20:04Chase AI · Tutorial
Claude Code Has Evolved
Why the skill backbone — not the dashboard — is where all the real value in a Claude Code Agentic OS lives.
May 14thThe argument in one line.
A folder of raw markdown notes becomes a living personal knowledge base once narrow agent skills automatically tag, cross-link, and rewrite it into wikis and visualizations, so raw capture can stay effortless.
Read if. Skip if.
READ IF YOU ARE…
- You take voice notes or web clips regularly but never organize them, and want an automated system to make that raw material useful later.
- You already use Obsidian, Notion, or plain markdown for notes and want to layer agent automation on top without switching note apps.
- You're comfortable pointing a coding agent (Claude Code or similar) at local files and want a concrete, skill-based workflow to copy.
- You want to explore your own notes and research the way you'd click through Wikipedia, following related links between ideas.
SKIP IF…
- You want a plug-and-play SaaS product — this is a DIY system built from markdown files, agent skills, and a cloud sandbox, not an app you install.
- You don't currently take any notes — the system only creates value once there's raw material to enrich.
The full version, fast.
Most notes rot in a folder because organizing them feels like a chore. The fix is to stop organizing and let an agent do it: capture ideas by voice or web clipper as plain markdown, then run a narrow 'enrich note' skill that tags each note from a controlled vocabulary, attaches a source URL, and links related notes. From there, pasting a single public gist (Andrej Karpathy's LLM wiki pattern) into a coding agent generates a full topical wiki — people, concepts, sources, index — from any pile of notes. A scheduled cloud loop (sync vault down, run the skill, sync back up, stamp an enrichedAt timestamp) keeps it current without a laptop staying open. The payoff is that once notes are just structured files, an agent can also generate HTML visualizations on request — a note-taking burndown chart and a graph view of linked ideas — turning a static note pile into something explorable.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 01:40
01 · Intro
Cold open with the 'junk drawer of notes' hook, framing the promise: agents can take unorganized notes and turn them into self-updating wikis and visualizations.
01:40 – 05:20
02 · Voice transcription tools
Demos his own note app Hubble, then free on-device speech-to-text tools Handy and VoiceInk as ways to capture ideas faster than typing, plus the Obsidian web clipper for pulling in web pages and full YouTube transcripts.
05:20 – 08:34
03 · Enrich notes skill
Introduces the 'enrich note' agent skill: adding tags from an existing controlled vocabulary, pulling a canonical source URL via web search, and linking related notes in WikiLink format, all run directly against markdown files.
08:34 – 11:30
04 · Build LLM wikis
Shows generating full topical wikis (index, people, concepts, sources pages) by handing a coding agent Andrej Karpathy's public 'LLM wiki' gist, applied across several of his real note vaults (AI hardware, a religious-history project, Claude Code observations).
11:30 – 16:02
05 · Automate everything
Builds a scheduled cloud automation using Oz: a Docker container syncs the Obsidian vault down via the Obsidian CLI, runs enrich-note or refresh-wiki skills, then syncs results back up, stamping processed notes with enrichedAt so reruns skip finished work.
16:02 – 18:37
06 · Visualizations
Generates two HTML visualizations directly from notes: a GitHub-style note-taking burndown/heatmap, and a space-themed force-directed 'thought constellation' graph showing topic clusters and links between notes.
18:37 – 19:12
07 · Conclusion
Recaps the homework (get notes into markdown, run the skills, get creative with visualizations) and plugs his own note-taking app Hubble as free and open source.
Lines worth screenshotting.
- Voice capture is roughly twice as fast as typing — about 200 words per minute spoken versus under 100 typed — making it the highest-leverage way to get raw notes down.
- On-device speech-to-text models have gotten good enough that a paid transcription subscription isn't necessary; free tools like Handy run local models with zero ongoing cost.
- A web clipper can pull a full YouTube transcript with timestamps straight into a markdown note, so a video's entire content becomes searchable raw material with one click.
- An agent skill instructed to be 'reluctant' to invent new tags — reuse the existing list, only add one when nothing fits — is what prevents a tagging system from sprawling into chaos.
- Because notes are plain markdown, no database, API, or custom connector is needed for an agent to enrich hundreds of files directly in a folder.
- A single public gist (Andrej Karpathy's LLM wiki pattern) is sufficient instruction for a coding agent to generate a full topical wiki — people, concepts, sources, index — from a folder of notes.
- A generated wiki can surface genuinely unexpected connections, like a jazz musician's commentary on AI showing up inside an 'AI and engineering' cluster, that a manual folder structure would never reveal.
- A wiki built entirely from raw notes is disposable and regenerable — it can be deleted and rebuilt any time the source material changes, since it isn't the source of truth.
- Stamping each processed note with an enrichedAt timestamp lets a scheduled agent skip already-touched notes, so recurring runs stay fast instead of reprocessing an entire vault.
- Obsidian Sync, at roughly $4/month, is a simpler way to give a cloud sandbox access to a local notes vault than setting up Git, though Git works if notes are already tracked there.
- Once notes and their metadata are just structured files, an agent can generate an HTML visualization directly from a plain-language request — no dashboard framework required.
- A GitHub-style burndown/heatmap of note-taking activity turns an abstract habit question ('am I writing consistently?') into something immediately visible.
Markdown notes plus narrow agent skills becomes a self-updating wiki.
WHAT TO LEARN
A folder of raw voice notes turns into a living, cross-linked knowledge base once a few narrow agent skills — enrich, build-wiki, automate — run against it on a schedule.
02Voice transcription tools
- Voice capture is roughly twice as fast as typing — about 200 words per minute spoken versus under 100 typed — so recording raw ideas by voice is the highest-leverage way to build a knowledge base.
- On-device speech-to-text has gotten good enough that a paid subscription tool isn't necessary; free options run local models with no ongoing cost.
- A web clipper converts any page — including a full YouTube transcript with timestamps — into markdown with automatic front matter, so raw capture stays fast.
03Enrich notes skill
- A single agent skill can enrich a raw note by adding tags from an existing controlled vocabulary, pulling in a canonical source URL via web search, and linking related notes — all from one instruction.
- Keeping an agent 'reluctant' to invent new tags — reuse the existing list, only add one when nothing fits — is what prevents a tag system from sprawling into unusable chaos.
- Because notes are just markdown, no database, API, or custom connector is needed — an agent can enrich hundreds of files by working directly in the folder.
04Build LLM wikis
- Feeding a coding agent a single public gist is enough instruction for it to generate a full topical wiki — people, concepts, sources, and an index page — from a folder of notes.
- A generated wiki surfaces genuinely unexpected connections that a manual folder structure would never reveal, because the agent cross-references content rather than following a fixed taxonomy.
- The wiki is disposable and regenerable — since it's built entirely from the raw notes, you can throw it away and rebuild it any time the source material changes.
05Automate everything
- A recurring knowledge base needs three ingredients: a synced copy of your notes in a container, an agent skill to run, and a way to push results back — everything else is orchestration.
- Obsidian Sync, at roughly $4/month, is a simpler way to give a cloud sandbox access to a vault than Git, though Git works fine if notes are already tracked there.
- Stamping each processed note with a timestamp lets a scheduled agent skip notes it already touched, so daily or weekly runs stay fast and don't reprocess the whole vault.
06Visualizations
- Once notes and metadata are just structured files, an agent can generate an HTML page from them directly — no dashboard tool or coding framework required, just ask for a webpage.
- A GitHub-style burndown/heatmap of note-taking activity turns an abstract habit question into something immediately visible.
- A force-directed graph view of notes and their links exposes topic clusters at a glance, which is a faster way to decide what to explore next than clicking through individual notes.
Terms worth knowing.
- Enrich note skill
- A named agent instruction set that adds tags, a source URL, and related-note links to a single markdown note, skipping notes that already carry an enrichedAt timestamp.
- LLM wiki pattern
- A publicly shared prompt (via a GitHub gist by Andrej Karpathy) that instructs a coding agent to generate a topical wiki — people, concepts, sources, and an index page — from a folder of notes.
- Frontmatter
- The YAML metadata block at the top of a markdown file (tags, source, enrichedAt, etc.) that agents read and write without touching the note body.
- Oz
- A cloud platform used in the video to run scheduled agent automations inside a Docker container, syncing a notes vault down, running a skill, and syncing changes back up.
- Obsidian Sync
- A paid (~$4/month) service that keeps a local Obsidian vault synced to the cloud, used here so a remote sandbox agent can read and write the same notes.
- Thought constellation
- The video's name for a force-directed graph visualization of notes and their links, styled as a starfield, with color-coded clusters by topic.
Things they pointed at.
01:19toolHubble.md
03:50toolObsidian web clipper
13:12productObsidian Sync
Lines you could clip.
00:44
“You don't have to pay a monthly subscription in order to use this. You have been lied to if you think you need to buy WhisperFlow or something like that.”
10:10
“The wiki is the IDE; the LLM is the programmer, the wiki is the codebase.”
17:10
“You're no longer constrained by the Notion or Obsidian or whatever other apps you've been using up until this point. Get creative.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
00:00Do you have a that you're way too lazy to go organize? Or maybe a bunch of meeting notes or Notion Pages or Obsidian notes? That's been me and everyone around me for the past, like, forever.
00:12But it's 2026. We have agents now. So I wanted to show you some simple workflows that can take those unorganized notes and turn them into wikis and visualizations that update themselves.
00:23So you can keep lazily taking notes and then pull out new ideas you didn't even know were in there without having to think about it. And, yes, if you're an Obsidian head, this one is for you. So to start, here's the system.
00:35We're gonna take all of our notes just using markdown files because agents love them. It's also easy to find note taking apps that let you edit markdown. I even built my own called Hubble that I'm gonna be showing you in a second.
00:46You start by just taking all of your notes inside of a raw folder and just whenever something comes to mind, write it down. And then we're gonna figure out how we can turn that set of raw ingredients into something a bit more enriched, adding tags automatically, and connecting related ideas based on the content.
01:03And then we're gonna take those raw ingredients and turn them into organized wikis using a set of agent skills. And then we're gonna go a little bit further and talk about HTML visualization. You may have heard about people on the Claude team doing this.
01:15You can do it for your notes too. Now to get those raw ideas down, you could just use any old markdown editor. This is one that I built called Hubble that gives you kind of an Apple Notes like experience, but writes to markdown on the other side.
01:28There's also Obsidian and a number of apps you may have heard of to do this sort of thing. And the goal is to just take notes whenever you want to. Give it a nice heading.
01:37Then for the content, yes, you can faithfully type everything down. But these days, I suggest just using a voice transcription tool either in voice memos on your phone or using one of these apps on your desktop.
01:50You can see right here it's transcribing everything that I'm talking about. And when I let go of the key, it's going to type it all out. In fact, everything that was inside of this note that I took after listening to a podcast about how Ferrari was founded, all of this was written using a voice transcription tool.
02:05It even generated all the line breaks and all the punctuations for me. It got the spelling of names right based on context. Voice descriptions come a very long way, and you also notice it takes zero effort to get ideas down this way.
02:19The average typing speed is, you know, below 100 words per minute, but speaking, you can easily get down, like, 200 words per minute or more. So unless you are a stupidly fast typist, voice is the most efficient way to just get ideas down. Now you don't have to pay a monthly subscription in order to use this.
02:38You have been lied to if you think you need to buy WhisperFlow or something like that. On device models have gotten crazy good at transcription. There are a couple tools that I'll recommend for this.
02:47One of them is Handy. This is a free and open source tool that I believe is platform agnostic, and you can just install it, and it sets up a local model.
02:57I believe it's NVIDIA Parakeet as of the time of this recording. And it's super fast. You just hold a key, you let go, and it transcribes on device.
03:05There's also this other app called VoiceInc. This is the one that I use because it also generates paragraph breaks, which I think is a little bit nicer, but it works exactly the same way. It's also open source.
03:15They charge a one time fee of, like, $20 to get updates. You might as well pay it to support the developer, honestly. And you get access to any AI models that you want to.
03:24So if you wanna plug in custom models, you can do that. I personally use the default. It also looks at your screen to understand what the transcription might be based on context clues, and it doesn't go off your device.
03:35It's so cool. So that's voice. The second way I suggest getting ideas down is to get a nice web clipper in your web browser.
03:42So this is a tool from Obsidian that you can use. You can see up here I have this app in Chrome. It's called the Obsidian web clipper.
03:50If you click on this, it will take the contents of whatever web page that you're looking at and turn it into some markdown that you can save locally. You can save it into Obsidian if that's your note taking app, but don't worry if you don't use Obsidian. You can just copy as markdown and then paste it into whatever app you're using.
04:07And you can see it also generates a bunch of front matter properties to easily find information later. Like, if you're referencing an author, it'll put it in there. It'll generate a little tag here.
04:16It'll also copy the source. So if you need to get back to the URL, that'll work.
04:21You get rich formatting. It's awesome. And it also works with YouTube transcripts.
04:26So look at this. If I click on the Obsidian button up here, you'll notice that it generates the full transcript of the video pulled in straight from YouTube with time stamps for everything. So you seriously don't miss a single detail.
04:38And if you're thinking that's too much data, I feel like I should just write down the things that I care about. Our goal here is to get down as much raw material as possible because we're eventually gonna feed this into an agent that generates wikis, that generates related ideas.
04:53And the more context that you can give it on all of the sources that you're pulling in, the better those outputs are going to be. So I encourage you to use web clippers. Use your voice tool and ramble for ten, twenty minutes.
05:06Get down as many ideas as you can to get the most out of this. And with that, let's go ahead and transition to phase two where we will start generating some metadata across all of our notes. So I'm gonna go back to that voice note that I mentioned earlier.
05:20And, ideally, we could have some more metadata around this. Like, it would be great if I could get the URL to the podcast, but it would be great if I could get some tags, especially tags that I use across a whole note taking system. Too lazy to come up with that, but, ideally, I would have one.
05:35And also some related links. Right? Like, if I listen to other business podcasts related to this, it should be linked so I can go follow the paper trail, like clicking into backlinks on Wikipedia.
05:45So I generated a skill to do exactly that. It's called enrich note inside of here. I will include links in the description if you wanna try out these skills.
05:54And it's pretty simple. We're just gonna ask it to enrich the note by adding those details I mentioned. So I have a tagging system that I'm tracking as markdown.
06:03I had an agent initially generate this tag list from all of the notes just doing a one time scan through the content. And from here, I instruct the agent to be reluctant to add new tags to the system because, otherwise, the agent constantly invents new tags, and you end up with a sprawling system that runs away from you.
06:23So do an initial pass, and then take all of those tags and apply them only when they're related to whatever the content is. Some of it is for specific topics. Others are for tracking the medium, like if it's a book, a podcast, or a video.
06:36And then pulling in the source using web search to find that podcast if it can, and linking related notes using the WikiLINK format. And it's pretty easy to ask an agent go off and do this because we're tracking all of our notes as markdown. We don't need any fancy connectors or MCP servers.
06:53We can just route to whatever that directory is using our agent of choice. I personally suggest using a low reasoning model for these skills so that it's token efficient, gets done nice and fast. And to start, I'm just gonna copy the path to this note, and I'm gonna ask the agent to enrich it for me.
07:09And we can see here it was very smart. It found a related podcast that I also took notes on recently, and it also did a web search to figure out what the acquired podcast URL was.
07:20If I head back over here, boom. Now we have some properties. So we have our podcast and business tags up there.
07:26It eventually will call luxury. I might tell it. That's a little bit excessive.
07:30Uh, we have the source as well as the URL. So if we wanna link into it, we can. And we also have a related note for this other podcast that we listen to about the founding of Walt Disney.
07:41Look at that. And as you can see, I ran this process across a whole vault of notes. I believe there's, like, a 100 or so notes inside of here.
07:49And all of them have these related links that I can go follow if I want to. Like, there was a short story that I read that links over to Egypt loops verification, apparently.
07:58So now I can go look at that note. If you're used to Wikipedia deep dives, this is . I personally love it too.
08:07Uh, here's another link about Claude. Uh, it's really fun to go around this sort of web of connected ideas. But as you can tell, you kinda have to be a nerd to actually click through and look at everything.
08:18This is really useful if you're good with, like, fuzzy finding and then clicking from one to the next to the next to find the related ideas, But it's a little more useful if we can get a bird's eye view across all of these notes to understand where to start, really. And with that, we can move on to the next phase in our journey, which is generating Wikis.
08:37So this is the idea that you can choose a topic that you're most interested in. For example, if you're researching AI hardware, you could point an agent at it and say, generate a Wiki around AI hardware.
08:47And the system I like to use is generate a map of all of the people that are involved, uh, all the concepts, all of the source material, and then an index page so I can click into there and explore it.
08:59To have an agent generate this, it's as simple as taking this GitHub gist from Andre Karpathy and saying, I want a Wiki like this guy's, but for whatever topic that I care about. So if you wanna grab a link to this gist, paste it in your agent of choice and get a Wiki. And here are a few of the Wiki pages that I've generated.
09:18One of them is pretty predictable, the latest in AI. I take all sorts of notes around that. So it generated an index page for me, and you can see all sorts of sources related to different topics that I'm researching.
09:29One of them was for a side project. One of them is around all the voice transcription tools, including the ones that I told you about. Took notes on that in there.
09:37One is about Clog Code specifically. One of them is about a certain content creator that choose Obsidian more. Uh, one of them is about AI handoff strategy, all sorts of things that we can pull out here.
09:49We also get some related concepts and some people that we take notes on. So there are some expected players inside of here.
09:56There's also, uh, some unexpected ones like Adam Neely. He's a jazz musician that I listen to who's talking about how AI is influencing the music industry. And then we have organizations since that was relevant to this topic, and you can click through and find out more for all of these.
10:12So it already took notes for me on who this guy is, which is awesome, and we can see all of the related ideas that this person has talked about. So I can click in and look at agent loops, for example. And you can take this all sorts of directions.
10:24Like, if you're in enterprise and you're generating sales leads and you need to remember all of the people and all of the calls that you have taken, well, this is gonna generate a very useful index for you without you having to lift a finger. That's great. It's also good for side interests as well.
10:39I've personally been listening to the bible in a year podcast, actually, but I generated a Wiki page just for that and any spare notes that I've taken. And this has been useful for me because if anyone's read sort of older literature, you know that some of the names can blend together.
10:55Same goes for researchers that you might run into or take notes on. So it's very useful to get an index of all of that with the correct spellings and background on each of those people. And here, generated some bespoke concepts that were specific to everything that I wrote down.
11:10So we have concepts. We also have all of our sources linked up here. And if we click through at the end, we also get our related notes.
11:17So, again, we're connecting related ideas. But now these are summaries of those notes that we were taking earlier. So they're much easier to read and digest.
11:25Now this is cool, but you probably don't wanna manually create these wikis every time you wanna go look at them. It would be more useful if as we take notes inside of that raw directory that we were talking about earlier, it just magically enriches for us while we sleep.
11:42Just have an agent wake up, enrich, and tag everything. And then if another agent come along and update whatever wikis are relevant to those topics, so every time you go look at that wiki, it's fresh. As you can imagine, that's where we bring in some AI automations.
11:56There are all sorts of services that you can use to do this. Some of them, like the Codex app, just run on your machine. So you have to have your laptop cracked open.
12:05So every time the agent wants to do something, there's some CPU so we can wake up and do that. Uh, but there are also cloud runners, and that's what I recommend personally. So we don't have to think about whether a computer's running.
12:16It just wakes up and does the work. And it's pretty simple to set this up. So if you want to set up a cloud sandbox, these are the basic ingredients that you're probably going to need.
12:25So you're gonna need some sort of container that'll be able to copy this full dataset that's on your computer into a separate sandbox. So you will need some sort of, like, file syncing service in order to make a Cloud Runner happen. I personally use Obsidian to sync all the markdown to a remote folder, and they have a CLI to make it easy for an agent to pull things down.
12:45It's called OB or headless Obsidian if you go Google it, and it gives you a single command called sync that will sync down whatever markdown is inside of the cloud into a local sandbox. So then the agent has access to all the markdown.
13:00It can iterate over it, and you can run all of those skills. Like, we can ask it to run enrich note for us across all of the notes that have not been enriched yet, or have it refresh the wikis for whatever topic that we're interested in. And then when it's done working, just have it run sync again to push everything back up, and you're good to go.
13:18Now one thing I'll mention is that Obsidian Sync is a paid service. It's $4 a month. It's nothing crazy.
13:25If you want to, you can also just use git for this sort of thing. So you could clone down all of your notes from a GitHub repository and then push them up at the end.
13:34This, of course, requires you to track all of your notes in Git, which in my experience is just more of a hassle than it's worth, but it is free. And if you're a developer, you're probably used to using Git anyways. I'm just gonna suggest Obsidian for this because you don't have to think about pushing and pulling.
13:51It's just a sync service that runs in the background. And here are some automations that I set up. This is using a cloud platform called Oz from the work team, and it's fairly simple.
14:00You can set up a Docker container with all the environments. This one just sets up a Docker container with the Obsidian CLI inside of it and also JavaScript because that's required to get the CLI to work. And then I include some instructions so that when an agent wakes up inside of this container, it knows what to do.
14:18So here it's telling me that Docker's gonna clone everything into this folder. You can run Obsidian sync, and here's some config to make sure other files get gobbled up inside of there. It tells it where the Wikis are and then gives some instructions on how to update it.
14:32Honestly, this was agent generated and could be shorter, but it does work for me in practice. The only goal is to sync it down, update the Wiki, and push it back up. I did another one over here for enriching notes as well.
14:44This one is backed by a skill that you can go find, and it just loops over all the notes looking for the special front matter property to make sure that it's not enriching something that was already enriched. It basically stamps it and says, I already did this in the past.
14:59So the next time the agent wakes up, it remembers, oh, I don't need to touch that. Another agent did it for me. And then at the end, it syncs everything back up.
15:07And we can look at one of the runs for that inside of here. Uh, Oz gives you a little preview in your browser. This is literally the warp terminal inside the browser.
15:16Kinda nuts. Uh, it ran the skill for us, and we can see everything that it did when it woke up. Here, it's editing some files.
15:25It enriched some talk resources that were untagged. It also found some of those notes earlier. Yes.
15:30I reverted them so I could demo for you. You get it. Uh, it found everything that was unenriched, and then it pushed it back up.
15:37Up. And down here, we can see the cadence. I set it up to run every day.
15:41You can also do it every week if you wanna conserve tokens a bit. I also included an auth token so that everything just authenticates and works. And then on the other side, you can wake up and look at your recent notes and see that they're all nice and beautifully tagged.
15:53You can look at your wiki and know that it is perfect and up to date. You don't even have to think about it. You just dump a bunch of raw voice notes inside of here, and you get beautiful wikis on the other side.
16:03Now the last thing I wanna talk about are visualizations because everything that I've shown you has been text based, which is fine. It's easy to follow a bunch of links down a rabbit hole to look at related people and sources.
16:16But what if we wanna get a little bit more creative? What if we wanna build full interactive apps around all of our notes? Well, that's easy to do because if you have agents at your disposal, you can also write some code.
16:28So I'm gonna show you how you might generate a visualization using HTML. So here's some examples that I've put together and started using. The first one is maybe you wanna generate a burn down chart.
16:38You know how on GitHub you can see all of your commits? What if you could do that for notes just to keep up a daily streak or something like that? So I had an agent generate one of those as HTML, and we can see that burn down chart right here.
16:51This gives us a in app HTML viewer so we can look at this burn down chart of everything that we've done. And through making this, I realized I am very inconsistent with when I write things down.
17:03I had a strong start for sure, so this was definitely helpful for me to see, yeah, maybe I need to pick up my habits a little bit more. Now the other thing I wanted to build was a graph view. If you use Obsidian, you know about graph views or something like that.
17:18Well, I wanted one of those too, but I wanted to get a little bit creative and just make it look like outer space. Because why not? We have agents.
17:25We can do whatever we want. Uh, so the first take of this was this thought constellation right here. And this has actually been way more useful than I expected to see the clusters of everything that I take notes on and where I probably need to write more down.
17:39So we can see AI and engineering over here. We can also see some hot spots on related ideas. So there was this one about Andrew Wilkinson and some of the notes that he gave about Claude code.
17:50We also see some other related Claude code observations clustered inside of there. In clicking in, we can see all the related nodes and ideas and just click through them.
17:59How cool is that? And we also see some related notes up here for scripture. And over here, just for stray thoughts.
18:05I am a very scattered person, apparently. Uh, we also see books and reading for all the books that I read, products and startups. You can ask the agent to change this however you want, of course.
18:15And I also went a little bit further and said, what if it was in space? Well, look at this. Now we're in space.
18:20We're also getting some nicer highlights to really tell as we're stepping through here. We can see, oh, there's that connected idea. Then we click on this guy.
18:27Oh, there's that connected idea. Again, you're no longer constrained by the Notion or Obsidian or whatever other apps you've been using up until this point.
18:36Get creative. Ask agents what are interesting ways that I can visualize all the stuff that I'm writing down. I'm sure you'll come out with something on the other side.
18:45If there's any homework I have for you, just take the notes you have and put them into markdown. You can ask coding agents how to do this. And then you can use the skills I showed off in rich notes, wikis, or go crazy coming up with visualization.
18:58Just ask Claude Code, generate an HTML file for me. And if you wanna try that note taking app I was showing you, it's called Hubble. I built it in my garage.
19:06It's free and open source, and it works on any platform. With that, like and subscribe.
19:11I'll see you in the next
The Hook
The bait, then the rug-pull.
The opening question — a junk drawer of unsorted notes everyone recognizes — sets up a demo built entirely around not organizing anything yourself: capture stays lazy, and agents turn the mess into self-updating wikis and space-themed knowledge graphs.
Named ideas worth stealing.
05:52concept
Enrich Note Skill
A single agent instruction: read the existing tag registry, reuse tags reluctantly rather than inventing new ones, attach a source URL via web search when available, and add WikiLink-style related-note links.
Steal forany personal notes folder, regardless of topic
08:55concept
LLM Wiki Pattern (Karpathy gist)
Paste a single public gist link into a coding agent and ask for a wiki 'like this one' for any topic — the agent produces an index page plus people/concepts/sources sub-pages from the notes it has access to.
Steal forturning any pile of research notes into a browsable reference
12:36model
Sync -> Run Skill -> Sync Back Automation Loop
- Sync vault down into a cloud container
- Run the relevant agent skill (enrich or refresh-wiki)
- Sync changes back up
- Stamp processed items so reruns skip finished work
The generic shape of every scheduled agent automation in the video — swap the skill, the loop stays the same.
Steal forany recurring agent job against a synced folder of files
How they asked for the click.
VERBAL ASK
19:08product
“if you wanna try that note taking app I was showing you, it's called Hubble. I built it in my garage. It's free and open source, and it works on any platform.”
Soft, single-mention product plug for his own app placed in the final seconds, right before the standard 'like and subscribe' close — not a mid-roll sponsor break.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
