Big Idea

The argument in one line.

The difference between a Claude install that does nothing and one that reads your inbox, manages your calendar, and works while you sleep is seven configuration steps — and most people skip all of them.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You downloaded Claude Cowork and stared at a blank screen with no idea what to do next.
You use Notion, Gmail, or Google Calendar as your daily operating system and want AI genuinely wired into all three.
You are a non-technical solo business owner, freelancer, or content creator who wants real automation without writing code.
You have tried AI tools and gotten generic context-free output and suspect the problem is setup not the model.

SKIP IF…

You are a developer looking for Claude Code or API access — this covers the Cowork desktop app only.
You already have a structured agentic workspace with memory files, context maps, and scheduled tasks running.

TL;DR

The full version, fast.

The gap between a Claude install and a working personal assistant is configuration, not technical skill. Seven layers close it: a dedicated workspace folder, global CLAUDE.md instructions that Claude reads at the start of every session, an About Me folder with memory and writing rules, app connectors for Gmail, Calendar, Notion, and Chrome, built-in skills for creating real files, custom plugins for specialist tasks, and scheduled tasks that fire automatically on a timer. Each layer compounds the one before it. By the end, the system triages email, writes in your voice, navigates your Notion workspace by context map, and runs a morning briefing before you finish your coffee — no code involved.

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 – 02:00

01 · Cold open and product pitch

Hook story (blank screen to full system), preview of seven steps, early-bird pitch for Cowork OS template.

02:00 – 08:00

02 · Step 1 — Install Claude desktop

Download from claude.com/download, set workspace folder, choose model (Sonnet vs Opus), live desktop-cleanup demo.

08:00 – 21:00

03 · Step 2 — Global instructions CLAUDE.md

Writing global rules; prompting Claude to generate its own CLAUDE.md; pasting into Cowork settings so rules apply across all sessions.

21:00 – 33:00

04 · Step 3 — About Me folder

Creating about-me.md, writing-rules.md, and memory.md; prompting Claude to research anti-AI writing style; updating global instructions to read all three files each session.

33:00 – 45:00

05 · Step 4 — Connectors

Claude in Chrome extension, Gmail, Google Calendar, Notion MCP; live inbox and calendar triage demo; Notion context map for token efficiency; drafting email replies.

45:00 – 51:00

06 · Steps 5 and 6 — Skills and plugins

Built-in skills (docx, HTML slides); creating outputs and projects folder structure; Anthropic plugins (Legal, Engineering, Apollo); specialist sub-agent builder; customer support plugin customization.

51:00 – 47:53

07 · Step 7 — Scheduled tasks and CASA framework

Weekly briefer and weekday inbox triage setup; Dispatch for mobile control; computer use settings; CASA framework close; CTAs for free guide, Agent OS, and channel subscribe.

Atomic Insights

Lines worth screenshotting.

The single most impactful setup step is writing a CLAUDE.md global instructions file — it is also the one most people skip.
Vague instructions produce vague output: the more specific your About Me files, the less Claude sounds like a generic chatbot.
A Notion context map stops Claude from burning through tokens hunting for the right page.
Claude Cowork only creates email drafts; it never sends unless you explicitly tell it to.
Scheduled tasks transform Cowork from a chat tool into an autonomous assistant that does work while you sleep.
The memory.md file is what makes Claude remember project state across sessions — without it every conversation starts cold.
Writing rules that explicitly list AI cliches to avoid are more effective than telling Claude to write naturally.
Plugins are not the same as skills: skills are tools Claude picks up, plugins are specialist agents with their own knowledge bases.
A permanently-on desktop computer is now a strategic asset — it is the machine your scheduled tasks actually run on.
Dispatch lets you control your desktop Claude from your phone, turning it into an always-available remote assistant.
60 percent of people who tried to build an agentic AI workspace reported being overwhelmed by their own information.
Ask Claude to write its own CLAUDE.md rules from your prompts — the output is usually better than what most people would write manually.
Connecting Notion via MCP gives Claude write access not just read access — it can update tasks, pages, and databases directly.
The workspace folder boundary is your safety perimeter: Claude can only see and touch what lives inside it.
Prompting Claude to research anti-AI writing style and encode it into a rules file produces more reliable voice-matching than vague style preferences.

Takeaway

Seven layers that turn a chat tool into a real assistant.

WHAT TO LEARN

Generic AI output is almost always a configuration problem not a model problem — and fixing it takes one morning.

02Step 1 — Install Claude desktop

Setting a dedicated workspace folder is a safety boundary as much as an organizational one — Claude can only see and touch files inside it.
Sonnet handles most tasks well; Opus is worth the token cost only for ambitious, multi-step projects.

03Step 2 — Global instructions CLAUDE.md

A CLAUDE.md file at the root of your workspace gives Claude standing instructions it reads at the start of every session, eliminating the need to re-explain your preferences each time.
Prompting Claude to write its own CLAUDE.md rules from a brief you describe usually produces better output than writing the file manually.

04Step 3 — About Me folder

The About Me folder (about-me.md, writing-rules.md, memory.md) is what makes an AI assistant feel personal — without it every session starts cold and context-free.
Explicitly listing AI writing cliches to avoid in a writing-rules file is more effective than telling Claude to write naturally.
The memory.md file should be structured so Claude appends new session learnings to the bottom and updates existing entries when they become stale.

05Step 4 — Connectors

Connecting email and calendar does not mean Claude can send anything; with proper global instructions it only drafts, and you approve before anything goes out.
A Notion context map prevents Claude from burning tokens hunting for the right page.
Context maps work for any tool with complex structure: Obsidian vaults, Google Drive folder trees, and local file hierarchies all benefit from the same approach.

06Steps 5 and 6 — Skills and plugins

Plugins differ from skills in a meaningful way: skills are tools Claude picks up when needed; plugins are specialist agents with their own knowledge bases that persist across all sessions.
The outputs and projects folder structure gives Claude a consistent place to save its work, which means you can find what it created without searching.
Custom plugins built with the specialist sub-agent builder can encode domain expertise into a reusable agent that anyone on your team can invoke.

07Step 7 — Scheduled tasks and CASA framework

Scheduled tasks are the leverage point most users never reach: a morning inbox triage that used to take 30 to 40 minutes runs automatically and delivers a summary before you sit down.
The Dispatch feature turns your always-on desktop computer into a remote agent you can direct from your phone.
Before building elaborate automation, consolidate your information into structured files: 60 percent of people attempting agentic setups report being overwhelmed by their own disorganized context.

Glossary

Terms worth knowing.

CLAUDE.md: A markdown text file placed at the root of a Claude Cowork workspace. Claude reads it automatically at the start of every session and treats its contents as standing instructions.
Cowork: The Claude desktop AI assistant product. It operates within a designated folder on your computer and connects to external tools via MCP connectors.
MCP (Model Context Protocol): An open standard that lets Claude connect to external services like Notion, Gmail, and GitHub. Each connector exposes specific tools Claude can call.
Scheduled Tasks: A Cowork feature that runs a configured Claude session automatically on a time schedule using all your connected tools and instructions, even while you are away.
Dispatch: A Cowork feature that lets you send instructions to your desktop Claude from the mobile Claude app, turning your home computer into a remotely controlled AI agent.
CASA Framework: Consolidated Agentic Systems Architecture — a four-phase model (Consolidate, Architect, Systematize, Activate) for building a fully AI-ready digital workspace.
Context Map: A structured markdown file that lists and describes all the databases or folders in a connected tool like Notion, letting Claude navigate efficiently without trial-and-error.
Specialist Sub-Agent Builder: A custom plugin that generates new Claude Cowork plugins with their own knowledge bases and skill sets — a tool for building specialist AI agents without code.

Resources

Things they pointed at.

02:02linkclaude.com/download ↗

07:30toolWhisper Flow (WisperFlow)

33:00toolNotion MCP connector

39:30toolApollo connector

38:30productAnthropic Legal plugin

41:30productbettercreating.com Specialist Sub-Agent Builder ↗

47:00productAgent OS and Life OS Notion systems

46:36linkCASA Framework free guide

Quotables

Lines you could clip.

18:36

“Do not build a spaceship when a bicycle will do.”

Punchy single sentence, a rule Claude generated for itself, lands as a standalone principle with zero setup.→ TikTok hook↗ Tweet quote

45:51

“What used to take thirty, forty minutes now takes five.”

Concrete time savings claim, no context needed.→ IG reel cold open↗ Tweet quote

19:58

“Vague instructions get vague outputs.”

Universal principle stated in five words.→ newsletter pull-quote↗ Tweet quote

01:20

“I cannot believe I am just telling everyone this for free.”

Classic pattern-interrupt hook that baits curiosity.→ 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

00:00So two months ago, I downloaded Claude Cowork to my Mac and stared a blank screen for about ten minutes before closing it. I had no idea what to do with it.

00:10Today, it reads my email, manages my calendar, writes in my voice, and connect to my Notion workspace and second brain. It can create PowerPoint presentations, clean up my desktop, and code entire websites for me.

00:24And with the release of scheduled tasks in Cowork, it even works for me while I sleep.

00:30The difference? Seven steps. And I'm gonna show you exactly what each one is.

00:35Trust me, by the end of this video, you'll be ahead of 99% of people using Claude and Cowork on their Mac. And it's only really a morning's work.

00:45No code, no tech background needed. If you can install an app and type a sentence, you can do this. Trust me, you have to try it.

00:53Even if you've never touched an AI tool, I don't know how to emphasize this enough. You will have a fully working assistant by lunch. And that safety question everyone has, well, should I really let AI touch my stuff?

01:05I'll answer that too properly so you can trust it. I can't believe I'm just telling everyone this for free.

01:11Once you've watched this video, if you'd like to have a fully ready made version of a Cowork OS system that basically does all of this and more for you and onboards you, leading you through the process, I've been developing my Cowork OS templated system.

01:28It's gonna build out the entire system for you with all the folders, plugins, and skills. I've created an onboarding coach, a system review, and first week guide, and it literally onboards you to Claude Cowork at a level that I'm gonna teach you how to do here and more.

01:46And I've designed what I'm calling my flagship feature, a system review that will help you improve Claude Cowork automatically using the system.

01:56So if you want to fast track to your best Claude Cowork experience, make sure you go and download Cowork OS. If you're watching this in the first couple of weeks of release, click that link below.

02:07You'll get early bird access to the template system from me. And otherwise, well, you can probably just download it now if you turn up a bit later. Enjoy.

02:17Step one, before you do anything else, you'll need get Cowork onto your Mac or rather the Claude desktop app. So you're gonna go to claud.com/download, and then you're gonna click whichever you're using.

02:31I'm a Mac user, so that's why I'm demoing. Download the Mac app, drag it into applications, and open it.

02:37That's it. Simple as that. And you'll see something like this.

02:41You need to sign up to a Claude account. If you haven't, it's free. But I actually recommend picking the pro plan, which is £15 in The UK a month, and that's a fantastic option.

02:54I'm actually paying this max. And I have to say, it's the best investment I've made this month. By far, it has 10 x or at least five x'd the work I've got done.

03:03So it genuinely is a very, good bet. And what you're gonna end up with when you're loaded up and have your account is something looking a bit like this. You'll see there is the chat feature.

03:13Now Now you're probably used to seeing this if you've used any LLM, and you'll have your projects if you already have a Claude account, which you can work with here. This is what I did years ago online, and all of my old systems are here.

03:24But we're not gonna worry about that. We're gonna go into the co work tab. And, of course, you have a very pleasing, easy to use interface with Claude Code if you want it.

03:34Now for those of you that don't know, Claude Co work is designed for your work. It works on your computer in a folder or in your folders, and it's actually a wrapper which is using the Claude code base model. Now the first thing you should do once you get into here is set it up in a folder.

03:52So we're going to select a folder to put it in. This is my current one. I've just got it in users, and I've called it Claude co work.

03:58But we're gonna choose a different folder. We're gonna go to documents, and I'm gonna create a new folder here and call it Claude Cowork demo, because I'm gonna show you how to set this up again.

04:10Let's create that. There you go, and open it. And I'm gonna allow.

04:14Now that means that Claude can only work in that folder. Now you should think of this folder as your shared workspace with Claude. Everything you want Claude to do in Cowork goes in here.

04:25Everything you don't stays outside, and that goes for what it can see as well. The next thing I think you should do is click here and select the model you wanna work with.

04:35I recommend turning on extended thinking. SONNET is good for most tasks. Opus is absolutely incredible and super powerful, but it will burn through more tokens.

04:45I'm on a max plan though, and I've been absolutely rinsing it, and it's absolutely fine. So depending on what you choose to pay, it's worth playing with this because it is amazing at doing a lot of different tasks. But, yeah, uh, this for faster quick answers.

04:57I recommend Sonnet to get started, Opus if you wanna do more ambitious projects, as it says here. We'll talk about these scheduled tasks later because you gotta stick around for that because it's probably the best thing. But there's a lot more we need to do before this truly becomes a personal assistant, and it's very easy to do.

05:13Before we move to step two though, let's try a simple task to demo Claude Cowork in action. And what I'm gonna do is I'm actually going to break my rule immediately.

05:23I'm gonna go to desktop, and I'm gonna allow that. Okay.

05:26Hey, Claude. I'd like you to go ahead and clean up my desktop or downloads files. Organize them into folders by category and rename anything that might need a better description.

05:37If it looks like I've already named it the file, leave it named as it is, and there you go. And, actually, I'm just gonna say desktop. Make sure you don't touch or do anything to the Claude Cowork demo folder.

05:49Now I could set this going, and it's gonna ask, are you absolutely sure you want to do that? I'm gonna say allow.

05:58Actually, before you do anything, don't mess with anything that's already within a folder, only files that are on the desktop. I'm just gonna clean that up so it definitely doesn't mess with things. So as you can see, you can also add a little extra comment in the queue and when it's ready to read it, it will also make sure it checks that.

06:17So it should leave everything that's in folders as they are and then just clean up these files for me. Got it. Only loose files leaving existing folders there.

06:27And here we go. Here's what it sees. Screenshots, screen recordings, FigJam PDF exports, documents.

06:35I'll also clean up these items. Now as Claude works, you'll see you have this progress view on the right hand side, and you can see the tasks it set itself. So it's really amazing.

06:46This is Opus giving itself a set of tasks and working through it. A really nice thing to do when you're using it is just open this up and see what it's doing. You'll see it's going through its list, following certain bits of information, and that can be a really nice way to work out how it's working.

07:02You'll also see instructions that it's read. It's got a Claude MD instruction file that's on the desktop with nothing in it, and we'll get around to that in a moment. And then context.

07:12This will show you any tools it's referenced, which we'll get onto in a little bit. Now if you're unsure about something like this, are you sure you wanted to delete something? Well, first of all, great.

07:21It's checking. But you can also go, what files are you asking to delete, please?

07:29And I can just reply to it like that, and there you go. It's really simple. You don't have to agree to anything you don't want.

07:35And immediately, everything cleaned up, labeled, and renamed as needed. So these reference PDFs have been adapted and dropped in.

07:44Fantastic. The other thing I truly recommend using something like this is get yourself a voice transcription or dictation or interface.

07:53I use Whisperflow. I've linked it below, and you'll see why because it just saves so much time. So there you go.

07:58That is what Claude can do. It can interact with your computer. But right now, co work doesn't know anything about you or how you work.

08:06It's just a blank slate. Step two is where the real setup starts. Step two is write some global instructions.

08:15This is the single most important thing you'll do in the whole setup, and it's really easy to miss. Cowork has a settings area down here.

08:23If you click on your name and go to settings, you'll see a a bunch of settings, and you can look through these, and we'll get into these in a moment. But this is where you can write global instructions.

08:33So if you did it here, this would be instructions for everything you do across chat, Cowork, and code, and you can absolutely do that. But I recommend potentially having one set of instructions for co work.

08:45So you click on the co work tab, go to global instructions, and we're gonna click edit, and we can create instructions in here. Now these are rules that it reads at the start of every conversation. Think of it like handling a new assistant a manual on their first day.

09:01And this is simply called a Claude dot m d file. Essentially, a markdown file, a simple text file that can be at a system level, and you can then add it to any folder to give that folder its own specific instructions as well. So you can either write your own or you could get Claude to help you.

09:19So let's do this. Hi, Claude.

09:23I'd like to create some global system instructions for you that you read every time you work with me no matter what folder within this main folder you are working within. Could you suggest what a solid system instruction would be? Now here's what I would want to put into it.

09:39I'd like to include in this a couple of suggestions. Who I am, like my job, that I'm a non tech user, and so I'd like you to always talk to me in non technical language and explain things as you go so I never get overwhelmed by technical information.

09:54I'm a content creator and theater director. Speak to me in plain English, never jargon. Be direct.

10:01Don't overexplain. If you can just do the thing, do it. So share with it also what you need help with.

10:07I work with email, Notion, Google Drive, and content creation. My preference is a British spelling, warm but direct tone. Challenge me if there's a better way.

10:16Uh, let's add some safety rules. Never delete, send, or publish anything without first checking with me, and any other recommendations you had for a really solid system instructions. I would also add that I want you to be honest with me.

10:28If you think there is a better way of doing things or that I'm risking over overcomplicating or going too far down a rabbit hole, you should flag it. So you can just talk to it like this. I'm gonna click let's go, and let's see what it does.

10:41So it's looking at this workspace and found an empty directory. So we're gonna create a file called claw dot m d, and it lives at the root of your workspace and picks up automatically every session.

10:52Great. So it's explaining everything to me because I've said I want that. That's what you need to do.

10:57Make sure you tell it that. Done. Here's here it is.

11:01Don't build a spaceship when a bicycle will do. Great. Clear recommendations for overlong prompt options.

11:07Scope drift check. No overwriting. Sounds good.

11:10Let me view the file, and then you can take a look at it here, which is great. So now if we go to the folder, we can see the Claude file, and there it is.

11:22Fantastic. Now one thing I recommend is this. Is it worth me dropping this into the Cowork global instructions window in settings for Cowork so it's pervasive across all folders?

11:38And here you'll see. Yes. Absolutely.

11:40Should we delete the old version if you're always gonna read the one in the global instructions?

11:46So you can have a little back and forth of it. Great. So what we're gonna do is we're gonna view the file, select everything in it by clicking copy, go to our instructions, co work, edit, paste it in, and save it.

12:03So that means Claude will always follow those instructions. And, of course, remember that you can add secondary Claude dot m d files in any folder, or you could set an instruction to say, when I say downloads, clean up my downloads like this. Simple as that.

12:17So now Claude Cowork knows your rules, but it doesn't know you yet. So my step three is to fix that where Cowork becomes a much more effective assistant with personal context, a memory, and a better writing style. These ones are where you suddenly become more advanced than most users.

12:35So step three is to create an about me folder. Take a little look at my current Cowork setup. You'll see I have this folder here, which if we click into it, we have a couple of other items in here.

12:47Now we're not gonna get into absolutely all of this because some of these are, like, for another video. And this folder is going to be where you teach co work who you are and how you want it to act. I have three files in mind.

12:59I have about me, which shares who I am and what I do. I have a memory file, which is where Claude will log memory of its conversations. And I have a writing rules file, which is gonna inform how it speaks and to avoid that AI writing style.

13:17So to show you how simple this is, let's ask Claude to make them for us while I explain to you and Claude what they should be. It really is that simple. So we're gonna jump back to Claude and say the following.

13:31Hey, Claude. I'd like you to create an about me folder within your coworker folder, which is gonna be the place that holds three key pages that will inform how you act every time we communicate.

13:43Once you've created them, I also want you to give me an update to your global instructions to ensure that you always read these three pages every time we start a session.

13:55I'm gonna keep speaking to it. File one is gonna be an about dash me dot m d.

14:03That's gonna include my name, what I do, what my business is, my tools, that kind of stuff.

14:09Let's make a naming convention for all MD files in the similar style to what I've written here. I think the about me MD could also cover my audience or customers, my current projects, everything a smart new team member would need to know about me on day one.

14:29Let's make this, and then we'll also make file two, which is gonna be writing rules dot m d. This is how I like things to be written. And to do this, I'd like you to go and research anti AI writing style in Wikipedia, find the page on that on Wikipedia, and then use it to create a broken down description of everything you should avoid when you write to not sound like an AI.

14:54Finally, I'd like you to create me a memory MD, which I want you to update to be a system that you write memories to.

15:03You append to the bottom of it, or update existing entries if it relates to them and is necessary. And this file should be the thing that prevents you from forgetting where we are with projects and what we're doing.

15:16You then need to reference all three of these pages for the global instructions to make sure that you know how to use them. Ask me any questions needed, but otherwise, go ahead and build those for me. Okay.

15:30Let's see what Claude does. Now this memory dot m d, this one's clever. It's a log where you will record global preferences, decisions, context after each session.

15:41So you make sure that your global instructions will remind you to always fill that out. That's gonna be really essential for Claude because it means that it's going to be always updated and clear on where it's at.

15:54So here we go. It's asking me some qualifying questions. Who will your main audience be?

15:59I'm gonna say it's gonna be all of these. Great. And then you can write something else in here.

16:06Why don't we do that? So now it's worth saying that Claude has a little bit of context about me already. Right?

16:14Because it's already connected into my workspace with connectors, but we'll talk about that a bit later. So what are your current active projects? I would say all of those.

16:27What's your core mission? And let's say that. Oh, actually, it's also to help business owners work with AI and to share intentional tech on the Better Creating channel.

16:38I have two channels, Systems Made Better and Better Creating. Better Creating is the tech channel for intentional tech, Apple products, desk setups. Systems Made Better is to help freelancers, solopreneurs, workers, founders, business owners leverage AI and Notion and other systems without being tech experts.

17:01I'm just gonna add that in. So you can see it's been working in a few different ways. And something that this system does is it can work with multiple agents at once, which is pretty crazy.

17:13So this is now working through the science of AI writing. It's dragging out key bits of information. It's building it all out.

17:20It's totally mad. Let's go and take a little look and see if anything's turning up. Well, we've got our about me folder and our about me file.

17:28We can take a look at that. And here it is. It's given quite a lot of information.

17:33Systems made better, better creating, core tools, how I think this is pretty cool. Right?

17:39It's giving a load of different information, and it's also obviously got a bunch of other bits of information I've already given it because I'm I'm trying to demo this, but you might have to tell it a few more things. Now we've got our memory MD, and there is session log, the active projects, and key decisions, preferences, append new entries below this.

17:58You can see that's all working really beautifully. And we've got our writing rules, and it's created a detailed writing rules, banned phrases, no superficial ink commentary.

18:08It's really cool. You can kind of dive into it, but, yeah, simple as that. I love that little bit of advice.

18:14Very powerful. So let's go and see what it said. Great.

18:17It's done it. 14 rules, system memory, and then we can go check them out.

18:21Excellent. Please tell me what I need to adjust in my global instructions, and I will paste them in.

18:29So don't forget, if you've put them into that setup, if you don't want to just leave them as you've got it in the folder, you need to go and do this. And look, it's really simple. It just writes the whole thing out.

18:38Non negotiables. Another example of how to keep things solid and safe, and these are at the start of every session. Read these.

18:46Should we move the required reading for every session to the top of the entry?

18:53Great. Yes. Go ahead.

18:58It's updating the global instructions. I can now just click in here, copy everything, and then we can just jump back into our settings.

19:07There we go. Save. And once you've got these set, you don't really need to change them.

19:12So the way to think about it is once the global instructions are solid and directing to these other documents, you won't have to touch them for a long time. There you go.

19:20Next time you start a conversation, it reads this file and remembers what happened before. It's like giving Cowork a notebook. It checks every morning.

19:29The more specific you are with these files, the better the results. Vague instructions get vague outputs.

19:34So let's start a new task, and let's say, tell me about how you are meant to work with me.

19:42Now what I should see it do is now read those instructions. Let's watch. Boom.

19:50You can tell it's worked because it's dropped in, that it's read all of that context. Wow. So in just a few minutes, you now have a version of Claude Cowork that knows your rules, knows who you are, and remembers what we've done together.

20:03You are now further ahead. But it gets even better. It's time to talk about customization, skills, and connectors.

20:12And let's start with connectors because connecting tools you already use transforms Claude into a whole new level of power. And the first connection is gonna surprise you with how much it can do.

20:25After that, I'll show you how we can download or create skills and plugins to supercharge Claude co work systems with specialist skills. So step four, you can access this either through settings or just click customize and click connectors.

20:42You'll see here I've already got a bunch of things connected. Notion, Slack, Superbase, and Vercel, they're connected to create websites.

20:49I've got make.com automations I can call upon. I'm using Google Calendar, Google Drive.

20:55It's amazing. Tools like Gmail, which we'll set up in a moment, which means that Claude can genuinely manage your emails. It's amazing.

21:02But first, we just wanna talk about Claude in Chrome because Claude can browse the web for you. This is one thing that's worth installing straight away if you're comfortable. It's a browser extension that lets co work actually browse websites, read pages, pull information, fill in forms, even click through multiple step workflows.

21:20Instead of copying and pasting URLs into the chat, you just say go read this page and it does it. It's amazing. So in here, click on Claude Chrome and you just enable it.

21:30Now you need to install the Claude in Chrome extension for the Chrome app. But if you just ask Claude to do that, it will do it.

21:38So actually, why don't we just say, I'm demoing how to set up Claude in Chrome. Could you model what you would do to help me install it? Or even just direct me to the link for it in Chrome.

21:53So here you go. This is the kind of thing you could do. Just ask it.

21:56We can then click, open the link, and there you go. You'll get to a page a bit like this, and you can just click install. So it is now up here.

22:04I have it, and it has a couple of things you need to be aware of. It's a beta feature. It has risks included.

22:10So this is up to you whether you wanna try it, but I am finding it's absolutely brilliant, and it will search the web for you. So why don't we do a little demo of this? Could you pull me up a popular Reddit thread on using Claude Chrome, please, in the Claude browser.

22:28So if we now go to Chrome, there you go. Here comes the tab.

22:36And for those of you that are worried about permissions, this is really useful to see. You can allow things that's due this time only. And there you go.

22:42It navigated for us. This isn't the perfect example, but you can often when you're building things out, see it do unbelievable things.

22:50This is a bunch of views that it's been working in in the past. You want to, you can just delete a group and it will just shut those down in case you don't need to see them. Yeah.

22:58So that's great. So you'll see that it was it was researching for me websites that I quite like the design of to inspire a website plan.

23:07Amazing. So back in connectors, we've got that set up. So if you're comfortable giving it browser access, install it now, thirty seconds and you're done.

23:14If not, no worries. Everything else works without it. It's a bonus, not a requirement.

23:19But I have to say it's been very helpful. Right. Now let's connect your actual apps.

23:23I've got all of these already connected. Let's do Gmail. There it is.

23:27It's as simple as this. Find the connector. You can search for connectors here.

23:32You can click add and browse connectors, and there's a bunch of them here. I mean, look at this. Essentially, what I recommend doing is getting on here and anything that you use, click the plus button and add it.

23:43It will always ask permission to use it. It is fantastic. It basically is just gonna be like what you need.

23:48I'm using a huge amount of things in here. I might even connect my PayPal at some point and see what that does, but that gets a bit scarier, doesn't it? But, yeah, look, Wix are in there.

23:56Marketplace, there's more here since last time I looked as well.

24:00So the list is growing exponentially. So let's do the obvious one, Gmail. I'm gonna click connect.

24:07It says, do you want to connect Claude to Gmail? Continue. I'm gonna select my email address.

24:14Continue. Now you can only connect one account these days. Hopefully, multiple accounts will come soon, but there you go.

24:23It's kinda done. And you can just decide within Gmail what permissions you want to give.

24:30Now for other systems such as Notion, there's lots and lots of permissions, right, that you could adjust depending on what you're comfortable with allowing it to do or if you want it to need approval or just it's blocked from doing. It's pretty simple to do. Connect that.

24:43I really recommend connecting calendar. That will allow it to search your calendar as well and check what's going on for you in the week and so on and so forth, create calendar events, all the rest of it. So why don't we test calendar in Gmail?

24:56I'm just testing your connectors. Please, could you check my inbox and my calendar and let me know anything that's important for tomorrow?

25:07So it's now loading tools. You can see them here. You can open them up and see what it's doing if you really wanna understand it.

25:14You'll see that it has loaded Google Calendar and Gmail. It's listing events, and here's the rundown. So let's just shut this up.

25:22Your morning kicks off with a weekly meeting with Lisa. Great. I've got partnerships, and then I'm going to the Adobe Create a Live event one till 07:30.

25:31I've also gotta go to selfages and return a top. There you go. And then a bunch of emails worth noting from different people, speaking at Notion AI labs for startups.

25:39We've got a check-in with my strategist, couple of people I'm working with. Look at that.

25:44And some brand collabs. I got I made five eighty three dollars on a pay for like commission. Love it.

25:53So why don't we just say, could you draft me a reply to Ben Farr based on the conversation? I will likely want to book a meeting with him. Let's ask him for availability.

26:05And because I've said book a meeting, my WhisperFlow has inserted my Calendly link. Really cool. So let's just see if that works.

26:15Now let's imagine that I was happy with that. You could just send it. But I'm actually gonna say this.

26:20This is great. Please put it in a draft in the thread ready for me to send. In future, just create the drafts in my email for anything I ask you to do, and I'll be able to check them there.

26:31You can just share me a list of what it is. Please update your memory or instructions as needed to make sure you do this.

26:38And in that way, we can give it feedback to update notes so that it always does these things. So the safety question everyone asks, should I let AI read my emails? Well, there you go.

26:48Two things. First, co work creates drafts only. It doesn't send unless you tell it to.

26:54Second, your global instructions, if you've created them like I have, already say never send without me checking them. But on that note, if you're feeling a bit more risk averse while it updates our memory and it's dropped the Gmail draft in, I recommend going into settings, going into privacy, and within this view, making sure that help improve Claude, sending sessions to anthropic AI models, I would turn that off.

27:18If you wanna be helpful, great, but this is a way to ensure that not many bits of information are dropped in. And for proof, here we can see Ben's email to me, and the draft is ready to go.

27:30Now on connectors, I do wanna talk about Notion because this is where connectors for me get super serious. This one is personal because Notion is where I run my entire business. If I jump over to Notion and we take a little look at it, this is where I keep projects, knowledge, meeting notes.

27:47I even have down here an agent brain section with specialist knowledge on things like YouTube strategy with a vector database of information on how to do better creation. Or, for example, an online business knowledge base that helps me with strategy around product marketing and brand development.

28:08If you connect it in the same way through the connectors, what happens next is different from email and calendar because Notion holds context, not just messages and meetings. So in this system, I have everything about my business.

28:21So get that set up, and then we're in a very different situation. If I'm gonna go back to my chat about the global instructions, can you search my Notion workspace for my agentic business method big knowledge project?

28:34It'll be in the project's database. So Notion uses something called an MCP server.

28:41This allows it to query, write, read, do loads of different things. I cannot tell you how powerful having this local connection to it is.

28:50Well, it's found it. Okay. So this is cool.

28:52Right? But it took quite a few attempts to find it. It has now found it, and it's gonna find information about it.

29:00But it's big. It's a detailed project page. This is your flagpitch product ecosystem in progress.

29:06We're going to be building all these things out. Great. These are all my current plans.

29:10Very exciting. Very exciting. But I would like to now be able to better understand the system.

29:17I'd like it to be able to read it more effectively. So if you're a Notion user, I recommend doing something like this. In my about me page, I'm adding a my context map.

29:26So what I could do is ask it to search this page by just clicking the copy the link to page, which has all of my databases within it, and then get it to create a context map of my Notion system. Look at that. So I've built that in.

29:40So I'm just gonna copy the one I've made, about me, drop it in. I've just dropped a my context map MD file in your about me folder, which gives a clear breakdown of my Notion workspace for you to more easily find things in the future.

29:56Please, can you share with me an updated Claude dot MD file so I can paste it into your global instructions referencing this so it can be updated and always looked at. So there you go.

30:07It's found that. It's a sat nav for your entire workspace. Fantastic.

30:12It loves it. It's loving its own work. Basically, I made it with it when it was a previous version.

30:18And then, of course, you would just take that, paste it into global instructions, and you're good to go if you're using that method. Okay. Following this, please could you check what tasks I have due this week in my tasks database?

30:30Let's now see if it does a bit of a better job. Now you can see the difference now. It immediately found the exact database, got a load of data back, and it's now running commands to give us the results.

30:42Now with a context map like this, even if you don't use Notion, make one for whatever files or systems you use and add it to your about me, and you will have co work connected to wherever you keep your plans, projects, notes, knowledge base. Combined with this about me folder, it understands who you are, how you work, and what you're working on.

31:02The other advantage of creating context maps for your connected tools is that it stops Claude burning through so many credits or tokens trying to find things. Awesome.

31:13So look at that. It's found my tasks. I've got 41 overdue tasks.

31:19I have not cleaned these up, but it might help me triage them. So this is a really brilliant way to clean them up. Would you mind triaging the overdue tasks or updating due dates for the next week or so based on priorities?

31:34And, you know, if you're using something like Google Docs, Obsidian, the principle works the same. Give co work access to where your information lives.

31:42So here is the view. It's cleaning this up at the moment. It's getting smaller.

31:48Smaller again. I mean, look at this. It's clearing everything up.

31:51It's now actually updating its memory because it learned about my project and product information. Very helpful. So that's connectors.

31:59You've now got a configured AI assistant that knows you, connects to your tools, and remembers your history. That's the setup most people never get to. But now, what co work can actually create?

32:11So step five is to use built in skills, specialized abilities that allow it to create specific file types.

32:20It can specialize in documents, spreadsheets, presentations, PDFs. It makes all of them. Don't think of AI as the tool.

32:27Think of it as the colleague using the tools themselves. Then your entire experience of AI and agent work will transform.

32:35Now there's loads of ways to create skills. If you go to customize skills and then plus sign, you'll see you can create with Claude, write skill instructions, and upload skills.

32:45And there's examples which you can turn on here and then customize to your own needs. For example, canvas design, for example, visual arts, PNG, and PDF documents, quite nice. We could just turn it on and you've got it, basically.

32:58It will do it when you ask it to do it. This is worth turning on, the skill creator, because it will help you create them. So for example, why don't we do this?

33:06Write me a one page project brief for the AgenTic business accelerator I'm planning as a Word document. I'm gonna do this in Sonnet to save some credits.

33:18You can see it's reading the doc x reference file there. It's loading its tools. It needs a bit more information.

33:24And what's great is it wants to be precise and clean. And it's using something called ask user question. Little tip, if you build your own skills, which you can just ask it to do using the skill creator skill, you can ask it to add in the ask user question.

33:38So let's call it a course program and a community. Oh, it's a live cohort as well as those two. Great.

33:45Internal just me. There you go. Clarity and road map.

33:48So it's just gone for a couple of questions. Let's see if it builds it. So here you go.

33:52It's writing it, generating a Word document, and then it'll validate and save to the workspace folder. We should see that turn up shortly.

34:00Now, obviously, you don't wanna sit here while it does things. Ultimately, once you give it skills which break down steps to larger tasks, you can set it going, leave, and let it do the work.

34:12It's absolutely amazing. And there it is. It just turned up.

34:15It will say you can view it, but why don't we just skip to the view, open this up, and if you'd like things to look nice, you can always say, make it look pretty.

34:26But look at that. It has done a very good job of creating me a quick one page document. Just these ideas of using these tools means that you can go really far really quick, but it gets a lot better than that.

34:39So let's close that up. Since we have limited it to the Claude co work demo area, we might want to now create a bit more of a folder system in our co work system for documents that it creates using those built in skills.

34:52Then I'll show you how to make your own skills. This was great. I noticed that you're just having to drop new items into the workspace folder that you're working in.

35:00Let's create a outputs folder, which is where you should put any outputs. And if it's for a new project, create a new project folder within that outputs folder.

35:11And I think we should therefore also have a projects folder in the main workspace, which is where you add memory and Claude MD files for specific projects.

35:21If you're unsure if I'm starting a new project, ask. Otherwise, you can create a project for new items with the output folder, subfolder, and the projects subfolder matching. Any questions, let me know.

35:35Otherwise, add these to our system. And I think for this task, we'll add it to Opus. Now the point of this is that you get a much cleaner look.

35:45I have my about me section, but I then have plot outputs where each output is within a project and projects where key projects have their own memory and project brief. This is so powerful because it means that you keep context running through each project.

36:04So this is the thing to think about. I've got plugins and skills and templates here. Don't worry about those.

36:09I think today, just focus on a projects folder and an outputs folder. Give your instructions clarity that it should put outputs into here and project working documents into projects. This essentially becomes a second brain folder system for your Claude co work account.

36:25There we go. Let's take a little look at it. And there we go.

36:28We've got our outputs and our projects. You'll see in here, we've got the Word document it created. It's just put it in there.

36:34And here, we've got the Claude MD for the business accelerator that's naturally built. Amazing. Didn't have to say much.

36:40And a memory where it remembers what happened. Oh my god. It's so clever.

36:45So, yeah, you really can't go wrong. Hopefully, by now, you're realizing how simple this is. It totally blows me away.

36:52I love this. Please, could you add an update for our global instructions and make sure they're updated? You've already done it.

36:59It's already done it. Look. There you go.

37:00It's way ahead of me. Brilliant. I've got a really cool skill here that I downloaded online, which is called front end slides, and this will create rich text HTML presentations.

37:12But let me just jump in therefore and show you in my outputs an example of this. So I created this slide using the skill I downloaded, and it's amazing.

37:23All I did was share a Notion script where I talk in another video about my CASA framework and using Notion and Claude as your system for organizing your business with AI, and it created this slide and I actually use it in that video that I've just recorded. Look at the animations.

37:41It's unique. It's beautiful. This is what's possible now.

37:45Kind of makes PowerPoint less useful because why don't you just get it to create something from a document? It's absolutely amazing.

37:52Well, hopefully, that gives you some inspiration about skills. They're like giving co work professional software training.

37:59It doesn't just write text. It creates properly formatted files that you can use and pass on to colleagues within your system.

38:08So think of built in skills handling standard, simple, singular jobs. But what if we could teach coworkers something no one else has?

38:17That's what plugins do, and it's where things get really impressive. So I'm gonna go to customize, personal plugins, add plugin, and browse plugins.

38:26And you'll see a bunch of plugins that Anthropic just offer already. This legal plugin is what took a ton of value off a bunch of companies when Claude announced it. Just add it, and you've got a legal assistant that can help you do all of this.

38:42Triage incoming NDAs, prep for meeting. I mean, it's it's amazing.

38:47Just add it and you've kind of got it. So I've got legal here. It'll do these things and I can click it to do certain things.

38:53I've got an engineering one which will help me review code and bugs, design systems and services architecture, like, it's totally amazing. Apollo, if you install that, it it can search for qualified leads for your business.

39:05And I've built my own plugin to build plugins. This is the specialist sub agent builder, and this builds custom AI specialist sub agents with searchable knowledge bases. And it turns your expertise into ready to install plugins that you can use in your system.

39:22This means that you can have something like my YouTube specialist, which is a strategist, which I can then feed information, whether that be a Notion database or built into the plug in, and it can help you.

39:34So this is how easy it is to add a plug in to your business. We're gonna browse plug ins. I quite like the idea of a customer support plug in.

39:40I'm gonna click install. I'm then going to where is it? There it is.

39:44Manage it. And I'm gonna click customize in the top right, and it just says customize the customer support plug in for me based on my company.

39:54Triage and manage customers inquiring about Notion templates, Claude code plugins, icon packs, and any of the products I sell on bettercreating.com.

40:10So it's now gonna read about who I am and what I'm doing and invoke the plugin customizer skill that it has. It's gonna find that plug in that we just built or added, and then just go and do it.

40:22I mean, that's kind of it. If you briefly come back, this is customizing this plug in beyond belief. It's absolutely fantastic.

40:28It's just asked me a bunch of questions. I will shortly have a customer support plug in that works for me, and I can go and update it continually. Amazing.

40:37As a recap then, think of skills as the tools already in Cowork's toolbox or specific toolbox items that you can add. Plug ins are like hiring a specialist who brings their own tools and methods.

40:50This marketing strategist, social repurposing agent, and YouTube specialist plug ins, they were built with my specialist sub agent builder. I've packaged it as a tool that can be used to build plugin specialists, and you can find it on bettercreating.com. No code required, step by step setup.

41:06So if you want to extend co work with your own specialist agent systems with their own knowledge bases, the link's in the description. Tools connected, skills working, plugins installed.

41:17There's one more level and it's the one that blows people away. What if co work could do things for you while you were not even there or asleep? Step seven, set up scheduled tasks.

41:30I think this is my favorite feature, the one that makes Cowork feel like it's actually working alongside you. Scheduled tasks run automatically on a timer.

41:38Everything you configured, your instructions, your connectors, your context, and they do real work on a schedule. It just means that you set them up and they are triggered. All you need to ensure is that your computer is turned on and has Internet connection.

41:51So this is an argument for essentially having this on a permanent desktop computer that you leave on. Here's how it works. Let me just show you a couple that I've created.

42:01This took me moments. Weekly briefer, I just set it up to connect to Gmail every Tuesday at ten, and you just write instructions.

42:10Please search my tasks and projects in Notion, Gmail calendar, and my hello at email, and share a overview of the week ahead. That's kind of it. Before doing any research, read these files about me.

42:21Yeah. Use the Notion MCP, blah blah blah.

42:24And then it just has a guide of style instructions, how it should report, and it sends me an email. And that's it.

42:29Why don't we run it now, and we'll see what we get. And if we go back into scheduled, we've also got a weekday inbox triage. This will triage my inbox a bit like I showed you earlier and give me a full overview of what that is to deal with.

42:46And that goes at 09:30. And we can actually see it ran. So let's take a little look.

42:51It found all of these items, blah blah blah, and it gives you a little update like that. Simple as that.

42:57And you have to work out what you want the output to be, where you want it to put it. But I would suggest something like set yourself of an outputs area with weekly briefing reports, and then it will just drop them in as an MD file, and you get your answers dropped in.

43:11And if you want it to format it and turn it into a Word doc, you can tell it to do that, a PDF. You can get it as a presentation whenever you like. Let's create a little example here.

43:20We're just going to go to scheduled, new task, schedule task only while your computer is awake, and you can do this to when enabled, Claude will prevent your computer from going to sleep. So if you have a laptop, maybe don't use this. So, yeah, that's all the more reason to own a desktop computer these days.

43:36And if you watch this channel, I cover a lot of Apple ecosystem and intentional tech. I think it might now be the time for a lot of us to consider a home computer and a laptop. And the home computer is where you set up all of this.

43:49Now, since I filmed this video, Claude has shipped endlessly. In fact, they've shipped two amazing new items, which just take scheduled tasks to the next level, dispatch and Claude computer.

44:02So as you can see now on the bar here, we've got our scheduled task that I was demoing before. If we now click into dispatch, you can see a couple of things are going on here.

44:11This allows you to essentially control your Claude desktop from your phone. This is why I'm going all in on a always on Mac on my desk.

44:23So I'm gonna have two Macs, one to travel with and one that stays in the studio. You can get Claude to prevent your computer from going to sleep, meaning that scheduled tasks will continue to run, but you can also allow it to work in your browser. Claude will take actions in Chrome without asking permission, including on websites you haven't approved.

44:40So you need to be okay with that. And then most importantly, you've got computer use. So if we open the settings on computer use, you can see you can give it permission to work within your computer.

44:54You can also deny it from using certain apps. So you can add a a kind of accessibility list. So there's quite a lot of granular control, and there's kind of various different things.

45:04But the most important part of this is dispatch. So that means with the Claude app, we can open Claude up. And if we go into the sidebar, you will see dispatch.

45:14And that allows me now to speak to Claude. Please, could you find out what's going on in my calendar today? So we're gonna send that, and now we should see, as you can see here, it is talking to the computer.

45:26You now have pretty much a kind of Open Claw style assistant on your desktop that you can control when you're out and about, and it will work on your computer and get back to you. And there you go. I've got a quick breakdown of my calendar.

45:39We could ask it to go and browse information, do research, or continue with projects using the plug ins and skills that we've created. So just imagine you have one of these running every weekday morning before you finished your coffee, coworkers, read your inbox, sorted everything by priority, drafted replies for the ones that need attention, you review it, tweak it as needed, and send.

45:58Essentially, a personal assistant you can reach at your office every day from your phone. What used to take thirty, forty minutes now takes five. What's the busy work that you do that you could get this system to automate?

46:10This could be a really great combination with Notion and a business account, with personal agent for working inside Notion, and this for automating your tasks across all of your platforms. So that's the seven steps, but there's a bigger picture that we need to address before you go. Everything we just set up, the workspace, the instructions, the about me files, the connectors, the skills, the plugins, and the scheduled tasks, that's the foundation.

46:34And it's more than most people will ever configure. But if you wanna go further, I've built a framework called the CASA method. It's a framework that maps out the entire journey to build an AI ready digital home, your consolidated agentic systems architecture.

46:51And yes, Claude Cowork did build this presentation. Most people get inconsistent results from AI. Scattered information, no context, generic output.

47:01For everyone that's downloaded my free guide to building out an agentic system for your work and filled out the form, 60% reported overwhelmed with their information. The number one issue with context was no clear path, and generic output was the thing that many were struggling with. I cover all of this on my second channel, systems made better, so make sure you should hit subscribe below.

47:21And in the description, you can find a link to my free guide for this CASA framework to consolidate, architect, systematize, and activate your AI workspace.

47:31And hey, if you're a Notion user, make sure you go and check out Agent OS and Life OS, which give you all of these systems for Notion so you can connect all of the dots. If that was useful, make sure you hit subscribe and don't miss the next one because I'm gonna be showing you exactly how to take this so much further. Drop your comments below because I read every single one, and I'll see you on the next one.

The Hook

The bait, then the rug-pull.

Two months ago the presenter downloaded Claude Cowork and closed it after ten minutes, stumped. By the time this video was filmed, it was reading his email, managing his calendar, writing in his voice, and running automated tasks while he slept. The gap was not talent — it was seven configuration steps that most users never take.

Frameworks

Named ideas worth stealing.

08:00concept

CLAUDE.md global instructions

A markdown file placed at the root of the workspace that Claude reads at the start of every session. Acts as a standing brief for every conversation.

Steal forAny workflow where you want consistent AI behavior across multiple sessions without re-explaining yourself each time.

13:00list

About Me folder 3-file system

about-me.md
writing-rules.md
memory.md

Three context files that give Claude a persistent identity layer: who you are, how to write for you, and what it learned last session.

Steal forAny knowledge-worker setup where personalization and continuity matter more than raw task speed.

35:00concept

Notion Context Map

A structured MD file listing all Notion databases and their purposes. Prevents Claude from hunting blindly through the workspace and wasting tokens.

Steal forAny tool with many databases or folders — applies equally to Obsidian, Google Drive folder trees, or local file systems.

46:36acronym

CASA Framework

Consolidate
Architect
Systematize
Activate

Consolidated Agentic Systems Architecture — a four-phase path from scattered information to a fully AI-ready digital workspace.

Steal forPositioning an AI consulting offer or course around a repeatable transformation framework.

35:00model

Outputs and Projects folder structure

A two-folder system inside the Cowork workspace: Outputs (where Claude saves created files by project subfolder) and Projects (where each project keeps its own memory.md and CLAUDE.md).

Steal forAny multi-project Claude Cowork setup where you want work organized without manual sorting.

CTA Breakdown