Big Idea

The argument in one line.

A personal assistant built from Claude Code automations across email, research, and content-repurposing can reclaim five to ten hours a week once the underlying skills sit on top of an organized, indexed knowledge base.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You run a small business, agency, or solo practice with inbound sales emails and manual proposal writing.
You publish content regularly and want it repurposed into a blog post, LinkedIn post, and tweet without extra work.
You already use Claude Code and want practical automation ideas beyond writing software.
You do daily research across scattered sources like X, GitHub, and YouTube and want it consolidated into one brief.

SKIP IF…

You're looking for a step-by-step technical build guide — this is a system overview, not a wiring tutorial for the automations shown.
You don't do any sales, research, or content work — none of the three buckets covered here will map onto your workflow.

TL;DR

The full version, fast.

Claude can run as a personal assistant once repetitive business tasks are split into three buckets: productivity and sales, research, and content. Email triage sorts inbox messages into leads, urgent, sponsors, meetings, and noise, drafts sponsor replies, and pre-researches promising leads before a discovery call; a parallel flow turns Calendly recap emails into branded proposal PDFs. A daily brief pulls GitHub trending repos, X activity, and YouTube trends into one report, and on-demand skills handle deep research and multi-video synthesis. A content-cascade skill rewrites new YouTube uploads into a blog post, LinkedIn post, and tweet automatically. None of it holds together without an organized Obsidian vault of raw, wiki, and output folders with index files that stop Claude from losing context as the system scales.

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 →

Chapters

Where the time goes.

00:00 – 01:07

01 · Intro

States the thesis — Claude as a personal assistant that automates the boring, non-decision work of running a business — and previews the three buckets covered.

01:07 – 06:39

02 · Productivity / Sales

Walks through email triage (leads/urgent/sponsors/meetings/noise), automated lead research and outreach drafts, and turning Calendly call recaps into branded proposal PDFs.

06:39 – 12:44

03 · Research

Covers the daily research brief pulling from GitHub trending, X, and YouTube trends, plus on-demand deep-research and a YouTube-to-NotebookLM synthesis pipeline.

12:44 – 17:15

04 · Content

Covers AI-assisted hook/title brainstorming trained on a reference creator's style, and a content-cascade skill that rewrites new videos into a blog post, LinkedIn post, and tweet.

17:15 – 22:06

05 · Obsidian + UI

Explains the vault/raw/wiki/output folder structure with index files that keeps Claude from losing context, and why a custom dashboard is needed to see everything the assistant is doing.

22:06 – 22:25

06 · Outro

Points viewers to the paid community for the exact setups and skills shown.

Atomic Insights

Lines worth screenshotting.

Splitting AI automation into three buckets — productivity/sales, research, and content — makes the workflow reproducible across almost any business, not just content creation.
An email triage system that sorts inbound messages into leads, urgent, sponsors, meetings, and noise can draft sponsor replies and pre-research leads automatically before a human ever opens the inbox.
A lead-scoring automation can pull a prospect's stated budget and timeline from a website form, verify the query looks real, then run a web search on the company before a discovery call happens.
Turning a Calendly call-recap email into a branded, presigned proposal PDF removes the single most time-consuming step in closing new client work.
A GitHub trending scraper broken into three windows — created this week, created in the last 30 days, and fastest-growing in 24 hours — separates brand-new tools from ones quietly gaining traction.
Comparing a creator's view count to their subscriber count is a better trending signal than raw views alone: 10,000 views from a 2,000-subscriber channel means something is resonating.
A built-in deep-research command can spawn up to roughly 100 sub-agents that gather information adversarially and cross-check each other, at the cost of heavy token usage.
Routing YouTube video URLs into an external synthesis tool avoids burning AI tokens on transcript processing, since another service does that work instead.
A content-cascade automation that checks twice daily for new YouTube uploads and rewrites the transcript into a blog post, LinkedIn post, and tweet turns one video into four pieces of content with zero manual touches.
Training an AI writing skill only works with a repeated human-in-the-loop cycle: give examples, get an AI draft, tear it apart line by line, and repeat roughly ten times before the voice sticks.
An Obsidian vault organized into raw, wiki, and output folders — each with its own index file — functions as a filing cabinet that keeps an AI assistant from burning extra tokens hunting for information.
Without an indexed folder structure, an AI system's cost per query rises and its accuracy falls as the amount of stored information grows.
A custom dashboard isn't a nice-to-have on top of an AI automation system — it's the only place all the scattered outputs, like metrics and reports, become visible in one view.

Takeaway

Turn repetitive business tasks into reusable Claude skills.

AI ASSISTANT SETUP

The busywork of running a business — sorting email, researching daily, and repackaging content — splits cleanly into automatable skills once it's organized into three buckets and backed by an indexed knowledge base.

02Productivity / Sales

An inbox can be automatically split into leads, urgent, sponsors, meetings, and noise, with each bucket triggering a different automated response.
A lead worth pursuing can be pre-researched — company background, stated budget, and timeline — before a human ever gets on a discovery call.
Post-call recap emails can be converted directly into a branded, presigned proposal PDF, removing the slowest step in closing new work.

03Research

A daily research brief pulls from a small, deliberately chosen set of sources — for AI news, that's X, GitHub, and YouTube — rather than trying to monitor everything.
Comparing view count against subscriber count is a sharper trending signal than raw view count alone, since it surfaces underdog creators breaking out.
Multi-agent deep-research workflows trade heavy token cost for adversarial fact-checking, so they're worth reserving for questions that actually need that rigor.

04Content

AI-assisted brainstorming works best as a back-and-forth sparring partner for hooks and titles, not as the final source of the words used.
Voice-matching a writing skill takes repeated iteration — feeding examples, rejecting drafts, and correcting them by hand roughly ten times before it holds up unsupervised.
A single piece of long-form content can be mechanically cascaded into a blog post, a LinkedIn post, and a social post without additional manual work.

05Obsidian + UI

An unindexed knowledge base increases both token cost and error rate as it grows, because retrieval falls back to brute-force search.
Organizing stored information into raw, processed, and output tiers, each with its own index file, gives an AI assistant a fixed place to look instead of guessing.
A dedicated dashboard is what makes an otherwise invisible automation system observable, since a terminal isn't built to be the place you check on your assistant's work.

Glossary

Terms worth knowing.

Email triage: An automated process that scans an inbox and sorts messages into predefined categories, such as leads, urgent, sponsors, meetings, and noise, so a human only reviews what matters.
Content cascade: An automation that detects a new video upload and rewrites its transcript into multiple pieces of content, such as a blog post, a LinkedIn post, and a tweet, without manual editing.
Deep research (agentic command): A built-in workflow that can launch many parallel research sub-agents to gather information from the web and cross-check it against each other before producing a synthesized answer.
Vault: A folder-based knowledge base used to organize an AI system's raw data, processed reports, and finished outputs so the AI can reliably find prior information.
NotebookLM: A Google tool that ingests source documents or transcripts and produces synthesized summaries, used here to condense multiple YouTube video transcripts into one digest.

Resources

Things they pointed at.

00:00productChase AI+ / Claude Code Masterclass ↗

03:09toolCalendly

07:34toolGitHub Trending

11:06tool/deep-research command (Claude Code)

11:19toolNotebookLM

Quotables

Lines you could clip.

00:00

“Claude should be running your life, or at least all the boring parts.”

cold-open thesis line, no setup needed→ TikTok hook↗ Tweet quote

11:54

“This is essentially a filing cabinet for Cloud Code.”

tight metaphor that sums up the whole vault section→ newsletter pull-quote↗ Tweet quote

11:59

“I don't wanna watch ten YouTube videos about a specific topic. Just give me the summary. Do the synthesis.”

relatable pain point framing for the research pipeline→ IG reel cold open↗ Tweet quote

18:23

“This is where you hear everyone talking about Obsidian and a second brain with Claude code.”

names the trend the video is riding→ TikTok hook↗ Tweet quote

The Script

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.

metaphor

00:00Claude should be running your life, or at least all the boring parts. And it's actually very easy to turn Claude into your own personal assistant, saving five to ten hours every single week just like I have. Now we've talked about creating your own Claude OS on this channel before, whether that's a web based setup like this one that includes voice or an Obsidian based command center.

00:20But in all those videos, we tended to focus really on system design, how we should create our skill architecture, and how we should set up the actual Obsidian vault. But today's video is gonna be a little different. This one is gonna look at my exact personal assistant setup.

00:35These skills and automations I use to run my business, my channel, my research, all of it, so you can get a good idea of what you need to do to recreate it yourself. Now when it comes to these custom fancy dashboards and command centers like you see here, we will get to that at the very end because there is some value in creating these when it comes to personal assistant setup.

00:54But the first thing we need to establish is what the heck is this personal assistant even going to be doing? What is the boring, redundant work that we can automate and turn into skills? Well, for me, there's really three different buckets that I operate in that has a lot of, like, sort of drudgery, right, that takes up an inordinate amount of time, doesn't really require a lot of decision making, but it's just like grunt work that I offload to Claude Code.

01:18And those three buckets are productivity and sales. I kind of put that in the same category. Research and then content.

01:24So the productivity and sales stuff has a lot to do with my AI agency business, and then the research and content kinda fall in sort of the social media sphere. Makes sense. Now the nice thing about this is a lot of what you're gonna see here, I think, does translate to almost everyone outside of the content stuff.

01:42I know a lot of people aren't creating content, but chances are you are working in some sort of sales or marketing environment where you have information coming in from clients or customers, and you need to respond to them, and you need to be able to do some sort of daily research to stay on top of whatever it is you're staying on top of.

01:57So a lot of what you see here, while it's specific to me, is something that can translate to you. Now what we will look at here is probably what's gonna give you the most immediate value, and this is sort of the productivity section. And it all spawns from this, the email triage automation.

02:11Now the email triage skill does the following. Every morning at 8AM, it takes a look at my Gmail inbox. It uses the Claude connector to do so.

02:20It takes a look at everything that's come in over the last twenty four hours, and it puts it into a series of buckets, either leads, urgent, warm, sponsors, meetings, or noise.

02:30Once it has all the emails broken out into those specific buckets, it then does different things depending on the bucket, specifically the leads bucket and the sponsors bucket. Now if it's a sponsor, Claude then automatically drafts up a reply with a boilerplate response and a link to my media kit saying, hey.

02:48Here's what the prices are x, y, and z. However, if it's a potential lead, we then begin a process that is a little bit more in-depth. So on my website, every lead fills out a form saying essentially their budget, what they're looking for, their timeline, all the standard stuff.

03:00All this information is picked up by Claude Code in this inbox triage. From there, what it does, if it thinks it's a real query, so the email makes sense, they actually spent time on their response, the budget aligns, and it wasn't someone just putting in random nonsense, Claude then kicks off a web search.

03:17So just some background on who their company is, takes a look at their email, basically tries to get me just some preliminary information so I know what I'm doing if I walk into a potential discovery call with them. From there, Claude then gives me a recommendation saying, here's what I found out.

03:31Here's what they wanna do. Should you move forward? Should you not?

03:34I'm still the arbiter here of whether we go forward or not, obviously. If I say, yeah, let's do it, it then creates a draft email with a response as well as a link to my Calendly so they can set up a meeting for me. If not, it just skips them.

03:48For all the emails and other buckets, if it's urgent, it's gonna give me a potential drafted reply. But for everything else, it's just kinda marking them, and what it also does is it gives me a full report, a written markdown file that goes into my Obsidian vault that acts as a summary that I can read in a glance. The report looks something like this and then also lists all the noise that I don't really care about.

04:05And so with the simple setup, I save a ton of time every single day because I don't have to read all my emails. I don't have to manually filter every single lead. And if it thinks the lead makes sense, it's already done some preliminary research and drafted the response for me.

04:17And this is something you can very easily adapt for your own purposes. What sort of emails do you normally get? How can you put them in different buckets?

04:24And then more importantly, what do you want Cloud Code to do with the information it has once it's sorted everything? Now the other half of the equation that saves me a ton of time is proposals and follow ups, and this is what I do with clients after we've already had the discovery call. So I use Calendly.

04:38So post discovery call, what I get in my inbox is a recap. Calendly automatically gives me a summary that includes all the key points of everything we've discussed. That recap then gets sent to Claude Code.

04:52Claude Code takes that recap and generates a branded PDF breaking down. Here's essentially the what we're gonna create.

05:00Here's the scope of work. Here's the pricing. Here's the form for you to sign.

05:04Let's go ahead and move forward. Now, obviously, I'm gonna take a look at all that, but that saves a ton of time versus, you know, having to manually create some sort of proposal.

05:12From there, generates a drive link that is sent to the client and the draft that includes the link itself. Now this is what the proposal looks like. And this is just a dummy version, so I don't show any, like, client information on here.

05:22But it breaks down the engagement, shows the timeline, and then shows the investment. And then at the very bottom, it just has some room for signatures.

05:30Now lastly, in regards to follow ups, that's a little more specific to some of, like, the AI audits I run. So I'm not gonna go deep into that here since it's kind of technical and really won't apply to 99.9% of you.

05:40But that, in a nutshell, is kind of the productivity sales side of my Claude code personal assistant. Again, it doesn't have to touch every single thing you do. But like you saw here, a lot of this is just boring stuff that otherwise I would be spending hours a day on.

05:55So what are you spending hours a day on or hours a week on that we can offload to Claude? You should be doing it. Now before we move into the research portion of my personal assistant, a quick word from today's sponsor, me.

06:09So I just released my Cloud Code Masterclass, which is the number one way to go from zero to AI dev, especially if you don't come from a technical background. We focus on real use cases like the personal assistant stuff we're talking about today, and I go over other AI tools. I also have a Codex masterclass in this community as well.

06:26So if you wanna get serious about learning AI, this is the place to be. I'll put a link to it down below. And everything you see here, my exact setups, the driver system, the Obsidian command center, can also be found inside this community.

06:39So now let's talk about research. Now, obviously, mine is gonna be all AI focused, and yours may or may not be. So what you need to keep in mind is the idea that all of this spawns from knowing where your information originates.

06:52In the AI space, that's really a handful of places. It's like Twitter, GitHub, and then sometimes things will merge on YouTube first.

07:00But kind of it. Those are kind of the big three. So that's where we're gonna go for information.

07:04Where you need to find your information for your niche, you gotta figure it out. I I I can't tell you that.

07:10So you gotta figure that out. But once you know that sort of source of knowledge, the setup is exactly the same for how I do it. So first things first, my main automation is my daily brief.

07:21Right? My daily brief goes out to those wells, those sources of knowledge, and figures out what the heck is going on.

07:27And so for me, like I said, that's really x, That's YouTube, and that's GitHub. Now the GitHub setup is simple enough for me.

07:35Here's one from last week. And what it does is it uses the GitHub API to find the top trending AI GitHub. And it breaks it out in a few categories.

07:42I look at, like, the top 10 trending for the week, so what was created this week. I look at the top five trending from those that were created within the last thirty days. And then I look at the fastest growing over the last twenty four hours and the last thirty days.

07:57So what's new and growing, and then what's kinda been around, and it's kinda shooting up the charts? This gives you, like, a good idea of, like, okay. What are people actually playing around with these days in the AI space?

08:06Because this changes literally daily, and it's impossible to, like, know about it unless you're actually looking at GitHub. And then we have Twitter, YouTube, and also a general web search.

08:14So I'm able to pull the big headlines. I can see what's trending on YouTube. Importantly, it shows me the creator and the views and their subs.

08:21So that kinda gives you a good idea. Like, I don't care if someone with a million subs gets 10,000 views, but if some dude with 2,000 subs has 10,000 views, then clearly he's talking about something that people care about. And then also, we look at Twitter.

08:33And beyond that, it does some basic analysis on, hey. Here's some content opportunities. But to be totally honest, when it comes to figuring out, hey.

08:41Here's what the research shows, and here's what we should actually talk about, AI is pretty hit or miss with that. What we really care about here is sort of just the raw data it's able to bring. And, again, what's the big sell here?

08:50The big sell is I don't have to do all that myself. Am I still manually gonna go on Twitter and YouTube and go through my subscriptions for, you know, five, ten minutes a day? Sure.

08:59Of course. But this stops that from being, like, a thirty minute process and saves me a ton of time, especially on the GitHub side. To do that manually would be terrible.

09:07End of note, this research daily brief and this email triage brief we spent a bunch of time on, these are actually combined into a single automation that I just call, like, my morning automation. And this is set up as a routine inside of Claude code.

09:22So it's a local routine, runs on my computer. As long as it's open, just fires off those automations automatically. Now as for the rest of the research side, I have what I call my XPulse or this gets kind of a pulse on Twitter, as well as my YouTube Dives, which are on demand, and Deep Research, which is on demand.

09:37Now the Twitter setup is something that is actually an app that I built that lives on Railway. So this lives on Railway. It's working twenty four seven.

09:47And what it does is it sends me messages on Telegram every hour or so showing me the top tweet in the AI space that's trending. And this is also tied to a lot of the major creators, especially ones who are unlike the Claw dev team and the OpenAI team. So if something is popping off, something big has opened up in the AI space, I know about it immediately.

10:08I don't have to sit there on Twitter refreshing all the time. So the first half, the daily brief and the expulse are you know, those are automated.

10:17These two are automated. I don't even have to think about them. These two, YouTube dives and deep research, these are specific on demand skills I can use whenever I find some topic that I care about and I wanna learn more about.

10:28Now, deep research is super basic. And by basic, I mean it's already a command set up inside of everybody's body code. So if you do forward slash deep research, that is a essentially preloaded dynamic workflow.

10:40It will spawn if you just let it go nuts up to, like, a 100 sub agents that do, like, adversarial, essentially, information gathering.

10:48Like, they'll go on the web. They'll find information. Then they'll test it against each other to see if it's right and do some synthesis versus your standard web search where you just send Claude code.

10:55You you say, hey, Claude. Go find me information about, you know, Fable five. It's just gonna send a couple sub agents to browse the web.

11:01This is that on steroids, but it eats up a ton of tokens, but it's something like a lot of people don't actually use. I also have what I call my YT pipeline skill.

11:11And so what that does is it then searches YouTube to find videos that are relevant to whatever I'm asking about.

11:20It then sends all those URLs to NotebookLM. NotebookLM itself then takes all the transcripts, does all the synthesis, and sends that to me.

11:29Now I do that using a specific CLI, which is the NotebookLMpy CLI, which actually gives Claude code essentially an unofficial API to NotebookLM. Again, being able to do that means I get all the power of NotebookLM, but it doesn't cost me tokens to have NotebookLM in Google servers do all that synthesis on YouTube transcripts, which, again, what are we doing here?

11:48We're trying to save time. I don't wanna watch 10 YouTube videos about a specific topic. Just give me the summary.

11:53Do the synthesis. Tell me what's the throughput between all 10 videos. Tell me where they aren't aligned, and then tell me why I should care.

12:00And you put all that together, and essentially what you get is a research setup that saves me undoubtedly five to ten hours by itself minimum from this.

12:11If I try to do all this research on my own, I'd be dying. You know? I would be seriously struggling.

12:15And I think for a lot of people, again, whether you're doing AI, um, based searches or not, I think the productivity side and this research side should really go hand in hand.

12:25There's a ton of places, I imagine, in your domain where the sort of emails you respond to, the sort of things you're creating are based on certain research. So why aren't we sort of combining these things? Again, especially if these are, like, a lower level, you don't really have to make decisions type tasks, and just hand it over to a personal system of Cloud Code and have it do that for you automatically.

12:43Now the last bucket here is content. And I'll spend the least amount of time here because I think this is the least relevant for most people. But just for those of you who might be doing content and wanna know how I do it, for me specifically, when it comes to scripting hooks and outlines, this is simply a single on demand skill.

13:01Now I really like Callaway, super good creator on YouTube. So I essentially trained Claude on a bunch of his videos to create those skills in terms of here's how you should do a hook, here's how you should think about transitions, and that sort of thing. But to be honest, I'm not someone who really does scripts, so this is just more of a brainstorming tool for me.

13:19But it is useful. I don't rely on AI to come up with the words that I say, but oftentimes, it's nice to have a back and forth where you have some semblance of an idea.

13:28You bring it to AI. It's been trained on certain skills, so it's not just giving you generic garbage. And through that back and forth, some ideas, you know, kind of coalesce for you, the actual human being.

13:38So in terms of hooks and outlines, that's how I approach it. I know some people have everything scripted, just not how I work. Then we have packaging.

13:45Again, similar to my hooks and outlines, I don't have it just create the titles or create the thumbnails. The thumbnails, I pretty much do all on my own. And then in terms of the titles, yes, it uses the same sort of, like, Callaway things where it's like, hey.

13:56Here's the sort of words that you use. It's sort of, like, psychological triggers, however you wanna say it. But at the end of the day, it's more of a back and forth.

14:02I have it look at what titles I've done well and why. And it just, again, if it's just you and a piece of paper and you're trying to come up with everything 100%, I think you're gonna struggle. And I think you also struggle if you just try to off offload all creative purpose to AI because it's just not that good at these things that require, quote, unquote, taste.

14:20But I think some sort of mixed approach helps a lot. But lastly, where AI really comes in handy for me is content repurposing.

14:28Now this goes beyond just, like, dropping a video in a folder and it puts it across a bunch of different platforms. What this does is every day, twice a day at noon and 8PM, it looks to see if a new YouTube video has been posted by me.

14:40From there, it runs this content cascade skill where it then fetches the transcript from that video, and then it rewrites a blog, a LinkedIn post, and a tweet based on that transcript.

14:55Now that skill took some time to get together because it's all about getting your voice correct, and your voice is your voice. It's unique. No one can tell you how to create a skill that's gonna, like, work best for that.

15:06My suggestion, if you wanna do something like this and have it do these sort of things, is you need to have examples of your own writing that you've done 100%. And what you do is you have a consistent back and forth, and it's kinda like a cycle. Right?

15:17So I give Claude examples of my writing. I then say turn that into a skill. Okay?

15:24It's then gonna run that skill, and it's gonna give you its example. And what are you gonna do?

15:30You're gonna destroy their example. You're gonna absolutely eviscerate it and say, this is wrong. This is wrong.

15:35This is wrong. This is right. Here's how we change this.

15:38So then you give them the updated example. It updates the skill. It gives you a new example.

15:43You destroy it. You do this over and over again for, like, 10 times, and then you continue to do it over and over again when you do it live.

15:50And over time, you will eventually get a skill that works for you. That's really the only way to do it.

15:57There's a huge human in the loop component if you're doing any sort of, like, humanizing or in your voice typewriting. So don't listen to anybody that says, hey. Here's just, like, just here's this one shot skill that's self improving.

16:07No. No. No.

16:08You have to be in the loop, and that's the way to do it. And so in this case, I did it for a blog because it's a little different voice for a blog versus a LinkedIn post versus a tweet. And to be honest, I'm very lazy in terms of actually posting these tweets.

16:21But LinkedIn and the blog post, totally viable. And again, it comes from a single YouTube video.

16:26I don't even have to think about it. It's automatically posting these things, putting it in my voice because otherwise, I just wouldn't do it. So that is the meat and potatoes of my personal assistant setup with Claude, the automations and skills that I actually use.

16:38Now what I think what I hope you got from that is that this is unique. Like, these are the things that I've identified that I don't really wanna spend time on. So I can spend my time doing things that I consider higher leverage.

16:49I don't wanna do emails. I don't wanna do a ton of research in the morning if I don't have to, and I don't wanna repurpose my content. I don't wanna have to manually write every proposal or filter every lead.

16:59Only you can identify what those sort of issues are for you. The point is once you identify them, you simply turn them into a skill. You make sure they work.

17:07You automate them. And you do that a few times, and all of a sudden, like, you are saving five to ten hours a week with that simple setup. Now to turn that into a more advanced coherent system, however, requires some sort of structure.

17:20Cool. We're creating all these briefs. It's doing stuff with emails.

17:22It's creating content. Well, like, how do I give it, like, a brain? How do I make sure it's able to reference things it's sent in the past?

17:29How can I take all these briefs? How can I put them on a single dashboard? That's where we start to get into, like, the CloudOS type videos and the CloudOS type lessons.

17:37And in this last part of the video, we'll touch on that briefly because that is something we have done plenty of content on in the past, and I'll link that here as well. Now this is where you hear everyone talking about Obsidian and a second brain with Claude code.

17:50Why do we care about Obsidian? Is Obsidian in itself necessarily changing the way Claude code works? The answer is not really.

17:58Right? Even something like this is just sort of just a fancy visual knowledge graph.

18:02But what Obsidian can do is allow us to organize everything that's created in our personal assistant setup, which over time creates a map that we hand to Claude code so we can quickly and correctly, truthfully, answer our questions.

18:19So how do we set up Obsidian in a way to do that? Well, luckily, it's pretty simple. The Carpathi method is the most common one you will see.

18:26I want you to imagine all of these things simply as folders. Okay? These are all just folders.

18:31At the top, we have the vault. This is the folder we have designated as the Obsidian vault. Underneath the vault as subfolders, we have multiple sections.

18:40We have the raw section, We have the wiki section, and we have the output section. We just need places to put information that makes sense. So raw is unstructured data.

18:49Like, hey. We had Claude code do a bunch of research on stuff. It hasn't done any synthesis.

18:53It's just dumping all that raw information here. Well, after it's, you know, gotten all this information, we probably wanna turn it into some sort of report. Well, that goes into the wiki section.

19:04And, hey. We have that report. We wanna turn this into a slide deck.

19:07Well, that goes into the output section. Simple enough. Now underneath all those folders, you're gonna have more subfolders because, hey, if I have a Wiki folder, but I have a million different reports, how does Claude code know where it needs to go to find information?

19:21Well, underneath every folder, we're going to have an index document, which essentially is like, you know, the table of contents for that subfolder that says, hey. Here's where this is. Here's where that is, etcetera, etcetera.

19:32So imagine we had, you know, essential reports about AI agents and RAG systems and content creation. And under those reports were even more detailed reports about specific things like autonomous coding and tool use patterns.

19:44So in those sub photos, what would you have? You would have index. So you kind of get where I'm going with this.

19:49We have things broken out into different subfolders that make sense depending on their use case. And in each subfolder, we have an index file that says, hey.

19:59Here's everything in the subfolder so that Cloud Code always knows where it's going. It can never get lost. Now when I say get lost, is Cloud Code actually getting lost?

20:07No. But if it's not organized, what happens is we're going to increase the token cost.

20:14Right? If we just rely on grep for everything, that isn't necessarily the most efficient way to do it. So if we don't have an efficient setup, it's gonna increase the token cost, and it's actually going to decrease the accuracy over time.

20:26As you get bigger and bigger file structures and you have more and more things in there, you need some form of organization. This is essentially a filing cabinet for Cloud Code.

20:37And, yeah, we get these cool knowledge graphs, but that sort of idea of the filing cabinet of the map is the real value play. And the second half of the equation is these sort of, like, command centers and dashboards. Like, cool.

20:46Cloud Code is my personal assistant. It's grabbing emails. It's doing research on contents, repurposing stuff.

20:51Like, where does that information go? Where does it live? Where can I see it?

20:55More importantly, can I see everything in one place? Yes. We can totally set that up.

21:00Now can that place be the terminal? Not really. That's, like, not the purpose of the terminal, and it's not even really the purpose of the Cloud Code desktop app.

21:09So when we talk about these command centers or we talk about some web app dashboard like this, that's the value. Right? It's the one stop shop that you really can't get anywhere else because it needs to be 100% customized.

21:20And you can really see that shine in this Obsidian base setup. Right? I have all my metrics here, my token burn.

21:25I have all these different automations that we've talking about are here. I can essentially run them at a click of a button. The audience section here is essentially a different breakdown of that morning report, and the same sort of stuff lives here in the web app version as well.

21:38I got the metrics. I got the skills. Down here, I have the documents so I can pull up those reports, and this is the exact same thing I would find inside of my Obsidian vault.

21:47Like, it's all connected. So these dashboards give you the observability. Obsidian is what allows you to keep it all organized, and it's these sort of skills and automations that are actually saving you time and doing the work as a personal assistant, as an executive assistant.

22:00And so you put that all together, and you have a way to make Claude actually work for you. Essentially, totally hands off. So that's where I'm gonna leave you guys for today.

22:08I hope that was somewhat eye opening in terms of how you should approach Claude as a personal assistant. If you wanna get your hands on the Claude code masterclass or the skills I talked about today or any of these sort of dashboards, my exact setups can be found inside of Chase AI plus. There is a link to that in the pinned comment.

22:22And as always, I'll see you around.

The Hook

The bait, then the rug-pull.

The pitch is blunt: Claude shouldn't just write your code, it should run the boring parts of your business. What follows is a tour of exactly which parts — sorted into three buckets, wired into real automations — that reportedly buy back five to ten hours a week.

Frameworks

Named ideas worth stealing.

00:54list

The Three-Bucket System

Productivity / Sales
Research
Content

Every automation in the personal-assistant setup is grouped into one of three domains, each broken down into its own set of skills.

Steal fororganizing any AI automation roadmap for a service business

02:10list

Email Triage Buckets

Leads
Urgent
Warm
Sponsors
Meetings
Noise

Every incoming email gets sorted into one of six buckets each morning, with leads and sponsors triggering their own automated follow-up flow.

Steal forinbox automation for any business fielding cold inbound

18:18model

Vault / Raw / Wiki / Output Structure

vault/
raw/
wiki/
output/

A three-tier folder hierarchy with an index.md table of contents in every subfolder so an LLM never has to guess where information lives.

Steal forany long-running Claude Code knowledge base

CTA Breakdown

How they asked for the click.

VERBAL ASK

06:08product

“So I just released my Cloud Code Masterclass, which is the number one way to go from zero to AI dev... I'll put a link to it down below.”

Framed as 'a quick word from today's sponsor, me' — a self-deprecating joke that softens a self-promotion into a bit, repeated more directly at the outro pointing to the paid community.

MENTIONED ON CAMERA

00:00productChase AI+ / Claude Code Masterclass ↗

FROM THE DESCRIPTION

PRIMARY CTAWhere the creator wants you to go next.

OTHER LINKSAlso linked in the description.

Storyboard