Whisper,
- 3:10- everything
- 3:11- processed locally on device, you can choose, like,
- 3:14- hey. I I wanna use Parakeet. But the
- 3:16- cool thing it does is it does
- 3:18- live processing, like live transcription,
- 3:21- and then it does post processing as well,
- 3:22- which is nice. So you can do that
- 3:24- thing where you're just saying, hey. I'd like
- 3:26- to make a list. Number one is this.
- 3:27- Number two is this. Number three is this.
- 3:29- And, normally, that would just be a big
- 3:31- flat wall of text going across. But if
- 3:33- you post process it and it gets smarter,
- 3:36- it actually renders, like, one, two, three, and
- 3:38- four into a list, and that's the text
- 3:40- that is expressed that you just dictated.
- 3:42- But I was thinking about that one the
- 3:43- other day. I was like, oh, this is
- 3:45- getting slower and slower. And I went and
- 3:46- looked, and I turned on, like, the pre
- 3:48- integration,
- 3:49- the Run It thing, the post integration.
- 3:51- So that one had to get kicked to
- 3:53- the curb. It was nice, but
- 3:55- it was it was slowing things down for
- 3:58- sure. Yeah.
- 4:00- All these things.
- 4:01- You know what other things well, nope. I
- 4:03- won't go down that rabbit hole. We can
- 4:05- talk about other things. But speaking of more
- 4:08- things we're running, you and I both started
- 4:10- playing with something within the last I don't
- 4:11- know when you started. I started, like, a
- 4:13- week and a half ago
- 4:15- playing with more things to run on our
- 4:17- computers.
- 4:18- Yes. And this is not maybe we should
- 4:20- preface this. This is not specifically, I would
- 4:22- say, Microsoft or March
- 4:24- Microsoft three sixty five or Azure episode. However,
- 4:27- it incorporates into our work with Microsoft three
- 4:30- sixty five and Azure and how we're starting
- 4:32- to track some of our resources. I have
- 4:33- a tie in I have a tie in
- 4:35- for you there. I'm actually gonna yeah. Do
- 4:36- you have a tie in for me? Yep.
- 4:38- Okay. So if you're listening and you're like,
- 4:40- there's not gonna be a tie in, Scott
- 4:41- has a tie in. I don't know when
- 4:43- it's gonna drop. So you better listen to
- 4:45- the whole episode
- 4:46- to figure out the tie in. Stick around.
- 4:48- Okay. I've been playing around
- 4:50- and been watching all these just videos fly
- 4:53- by and articles fly by
- 4:55- around the concept
- 4:57- of LLM wikis
- 4:59- and
- 5:00- leveraging
- 5:01- LLMs
- 5:02- to
- 5:03- manage a wiki resource for you. And, specifically,
- 5:06- like, I've been doing this with Obsidian Vaults.
- 5:08- So Obsidian Vaults work very nicely here. They're
- 5:11- marked down plain text. LLMs process plain text
- 5:14- pretty darn well. They understand markdown pretty well.
- 5:17- It's kinda, like, a a de facto standard
- 5:19- for authoring these things. Right? All the skills,
- 5:21- agents, you write everything. Everything you do is
- 5:23- in markdown.
- 5:24- So an LLM Wiki, if
- 5:27- you have a compendium of knowledge that you
- 5:30- wanna put together, and that could be around
- 5:32- anything. It could be maybe around a topic
- 5:34- that you're researching.
- 5:36- So I don't know. Let's say you want
- 5:38- to
- 5:40- in M 365 Land, do you want to
- 5:42- go explore the current single the current licensing
- 5:44- constructs and compare this e SKU to this
- 5:48- business premium SKU or to this add on?
- 5:50- You could go and take, like, all those
- 5:52- pricing pages and all the documentation around that,
- 5:56- and you could put it in just a
- 5:58- big dumping ground of markdown, just a raw
- 6:01- source. And then you can have an LLM
- 6:05- reason across that data
- 6:07- and extract information around it. So understand
- 6:11- what was the source data that came with
- 6:12- it, maybe what are the general concepts around
- 6:15- it, entities,
- 6:17- all these different
- 6:18- sorts of things that are out there.
- 6:21- So Andrew Kapathy, who's kinda, like, one of
- 6:25- the original or or original researchers over at
- 6:28- OpenAI,
- 6:29- I believe he's also the person who coined
- 6:31- the term vibe coding. So he's kinda pretty
- 6:33- close to
- 6:35- it. He's been in this space for a
- 6:36- little bit? Been in this space for a
- 6:37- little bit. Was recently hired by Anthropic
- 6:40- post
- 6:41- getting out there and putting some of these
- 6:43- things out. But I'll have a link in
- 6:46- the show notes for everybody to
- 6:50- a gist
- 6:51- that he posted.
- 6:53- And
- 6:54- and this gist
- 6:56- is just an LLM Wiki.
- 6:58- So you go and kinda read through this
- 7:00- thing, and it's just llm-wiki.md.
- 7:04- And
- 7:05- the core idea
- 7:06- of this whole thing, like I said, is
- 7:08- you kinda have a big dumping ground.
- 7:11- You could think about it as maybe, like,
- 7:13- a folder that you called raw or you
- 7:15- called ingest,
- 7:17- you called it clippings,
- 7:18- something like that.
- 7:20- And what you do is you keep putting
- 7:22- articles in there, reference materials,
- 7:25- but you put it in as markdown.
- 7:28- So this whole thing just becomes
- 7:30- a big bank of markdown that's sitting out
- 7:32- there. And
- 7:34- you direct the LLM that anytime
- 7:37- a new thing shows up in my raw
- 7:41- directory
- 7:42- to reason over it and then extract that
- 7:45- information, extract entities, extract core concepts, tell me
- 7:49- what the source of it is, and put
- 7:51- it all together.
- 7:52- And this is the nice thing about something
- 7:54- like Obsidian. So Obsidian, a personal knowledge management
- 7:57- tool, PKM,
- 7:59- But Obsidian
- 8:00- does
- 8:01- Wiki style links, so you can have all
- 8:03- these things start to relate to each other.
- 8:05- And then you get the kinda graph based
- 8:08- views that that come along with that as
- 8:11- well.
- 8:11- So it takes almost nothing to wire it
- 8:14- up, first of all, to get started. So
- 8:16- it's pretty much go out and take this
- 8:18- gist,
- 8:19- and, yes, you need to have access to
- 8:22- some kind of LLM interface. Now that LLM
- 8:25- interface
- 8:26- could be hang with me here. Could be
- 8:29- quad code, could be Gemini CLI. Well, actually,
- 8:32- Gemini do CLI just went away. Anti anti
- 8:34- gravity, but there's also anti gravity CLI.
- 8:37- It could be GitHub Copilot CLI, codex,
- 8:41- whatever that is. Or if you have access
- 8:44- to, I don't know, say, Copilot Cowork
- 8:48- and in in and the Frontier agent because
- 8:50- you're in the Frontier program,
- 8:52- you can also run an LLM Wiki inside
- 8:54- of Copilot Coburg. Have you tried this? I'm
- 8:56- curious to see. Did and it works? Oh,
- 8:59- yeah. It opens up a whole interesting world
- 9:01- of possibilities here. Hold on. Stick with me.
- 9:02- Now you have me curious.
- 9:05- If you're managing file storage today, you've probably
- 9:08- felt the pressure. Refresh cycles are harder to
- 9:11- plan, costs and complexity are increasing, especially as
- 9:14- file workloads scale across regions.
- 9:17- That's why Microsoft customers trust Nasuni. As an
- 9:20- Azure certified software,
- 9:22- Nasuni provides a global file system designed for
- 9:25- enterprise scale environments, keeping file data secure, governed,
- 9:29- and accessible across Azure environments without added infrastructure
- 9:33- complexity.
- 9:34- It's the foundation to keep your unstructured data
- 9:37- stable even as your infrastructure evolves. To see
- 9:40- how much you could save while keeping your
- 9:42- initiatives on track, visit nasuni.com/tco.
- 9:48- That's nasuni.com/tco.
- 9:54- So let's talk about wiring all this up.
- 9:55- So so you basically go out the and
- 9:59- I've done a couple of these now, like,
- 10:01- and been kinda playing around. And you've been
- 10:03- playing around with one to build, like, a
- 10:05- basically, an Obsidian Vault for us around the
- 10:07- podcast. We can go back and say podcast
- 10:09- stuff in. It's fascinating. When did we talk
- 10:12- about this? On what date this happened? How
- 10:14- many times did we have this guest on?
- 10:16- And now you've got all that information in
- 10:18- text in one place. Show me all the
- 10:20- episodes where we talked about a specific topic.
- 10:22- Like, it's been really interesting. Go show me
- 10:25- what topics. We've pulled analytics into it. Go
- 10:27- show me what topics
- 10:28- tied into
- 10:30- analytics. Yeah. All the things. Go take this
- 10:32- gist, and you can literally copy and paste
- 10:35- it. I don't know if this is how
- 10:36- you started with the one that you worked
- 10:38- on. I'm curious how you started, and I'll
- 10:40- tell you how I started. So I took
- 10:41- this gist, and I copied and pasted it
- 10:45- straight into a CLI. I took it and
- 10:47- put it right into actually, you know what
- 10:49- I did? Because I wanted to do this,
- 10:51- like, a freeway that anybody could get started
- 10:54- with. So I did it in cloud code,
- 10:56- but I did it in cloud code tied
- 10:58- to
- 10:59- OpenRouter
- 11:00- and the free models in OpenRouter. So I
- 11:02- wasn't using a paid anthropic or cloud plan.
- 11:06- I wasn't using, like, pay per token. I
- 11:08- did all this stuff for free. You could
- 11:10- do it free for free with anti gravity
- 11:12- and things like that. It gets a little
- 11:13- harder to do free later, like, once the
- 11:15- vaults get a little bit bigger, but you
- 11:17- can totally start out Really hard to do
- 11:19- free later. But you can start out for
- 11:20- free. So you take this gist
- 11:23- and you put it into your LLM interface,
- 11:26- in this case, let me say, like, Claude
- 11:27- code. Just paste it straight into the terminal
- 11:30- and add a line at the end
- 11:33- that says, I want you to turn this
- 11:36- into
- 11:37- a context file for this vault, or, like,
- 11:40- I want you to create the second brain
- 11:43- for Scott that manages
- 11:45- everything he finds on the Internet wants to
- 11:46- read about later. Right? So you give it
- 11:48- a little bit of context. That's it. You
- 11:49- just put in that one little sentence at
- 11:51- the end, fire it off. It will create
- 11:54- a Claude in the case of Claude, it
- 11:56- will create a claude dot MD. You did
- 11:58- this in GitHub Copilot. It'll create you, like,
- 12:00- a Copilot instructions dot MD. I recommend doing
- 12:04- a claud dot MD as a context file,
- 12:05- and then you can just pull it from,
- 12:07- like, LLM to LLM because all the various
- 12:10- or, I guess, from agent to harness to
- 12:13- harness. Because you can take a claud dot
- 12:15- MD and use it in claud. You can
- 12:18- take a clog dot m d and use
- 12:19- it in anti gravity CLI. You can take
- 12:21- it using GitHub Copilot CI. So so that
- 12:23- context file becomes very portable. So you do
- 12:26- that, and that's where you
- 12:28- start. And then it's gonna create your, in
- 12:31- my case, a clog dot m d.
- 12:33- And in my Cloud dot md, I said,
- 12:36- okay. I'm gonna go do a little bit
- 12:37- of light iteration here on some of the
- 12:40- instructions
- 12:41- that that I want this thing to have,
- 12:43- right, and have for its context.
- 12:45- So that was I've got a root folder
- 12:48- called clippings.
- 12:49- I believe in the original gist, it's called,
- 12:51- like, raw. I don't know. It didn't sit
- 12:53- well with me. I was like, yeah. I'm
- 12:54- gonna put Obsidian clippings in here, like Obsidian
- 12:56- web clipper, so I call mine clippings. And
- 12:58- then I added a couple different Wiki directories,
- 13:00- so it automatically does one for it does,
- 13:03- like, Wiki slash sources,
- 13:06- entities, Wiki slash concepts.
- 13:08- I wanted Wiki slash people because I knew
- 13:11- I was gonna be putting things in there
- 13:12- about, like, the podcast and you and I,
- 13:14- so I wanted it to extract, like, Ben
- 13:15- and Scott or Ben and Scott together,
- 13:18- kinds of things. And then
- 13:20- all you do is go out and start
- 13:22- putting some markdown in that in my case,
- 13:25- my clippings directory. So the way I did
- 13:27- that is with Obsidian Web Clipper. So Obsidian
- 13:30- Web Clipper is available on every platform. It's
- 13:32- there on iOS and Android, so you can
- 13:35- use it on your mobile. It's there
- 13:38- on Edge and Chrome and Safari.
- 13:40- So any kind of thing that you end
- 13:43- up looking at on the web, all of
- 13:45- a sudden becomes a resource that you can
- 13:47- take and you could say, hey. I wanna
- 13:49- reason over this later in my Wiki. So
- 13:52- maybe you start small. Go grab, like, five
- 13:54- articles that you're interested in and put them
- 13:57- in there. And then all you do is,
- 14:00- say, ingest my clippings or ingest my raw
- 14:03- stuff, depending on how you set up that
- 14:06- directory and how you framed out your ingest
- 14:09- flow. And then it'll go, and it'll just
- 14:11- build this stuff out. I went and iterated
- 14:14- over mine a little bit. So I I
- 14:16- because I was using
- 14:18- Obsidian Web Clipper uses a template when it
- 14:21- loads data in, so I wanted
- 14:24- all my reasoned over data to also follow
- 14:27- a similar template. So it had a title,
- 14:29- a publish date, a create a date, tags,
- 14:32- reference resources, things like that. So I actually
- 14:35- created a template dot m d file, and
- 14:38- then I went back and I modified my
- 14:40- cloud context. And I said, okay. Hey. When
- 14:43- you ingest things,
- 14:44- make sure that the outputs
- 14:46- use this template. Super easy. Right?
- 14:49- So and you just start right there.
- 14:52- And
- 14:54- you start to get this little bit of,
- 14:55- like, this is interesting.
- 14:57- And then you start wondering what else you
- 14:59- can do with it. But it runs away
- 15:01- from you very fast because
- 15:04- you're ingesting text and maybe not just a
- 15:07- couple lines of text, but let's take that
- 15:09- m $3.65 pricing example. Right? I I wanna
- 15:12- build something that can go reason over that.
- 15:14- You're going and grabbing every pricing page, potentially
- 15:16- the images off those pages and doing, like,
- 15:19- OCR and computer vision over them. So it's
- 15:22- pages and hundreds of kilobytes of text that
- 15:24- you're putting in, which is heavy on the
- 15:27- token side, both on input and output. Yes.
- 15:30- So this is where something like maybe like
- 15:32- OpenRouter and a free solution
- 15:34- starts to run away from you a little
- 15:35- bit. So I did OpenRouter. I got by
- 15:38- with maybe, like, my first 10 clippings in
- 15:41- doing this. And when I ran out of
- 15:43- stuff in OpenRouter, I was like, well, I
- 15:44- just wanna see, like, how far I can
- 15:46- push this free thing. So I went over
- 15:48- to Anti Gravity, which was recently released, and
- 15:51- that uses Gemini and Gemini Flash under the
- 15:54- hood. Let me tell you, anti gravity burns
- 15:57- some tokens. I think I burned through all
- 15:59- my tokens in anti gravity, like, on my
- 16:00- after, like, my first two ingests,
- 16:02- and it burned them for, like, six days.
- 16:05- You have no more usage for the next
- 16:06- six days.
- 16:08- But what you do there is
- 16:10- LLMs,
- 16:11- they really like to overthink things. Right? And
- 16:14- so they'll start to do things like, you'll
- 16:16- notice these patterns when, in my case, I
- 16:19- was saying ingest my clippings.
- 16:21- I noticed every single time, no matter
- 16:24- which harness I was using or which model
- 16:27- I was using, all of them were doing
- 16:29- similar things. They would say, like, oh, well,
- 16:31- I need to go figure out how to
- 16:34- list your ingest directory. So every single one
- 16:37- was saying, thinking thinking thinking. Okay. I'm gonna
- 16:40- list your your ingest directory. So I'm gonna
- 16:42- go do, like, an lsstar.md
- 16:45- kinda thing. You can automate all that stuff.
- 16:48- You don't have to have the LLM do
- 16:49- it. So the trick is you come back
- 16:51- or you or you don't have to harness
- 16:53- and the model do it. So you come
- 16:54- back and you say, hey. I've got this
- 16:56- context file now, and we've come up with
- 16:58- a workflow. How can I automate this workflow?
- 17:00- Can we make parts of this workflow scriptable?
- 17:03- Just ask the question.
- 17:04- And in my case, like, the one that
- 17:06- I've been really living in, I called it
- 17:08- my brain bank. I said, hey. This is
- 17:09- Scott's second brain.
- 17:11- So it created me a Python script called
- 17:13- brain dot p y.
- 17:15- And off brain dot p y, it created
- 17:17- a number of subarguments.
- 17:19- So it took my ingest flow, and after,
- 17:22- like, three ingests and it had all the
- 17:23- context, and it kinda knew what it was
- 17:25- doing repeatedly, it built me a brain dot
- 17:27- p y,
- 17:29- slash
- 17:29- ingest or, like, dash ingest command. And then
- 17:33- I said, oh, hey. You're doing this. Why
- 17:35- don't you build me, like, a like a
- 17:37- linting command to make sure that you're always
- 17:39- matching up to my template? Because I used
- 17:41- to use spinning a lot of tokens trying
- 17:43- to go back and read that template dot
- 17:44- MD and then get it back rear. So
- 17:46- it made me brain dot p y lint,
- 17:49- made me brain dot p y organize,
- 17:51- made me brain dot p y,
- 17:55- clean up any old clippings, maybe, like, organize
- 17:58- them into subfolders,
- 18:00- things like that, link them back to sources.
- 18:01- So over time, I've been on this journey
- 18:03- of saying, hey. All the stuff that you're
- 18:05- doing that I see you're repeating yourself doing,
- 18:08- but it's you as, like, the LLM thinking
- 18:10- about it, just go slam that in my
- 18:12- script. Just keep putting it in there, putting
- 18:14- it in there, putting it in there. So
- 18:15- then you're not burning all the tokens. Like,
- 18:17- it doesn't have to and then you tell
- 18:19- it, go update my context file. So if
- 18:21- I look at how I started from, let's
- 18:23- say, like, my ingest flow in my claw.md,
- 18:27- originally,
- 18:28- that was
- 18:30- something like
- 18:32- go
- 18:33- it said, hey. Your vault structure for clippings,
- 18:36- things like that. So my operations for ingest
- 18:38- was, like, step one, go list this directory.
- 18:42- Now step one has turned into research.
- 18:45- Run Python three scripts
- 18:47- brain dot p y scan ingest,
- 18:50- and then
- 18:51- take that output and then go check this.
- 18:54- Go run brain dot p y again, get
- 18:56- this list output, compare those two things. So
- 18:59- that way, it doesn't have to, you know,
- 19:01- start immediately every single time enumerating a whole
- 19:04- directory and running over it
- 19:06- and being very heavy. So you can start
- 19:08- to very quickly, like, drive it down into,
- 19:10- like, scriptability and less and that was the
- 19:12- part that I kind of enjoyed. Right? Like,
- 19:14- IT pro hat. Like, hey. Let's go get
- 19:16- this going. I think you were doing some
- 19:17- PowerShell scripts and things on your side,
- 19:20- on the Windows side. But you just kinda
- 19:23- keep going down that path, going down that
- 19:24- path, going down that path. And then
- 19:27- because you're doing this all in an Obsidian
- 19:30- markdown style, now you go and open it
- 19:32- up in Obsidian. Just go open your project
- 19:35- directory and what you're doing in there,
- 19:37- and you'll get access to
- 19:39- all the features of Obsidian. So that includes
- 19:43- tagging. It includes searching. It includes the PKM
- 19:46- with the knowledge graph.
- 19:48- Obsidian's
- 19:48- like a canonical markdown editor. It's really darn
- 19:51- good at it. You and I had been
- 19:53- chatting about one of the ways we do
- 19:55- research for the podcast is by following a
- 19:57- bunch of RSS feeds. What's the easiest way
- 19:59- to ingest the RSS feed? Well, I've been
- 20:01- going through, like, Feedly or Reader and opening
- 20:05- it in a website and then hitting the
- 20:06- Obsidian web clipper and doing things like that.
- 20:09- There's this really cool plugin for Obsidian called
- 20:11- RSS dashboard where you can just follow an
- 20:13- RSS feed. And as you follow, it's
- 20:17- it and now because it's an Obsidian vault
- 20:19- and it's just marked down, and I can
- 20:21- have it in my OneDrive
- 20:22- or you can have it in your Google
- 20:23- Drive or whatever,
- 20:25- It's actually kinda like Feedly, and that suddenly
- 20:27- it's a cloud sync service. Right? It's following
- 20:29- you around in your configuration.
- 20:31- But RSS dashboard is really cool because every
- 20:34- feed article that comes through, you can just
- 20:37- click a red or an unread button to
- 20:40- to flag it or get rid of it.
- 20:42- But if you click save on it, it
- 20:44- has a really nifty templating system. So I
- 20:47- was able to take my template dot m
- 20:49- d, which, again, was just based on an
- 20:51- Obsidian web clipper kinda default template. But I
- 20:54- was able to make that my template for
- 20:56- RSS dashboard
- 20:57- and have it every time I saved an
- 20:59- article from RSS,
- 21:01- it saves it as markdown over in my
- 21:04- clippings directory automatically. Right? Like like like, yes,
- 21:07- I have to click save on the article,
- 21:09- but it gives me this kind of, like,
- 21:10- one stop shop
- 21:12- ingest flow to get those things kind kinda
- 21:15- through.
- 21:16- So really interesting way to go and kind
- 21:20- of reason over personal data,
- 21:22- really any kind of project that you're doing.
- 21:24- I I would recommend constraining this stuff. Think
- 21:26- about it maybe like, if you're coming from,
- 21:28- like,
- 21:29- if you're coming from Claude Land, maybe think
- 21:31- about it as, like, project based. Or if
- 21:33- you're coming from, like, the cowork side of
- 21:35- the house,
- 21:36- maybe you use, like, notebooks in cowork, things
- 21:39- like that. Like, let me group these documents
- 21:41- together so that I can go do some
- 21:42- research on them kinda thing. I do find
- 21:45- that, like, I I I think that they
- 21:46- do a lot better when it's kind of
- 21:49- same knowledge area to reason over rather than
- 21:52- having a whole bunch of different things there.
- 21:54- Because, again, you could see eventually, you get
- 21:56- back to, like, really heavy context windows and
- 21:59- and things like that that happen along the
- 22:01- way. But so let's talk about now hey.
- 22:04- Great. You did this in our RSS feed.
- 22:06- Or, alright. You told me that, like, I
- 22:08- could dig an article from the web, and
- 22:10- I can clip it in here. Sure. Whatever.
- 22:12- Copilot cowork.
- 22:13- And Copilot cowork,
- 22:15- the Frontier agent. So the Frontier agent that's
- 22:18- out there has custom skills. So you can
- 22:21- take
- 22:22- the very same
- 22:23- kind of flow
- 22:24- around
- 22:25- having a context,
- 22:27- in this case, like, the claw.md.
- 22:29- You can actually take everything that's in your
- 22:31- claw.md
- 22:32- and put it in a skill.md
- 22:34- inside Cowork. And you go, oh, well, that's
- 22:36- not helpful. Like, you're all you're doing is
- 22:38- telling me that I'm just using co work
- 22:39- to process my clippings and things like that.
- 22:42- I actually recommend going a step further. Create
- 22:44- yourself skills
- 22:46- that do things like
- 22:48- take take the emails in my inbox
- 22:51- and
- 22:52- save them into my ingest directory as markdown,
- 22:55- and it'll take all your emails for that
- 22:57- day. Like, I've I've somebody on my team
- 22:59- did this, and it's super cool. He has
- 23:01- it so Cowork
- 23:03- looks at his inbox every day. It takes
- 23:06- all the items in his inbox,
- 23:08- and it turns them into markdown. And then
- 23:10- once they've been turned into markdown, they just
- 23:12- get marked as read and archived because now
- 23:14- they're sitting over in markdown, right, just in
- 23:16- in his OneDrive.
- 23:17- And then he runs an LLM Wiki style
- 23:20- kinda thing on top of it to
- 23:23- extract what are his priorities for the day.
- 23:25- Maybe he got an email from me or
- 23:26- he got an email from a coworker or
- 23:28- he's got specific projects that he flagged, things
- 23:31- like that. And then he has this set
- 23:34- of skills that iterate over it, and I've
- 23:36- been playing with this too. It actually works
- 23:38- really darn well.
- 23:39- You can take those things,
- 23:41- reason over them, create your own personal knowledge
- 23:44- management vault just inside of your OneDrive for
- 23:47- business. Obsidian's free. You're effectively getting cloud sync
- 23:51- because you got markdowns. So if you're in,
- 23:53- like, you're in, like, oh, hey. I need
- 23:54- to get to this on my work phone
- 23:56- or things like that, run Obsidian on your
- 23:58- work phone. Just point it at your One
- 24:00- OneDrive directory. It's okay.
- 24:02- It it'll all work. And then you get
- 24:04- to start thinking about, like, richer functionality within
- 24:07- Obsidian. So this was the thing that unlocked
- 24:09- it for me in the Copilot Cowork scenario
- 24:11- was not just, like, reason over what happened
- 24:14- today and maybe give me a, like, a
- 24:17- table. Like, tell me, like, the top three
- 24:19- people who emailed me or things like that.
- 24:21- But Obsidian has plugins
- 24:24- and
- 24:26- Wiki style markdown for things like task lists.
- 24:29- So you can actually have it, like, take
- 24:31- if you got 50 emails last night and
- 24:34- you walk in the morning,
- 24:35- you can have it take those 50 emails,
- 24:38- reason over them, do whatever it needs to
- 24:39- do, and create, like, your to do list
- 24:41- for the day. And then you actually get,
- 24:43- like, a clickable to do list, not in
- 24:45- Planner or anything like that. Like, you can
- 24:47- just put it in Obsidian and have it
- 24:49- there and ready to go, and it's very,
- 24:52- very portable. And because Copilot Cowork
- 24:55- has access to, like, your team's messages and
- 24:57- your inbox and all these things, You can
- 24:59- have it, like, mark your email as read
- 25:01- and archive it. You can have it organize
- 25:04- your email for it as it's doing this
- 25:05- extraction and pushing things out. You might have
- 25:08- it like, in my case, I get a
- 25:09- lot of, like, junk mail and newsletters.
- 25:12- So no matter how hard I try and
- 25:13- filter those things out, some always make it
- 25:15- into my inbox or my inbox other things
- 25:18- like that. So I can have that flow
- 25:20- come back and
- 25:22- clean that up as well. Right? Like, say,
- 25:24- hey. It notices there's a newsletter there. I
- 25:27- don't have to worry about creating a custom
- 25:29- rule for it anymore. I've just got Cowork
- 25:32- dumping that stuff off to a newsletter directory.
- 25:34- And then part of my ingest flow, because
- 25:37- it's still ingesting those things into my Wiki,
- 25:39- is telling me what newsletters I got that
- 25:41- day just in case there was something there
- 25:42- that I wanted to read or or things
- 25:45- like that. So you can absolutely take this
- 25:47- concept
- 25:48- and pull it over to
- 25:50- the world of things like Copilot, Cowork,
- 25:53- and all of that. Or if you wanted
- 25:56- to yeah. I've seen a couple folks do
- 25:58- this with
- 26:00- just something like GitHub Copilot and WorkIQ
- 26:04- or
- 26:04- Copilot
- 26:06- CLI,
- 26:07- Claude with the, like, the WorkIQ MCP and
- 26:10- things like that as well. The problem with
- 26:11- the WorkIQ MCP is I've had way better
- 26:14- luck with, like, Copilot coworkers saying, actually, like,
- 26:16- move this message from here to here. WorkIQ
- 26:19- is, like, glorified graph reader. So you end
- 26:22- up in some weird throttling scenarios and other
- 26:24- things that CoPilot CoWorks seems to generally skirt
- 26:28- around.
- 26:30- You're out of time. CoPilot needs to be
- 26:33- deployed ASAP, but is your tenant really ready
- 26:36- for it? Years of data, permissions, and users
- 26:39- lurk in its shadows, ready to be exposed
- 26:42- by AI.
- 26:43- Sharegate Protect sees it all, so you can
- 26:46- find the exposure risks, fix them fast, and
- 26:49- deploy AI with confidence.
- 26:51- Microsoft
- 26:52- three sixty five governance. We got this. Learn
- 26:55- more at sharegate.com/governance.
- 27:03- Have you tried it the other way where
- 27:05- once you have your Obsidian Vault and you're
- 27:07- syncing it down with OneDrive,
- 27:09- you save articles
- 27:11- into your raw folder in OneDrive
- 27:14- and have Copilot Cowork actually ingest all the
- 27:18- stuff you're saving locally
- 27:19- Yeah. And then just use
- 27:21- so you use Copilot Cowork for everything. Yes.
- 27:24- Use Copilot Cowork as the harness. As the
- 27:27- LLM, the harness. Yep. Where I said, hey.
- 27:29- When we started in the beginning, I said,
- 27:31- hey. I was doing this in Claude with
- 27:32- open router. Replace Claude and open router with
- 27:35- CoPilot CoWork,
- 27:37- the Frontier agent. And then you just switch
- 27:39- from doing everything in the CLI to going
- 27:42- into Cowork and starting
- 27:44- conversations
- 27:46- in Cowork in the browser for all the
- 27:48- ingesting and
- 27:49- processing that raw data. Yes to starting. But
- 27:53- once you're like
- 27:54- let's do, like, the crawl, walk, run thing.
- 27:56- So I think crawl is us walking in
- 27:58- there and typing all the time and kicking
- 28:00- these things off. Once you have a refined
- 28:02- workflow, Copilot Cowork lets you schedule this stuff.
- 28:05- And if you're structuring it in a way
- 28:08- where you're being principled, like in my case,
- 28:10- with my local
- 28:12- vault here that I was using for, like,
- 28:14- ingesting RSS feed articles and things like that.
- 28:17- So in that one,
- 28:19- I'm sitting in,
- 28:21- I'm sitting in a world of that workflow.
- 28:24- That's really like an on demand thing. Right?
- 28:26- Like, I don't always have clippings in there.
- 28:27- I don't need it to be looking once
- 28:29- an hour or anything like that. Emails are
- 28:31- coming in all day. So I can have
- 28:33- that skill
- 28:34- run every hour and do that ingest
- 28:37- hourly. So I don't even have to live
- 28:39- in Outlook. I can just go live in
- 28:41- Obsidian. It's kinda kinda kinda nice. The other
- 28:44- thing too is
- 28:45- I talked about the concept of, like, taking
- 28:48- some of the heavier operations and going and
- 28:50- putting those in scripts and things like that,
- 28:52- you can still do that in co work
- 28:53- because co work can execute local flows.
- 28:57- Or the other approach that you could take
- 28:59- would be
- 29:00- to really think about maybe rather than, like,
- 29:02- scripts and functions
- 29:04- to just skills and functions.
- 29:06- So having kind of the building blocks sitting
- 29:08- inside of composable skills
- 29:10- that then you can stack per scheduled task
- 29:13- or whatever you wanna do on on on
- 29:15- the CoWork side. Interesting. There was another question
- 29:17- I was gonna ask about this, and I
- 29:19- can't remember what it was. Adjusting, scheduling,
- 29:22- oh,
- 29:23- have you compared
- 29:25- so this has always been a debate too,
- 29:27- right, is how
- 29:28- Copilot does in terms of responses versus something
- 29:32- like Claude or
- 29:34- any of those. Have you compared the LLM
- 29:36- that gets built or the Wiki that gets
- 29:38- built by Cowork versus a Wiki that gets
- 29:40- built by Claude
- 29:42- just to see
- 29:43- side by side
- 29:45- how I know CoWork is different because a
- 29:46- lot of it I don't know that there's
- 29:48- as many system prompts injected into
- 29:51- CoWork
- 29:51- given that things like DLP don't even work
- 29:54- in it versus the normal Copilot. I feel
- 29:57- like the system prompts are are still there.
- 30:00- You do have
- 30:01- depending on your tenant configuration,
- 30:03- you do have model selection or at least
- 30:06- model I would call it model selection light.
- 30:08- Like, in my case, I can choose a
- 30:10- couple variants of GPT
- 30:12- and a couple
- 30:13- anthropic models.
- 30:15- You do get much richer model selection when
- 30:17- you're over on a CLI
- 30:19- or, like, a GitHub Copilot chat kind of
- 30:22- experience.
- 30:23- But the nice thing is
- 30:25- you're talking one vault that just lives out
- 30:27- or a series of vaults. It's all just
- 30:29- text, right, that that lives out here in
- 30:31- a side card. So you're not restricted
- 30:34- to
- 30:35- just Copilot co work or just GitHub Copilot
- 30:39- CLI. You could augment this. Yeah. You can
- 30:41- augment this and kind of go in. So,
- 30:43- like, for me, I tend to live in
- 30:44- the terminal.
- 30:46- So I'm not going in, like, during the
- 30:49- day. I don't actually, like, worry about going
- 30:52- into cowork and to ever typing a question
- 30:55- in a cowork. Cowork's a harness for me
- 30:57- to go and run skills, do the extraction,
- 31:00- kinda Right. Compound value and compound knowledge over
- 31:03- the course of a day, a week, a
- 31:05- month, things like that.
- 31:07- But when I'm going in and I'm going
- 31:09- across
- 31:10- and I'm saying, hey. Look at this stuff.
- 31:12- I'm typically doing that through
- 31:14- the CLI or, like, just just, like, GitHub
- 31:18- Copilot chat, things like that. Now the other
- 31:21- interesting thing that I found here is
- 31:24- it starts to make the data that you
- 31:27- extract
- 31:28- a little bit more deterministic.
- 31:30- So
- 31:31- it's a weird black box when you go
- 31:34- into Copilot
- 31:35- or you go into co Copilot cowork, and
- 31:37- you say, take a look at all my
- 31:39- emails from today
- 31:41- and
- 31:42- extract StackRank my top three priorities. Right? You
- 31:45- never know. Is it actually gonna search all
- 31:47- my emails for the day? What is what
- 31:48- does it consider an email for a day?
- 31:50- Is it only looking at my inbox? Is
- 31:52- it looking at my inbox plus my inbox
- 31:54- other? In my case, I run inbox inbox
- 31:56- other, and I do the Scott Hanselman inbox
- 31:58- cc things. I got I really got three
- 32:00- inboxes. Right? So I never know, like, what
- 32:02- it's looking at and what it's doing. But
- 32:04- by extracting the data and pulling it out
- 32:07- and sidecarring it, when I say go look
- 32:09- at today, today is a very real thing.
- 32:12- It's a discrete set of data that sits
- 32:14- out there and is available.
- 32:17- And and I found that accuracy has kind
- 32:20- of improved there. I have, like, far more
- 32:23- confidence that it
- 32:25- can can do what it needs to do.
- 32:27- Have you run into
- 32:29- anything,
- 32:30- like, using co work to your point earlier
- 32:32- about it can be expensive? Sorry. I think
- 32:34- I have a meeting up.
- 32:41- This episode is brought to you by Trusted
- 32:43- Tech. If you're managing Microsoft three sixty five,
- 32:46- you already know licensing isn't simple.
- 32:49- E three versus e five, the brand new
- 32:51- e seven licenses,
- 32:52- Copilot security bundles, and with price changes coming
- 32:55- in July
- 32:56- 2026, the stakes are getting higher. Trusted Tech
- 32:59- helps IT leaders make sense of their Microsoft
- 33:02- three sixty five environment with a free licensing
- 33:05- consultation.
- 33:06- Their engineers analyze what you're licensed for and
- 33:09- what you're actually using and where you can
- 33:11- optimize, whether that means consolidating
- 33:13- security tools, preparing for Copilot, or eliminating wasted
- 33:17- spend. Plus, they offer proactive
- 33:20- and reactive support, which has been awarded yet
- 33:23- another solution partner designation from Microsoft
- 33:26- for support services.
- 33:28- If you want a clear, data backed plan
- 33:30- that has helped over 7,500
- 33:33- subscribing
- 33:34- customers
- 33:35- save up to 20% on Microsoft three sixty
- 33:37- five before your next renewal,
- 33:39- visit trustedtech.team/mscloudprom365
- 33:46- and schedule your free consultation
- 33:48- today.
- 33:54- Okay. I have a meeting at four. Where
- 33:56- you run into some of that quotas and
- 33:58- stuff? Like, I did mine. I did the
- 33:59- podcast. It was granted, it was 400 and,
- 34:03- what, 28 episodes.
- 34:05- I had all the transcripts,
- 34:07- so it was some big text documents, like
- 34:10- thirty minutes of talking turns into a lot
- 34:11- of text. And I was using
- 34:14- Claude, and I was doing the
- 34:17- Claude Max
- 34:19- five x. So this is the $100 a
- 34:21- month plan on Claude, and you get, like,
- 34:23- a five hour quota. When I started ingesting
- 34:26- podcast episodes,
- 34:28- and even I experienced this even. I tried
- 34:30- to put a bunch of stuff in Python,
- 34:32- do as much as I could still burn
- 34:33- through my five hours of quota in, like,
- 34:35- twelve minutes.
- 34:36- It was crazy how quick I burned through
- 34:39- tokens.
- 34:40- Did you
- 34:41- how much did you test cohort to see
- 34:43- if during the ingest or have you noticed
- 34:45- do you start getting
- 34:47- throttled or limited where it's like, okay. Go
- 34:50- use something else to your point for that
- 34:52- initial if you're gonna ingest a bunch of
- 34:55- stuff, or do you have to start being
- 34:57- careful about how often or how much you're
- 34:59- ingesting yet?
- 35:00- I can imagine it might you might start
- 35:02- getting limits eventually. Yes. You do have to
- 35:05- be careful. So so I think the
- 35:07- interesting thing here and the callout that I
- 35:09- keep coming back to is
- 35:11- and you're nailing it, is the context windows
- 35:14- and the token usage, both on inputs and
- 35:17- output, does get very heavy.
- 35:19- So
- 35:20- I I think this is someplace
- 35:23- where I would be inclined to think that
- 35:25- folks in our community would be able to
- 35:27- wrangle this pretty well given kind of their
- 35:30- backgrounds as
- 35:31- developers,
- 35:32- IT pros, more experienced end users. You have
- 35:36- to think a little bit like that scriptwriter
- 35:38- or that programmer.
- 35:40- So your job isn't just
- 35:42- to have a big context window that you're
- 35:44- firing off every single time. It's kinda like
- 35:46- ruthless optimization.
- 35:48- Get this down to here. Get this down
- 35:49- to here. Get this down to here. So
- 35:50- so what I realized was
- 35:53- when I went and did this on my
- 35:55- own, just kinda same thing over in, like,
- 35:58- Claude with open router, blah blah blah, and
- 36:00- I too had to switch over to my
- 36:01- paid Claude plan. Don't get me wrong. Like,
- 36:04- it it it didn't get me all the
- 36:05- way there. It's expensive.
- 36:07- I think one of the reasons it didn't
- 36:08- get me all there all the way there
- 36:10- because I didn't think about, like, a programmer
- 36:12- from the start.
- 36:13- And I I just kind of approached as,
- 36:15- like, the spirit of what mister Kabathea put
- 36:18- out there of, hey. Here's an LL on
- 36:20- Wiki, and you give it a simple set
- 36:21- of instructions,
- 36:22- which I I recognize a lot of these
- 36:25- folks who are putting these things out there.
- 36:26- They do live in a world of unlimited
- 36:28- tokens. Right? They work at Anthropic, and they
- 36:31- work at OpenAI
- 36:33- and things like that, and not all of
- 36:35- us have that luxury.
- 36:37- So you do have to kinda think about
- 36:38- that and start to refine things. I have
- 36:41- noticed
- 36:42- I I will say so I I mentioned
- 36:44- I've built a couple of these now.
- 36:47- So the first one, I really went down
- 36:49- that refine refine path. Now I have this
- 36:53- really solid
- 36:54- templated thing that already has the scripts from
- 36:57- day zero. So when it starts,
- 37:00- it's already starting in a more optimized state
- 37:03- rather than having to start from scratch every
- 37:05- single time. So you kinda carry that knowledge
- 37:07- forward, and that helped me a whole bunch
- 37:10- when I was thinking about the Copilot Cowork
- 37:13- side of it. And then it really started
- 37:15- to click for me when I saw one
- 37:16- of my coworkers
- 37:17- who was in the mixed modality of Copilot
- 37:20- Cowork
- 37:21- for things like email extraction,
- 37:23- but then they were on they were still
- 37:25- on the CLI and doing, like, native Obsidian
- 37:27- Web Clipper things and all that stuff. Right?
- 37:29- Like, they were kinda using multiple tools
- 37:32- all to have the data coalesced
- 37:34- into
- 37:35- a single source.
- 37:37- And, effectively, like
- 37:39- I mean, Obsidian's markdown, you can make it
- 37:41- look pretty, all those things. It has checkboxes
- 37:44- and tables, and you can put images in.
- 37:47- Like, is it as rich text as something
- 37:50- like OneNote is? Absolutely not. But when you
- 37:53- start to get this flexibility of having, like,
- 37:55- multiple ingest points, multiple output points,
- 37:59- and kind of the capabilities that come with
- 38:01- the plugins in Obsidian,
- 38:03- all for free,
- 38:04- it kinda just works. Right? It's a very
- 38:08- it's it's a very cool system.
- 38:10- And once you start to go down this
- 38:12- path, like, for me, I, like, I'm one
- 38:14- of those people, like, I hate to admit
- 38:16- it, but I can live in email. Right?
- 38:18- It's pretty easy to get 50 emails overnight,
- 38:20- then you just get stuck in it. You're
- 38:22- like, I don't really live in email as
- 38:24- much anymore.
- 38:25- Email has just turned into I've got this
- 38:28- other thing over here that's taking a look
- 38:30- at it. I had to refine it. I
- 38:31- had to take some time put time into
- 38:33- automating that process and
- 38:35- automating myself out of that job of living
- 38:38- in email,
- 38:39- but it
- 38:40- it helps on the other side. And I
- 38:42- think, like like, the thing that really, like,
- 38:43- kicked my butt was
- 38:45- this team teammate of mine who,
- 38:49- he lives in the chaos that I live
- 38:51- in
- 38:52- and the rest of our team lives in.
- 38:54- And he has this
- 38:56- super awesome
- 38:58- I hesitate to call it a demo because
- 38:59- he showed his real workflow for how he
- 39:01- manages all this stuff. He put it all
- 39:03- in a Git repo for us so we
- 39:04- could just pick up skills
- 39:06- and run with them. And he's one of
- 39:09- the first people that I'd seen who successfully,
- 39:12- like,
- 39:13- navigated
- 39:14- inbox zero
- 39:15- at Microsoft
- 39:16- without feeling like he was missing out on
- 39:19- something or feeling like you're not just doing,
- 39:21- like, the
- 39:22- control a, delete, or control a, archive, and
- 39:25- I'll hope they email me again tomorrow
- 39:27- kind of thing.
- 39:29- It it was also very interesting to me
- 39:30- to see the way that
- 39:31- this individual managed tasks in Obsidian, like, how
- 39:35- they weren't trying, like, a forcing function of,
- 39:37- like I've seen a lot of people use,
- 39:38- like, Copilot cowork and and integrate it with
- 39:41- Planner and things like that. They put together
- 39:43- their Kanban boards, and I I feel like
- 39:45- the process runs away from you. It gets
- 39:47- a little too fancy at some point. I
- 39:49- I think
- 39:50- we're much better served when stuff becomes nice,
- 39:53- simple. Like, here's a to do list. I've
- 39:55- got a checkbox. I check that thing off,
- 39:57- and it gets a strike through it.
- 39:59- And you get to feel good about that.
- 40:01- Like like, there's a little there's a little
- 40:03- rush of endorphins,
- 40:04- right, when you when you check that checkbox
- 40:07- and have that
- 40:08- sense of accomplishment that's out there. And because
- 40:11- this stuff is all text, you can do
- 40:13- things because, again, LLM Wiki style Obsidian, you
- 40:16- don't even have to check the checkbox.
- 40:19- I would imagine I haven't done this, but
- 40:21- I think this would be totally doable. Like,
- 40:23- I could go into GitHub Copilot CLI, and
- 40:25- I could say,
- 40:27- my my task for today
- 40:29- to email Jack is done. Can you just
- 40:31- check that off for me? And I would
- 40:32- go check it off for me. Right? I
- 40:34- don't have to
- 40:35- worry about actually even opening a vault or
- 40:37- looking at anything and
- 40:39- running that way.
- 40:41- Like, the the elephant in the room, which
- 40:43- I don't think anybody could address at this
- 40:45- point,
- 40:46- and, I mean, it it is there, is
- 40:50- your like, this stuff's all plain text. Right?
- 40:52- You are extracting it out and and kinda
- 40:55- putting it down to your,
- 40:56- your OneDrive and things like that. So stuff
- 40:59- like
- 41:00- IRM and MIP
- 41:02- really don't apply to markdown files.
- 41:04- So it certainly, like, don't break,
- 41:07- like, organizational constraints
- 41:09- or things like that. Pay attention to sensitivity
- 41:11- labels and all that stuff. Those types of
- 41:13- things? Well, Cowork doesn't pay attention to sensitivity
- 41:16- labels, so why should I? No. It does.
- 41:18- It can't read any MIP file. It's kinda
- 41:20- useless. So Yeah. If you're actually encrypting it.
- 41:23- Yeah. No. I'm gonna have to play with
- 41:24- this too because I'm doing all my meeting
- 41:26- I'm doing this with meeting notes, and I
- 41:27- created a Wiki for meeting notes. So, like,
- 41:29- meeting notes out of people, so I have,
- 41:31- like, people and topic and all of that
- 41:33- meeting notes. It would be really interesting to
- 41:36- take this into Cowork and actually, like, now
- 41:37- I can correlate meeting notes with emails, with
- 41:41- all of that. How about just meeting notes,
- 41:43- but meeting notes with transcripts? So go create
- 41:45- yourself a skill, load it into co work,
- 41:48- have it extract your meetings transcripts
- 41:50- from the prior day, ingest them in, and
- 41:53- then go run this over them. Right? Like
- 41:54- like, that's just a super Yeah. I see.
- 41:56- I'm already doing that part of it. Now
- 41:58- we just have to tie in the emails.
- 41:59- So it is. This it's it is definitely
- 42:02- interesting to start tying all this together.
- 42:05- Again, those are the two I've done is
- 42:07- emails. I did the podcast.
- 42:09- I've started to do just references, but I'm
- 42:10- gonna go play with blog posts and figure
- 42:12- out how to incorporate some of that other
- 42:14- stuff in with Cowork. But I just gotta
- 42:16- chat that my other meetings started. World of
- 42:18- possibilities here. Absolutely. So I'll put a bunch
- 42:21- of links in the show notes to kinda
- 42:23- the original gist and things that came out
- 42:25- here. Yeah. There's some good YouTube videos out
- 42:27- there we can post in the show notes
- 42:29- too,
- 42:30- walking through how different people have done this
- 42:32- and showing how some of it's set up
- 42:33- too. I haven't seen a lot of people
- 42:35- who have done
- 42:36- Copilot Cowork. There's a lot of Claude. There's
- 42:39- a lot of good reference implementations.
- 42:41- Yeah. Claude's out there. Super easy to, like,
- 42:44- Google, Bing, DuckDuckGo, Kagi, whatever your search engine
- 42:47- choices.
- 42:48- I will put out some examples for those
- 42:50- who are in the Microsoft ecosystem.
- 42:52- So So there's a pretty good one out
- 42:53- here that I'll put in. It's called the
- 42:54- GitHub Copilot Wiki, an AI powered second brain
- 42:58- template.
- 42:59- So it that's it's kinda like a forkable
- 43:01- template that you can start with. So it
- 43:04- already has the wiki structure.
- 43:07- It has a template dot m d.
- 43:10- It has a lot of these constructs, right,
- 43:12- that I talked about here, and it can
- 43:14- just kinda, like, bootstrap you if you didn't
- 43:16- want to
- 43:18- if you didn't wanna start start for yourself
- 43:24- links a template too. So we'll drop those
- 43:25- in the show notes. Cool. Perfect. Well, thanks,
- 43:28- Scott. Yeah. Enjoy your weekend, and I'm gonna
- 43:30- go jump on my next meeting. Perfect. Me
- 43:32- too.
- 43:34- Life life is on. Alright. Thanks, Ben. Alright.
- 43:36- We'll talk to you next time. Yep. Bye
- 43:37- bye, Scott.
- 43:40- If you enjoyed the podcast, go leave us
- 43:42- a five star rating in iTunes. It helps
- 43:44- to get the word out so more IT
- 43:46- pros can learn about Office three sixty five
- 43:48- and Azure.
- 43:49- If you have any questions you want us
- 43:51- to address on the show or feedback about
- 43:53- the show, feel free to reach out via
- 43:56- our website, Twitter, or Facebook.
- 43:58- Thanks again for listening, and have a great
- 44:00- day.