CLAUDE COWORK FULL COURSE (2+ Hours)

In this course, I'll take you from a complete beginner to becoming an expert in cloth co-work. Cloud Co-work is quickly becoming one of the most important tools to master for knowledge workers. Kind of like cloud code already

workers. Kind of like cloud code already is for engineers. With co-work,

nontechnical professionals and business owners can start using AI agents to automate real day-to-day tasks across marketing, sales, operations, or any other business department. And when

deployed and used across a business, it can entirely transform the way a business operates. So, in this video,

business operates. So, in this video, I'll cover everything you need to know to leverage this powerful tool for yourself or your business. So, whether

you're a founder, a marketer, someone in sales, or just a professional who wants to get way more done and stay ahead in AI, this is the course for you. So,

here's what we're going to cover. First,

I'll show you quickly how to set up co-work and give you a quick overview.

Then, I'll cover file access and projects, which are important new concepts to understand in order to use co-work efficiently. Then, I'll cover

co-work efficiently. Then, I'll cover connectors, MCP, browser use, computer use, and dispatch. After understanding

these fundamentals, I'll go in depth in what I believe is the most important feature to master in co-work, which are skills. I'll show you what they actually

skills. I'll show you what they actually are, how they work, when to build them, but more importantly, how to actually build, test, and improve them efficiently, which is the key to making co-work actually useful. Then we'll

cover plugins, uh, show you what they actually are, how to leverage them to give cloud specific expertise. I'll show

you how to build your own ones, and why plugins could mean a fundamental shift in the SAS landscape. Then I'll cover using and building agents in Cloudco in order to run bulk tasks and complex

workflows autonomously. Then I'll cover

workflows autonomously. Then I'll cover scheduled tasks and scheduled skills that allow your AI agents to run automations and tasks autonomously and show you some practical use cases to automate some operational tasks. And

then after you understand all the major feature, I want to make this uh practical and show you how to use co-work and skills to automate a significant part of your marketing tasks uh from copywriting to ideation,

research, analytics and visual asset generation. I also cover how to use

generation. I also cover how to use co-work to automate uh sales tasks like prospecting outreach research qualification, uh call preps and follow-ups. And lastly, I'll show you

follow-ups. And lastly, I'll show you how to set up what I call AI operating system or a second brain inside of co-work that gives you and your team persistent context and memory across

across every chat you open. And this can scale across your team and makes co-work and all of the features that are covered here above far more powerful. I'll also

add the timestamps with the chapters here below. So, if you're already

here below. So, if you're already familiar with some of these concepts, you can skip to the part that's most relevant to you. Now, Cloud Co-work at the time of filming is currently available on the Cloud Pro team or enterprise account. It's not available

enterprise account. It's not available on the free tier yet. And second, it's available on the Cloud desktop app, not in a browser yet. So, it's desktop only.

So, once you've downloaded Cloud Co-work and you're minimum on a pro account, you'll see a second tab here next to chat, which is called Co-work. Now,

although the Cloud Co-work interface might look very similar to the original uh chat interface, Co-Work is actually far more powerful because under the hood, co-work is actually an AI agent, it can do much more than just questions

and answers. You can see that we also

and answers. You can see that we also have the cloth code tab which behind the hood works very similar to co-work but has some extra features that make it even more powerful. But this mostly is

for engineering tasks. For most

non-engineering tasks, co-work is really all you need. Has a more intuitive UI and for businesses it can be a superior setup because it has permission settings and collaboration features built in to

safely roll this out and start using these agents across your company. Also

once you start understanding cloud co-work and the concepts cloud code becomes a lot more intuitive too because much of what we're doing is exactly the same. So once you start using co-work

same. So once you start using co-work and more and you want to use some of the extra capabilities using cloud code becomes easy and I'll show you some examples of this later in this video too. Now cloud co-work was only launched

too. Now cloud co-work was only launched a couple of months ago and entropic is trying to make cloud co-work the cloud code for nontechnical knowledge workers.

So this tool is and will become more capable by the week. So at the time of you watching this video some things might already look a little bit different in the UI. However, these

fundamental concepts that I will cover will still apply now. And the first thing you'll notice in coowwork is that you can give clot access to a folder on your computer here. This means a couple of things. First, you can give clot

of things. First, you can give clot access to real context about your business or your project with a single single folder select. For example, here I've selected a folder um with my

YouTube transcripts, a document on my ICP and my YouTube strategy. So anytime

I want to use claw to help me with any YouTube related task like research, ideation or helping me write the script, I can just give it access to that folder and it will already have all of the context without me copying and pasting

all the time. So if I now ask Claude something like can you help me ideate on some new YouTube videos based on my uh YouTube strategy.

So you can see Claude instantly has access to all of these context documents like the YouTube strategy and ICP. It

checked what videos I've already up published this year and based on that it gives me 10 actual relevant video ideas according to my strategy. Clot can now control your computer etc. But second in

contrast to previous projects in chat clot doesn't just read these documents here. It can also create edit update any

here. It can also create edit update any file directly into that same folder including presentation CSV sheets Google docs and other file formats. So if I actually generate a script or a

presentation for my video, I can directly save them in the folder. And

also when I make changes to my YouTube strategy, you can directly update the strategy doc. And this is really why

strategy doc. And this is really why this feature is so important because no matter how good these N&M and these AI agents become, context is really the key to get good outputs with AI and file access will give cloud co-work

persistent context around your project, your business, and your tasks in every new chat you start. Now later I'll show you plenty of examples of what context documents to use and when but just understanding the importance of this

feature is enough for now. Now let me quickly cover projects which is a new updated version of the old claw chat project. Now you can imagine if you have

project. Now you can imagine if you have a lot of different projects or tasks or department in your business and for a specific type of tasks you want to have a specific type of context you can make

a project for each. For example, for my YouTube related tasks, I've set up a project here that's connected to my YouTube folder with all of these YouTube related uh context documents. And as you

can see, this YouTube project is already connected to my YouTube folder here. So,

in this project, I don't even have to manually select the folder each time. I

can also see all of the chats I had around YouTube tasks organized below.

And besides that, every project can now also have specific memory. So, I can tell CL something to memorize that's specifically relevant for this project.

For example, here I have a rule to always mimic my tone of voice from the transcript when helping me write scripts. And you can do this by just

scripts. And you can do this by just telling claude, please help me memorize this. And it will automatically save it

this. And it will automatically save it there. You can also see scheduled tasks

there. You can also see scheduled tasks related to this project, which I'll cover later in this video. And we can give it specific instructions more or less like we did uh with custom GPTs or oldcloud projects like a system prompt.

So how do you set up these projects? You

can just go here to the sidebar, click on project, create new project, and there you'll get three options. You can

go and start from scratch where you add the contacts manually. You can import it from an old Claude chat project or you can use an existing folder which in my case I would do. I would then select the YouTube folder on my computer, add the

name and some instructions for Claude on how to work inside of this project. For

example, you're my YouTube assistant.

You will help me with any YouTube related tasks from ideation to packaging to introw writing to helping me with scripts, presentations uh and more. I

would give it some context on the context files it has access to. You'll

have access to uh my YouTube folder with all my past YouTube transcripts so you get an idea of what my last videos were about and in order to mimic my tone of voice when you use it for script writing. Also, you have access to my

writing. Also, you have access to my YouTube strategy doc and ICP document.

And I could also add an instruction like any asset we create in these chats like presentations, scripts, please save them directly into the folder. We can add any extra files we want and then we just create and we have our project set up.

Now again, this is going to require a little bit of effort when you just get started and figuring it out, but these are really important features to start using and experimenting with because this will make the outputs of your AI, however good these models get, far

better and easier. Now, all the way at the end of this video, I'll take this up a notch and show you how to set up an entire AI operating system where large amounts of context like you can see here, hundreds or thousands of documents

can be stored and used in cloud co-work and can even be scaled across your team.

You can see I have hundreds of files here in what I call our Ben AI OS that is shared across all my team members too. So everyone's AI agents on co-work

too. So everyone's AI agents on co-work have access to all relevant context documents around our business. But for

now, don't worry about that yet.

Experimenting with this is already a big step. I'll explain exactly how to set up

step. I'll explain exactly how to set up an AI operating system at the end. So

now that we've covered the setup and file access and projects, the next thing that makes Cloud Co-work powerful is being able to connect it to the internet and any software we use. Now connectors

might sound simple but it's actually important to understand when to use each one of these uh in order to make cloud efficiently use the internet and your tools. Now first we have the built-in

tools. Now first we have the built-in connectors from cloud co-work which we can find here in customize and in connectors here you can see I already have it connected to my Google inbox uh fireflies for my meeting transcriptions

my CRM slack etc. And this allows cloud now to take actions and get data directly from these softwares. For

example, I can now go into that YouTube video ideation chat and say something like add idea four and five into my notion YouTube tutorial pipeline.

And now directly updated my notion pipeline with the two ideas and a description. Now this becomes especially

description. Now this becomes especially powerful with skills and automations which I'll show you later. But first,

how do you install them? Now if if cloud has a native or built-in integration, it's really easy. We can just go here to browse connectors and for example notion we can just see if it appears and then usually it's just all you need to do is

log in and it's connected. Now cloud is expanding these built-in integrations quickly but the reality is that many many softwares are not in this connector list yet but this doesn't mean we cannot

access them. So the second best option

access them. So the second best option you have is to use an MCP server. Now

all of the connectors that you're seeing in here are also MCP servers under the hood. And all MCP is is basically a

hood. And all MCP is is basically a bundled set of all the API calls or actions for a tool into one package that we can easily give to this AI agent incoork without installing all of these

actions or API calls separately. For

example, you can see that the air table MCP has many of these different API calls all installed through the MCP server directly. Now, what I see most

server directly. Now, what I see most people do when their connector is not listed in the connectors in cloud is they start using the cloud browser to access the software. Now I highly recommend exploring two other options

before doing that. And why is that?

Because browser use is actually an extremely inefficient way to access softwares or tools. It's tokenheavy,

errorprone, and pretty expensive. So the

second best thing you want to explore is to explore if they have an MCP server.

So the best thing you can do is just go to Google. You just type in your tool

to Google. You just type in your tool together with MCP server. For example,

uh web flow MCP server.

And if your tool has an MCP server, usually the documentation page will pop up. For example, here. And in these

up. For example, here. And in these documentation pages, almost always you have the instructions on how to set it up inside of Cloud Desktop. For the Web Flow one here, for example, you have installed the MCP server. Now, there's

generally two ways you might have to set up this uh external MCP server and connect it to cloud uh co-work.

Sometimes you'll get a remote server URL. In this case, no. But if it does,

URL. In this case, no. But if it does, you can copy that URL. Then you can just go here to connectors and then um click the plus icon and add custom connector.

There you can add the remote MCP server URL with some advanced settings which will usually be listed in the documentation. Now you might not always

documentation. Now you might not always be able to install it like that. So the

second way is by opening up the configuration file. We do that by going

configuration file. We do that by going to the settings. Then we click here below on developer. We click on edit config and there we'll get a JSON file which

you can open in any text editor. In my

case, it will open in an ID. You can

just then take that config file or that JSON, throw it into any AI chat and go, can you please update this JSON with the new Web Flow MCP server.

You then add in the instructions from Web Flow in this case.

Cloud then gives you the updated JSON, which you can paste back in to the config file. You save that and when you

config file. You save that and when you reinstall Cloud, you will have the Web Flow MCP installed. Now, but what if your software doesn't actually have an MCP server? The next best thing to do uh

MCP server? The next best thing to do uh is to actually build your own MCP server. Cloud has a built-in feature or

server. Cloud has a built-in feature or skill that helps you build an MCP server for a software or a tool um that doesn't have it natively yet. Again, you can use a browser, but it's going to be more

efficient. So, if it's a software you

efficient. So, if it's a software you want to access a lot through co-work, highly recommend exploring this option first. For example, my community is run

first. For example, my community is run in Circle and Circle doesn't have an MCP. So, what I did is I just asked,

MCP. So, what I did is I just asked, "Lot, use the MCP builder skill." And

then I asked, "Can you build me an MCP server for Circle.so?" And it then build me an entire MCP server from scratch for Circle and gave me a step-by-step breakdown on how to install it, which will be very similar to what I showed

you before. Now, let's say your software

you before. Now, let's say your software doesn't even have APIs, let alone an MCP, or it's just a tool that you want to access maybe once or twice, then you don't want to actually go through setting up this MCP, of course, and that's where you can use the Cloud

browser. Now generally you can just tell

browser. Now generally you can just tell cloud to use the browser uh or generally when you tell it to access a software it doesn't have a connector on MCP connection with yet uh it will use the browser automatically. And lastly we

browser automatically. And lastly we also now have cloud computer control.

This is where we can actually let cloud not only access our browser but our entire computer. Now this is the most

entire computer. Now this is the most inefficient and tokenheavy way to access external softwares. Um and because we

external softwares. Um and because we can now already give cloud access to files on our computer and we have all these ways to connect it more efficiently to tools in the internet.

there are not a lot of use cases you want to actually use this for in my opinion. Maybe if you have some local

opinion. Maybe if you have some local software with specific workflows. Um

this is something you might want to explore, but again it's going to be very inefficient, slow, probably pretty expensive and tokenheavy. Now if you want to use it, you have to go to the settings first and turn it on. And then

here under the settings, the general settings in the desktop app, you have to turn on computer use here. And the way I use it is by just telling cloud that it has to use it. for example, update this

in my anti-gravity ID using cloud computer control with this new JSON.

You can see it's now taking control of my computer.

Now, it's doing this based on vision.

That's why it's so token heavy and sort of inefficient and it will be also be slow as you can see. So, you see it says I can see the cloud desktop config files already open and has the updated JSON.

So, nice feature. There might be some use cases but in practice you want to go with the other options first. Now one

more uh thing I would recommend for any connectors always put put in the settings on always allow as long as that's safe of course but it will prevent you from uh giving permission all the time. Now one more thing I want

to add is internet use let's say for uh research tasks etc. Clot can now read most links you give it, right? Any

website link you give it, it can uh read it. But some websites will have

it. But some websites will have restrictions. So sometimes clot can't

restrictions. So sometimes clot can't access all website. Now the most common ones where you'll see these restrictions are social media sites like LinkedIn, Instagram, Facebook, etc. They can't directly be accessed by cloud by just

giving Claude a link. Now here again we have two options. Use the browser to access these social media pages or pages that CL can't access. But the second option again is the preferred option which is a far more efficient option is

to use a tool called appy. And ampify is basically a scraping connector that can access all of these social media and any website that can't be scraped directly by claude. So here you can see we have

by claude. So here you can see we have appy with lots of different scrapers to get data from hard to scrape websites.

So if you're planning to get data from these types of websites often I highly recommend installing appy. Later in this video at the marketing uh use cases part of this video I'll show you step by step

how to set up this uh appy connector.

Now lastly we have another feature here that wasn't available in cloud chat of course which is called dispatch. Now

dispatch basically means we can now use cloud uh co-work directly from our phone. All we uh have to do is install

phone. All we uh have to do is install the cloud app on our phone. We then

connect it to the same account. We allow

for dispatch and then we can trigger cloud co-work and our computer from our phone. And this is of course really

phone. And this is of course really powerful because these this allows these AI agents to start working and doing tasks on our computer and in our softwares even when we're not in in front of our computer. For example, here

through my phone I said check my emails of today. Do I have anything important?

of today. Do I have anything important?

And then it went through my email inbox to give me a summary on my phone. Now

this becomes especially powerful when you're a bit more advanced and have automations and skills set up in cloud co-work. But it's good to know that this

co-work. But it's good to know that this feature exists. So now that you

feature exists. So now that you understand the fundamentals of cloud co-work with file access projects and connectors, let me cover the most important thing to get good at in cloud co-work skills. Now this guide should

co-work skills. Now this guide should help you whether you work with cloud code, co-work, open AAI, Google as they all have integrated uh skills by now.

Now before we dive into what skills are and how to build them effectively, let me quickly explain why learning skills is a big deal. Now, we're all seeing that these AI agents are getting really powerful, but no matter how good they get, they still need that specific those

specific guardrails, context, and SOPs around all the unique ways you and a business do things and use software.

Now, we've tried to solve this before.

Of course, we have projects and custom GPTs where you set a system prompt and add context files, but generally the problem with these is they're isolated.

You're hopping between different windows. They don't really self-improve

windows. They don't really self-improve and can't handle a lot of context. On

the other side, of course, we have NAND and automation platforms that hardcode the guardrails and make a system deterministic. And for fully

deterministic. And for fully deterministic non-human in the loop workflows, these are still great, but most day-to-day work isn't actually deterministic. Uh the processes aren't

deterministic. Uh the processes aren't always clearcut. It's context dependent.

always clearcut. It's context dependent.

It requires judgment, and that's why human in the loop is often really important to get good outputs from AI.

And skills sit in the middle. They're

essentially instructions for an AI agent on how to do a specific process. They

can self-improve. They have human in the loop. But unlike a project or custom

loop. But unlike a project or custom GPT, thousands of skills can be accessed by the same agent. And the game changer is that they can be created and updated by simply prompting. And that means

anyone can start automating their tasks and workflows through a single interface on cloud code, co-work. They're also

sharable across your business. So one

person's process and domain expertise can instantly be used by the entire team, which of course has huge implications for onboarding consistency and how a company operates. And if

thousands of these specific capabilities can be given to one very capable general agent, it seems likely uh that AI agents will slowly become the single interface for doing work. Now, of course, we are

very early with this, but I predict that building the skill infrastructure for these agents to do specific tasks well will not only allow you and your company to become far more productive, it will

also become monetizable. Sort of like a new software layer. But just like with software engineering, prompt engineering, and AI automation engineering, you need to actually get good at it. Even though anyone can create skills through prompts, the same

skill can produce completely different outputs depending on how you build them.

For example, this is an infographic that I got out of an infographic skill that I quickly created. And this is the output

quickly created. And this is the output of the same input of an infographic skill I put a lot more effort in. As you

can see, it's a lot clearer and a lot more aligned to my brand. I'll show you a full demo of the infographic skill later in this video. Also, if you want access to that infographic skill and all of the other skills me and my team are building out, you can also check out my

AI accelerator where we uh list them all for you to download or customize. Now,

before showing you how to build these types of skills, it's good to understand what skills actually are. And simply

put, agent skills are folders of instructions, scripts, and resources that agents can use to do things more accurately and efficiently. And at the core, we have the skill.md file. You can

basically see this as an instruction on how to do a process. Think of it like a system prompt, a cloud project or a custom GPT. But besides that process

custom GPT. But besides that process instruction, we can have additional instructions on how and when to use knowledge files, how and when to use tools, how and when to spin up sub agents and how and when to use code

executions. So the skill MD you can

executions. So the skill MD you can basically see as the SOP. Now that skill MD is just the core. What makes skills more powerful are those additional reference files or the context we add.

Now what kind of reference files can we have? Now firstly we can build really

have? Now firstly we can build really simple skills with no reference files at all. It's just the skill MD or process

all. It's just the skill MD or process instruction. For example, we have an

instruction. For example, we have an sales account research skill here from Enthropic which has no additional reference files just the skill MD here.

And the skill MD as you can see includes an execution flow. Right? Step one parse request, step two web search, right?

Step three enrichment and step four CRM check. But commonly when we build our

check. But commonly when we build our skills we want to add some reference files to provide the skill with more information. Firstly, we can have text

information. Firstly, we can have text files. For example, common ones you

files. For example, common ones you might want to include are example outputs, style guides, context around your ICP, uh your background, etc. For example, in my newsletter writer skill,

you can see I have multiple reference files here to give the skill more context on my background, what we do, my ICP and voice personality. Another

common one to use is an MCP instructions file, which basically lays out to the agent how to use a specific tool efficiently in this specific skill process. for example, the way you

process. for example, the way you navigate or use your CRM for this specific task. Now, don't worry, your

specific task. Now, don't worry, your agent can build these MCP documents.

Now, a second type of reference file we can have in there are assets or non-ext files like images, presentations, videos, or even uh binary, for example, to give it good output examples of a

presentation layout that you want. And

then lastly, reference files can even include code scripts, Python or JS functions that can take actions like doing API calls in a software or executing functions. For example, my

executing functions. For example, my infographic skill, we have a script to do an API call to the nano banana Google API call. So you can see that these

API call. So you can see that these skills can be very simple or can become quite complex and can look a lot more like a software except skills of course are software for AI agents. But how are

we able to give one AI agent access to thousands of these skills, feed it all of these documents without overloading it with context? Now this is done through a process called progressive disclosure. Now it sounds complicated,

disclosure. Now it sounds complicated, but it's not. Basically when we create or upload a new skill and give it to our AI agent include co-work or code only the metadata which is the description and the name are stored in the agent

memory. In my infographic skill this is

memory. In my infographic skill this is the name and this is the description and this is how your agent knows when to trigger or use a specific skill. And

then only when the skill is triggered the skill MD or the process will be loaded into the context window of the agent to understand how to execute on this specific process. And only when the

skill instructs to use a reference file, it loads the reference file into the context window. And because of this

context window. And because of this progressive disclosure of context, we can give one agent access to thousands of potential skills. Now, skills and plugins are relevant for any industry, business department, and any niche

workflow. And we're seeing three layers

workflow. And we're seeing three layers of skills and plugins appear. Uh we have the general plugins and skills, for example, the ones built by Enthropic or OpenAI. Um but we're already seeing

OpenAI. Um but we're already seeing marketplaces of skills like skills MP smithy um where people are building and potentially selling them and potentially SAS businesses will start creating their

own plugins. Now as these will have to

own plugins. Now as these will have to be general purpose skills in order to be relevant to multiple industries and businesses. Companies will want to build

businesses. Companies will want to build their own ones based on their unique processes or customize some of these general skills. But even within a

general skills. But even within a company it makes sense for different employees and different people to customize skills to their specific way of working and their specific workflows.

For example, an entropic or third party, let's say sales outreach skill can be useful, but a company probably wants to customize and improve it by adding uh the company brand tone, the ICP and the

business context. And one uh sales rep's

business context. And one uh sales rep's copywriting style will be different than anothers. So, they can even start

anothers. So, they can even start customizing uh those skills for their specific copyrightiting style. For

example, of course, if you as an individual start to get good at it and you have specific domain expertise and build a skill out of it, you can potentially monetize it by providing it to the general public. So how do we actually build these skills ourselves?

Now we can first of all customize entropics or the third party provider skills. There are multiple of these

skills. There are multiple of these marketplaces now. But the real unlock is

marketplaces now. But the real unlock is really building them yourself because every person and every company has their specific ways of working and their specific expertise. Now there are two

specific expertise. Now there are two main ways to build your own. First way

is you do a task once manually with an AI agent and then ask it to save it as a skill and the second way is to instruct it on how to build the skill right away.

For example, here I downloaded a third party provider, Meta Adale, from one of the marketplaces, imported it, and used it here in my own account. And then I asked it to use this skill combined with

my specific knowledge sources to create a new skill that helps me create ads.

And then it created a new skill which I can add to my library. Customizing

Entropics built-in plugins and skills is also very easy. You can just go to skills, click on edit, and customize with claude or even the plugins by just clicking here on customize. And here I created my own skill by just prompting

it and giving it some reference files.

Now, whatever the method, applying some best practices is important. So, let me start by saying that building good skills is kind of like an art. It's the

art of putting your domain expertise of a process into something productizable and usable for an agent. And the more I do this, the more similarities I see between skill engineering and software engineering, but it's sort of like

software engineering for an AI agent.

Uh, but you have to think about UX, right? When to add in human in the loop.

right? When to add in human in the loop.

You have to think about context engineering, right? which balance of

engineering, right? which balance of context uh produces the best outcomes.

You're adding features, removing things, handling edge cases, etc. The beautiful thing with skills is you can easily update them with prompt. So skills are never finished and what I've learned is the more you use them, the better they get. I've probably updated and iterated

get. I've probably updated and iterated on my infographic skill around five times to get to the outcome that I get now. Now, the first step is the step

now. Now, the first step is the step that most people skip, but what makes the biggest impact uh before prompting to build or customize a skill, you really do want to take a step back and think about the ideal step-by-step

process to get to a good outcome. You

want to think about uh what knowledge sources or additional information you can give your AI agent in this skill to do a better job. Now, generally some context or reference files that I recommend you have, especially if you're

going to use these skills more for marketing or sales related uh skills is what we do basically a description of your business or description of your ICP of your voice personality if you're

using this for uh LinkedIn etc. Newsletter strategy in this case I have this also for my YouTube for my LinkedIn etc. And also writing framework these are some of the documents that I reuse across different skills too. So I highly

recommend this. Now if you don't have

recommend this. Now if you don't have these reference files yet, you can do it together with cloud. For example, here I asked it, I want to create a YouTube strategy document that I can later give when I'm building skills for research, ideation, scripting, etc. And then I

basically ask it to ask me some questions about my YouTube strategy. And

through the planning mode, as you can see here, ask questions, etc. You'll get more and more context. And at the end here, you can see club generated a YouTube strategy document for me, which I can then reuse in multiple uh YouTube

skills that I create. And once you have a few of these, building skills will get a lot more efficient. Now once you have that, of course, you want to think about which tools or software the agent needs in order to uh do this task. And then if

possible, you always want to prepare good output examples. And this is generally the thing that impacts performance the most. Now once you have that sort of clear, we move to the building, which of course in this case is prompting. Now I just go through a

is prompting. Now I just go through a framework that I like to use, which is not hard science, but I think can help you uh to think about what to include to get to a good first outcome uh of your skill. Now first you want to define the

skill. Now first you want to define the name of the skill and how the skill should be triggered. So in this case the name of the skill should be infographic generator and should be triggered anytime a user mentions he wants to generate or create an infographic and

this is what the agent uses to write the meta description. Then just like we're

meta description. Then just like we're prompting you want to define the goal or the objective for the skill. Uh it can be quite short because we dive a lot deeper into the process later. Now in

this case an infographic skill that creates quality infographics according to my brand style for LinkedIn and newsletters. Then we want to give claude

newsletters. Then we want to give claude context on connectors, APIs or MCPS that it needs uh to use in this skill. Now if

there's a specific table, maybe a page or a process it should take in that software for this uh skill, you can also describe that here. And then you want to lay out the process uh step by step.

This is one of the most important things and it's good to actually spend some time on it. Now a good way to think about what to include in each step is what does it need to do? First of all, when do you want a human in the loop?

What kind of human in the loop do you want? With the dynamic QA boxes of

want? With the dynamic QA boxes of Cloud, we can now choose between check boxes, open field, single select, etc., which kind of becomes like UX design and four, what additional context should be

used to do this task better. If it's

very short, you can add it in the prompt. But if you have a specific file

prompt. But if you have a specific file it needs to read, you want to specify that for each of the step. Even if you don't have the reference file yet, you can just paste the context here in the prompt and ask it to create a reference

file for that scale with that context.

Generally, you want to try to keep the skill MD very clean and focused on the process. Any additional information

process. Any additional information should be in the reference files. That's

how your skill is going to perform a lot better. And lastly, you want to think

better. And lastly, you want to think about what you expect as an output for each of the steps. Something that I like to do and has made a big impact on my productivity is when you have human in

the loop steps, ask clot to always give you multiple variations or options which you can choose from instead of just oneoff outputs. So, I usually say

oneoff outputs. So, I usually say something like that. It should give me five different ways an infographic could visualize this idea or concept. And then

lastly, you want to think about rules for this skill. Here you can basically predict what could possibly go wrong when executing this skill and write it down as a rule. Now, this is also the section that you will continuously

update when improving the skill. And two

tips here in the rule section is first of all to double down on instructing it to use the knowledge files as obligatory steps in the process when needed.

Otherwise, I I notice it tends to skip over some. And second, I usually double

over some. And second, I usually double down on the multiple variations with human in the loop. And lastly, a really good thing to include is progressive updates. Basically, instruct the skill

updates. Basically, instruct the skill to be automatically updated and improved when when using the skill so it becomes self-learning. For example, every time

self-learning. For example, every time it define a clear thing not to do anymore in this skill, update the rule section. And what can be very powerful

section. And what can be very powerful too is instruct in the process that if a user approved the final outcome to save that as a good example so it learns and builds more data around what good looks

like automatically. Now let me show you

like automatically. Now let me show you a quick demo of how my infographic works so you get a better idea of what you can get out of this. So I'm in this case I'm using the code tab here because uh with API calls it can be a little bit easier.

Uh, so I can just trigger it with the slash command. And this is already an

slash command. And this is already an updated version from this initial prompt that I gave it. First, it asked me what content should I turn into an infographic. So I'll just paste in a

infographic. So I'll just paste in a part of my video where I talk about what skills are. A later update I made from

skills are. A later update I made from this first prompt is to ask it first for which platform it is to understand which format it has to generate it in. So in

this case, LinkedIn. And as you can see, it uses these QA boxes because I instructed it to do that. And then what type of visual should it be?

Infographic. Then I later adjusted it and iterated on it to get a checkbox out. So I can actually start generating

out. So I can actually start generating multiple variations. But here it

multiple variations. But here it suggests me what we should visualize.

This case I'll go with anatomy of a skill. And then it suggests me ways to

skill. And then it suggests me ways to visualize that concept. I can choose what makes most sense and generate those. I'll just do all of them.

those. I'll just do all of them.

Now again the first time I launched this prompt, it wasn't this good yet. I had

to iterate three or four times adding extra rules etc. So be claim an iterative process and it's really about that the more you use it the better it becomes but these are quite powerful and

really follow my uh brand and guidelines and you can see I added this box at the end if I click keep all finalize it will save it as the good output uh examples so it gets trained on what good look

looks like now building your skills through this framework will allow you to get to good skills a lot faster but to get to really optimized high performing skills we need to efficiently test and improve them that's why in the next

section ction, I'll teach you how to use CLOT skills 2.0, which is CLA's built-in evaluation and testing feature that can help you optimize your skills really fast. So, what are these skills 2.0?

fast. So, what are these skills 2.0?

Basically, Enthropic just updated their skill creator skill. You can see that it now has folders with Eval Viewer agents that can analyze, compare, and grade,

and scripts for benchmarking and report.

Now, you don't have to understand what's actually in there, but this basically allows us to run automatic tests on our skills in order to improve them faster.

For example, here I ran a test on my YouTube to newsletter repurposing skill.

It ran five test variations and automatically scored performance based on my criteria like word count, m dashes, and if it included personal stories and then automatically scored

the performance of each of the test. I

can also check the outputs from the test or the eval here in this report. And

based on the performance results and my feedback which I can give here for all the different variations, it can optimize a skill far faster and drastically improve the way we build good skills. Now although this is a

good skills. Now although this is a great feature, it is actually really important to understand how to use this efficiently. Now before showing you a

efficiently. Now before showing you a full demo of how to use them and all the use cases for them, it's important to understand how they actually work. And

as I showed, they've basically just updated the existing skill creator skill to now include these E files, which basically just means testing. So if

you've already created a skill, you know, you can just build a skill by prompting it. And this will still work

prompting it. And this will still work the same way with these new skills. But

now after you've created a skill, Cloud will automatically ask you to test your skill. But instead of just running one

skill. But instead of just running one test and you evaluating the output, it can run multiple tests at the same time and focus on testing specific things and give benchmarks or scores for them. For

example, you can test on speed, token usage, output quality, uh tool call use, let's say copyrightiting style, but really anything you want to test it on.

It will then give you that data in a nice structured document that I showed you for you to analyze. And based on the results, both good and bad, and your feedback, it can easily update and optimize the skill very fast. And if you

do a few of these iteration loops, you can get to really good skills really fast. Now, besides this, there's another

fast. Now, besides this, there's another new thing we can do that can help you optimize the skills you build even more, which is through AB tests. And in AB tests, you're basically testing

different versions of skills to see which one works best. You could also test your current skill against non-sklls to see if it actually performs better than if you wouldn't have a skill

at all. But Entropic mostly recommends

at all. But Entropic mostly recommends to do this when a new model is released, like let's say for example, Opus 4.7 or 8. um because your skill might become

8. um because your skill might become irrelevant when models become more powerful. So let me show you some

powerful. So let me show you some examples of how to create these skills, test them efficiently with both evals and AB test to get to good skills fast.

Now just show you a full process of building and testing a skill to get to good skills fast. Now here you can see that I built my YouTube and newsletter repurposing skill from scratch. Now I

did this using a prompt format that I covered in my last video on skill building to get to a good first version of the skill. Now, this is not hard science, but it's a good framework uh to go through to make sure you include all

best practices when creating a new skill. Now, if you want to break down of

skill. Now, if you want to break down of the entire framework, you can you can also check out my last video. Well, I'll

go through it uh very quickly here, too.

Now, even though we can prompt it really quick to create a skill, actually thinking about it and putting some effort in is really important because the more accurate and precise you are, uh both when creating a skill and testing the skill, um the more efficient

this process and the better your skill is going to be. So in this case I'd said create a skill according to the following guidelines. Then I gave it a

following guidelines. Then I gave it a description of what the name of the skill uh should be and how uh the skill should be triggered. So in this case the name of the skill should be YouTube to newsletter. It should be triggered

newsletter. It should be triggered anytime a user mentions that he wants to repurpose a YouTube video into a newsletter. Then I gave it the main goal

newsletter. Then I gave it the main goal for the skill. In this case a skill that repurposes a YouTube video into a full ready to publish newsletter issue within written in my voice and style. Then I

covered the connectors that it has to use in this skill. In this skill, you need to use the YouTube transcript MCP from Ampify to extract the video transcript. Also, I specified the

transcript. Also, I specified the reference files it has to include in the skill. In this case, these are the

skill. In this case, these are the reference files uh that I find get my copywriting skills to way better outputs. Like for example, the what we

outputs. Like for example, the what we do, which is basically a description of my business, my ICP, my voice personality, a newsletter strategy doc, a writing framework, and newsletter

examples, which is one of the most important for copywriting skills. So

these skills actually follow your tone of voice and then I laid out the entire process it has to follow to get to that final outcome. So in this case ask the

final outcome. So in this case ask the user for the YouTube video they want to repurpose extract the video transcript using ampify then read the reference files analyze the transcript and suggest

five newsletter angles. In this process I also define where I want to have human in the loop. For example here present the suggestions using the QA box and where I want the output to be. In this

case, save the newsletter as a word document. And lastly, one section I like

document. And lastly, one section I like to include in the skill prompts is the progressive updates. Whenever a user

progressive updates. Whenever a user specifies specifically not to do something anymore, it should automatically update the role section in the skill MD. And this is basically how your skill can become self-learning because anytime you can give it quick

feedback, it can automatically update the skill. So, it won't do that next

the skill. So, it won't do that next time. Now, with this prompt and these

time. Now, with this prompt and these reference files, it started creating the skill as you can see here. Now, of

course, this was already possible before the update of the skills, but now you see Claude asked me, uh, want me to run some test cases to make sure it performs well or are you happy to try it out directly? So, in this case, I said yes,

directly? So, in this case, I said yes, okay, please run some tests. And now it started doing an eval. So, you can see it says, let me set up uh test use cases and run them. I'll create three realistic tests with different YouTube

videos. It then basically spins up three

videos. It then basically spins up three different sub aents to run these tests in parallel. and then it started

in parallel. and then it started evaluating the outputs based on uh the criteria as you can see here. Now this

is the point why we need to actually understand how to use this efficiently because in this case I didn't specify what the testing criteria uh should be or what to optimize for. So cloud

basically came up with the criteria itself probably with some information from my reference file. So in this case it defined these criterias right a PS section with a soft pitch to the to my

AI community a signature sign off with Ben a word count range and if it actually produced a docs file now and of course this can be useful to sort of test if your skill actually is

functioning or or working. But if you actually want to optimize it we want to be much more precise on what to optimize for and the criteria to test on. And

I'll show you an example of that in a second. But as a result here we get uh

second. But as a result here we get uh of course that report where we can basically see uh the full execution of each of the tests. Uh so we can see the prompt here that was used in this

specific test. The steps that were

specific test. The steps that were completed. So in this case video

completed. So in this case video collection, transcript extraction, reference file analysis. So basically

checked if it actually followed the step-by-step process instructed in the skill. Now in this case it did it well.

skill. Now in this case it did it well.

And then it also gives us the full output of the newsletter. We can also see the formal grades of the criteria it test for here. In this case, zero failed. And we can give feedback on each

failed. And we can give feedback on each of the specific test results here. And

if we click on next here, we can go through all the different variations of test tests and give feedback. And if we want to optimize the skill, we can basically just tell Claude to either do it based on the results of the test he

run or combine it with together with our feedback. So if you add feedback to each

feedback. So if you add feedback to each of the tests, you can just click copy here and add that then to the chat and it will have more context on how to optimize this. Now again giving a basic

optimize this. Now again giving a basic prompt like I did like saying uh run some tests may be useful again to check if if it's actually functional. But if

you actually want to optimize it uh for specific criteria, we have to be much more specific. So what I recommend you

more specific. So what I recommend you at least include when you prompt clot to run an EVA or a test is to um define what you're going to optimize this test for. And it's really important to choose

for. And it's really important to choose one here. Don't try to optimize five or

one here. Don't try to optimize five or six different things at the same times because there there will be too many variables. So you want to test one thing

variables. So you want to test one thing at a time. Optimize one thing at a time.

And also you want to define the criteria for the evals. And optionally you can define how the test should be done. So

for example in my case I said let's run a new test and we're going to optimize the test for matching Ben's copyrightiting style and voice as closely as possible. Don't use the current test. It should be a new new

current test. It should be a new new test. Now the criteria for the evals are

test. Now the criteria for the evals are how clo closely does it follow Ben's example references second does it have m dashes third the length of the newsletter and fourth does it include

personal stories or references from Ben's personal background reference file and then I defined how the test should be done because of course I don't want this to run on five different YouTube videos I want to see different outputs

based on the same YouTube video so we're going to use only one YouTube video to test five different variations and you can also define how many test variations you want the eval or the test to run. So

in this case I specified five. Then it

ran the test again and in this case it ran the test based on my criteria. So it

ran it on a style match and you can see two out of five failed uh word count and personal stories and you can see one out of five failed. And you can see this is already a much more useful test result.

And based on these test results and evals, I can tell Claude to optimize for um not failing on the personal stories on the style match and it will already do a pretty good job in optimizing my

skill. But of course again I can go into

skill. But of course again I can go into the review and give it extra feedback.

Just copy it, add it here and it can update the skill to make a better version.

Now lastly, let me show you a quick example of how and when to use AB tests where we actually test different versions of skills against each other.

Now this is generally what you want to do only when you already have a functioning and sort of good performing skill. And these AB tests basically

skill. And these AB tests basically allow you to optimize functions of skills even more. Now the way we do it is by just telling Claude to run an AB test on the skill. For example, here I

said use the creator skill to run an AB test on the YouTube to newsletter skill so that we optimize the skill for speed.

It can't affect the step-by-step process. The process still has to be

process. The process still has to be there and the reference file still has to be read. It has to use uh this YouTube link for the test. Now, of

course, you can use these AB tests to optimize for speed, token usage, output quality, again, really anything you can imagine, but in my opinion is much more focused on really trying to get from a

already good skill to a great skill or a more efficient skill. What it does in this case is it basically spins up a sec second skill, a version B that in this case is the leaner skill. It cuts away

some of the context that it thinks is unnecessary to make it run faster. But

it basically comes up with this idea of this separate version or this type B version of a skill itself. Also, it

doesn't have to be just one. Um, you can also tell it to spin up six different versions of scales. Then it ran the tests and the results are actually that my original skill used 93,000 tokens and

took 204 seconds and the optimized new version only took 77,000 tokens and was a lot faster with 160 seconds. But what

you can see is that it failed at transcription extraction and the word count was too short. And then at these AB tests, we also have in the report a benchmark tab here where we can compare

the outputs and see how the how the two uh skills performed against each other.

I can say in this case it calls it without skill, but this basically means the original skill. Of course, it was a lot faster, but it filled in the transcription and um the word count.

Claude then also gives you a recommendation on what to do based on the results. So in this case, the speed

the results. So in this case, the speed optimization are solid and worth keeping. The two failures are likely due

keeping. The two failures are likely due to the transcript tool behaving differently for each agent which ran the different tests, right? Not the skill uh changes themselves. A rerun would

changes themselves. A rerun would confirm this. I actually later reran it

confirm this. I actually later reran it and it actually worked. I didn't see a significant difference in the outputs.

So I updated the scale and now it's a far faster scale at a lower token usage.

Now again, you can of course use these AB tests for many more things. For

example, I also did an AB test um for context engineering because I know that for these copywriting skills uh and automations, one of the most important impacts on the output quality is which

balance of reference files or context files that you give it. So for example, you can also run an AB test to define if a skill with or without a specific reference file is more efficient or not.

So in this case, for example, I said use the skill creator skill, which is important to mention uh to do an AB test on the YouTube newsletter skill. I want

to run one skill that uses all the eight reference files and one skill that uses all of them, but not the voice personality reference files. I want to assess if the voice personality file actually improves the cop copy or harms

it. Use this YouTube link for all the

it. Use this YouTube link for all the tests. And then again, in this case, it

tests. And then again, in this case, it created one new skill without the voice personality reference file. ran the test on the same YouTube link and give me an output. And in this case, of course,

output. And in this case, of course, it's a very subjective output. So, I'd

really need to read both of these newsletters to make sure that one is better than the other or if there's really a difference and then maybe optimize the skill accordingly. Now,

these reference files that I use for copyrightiting styles actually come out of a lot of experimentation and testing that I've done before last year uh through prompt methus, which is basically a context engineering or

prompt engineering IDE. So, if you're building any copyrightiting skills, I highly recommend implementing those context files that I mentioned to get a good match of your tone uh of voice and style. Again, you can find all of these

style. Again, you can find all of these reference files uh and adapt them easily by uh just going to my AI community or AI accelerator. I have all these

AI accelerator. I have all these reference files, all the skills we're building now, including the one I showed you. Uh we have one-on-one unlimited uh

you. Uh we have one-on-one unlimited uh live tech help and multiple weekly Q&As and workshops where we dive a lot deeper into these tools. So, if that's interesting to you, uh definitely check it out. But we also have blueprints on

it out. But we also have blueprints on how to get your first customers in AI if you're interested in building your own business. So now that you understand

business. So now that you understand everything you need to know about uh skills, let me move to the next topic, which is the obvious next step, understanding plugins. Plugins are

understanding plugins. Plugins are basically a bundled set of skills connectors and agents to essentially add another layer of complexity on top of skills. They also allow you to easily

skills. They also allow you to easily share them across teams and externally.

And you can import external plugins into your cloth coord. As you can see, Enthropic has already built some of these plugins for us across all business departments. So, we have sales,

departments. So, we have sales, productivity, product management, legal, finance, uh customer support, etc. Now, the idea from Entropic is to basically create these plugins packaged into job departments. So, each plug-in basically

departments. So, each plug-in basically becomes a specialist in that job function. For example, we have the

function. For example, we have the customer support plugin here with specific connectors to customer service platforms and skills on how to do specific tasks relevant to customer

support like customer research, response drafting and escalation. I can use these plugins and skills by just adding a slash go to plugins and then for example go to sales and go to the call prep

skill. But I can also just tell Claude I

skill. But I can also just tell Claude I want to prepare for a sales call and it would automatically assume I want to use that skill. Now, these entropic plugins

that skill. Now, these entropic plugins are great, but as every company or person is unique, I think the real unlock is by creating or customizing them for yourself. Now, before showing you some examples and how to set them up, let me quickly explain what exactly

these plugins are, why they can be more powerful than just skills, and how these might fundamentally change the SAS landscape and the way we work. Plugins

are essentially packaged or bundled sets of skills, commands, agents, and connectors. And this basically means and

connectors. And this basically means and does three things. First, it adds a layer of complexity on top of skills because plugins can include multiple skills. It can include commands which

skills. It can include commands which serve as workflow triggers. They can

trigger multiple skills in sequence to automate more complex tasks. It can have specialized agent teams and a preset set of connectors. So, we can add a lot more

of connectors. So, we can add a lot more functionality to these plugins. Second,

through plugins, these bundled packages can easily be shared and divided between company departments. Sales teams can get

company departments. Sales teams can get a sales plugin. Marketing teams can get a marketing plugin. Each with the connectors and skills that each department needs. And thirdly, these

department needs. And thirdly, these plugins are also becoming uh versionable uh skills too by the way, which means they can be updated at any time across any account that uses them. And because

of these characteristics, plugins start to look a lot more like software of course or SAS. And I predict that SAS will start to launch their own plugins with their specific functionalities.

Through cloud co-work anyone now without any technical understanding in a business can start automating their day-to-day tasks and workflows across softwares through skills and agentic workflows and they can be easily made

and improved without any technical understanding by just prompting it. They

also become very easily sharable and usable by anyone in the business. For

example, my developer can now use my uh LinkedIn writer skill to write a LinkedIn post according to my specific workflow with my domain expertise and knowledge sources embedded and any new employee like this will be able to do

the job 10 times faster. And this setup could potentially of course be much bigger than just automating a few tasks.

We are still early with this. We're

definitely not there yet. But of course the potential for cloud co-work or an AI agent is to become the main interface for a person or a business to do work.

Most of us jump between uh 15 different tools and softwares every day, each with their own interface, with their own learning curve. But with plugins, cloud

learning curve. But with plugins, cloud can basically be in the middle of all of that. You talk to one interface only and

that. You talk to one interface only and cloud can access all of these softwares through plugins and connections. I

predict that cloud co-work specifically will uh slowly but more importantly safely start to add the most popular features from cloud code and open cloud uh like persistent memory proactive

messages maybe multi- aent teams to make it actually usable for an average business and a non-developer and of course SAS companies can be threatened by this when anthropic launched plugins

companies like Salesforce service now and Adobe all saw their stock drop uh because of course if the main interface for work really is slowly becoming agentic through cloud Google or OpenAI, it probably becomes a superior interface

instead of hopping between different tools all the time. Now, are all of these SAS going to be wiped out? Nobody

really knows. I think we're still very early, but one potential outcome could be that uh these SAS businesses will build their own plugins. Some SAS

companies are already planning to do this. Um, but maybe plugins can

this. Um, but maybe plugins can potentially become the way a SAS survives. They can build plugins with

survives. They can build plugins with their unique workflows, tools, and functions and potentially plug them into an Enthropic or OpenAI plug-in marketplace. And I potentially see three

marketplace. And I potentially see three types of plug-in appear. Um, first we'll have the Entropicuilt plugins. Right

now, Entropic has released these open source plugins for the most common business departments, but they've already confirmed that organizationwide sharing and private plug-in marketplaces are coming soon to make them more

accessible to businesses. Then second, I think we'll see uh thirdparty provider plugins coming up. Maybe SAS companies will start building their own plugins.

We'll probably see other builders providing plugins that could potentially become monetizable. And then lastly, of

become monetizable. And then lastly, of course, we will have loads of customuilt plugins because every business has unique workflows and and ways of doing work. We've been doing this internally.

work. We've been doing this internally.

We've building out our own Beni marketplace with plugins across each of our company departments. And that's

really what I think you should be exploring right now, building your own custom plugins. So how do you set them

custom plugins. So how do you set them up and customize them for yourself? Now

there are two main paths. First we can use the existing anthropic and thirdparty provider skills and uh plugins and customize them for our unique workflows and and ways of working. Entropic made this really easy

working. Entropic made this really easy to do because in the skills we can basically just use a skill and ask cla to adapt the skill. For example, in the draft outreach message, uh we can uh

tell Claude inside of the window, please adjust, uh the draft outreach scale and app, let's say some examples of your outreach templates. You can also easily

outreach templates. You can also easily customize an entire plugin here from Entropic by just going here to the plug-in and clicking customize and you'll see that you can just talk to

Claude and customize this specific sales plugin for your company. Now, using and customizing these pre-built ones uh is definitely useful, but I think the real unlock is by building your own. And the

way I recommend doing it is by creating skills first. Once you have a set of

skills first. Once you have a set of skills, you can bundle them into a plug-in. And then, if you want to

plug-in. And then, if you want to automate more complex workflows, you can chain multiple of these skills together into a gentic command. So, how do you set up these skills? Again, really easy.

You can either just tell Claude, can you help me create a skill that helps me with my YouTube packaging? I lay out the process of my YouTube packaging with some additional documents or references.

and Claude then creates a skill that we can add to our library. And the second way is to use some of your old projects or custom GPTs and transform them into skills, which I've been doing a lot.

Again, you can just throw in the system prompt and any specific documentation you had in your project and CLA can build a skill out of it. Now, if you've already built some skills and you want to bundle it into a plug-in and maybe add some agentic commands to it, uh

again, you can do it really easily in Entropic. You can literally just ask

Entropic. You can literally just ask clot, can you create a plugin for me?

Then you can define which skills, connectors, etc. you want to include and Cloud can build the plugin for you. So

you can see it built the entire plugin here and then I can just click add to plugins. Now an important note here is

plugins. Now an important note here is that all of these plugins or skills that you create on your own cloth co-work account will only be locally saved. So

if you want to use this across your business, other people will not have access to it yet. So the easiest way to do that is by just ask to give you a zip file so I can share the plug-in across

my team. Right? Cloud will then give you

my team. Right? Cloud will then give you a zip file which you can open, share to other people and other people can then go into their plug-in section and upload that zip file here. Now I think cloud

coork is going to make this sharing internally of plugins and even externally a lot easier very soon. But

for now if you want to internally or even externally want to share these plugins or even marketplaces which are groups of plugins uh you'd either have to do it through zip files or through GitHub. You'd most likely have to set up

GitHub. You'd most likely have to set up a marketplace together with clot code.

Now, uh, Entropic released a full guide on how to do this, which I'll make sure to link in the description below, too, if you're interested in actually deploying it maybe on a marketplace or really companywide through a GitHub link. But again, I think Cloud Co-work

link. But again, I think Cloud Co-work will make this a lot easier very soon.

So, now that you understand plugins, let me cover another major feature in both these plugins and in Cloud Co-work, which is using agents. Agents in Cloud Co-work allow us to run bulk tasks in

parallel and allow us to build more complex automations. So, in this next

complex automations. So, in this next section, I'll show you how they actually work and how to use them. Now, before

explaining exactly how these agents work, let me show you a quick example of what they can be used for. For example,

here I gave Claude a lead list of more than 150 marketing agencies, which we got from Apollo, a lead database. And

then I let it qualify the list with sub agents first according to our ICP criteria, which is a marketing agency providing SEO services and located in the US. And then you can see it launched

the US. And then you can see it launched 15 sub aents to qualify all 150 leads in parallel. Each agent will use web search

parallel. Each agent will use web search to verify the company actually offers SEO service and is US-based. So here it launched 15 sub aents in parallel and then identified that only 82 were

qualified. And because these sub aents

qualified. And because these sub aents work in parallel, this only took 2 minutes. After that, it went to the next

minutes. After that, it went to the next step in this workflow which is enriching the leads with more uh information. So

it launched another 18 sub agents, 17 lead researchers, and one LinkedIn scraper. It then added that

scraper. It then added that qualification status and lead intelligence to the CSV sheet. And then

it spun up another 17 agents to write personalized outreach messages based on the lead intelligence data. The

personalized outreach messages are also added to the sheet. So as you can see here in the same chat include it qualified 82 out of 150 leads. It

enriched 82 leads and then wrote 37 icebreaker messages. And this is really

icebreaker messages. And this is really the power of agents because this was of course very difficult to do before we could use sub agents. And now that we have plugins, each one of these bulk tasks can be saved as skills. Multiple

of these skills can be strung together and combined into an agentic workflow through commands. And we can even save

through commands. And we can even save the specific agent instructions in the agent section. So you can see in our

agent section. So you can see in our sales plug-in here, we have specific skills for each of the tasks in the workflow for lead qualification, lead intelligence, and email personalization with specific instructions on how we

exactly do these processes. For example,

in the lead qualification skill, we have our ICP criteria, the qualification process it has to go through, and instructions on how and when to use a sub agent. Here the instructions for

sub agent. Here the instructions for what these sub agents have to do are saved here in another tab, which is called agents. And here, for example, we

called agents. And here, for example, we have the lead qualifier sub agent with a specific instruction on what these sub aents need to do. We've combined these three skills into one command to trigger

all of these skills in sequence together with all the agents with just one slash command. So you can see the command we

command. So you can see the command we set up for outbound pipeline is just a sequence of steps to take and skills to use in order to execute a more complex workflow. And this allows us to run this

workflow. And this allows us to run this multi-step workflow with sub agents with just one command here, which is the command outbound pipeline, which I can

trigger simply by going to plugins sales and outbound pipeline. Now, this might look a little bit complicated to you, but it's not. And I'm going to show you how to set this up yourself from scratch. Before agents were available in

scratch. Before agents were available in cloud, bulk processing was difficult for many tasks because we essentially only had one agent to process each task separately. This overloaded the context

separately. This overloaded the context window, made it extremely slow and often just not viable to do for these bulk processes. Well, Cloud can now spin up

processes. Well, Cloud can now spin up these sub agents which can then process tasks in parallel which just means at the same time. And each of these sub agents can also use tools and skills.

For example, our lead intelligence sub agents uses a LinkedIn scraper to scrape LinkedIn data. So if we tell Cloud

LinkedIn data. So if we tell Cloud Co-work to research 100 leads for example, he could spin up 10 sub agents each researching 10 uh leads um through

let's say the LinkedIn scraper and and other tools and then give a summary back of the lead research to the main agent.

Now why is this a big deal? Because the

main agent will not be bloated with research context just a summary and context is one of the most precious resources for these agents and therefore it will drastically improve the performance first of all and because sub

aents can run in parallel a task like this one in example can be performed theoretically 100 times faster and because of this we can actually start using clocker work for these bulk task executions for deep researched tasks but

also we can start using it for these autonomous multi-step workflows and these three things together make it a lot more powerful and it becomes much more like an automation tool uh comparable to nadan or make.com. Now,

we're not completely there yet, which I'll get to later in this video, but because this is now possible, hundreds of new use cases pop up for automation right inside clockwork. So, you can think of, for example, creating bulk ad

variations, writing bulk uh product or SEO descriptions, bulk email personalization. and we can start doing

personalization. and we can start doing data analysis on thousands of transcripts for example and pull all the research data back into a presentation.

Uh we can think of doing SEO audits and of course we can also combine these new use cases together to do autonomous multi-step automations in bulk like I showed you in my example. Now, before

showing you how to set it up and how to do this yourself, there's one more thing I want to clarify because clot code now also came out with a new multi- aent feature. But this one is different from

feature. But this one is different from the one we have available in cloud co-work. In clot code, we can now even

co-work. In clot code, we can now even run multi- aent teams that can communicate between each other. While in

co-work, they are all isolated sub aents. Now, you can imagine that when

aents. Now, you can imagine that when they can communicate with each other, it really becomes an even more powerful setup because they can work together towards a shared goal or a to-do list.

uh Entropic used this setup with this multi- aent team uh to build an entire C compiler in 2 weeks which is something extremely hard to do in engineering and would normally take a large team of engineers to do for months. Now this is

really really powerful but I see it mostly as an engineering feature.

Multi-agent teams are good for tasks where agents have to work together on one shared goal or uh work together on a to-do list while sub agents are preferred when each one can work in

isolation and for most day-to-day office related work and tasks in marketing, sales, customer support, etc. The sub agent setup is usually enough and can actually be preferred setup because it

uses less tokens and it's really all we need for most of these isolated tasks like lead research, analyzing data, etc. Now, even though sub agents use less tokens than agent teams, they still use a lot of tokens. So, you that's one

thing you want to keep in mind and is still a li limitation. It can use quite a lot of tokens when you spin up these sub agents. Now, of course, this going

sub agents. Now, of course, this going to get cheaper and cheaper, but uh if you want to start doing this, you probably want to be on the clot uh max plan uh to not be hitting your limits all the time. And second, what I've

noticed is for really large bulk tasks, this is probably not a setup you can use yet. I've seen this work best uh for

yet. I've seen this work best uh for batches of 100 to 200. More than that, it's going to get complicated, take a long time, cost a lot of credits. That's

why for large bulk workflows, especially ones that are non-human in the loop, an automation platform like naden or make.com is still a better option to automate those types of processes. So,

knowing that, let me show you how to use these agents in cloud co-work. So, first

I'll show you how to make sure cloud uses these parallel sub aents because there is a specific way you need to prompt it. Then I'll show you how to

prompt it. Then I'll show you how to save these tasks as skills and agent files. And then I'll show you how you

files. And then I'll show you how you can build these commands and add them all into a plug-in that you can easily share with your team or with other people. So let's say we're going to

people. So let's say we're going to replicate that same uh use case as before, but in this case, I don't have the skills yet. Then give it the CSV sheet of the leads. And then I would prompt it with something like this. I

want to qualify these leads uh against my ICP criteria. My ICP criteria are that we only work with marketing agencies who provide SEO services and

are based in the US. Please spin up 15 parallel sub agents that each research 10 leads in this list and make sure they're qualified leads. Based on the

summaries you get back, uh you deliver an updated CSV sheet and add a column with the qualification status.

Now, specifically mentioning that it has to spin up these parallel sub aents is important because sometimes if you don't say it, it will actually run sequential sub aents, meaning that they'll be running one after the other, which means

it's going to take a long time and sometimes it won't even run sub agents if you don't specifically tell it to.

So, you can see it already made a plan here, right? Launching 15 parallel sub

here, right? Launching 15 parallel sub aents, collect results, and generate updated CSV.

So you can see it's now spinning up these batches of sub aents to process and qualify all of the leads. And you

can see each of these sub agents does multiple queries on the internet on each of the leads to find more uh information about them. So it's now done the

about them. So it's now done the qualification on 150 leads, gave me a CSV sheet which I can throw in sheets and now we have a new column with qualification status and a qualification

reason column. Now, if this is a

reason column. Now, if this is a repeated workflow that I would do, I could already save this as a skill by just telling club, please save this as a skill. So, next time I don't have to

skill. So, next time I don't have to give all of these instructions. It will

already know exactly what to do with a lead list it will get. But I'll show you one more example and then we're going to build the entire plugin right away. So,

we're going to build multiple skills, commands, and agents into a plug-in right away. So, let's say now we want to

right away. So, let's say now we want to enrich the leads a bit uh so we can use that enriched data later for personalized outreach. Now, for this

personalized outreach. Now, for this task specifically, and generally if you're going to enrich leads, I'd highly recommend installing the Appify MCP.

Now, if you don't know Appify yet, Appify basically has thousands of scrapers, including scrapers for hard to scrape websites like LinkedIn, Instagram, Facebook, and through the

Appify MCP, we basically get instant access to all of these social media and websites that Claude cannot access. Now,

you can just make an account uh for free here on Appify. uh you get some credits.

Um but if you're going to use it a lot, uh I recommend upgrading and it will cost you uh $29 a month. Now, once

you're in your account, you need to get your API key, which you can get here by uh going to settings and API and integrations. And here you can copy your

integrations. And here you can copy your API key. Then include you can go to the

API key. Then include you can go to the connectors in the settings and click connectors and you go to browse connectors and you look for ampify.

You'll see that appear. Now, mine is already connected. Um, but once if it's

already connected. Um, but once if it's not connected yet, uh, it will ask you for the API key and enable tools, which basically specify which scrapers from ampify we're going to use. Now, the

easiest way to do this is by going to mcp.appify.com.

mcp.appify.com.

And here you can basically specify all the scrapers you want to give access to cloud. So, first I'd recommend selecting

cloud. So, first I'd recommend selecting all here. Then you can click here on add

all here. Then you can click here on add or remove actors. And here you can specify all the ones you want. Let's say

for example, LinkedIn mass LinkedIn profile scraper and let's say a LinkedIn post scraper. This one for example. So

post scraper. This one for example. So

we actually know the last uh posts that our leads have posted. So we can use that as a personalization. So once I've selected, you can select way more. Uh

you click on save changes. Then you go here below and then you take this part after the equal sign. You copy that. You

go back to cloud and that's what you add here in the enable tools and you click on save. Sometimes you have to restart

on save. Sometimes you have to restart your cloud co-work for ampify to actually be accessible. Uh but then you should have appy installed. And now I can say something like please spin up 10 parallel sub agents that enrich seven

qualified leads each. They should find the lead's last LinkedIn post uh the company size and the company description using the appy scraper specifically the LinkedIn scrapers to get these details and then you uh should add these data

points to the CSV sheet so we can later use that data for personalized outreach.

Again, I specified the use of parallel sub aents and also how many leads each sub aent should process. In general,

what I've seen is the best is to have between five and 15 uh leads or or pro tasks per sub agent. Otherwise, it could take a lot longer if you do more. And if

you spun spin up too many sub agents, uh it also runs into complications sometimes in my experience. So, you can see it's spin up again 10 sub agents.

So, you can see each of the sub agents is using a LinkedIn profile scraper. So,

it's now enriched the CSV with the last LinkedIn post if they had one. Company

description and company size. So, you

can see we get the last LinkedIn post here, the enriched company size and the company overview. Now, of course, now we

company overview. Now, of course, now we can do another skill to write personalized outreach messages based on this enriched data. But I think you get the point. So let me show you how you

the point. So let me show you how you can now convert this into skills agent docs so you don't have to repeat what these agents have to do and commands to string multiple of these tasks together into one workflow. Now the amazing thing

include is that we can build these skills right from prompt. Now because

we're going to create multiple skills we're going to create commands and agents. I would advise you to add them

agents. I would advise you to add them all into one plugin. So it's all stored inside of the same plugin which then can also easily be shared. So, I could say

something like, uh, for example, please create a new plug-in that's called Ben's outbound uh, sales process.

Inside of that plug-in, I want to create one skill for the exact way we did the lead qualification and call it Ben's lead qualification skill. Make sure to also create an agent.m MD. And the skill

should refer to the agent.md

instructions on what the agents need to do. Also, create a skill for the exact

do. Also, create a skill for the exact way we did the enrichment process. and

call it Ben's enrichment skill. In that

one, also make sure to create a separate agent MD for the enrichment process. In

both of these skills, it should clearly be instructed to spin up parallel batches of sub aents just like uh we did in the task here above. Again, this is important to specify also when you're

creating the skills or the plug-in um because then it it's clearly stated in the skill to use those parallel uh sub aents. Of course, we can also combine

aents. Of course, we can also combine this into a command to trigger both of these skills or tasks at once with one specific command. So, for example, I

specific command. So, for example, I could say something like also, please create a command MD where I can trigger both of these skills uh in sequence when

I trigger a slash command uh with called outbound enrichment. This command uh

outbound enrichment. This command uh should ask the user for a CSV sheet with leads and it should then automatically use the lead qualification scale and the lead enrichment uh scale in sequence to deliver a fully enriched CSV to the user

at the end.

All of these new documents and files should be stored in the same plug-in.

So you can see now starts building the plug-in. It makes to-do list. Click

plug-in. It makes to-do list. Click

create plug-in directory. Create lead

qualification scale. Create lead

enrichment scale. Create outbound

enrichment command. So now created the entire plug-in with two skills, lead enrichment and lead qualification, one command outbound enrichment. As you'll

see, it already includes all the instructions on how to run this workflow, right? So first use the lead

workflow, right? So first use the lead qualification, then the lead enrichment.

We have the agents, the enrichment agents, and the qualification agent.

Now, sometimes it gives you a button which is called install. So you can install as a plug-in right away.

Sometimes it don't doesn't work in my experience. So you can just ask it to

experience. So you can just ask it to give you the plugin as a zip file. This

then you can use to share to other people or to import it yourself if you don't have the install button. And then

you can just go to your plug-in section, go to plus and click on upload plugin.

And then you can drag that in and upload. And now you can see I have

and upload. And now you can see I have my new plugin with commands, skills, and agents. You can also upload these

agents. You can also upload these plugins onto a GitHub account. So you

can get a link, a versionable link that you can share with other people. And now

that I have these commands and skills, I can of course trigger them anytime I want to do this task again right inside cloud. So I can just go and slash

cloud. So I can just go and slash command, go to plugins and trigger the outbound enrichment command. Now the

last major feature you really want to understand is scheduled tasks. Schedule

tasks allow you to schedule prompts and skills. And it's a game changer because

skills. And it's a game changer because of course it now allows clot co-work to actually automate some of these tasks and skills on autopilot without you being there. As you can see here on the

being there. As you can see here on the left, we now have an option called scheduled where I can add a new task, define the name of the scheduled task, the description, the prompt that I automatically want to send, the time

interval or frequency when we want to send it, the model we want to use, and the folder we give Cloud access to. And

essentially when we do this all cloud does is it automatically sends a prompt to a new cloud session at the defined time interval. So this can be every

time interval. So this can be every month, every week, every day, every 10 minutes, whatever you define. Now this

is cool, but the real unlock is scheduling not just prompts but scheduling skills. Because with skills,

scheduling skills. Because with skills, we can really start automating our repetitive and day-to-day tasks. For

example, I already have scheduled skills set up for daily email categorization, daily task reviews, post meeting to-do updates, automatic follow-ups on failed

customer payments, and monthly accounting. And to schedule a skill, all

accounting. And to schedule a skill, all you need to do is add in the prompt that it has to use a skill at a specific time. For example, on the daily email

time. For example, on the daily email check, I have checked my unread emails for relevance. You're my personal

for relevance. You're my personal assistant. Uh, use the daily email

assistant. Uh, use the daily email skill. You see that this repeats every

skill. You see that this repeats every day at 5:46 p.m. And here on the left in the scheduled section, we can see all the scheduled task runs for all of the different tasks. So, if I go to the

different tasks. So, if I go to the daily email check, you can see that it got triggered automatically at 5:46 p.m. It then read the skill to know what to do and then

started going through my Gmail inbox and prioritizing action items. Now, one limitation that we still have is that this will only run if your cloud desktop is opened. If it's not, it will not run

is opened. If it's not, it will not run the scheduled tasks and it will run them immediately whenever you open the cloth desktop or clot co-work app. Because we

can now save our processes as skills, give co-work access to our tools, our browser and folders on our computer.

This really becomes like a powerful automation platform. Now, before showing

automation platform. Now, before showing you the other use cases that I've been using it for, uh I want to show you quickly there are two other ways to schedule skills or tasks. You can just go into a chat and go slash schedule.

And here we can basically type in what kind of schedule task we want to create.

So just say every day at 6:00 a.m.

schedule my email categorization skill.

And then the third way is by just telling Claude whenever you're creating a new skill to schedule the skill at specific time interval. So for example in my monthly accounting skill I said I need your help in creating a new skill.

I laid out the process of doing my accounting and then I said you're going to run this on the first of each month at 9:00 a.m. And once you do this it will automatically save it here to the scheduled section. Now, let me show you

scheduled section. Now, let me show you some examples of what I've been using it for. A really good one that's already

for. A really good one that's already making an impact on my revenue is an automatic failed uh payment follow-up.

So, this one basically runs every day.

Then, it checks first all the failed payments from uh yesterday, then reviews the previous email threads with these customers to make sure we didn't already send them a message, then categorize the payment fail failure, and depending on

that, it crafts an email according to the failure type. then saves that as a draft email in Gmail. So I can still actually check before sending it out and then report the results. So you can see

for this date it found two failed payments. It crafted email copy and

payments. It crafted email copy and added it as a draft in Gmail. Another

one is to automatically analyze um the transcripts from our team meetings on a daily basis and then create the tasks for all our teams, give us an overview of the meeting and prepare the next

day's call and agenda automatically inside of notion. So you can see this runs on a daily basis. It analyzes the transcript from fireflies. It then

create tasks, updates our processes and gives an overview of the meeting and then adds that to notion. As you can see here in my notion, I got the meeting overview, key decisions made and the action items. So this is a completely

autonomous automation that runs in clockwork. But even for non-aututonomous

clockwork. But even for non-aututonomous automations like my accounting skill where I need human in the loop, this is still very useful because it basically also is a reminder for me to uh to use

this every uh first date of the month to go through this process and it already can do a lot before I jump in to add some human in the loop. For example, it already goes through my Gmail, my

Stripe, goes through my browser, can update the spreadsheet, and whenever it needs some more information, it can just ping me with this blue dot here. I can

steer it or give it some extra information and it can finish the process. Now the outcome I can't really

process. Now the outcome I can't really show you because this is of course private information but this skill uses of course Gmail, Stripe, web search and even our uh Chrome browser to actually

up update the spreadsheet uh inside of Chrome. The last one is very simple

Chrome. The last one is very simple which basically checks my task for the day and prioritize according to importance and also takes into account what I've done yesterday and the day before. But I think you get the idea and

before. But I think you get the idea and I think this is a really powerful upgrade for clock coork because this really starts looking like an automation platform. We can of course also start

platform. We can of course also start using this for notifications. For

example, let's say in the um email scale, I would add that if there's a high priority email, it should send me an urgent notification through Slack.

You can also use this basically as a trigger for when something else happens in another software. Let's say, for example, you want to send a thank you email whenever a new customer has paid you. You can basically just run a

you. You can basically just run a scheduled task or skill that checks every hour if there were new payments and if so, send an email or thank you email to that customer. What we don't have yet, of course, is instant triggers

because we don't have web hooks yet, but we can basically get around that by just letting it do something on a very regular basis. And lastly, you can also

regular basis. And lastly, you can also schedule commands. If you don't know

schedule commands. If you don't know what commands are, commands, you can basically combine multiple skills into a bigger workflow to automate even more complex tasks. For example, in this one,

complex tasks. For example, in this one, we have a command outbound pipeline. And

this one includes three different skills. Lead qualification, lead

skills. Lead qualification, lead qualifier, and personalized outreach skill. We can also run a task to trigger

skill. We can also run a task to trigger a command which will then trigger multiple skills in sequence. So, we can really get to more and more complex automations. So, now that you understand

automations. So, now that you understand all the major features and the important concepts in co-work, let me show you some actual real marketing use cases in coowwork. from copywriting to visuals to

coowwork. from copywriting to visuals to data analysis and research. I'll set up a skill entirely from scratch so you can see my thought process when I'm building these types of skills and I'll show you plenty of examples that can hopefully

give you some inspiration on building your own marketing skills. Let's start

with copyrightiting. Now, I'll show you an example of how I would use co-work to automate the process of writing a newsletter in my tone of voice, but more importantly, how we would build a skill out of that. Now, there are really three

main ways to build skills or automations around your specific workflows. The

first one is by customizing Anthropic's built-in one. So we can go here to

built-in one. So we can go here to customize browse plugins and there we'll see Enthropic and partner plugins. So

let's say for example we go to marketing we can install it here and now we have a new one here by entropic and here we'll have all the skills of entropic. There

are also marketplaces where people list their own skills for you to download like smiththery or skills MP. But

because in general these are more generic, they could be a good starting point. But I strongly believe that you

point. But I strongly believe that you should learn how to build your own. And

to build your own, there are really two main ways. First, we can just lay out

main ways. First, we can just lay out the entire process right away through cloth and tell it to build a skill or an automation around it. Now, this is mostly useful if you've already have a very clear idea on a specific process

and you have done it before with AI. For

example, I built many skills like this by converting some of my old projects or custom GPTs which were already proven processes with AI and telling clot to build a skill out of it by just giving it the system prompt. And then the

second way which I think for the majority of use cases and people is the way you should do it is by doing the process once together with clot manually here. Claude will then have all the

here. Claude will then have all the context around how you do this process and then you ask cloud to build a skill out of it. Now, one important thing to keep in mind before we start is that the key to working with co-work is that you

go into these chats with the intention of building a skill out of it. And this

means it will likely take you a bit longer than if you would do this maybe even without AI or even if you would just use AI to do this task once. And

that's because we're essentially trying to build an automation around it. So go

into this with the framework of I'm onboarding an intelligent intern once.

That might take me a little bit more time, but once I've done it, this process is forever forever out of my hands. Because once you've put in the

hands. Because once you've put in the effort to actually do this a few days, you could have automated a large percentage of your day-to-day tasks and workflows. So, let me walk you through

workflows. So, let me walk you through an example of how I would do this.

First, you want to get in the habit with co-work to always select a folder to work in. Why is that? Because any assets

work in. Why is that? Because any assets you're going to generate, etc., you'll be able to access easily on your computer and can be saved in that specific folder. So, let's say we're

specific folder. So, let's say we're creating a new one.

I open it. It can now access this folder and save things to this folder. So I can say for example, I want you to help me write a newsletter based on my last YouTube video. I'll give it a link of my

YouTube video. I'll give it a link of my last YouTube video. Now it doesn't have access to the transcript here. So in

this case, I would say use the Aify MCP YouTube scraper to get the transcript and output it here.

Now, if you don't know Appify, Apify is a really important connector you would need, especially for marketing use cases. Appify basically list dozens of

cases. Appify basically list dozens of these scrapers like YouTube transcription scrapers, Instagram scrapers, LinkedIn scrapers, etc. to get data from hard to scrape websites. So,

this is one of the most important connectors you want to set up in your Clarkwork account for marketing use cases. How do you set this up? You can

cases. How do you set this up? You can

go back to cloud. You can go to settings and there you'll find a tab called connectors. Here you can go to browse

connectors. Here you can go to browse connectors and you can search for appy.

You can see that mine is already connected but here you can install it.

All you need is to set up an appy account. You can start for free. Once

account. You can start for free. Once

you have created an account all you need to do is go to settings and copy your API token. You go back and in ampify it

API token. You go back and in ampify it will ask you to add your API token and to add your enabled tools which basically means your uh scrapers. So how

would you add those? Really easy. You go

to MCP once you have an account mcp.appify.com.

mcp.appify.com.

Here you can select all the scrapers you want. For example, in my case, YouTube

want. For example, in my case, YouTube transcription scraper. You click that.

transcription scraper. You click that.

You save changes. You click here on select all. Select all.

select all. Select all.

Select all. And select all. And then

here in the MCP server URL, you copy everything that's after the is symbol.

You copy that. You put it in here. And

that's how we let Claude know which scrapers we want to give it access to.

You click save and then you can here go here on always allow. This is how Cloud will not ask you permissions every time it uses a scraper. So if we go back to the chat, we can now see that we got the full transcript. Now this is an example

full transcript. Now this is an example of what I was saying. This might take a little bit longer the first time. We

have to set up Apify MCP. While I could just download the transcript myself and put it in here, but the point is we're going into it with the intention of building an automation. So we want Claude to be able to do it himself through this appy scraper instead of us

copying and pasting all the time. So the

next step in the process that I like to follow especially for these copyrightiting tasks is the following.

This is a long transcript. So there's

multiple ideas and angles in it. So I

want first to define what angle we're going to take for my newsletter. So I

would say something like based on my ICP and newsletter strategy document, suggest me five potential angles for this newsletter based on this YouTube video.

Now, if you don't know it yet, I'm using Whisperflow to talk to AI, which I highly recommend to install. Now,

because we're going to define these angles, this is where I want to start giving Claude more context on my audience, my business, and my strategy.

And that's why I'm saying here um based on my ICP and newsletter strategy document for copywriting specifically, having these documents or context for AI is really the key to getting good

output. Across all my copywriting

output. Across all my copywriting skills, I have around six or seven documents with extra context that I've seen produce the best outputs for copywriting tasks. So, you can see I

copywriting tasks. So, you can see I have a folder on my computer with all these context files, right? By profile,

background, what we do, my ICP, voice personality, newsletter strategy, pain points, my audience, writing framework, newsletter strategy, etc. Now, how do I build these? I basically build them

build these? I basically build them together with AI. So I just told it I want you to create a YouTube strategy document that I can later give uh to you when I'm building skills for research, ideation, scripting, etc. So can you

help me go through some questions that you want to ask me in order to get that document straight? Now I've pretty

document straight? Now I've pretty extensively tested also last year while building NAN automations what balance of context and what types of files we need in order to get the best outputs for uh

copywriting. So I'm going to make this a

copywriting. So I'm going to make this a lot easier for you. All of the files that I use for copyrightiting and all of the others which I'm going to show later in this video for visuals etc. I'll put for free in the link in the description below. All you would have to do is

below. All you would have to do is download those, open a new chat session with Claude, give it a context file and then say something like this document is from someone else. I want to use this same format but adapt it to me and my

brand strategy. Please analyze the doc

brand strategy. Please analyze the doc and ask me questions so you can build me a similar doc for me and my brand. Club

will then start asking you some questions which you can just answer and then you'll have these documents yourself. Now again, this is going to

yourself. Now again, this is going to take a little bit of effort. This is

going to take you half an hour to an hour, but I can tell you these types of files I've used across many different skills and I've used for over the last six months, not only in skills, but also in my automations. This is definitely

worth putting the time in because the output quality of your skills is going to be 10 times better. So, in this case, I would give it those two files, ICP and newsletter strategy. And by the way, for

newsletter strategy. And by the way, for me personally, what I uh prefer to do with these types of skills where there is human in the loop is to always give uh multiple potential suggestions from AI instead of one-off outputs because

this allows me to choose from multiple options instead of uh iterating on one-off outputs all the time. So this

case, we got the new SAS layer nobody is building yet. Stop building workflows,

building yet. Stop building workflows, start building skills. While your AI agent is only as good at as its skills, the skill level that will separate AI winners from AI tourist, I train my AI agent to get better every time I use it.

And although I'm not a fan of the sentence here, I think this is the strongest angle. Skill engineering is

strongest angle. Skill engineering is where prompt engineering was in 2022.

Now I don't want it to write the newsletter right away. I want to have one more step in between. So let's say something like let's go with angle four.

Now let's first define a highle outline of sections that we can cover in this newsletter. Give me three suggestions on

newsletter. Give me three suggestions on highle outline of sections we uh can cover. But before doing that, please

cover. But before doing that, please read my newsletter example doc to get a better understanding of how my newsletters are normally structured.

Again, I give it more context on how I normally write my newsletters, how they are structured, etc. by giving it simply a document with some examples of my past newsletters. Now, this is one of the

newsletters. Now, this is one of the most important documents you can have for any copyrightiting skills. Just

giving examples of good outcomes or outputs of your copywriting style, right? So, you really want to prepare

right? So, you really want to prepare that uh to get good copyrightiting skills. So, you can see outline A, the

skills. So, you can see outline A, the skill gap story, personal narrative, so vulnerable opening. Outline B is

vulnerable opening. Outline B is software is being rewritten. A bold

opener, skills and plugins are new software layer. And outline C, before

software layer. And outline C, before after contrast.

Let's go with C.

And now we're going to actually write the newsletter. And here is where the

the newsletter. And here is where the reference docs become really important.

Now let's actually start writing the newsletter.

First, I want you to follow the exact style and tone of voice of the example documents. Almost mimic it. saying

documents. Almost mimic it. saying

almost mimic it in my experience uh really works to get um the output even closer to your tone of voice. And I

would usually say something like this too. Also almost mimic the sentence

too. Also almost mimic the sentence structure, sentence length and styling from the examples.

I want you to base claims based on what I said in the video, not make them up yourself. Now, another really important

yourself. Now, another really important reference dock to have for copywriting tasks or skills is a voice personality.

And in your voice personality guide, it basically lists out uh your tone of voice attributes, the message people should carry away, the key misconceptions my audience have, my

signature phrases, etc. Again, this one is available in link in the description below so you can adapt it to your tone of voice and style. But this really helps in my experience to get that tone of voice close to yours. So I also say

something like, I want you to closely follow the voice personality guide when writing the newsletter. Then another

reference doc I very commonly use is a personal background document. This

basically lays out my entire life story uh in like four or five pages. What I've

done in the past, what I've failed at, interesting stories or books I've read that have inspired me and things like this. And this is really important

this. And this is really important because it can make personal uh reference stories from this document.

So, I'd say something like, "Also, read my personal background dog to get more context on me, my background, and potentially some stories you could use if appropriate in this newsletter." And

then lastly, I would want to say also reread my ICP document and my newsletter strategy document to know the strategy and my ICP in detail. And then lastly, I would say something to make sure it really reads those files is before

outputting the newsletter, please review and check the newsletter output across all the above references and guidelines to make sure it adheres to my tone of voice and style. I'll then give it all the reference docs, ICP it already had.

So the vers personality, my profile background, and then I forgot one more thing. I didn't give a document on

thing. I didn't give a document on actually what my business is about, which is another important reference doc. And I would say something like to

doc. And I would say something like to get more understanding around my business and to create a good call to action, please read the what we do doc.

So I'd give it access to that one too to understand what my business is about. So

now we can fire that off. And again, as you can see, I put quite a lot of thought generally when uh working to together with cloud co-work. But the

whole point of this is once I put some thought in and this actually works well, I have a skill which I can then reuse over and over again. So I never have to give it this context again or start instructing it with these specific

things again. So it's important to put

things again. So it's important to put some effort into into actually making good skills because anyone can build skills now uh in in 3 seconds by just telling claw to build a skill but there are going to be very few people who actually build good skills and those are

the ones who are really going to leverage this software to become a lot more productive. So now we got the

more productive. So now we got the output skill engineering with subject line options. I ran the same input

line options. I ran the same input through different AI skills last week.

Same content same model same prompt. The

first output, generic, flat, something you'd scroll past without thinking twice. The second, clean, on brand, the

twice. The second, clean, on brand, the kind of infographic that actually makes people stop. The only difference between

people stop. The only difference between those two outputs was how well I built the skill. Now, I can already tell you,

the skill. Now, I can already tell you, and if you know my newsletter, this is the kind of style that I use. Short

sentences. I start usually with a personal story. It's structured in a

personal story. It's structured in a very similar way with these headers. It

compared it to prompt engineering, so it follows the structures of my newsletter very well. it

very well. it usually I end up on an inspirational note which it did in a similar way. The

skills and plugins that make agents do specific work 1% are building the infrastructure layer. The 1% have an

infrastructure layer. The 1% have an advantage that compounds every time they use it. And usually I end it with a call

use it. And usually I end it with a call to action to my uh community which in this case it did. Ben the accelerator is $97 a month less than 1 hour of mediocre consulting. You get 100 plus battle

consulting. You get 100 plus battle tested templates all the skills and plugins we build weekly Q&A. So it has that context around what my business does. Now this would be a good enough

does. Now this would be a good enough output for me to build a skill around.

But I would add one more step to the process which is please give me 10 options for subject lines based on the example document where you can find how I write my subject lines. Remember that

most of my subject lines are in small caps to make them look personal and usually get higher open rates. So do

that as the last step and then we're going to create a skill out of this. Now

I'd say these subject lines are too long and that's the whole point. Any

corrections we give here during this initial process will be saved in the skill later so he won't make that same mistake again. I want my subject lines

mistake again. I want my subject lines to be between three to eight words.

These are too long. So good. We got the final newsletter. And now of course

final newsletter. And now of course we're going to build the skill out of it. Now it's also important how to

it. Now it's also important how to instruct it to build the skill out of this process. Okay. Now let's build a

this process. Okay. Now let's build a skill out of this. The name of the skill should be Ben's newsletter writer. It

should trigger every time that I mention that I want to write a newsletter. It

has to follow the exact step-by-step process that we just did above. It

should include all the reference files that I've added in this chat. And

besides this process, I want to add one more step all the way at the start, which is asking me what the source of the newsletter is going to be. They're

going to be three sources. It can either be from an idea or insight. It can be from an article or it can be from a YouTube video. It should ask me which

YouTube video. It should ask me which one it is. If it's from an article, it should use the API web scraper to scrape the article to get the full context of the article. Now, some best practices

the article. Now, some best practices that I recommend you you prompt it here when you're creating the skill is first of all, you want to double down on the reference files. Otherwise, it might

reference files. Otherwise, it might skip over that in the skill. So, I want to make it obligatory in the skill uh to always read those reference files at the exact point we did this in the above

process. It cannot skip over that.

process. It cannot skip over that.

Another one that's good to double down on is any corrections that are added in this process above. Um, please add that those as rules in the skill MD so it doesn't make that same mistake again.

Lastly, what we can do with skills is make them self-learning. Also add a progressive update feature in the skill where every time a user when using the skill says to not do something anymore, automatically let the skill update

itself with a new rule. And this is essentially how we make um the skill almost self-learning because it can automatically update when you say it to not do something. So from that, Glad

will start building the skill. And

again, a skill really is never finished.

We can still update it every time we use it. And and that's really the point of

it. And and that's really the point of it. The more we use it, the better our

it. The more we use it, the better our skills get. But this is the way to get

skills get. But this is the way to get to a good first output that becomes reliable to actually reuse over and over again and to improve over time. Now you

can see it now reads the skill creator skill to build a skill out of this. So

now it's built the entire skill. As you

can see in the skill MD, it's basically an overview of the skill and the process it has to follow. So you can see identify the source, read reference files, suggest five angles, read more

reference files, suggest three outlines, write newsletter, and suggest suggest 10 subject lines. And as you can see, it

subject lines. And as you can see, it also stored all of these reference files into the skill. So next time I use this, it will already have access to those. I

can now just click copy to your skills.

And now these skills will be saved into my account. So if I go here to

my account. So if I go here to customize, you'll see that it appear appears here in skills. And now all I have to do is if I want to use this automation or skill again is by going through a slash command or just saying

please I want to write another newsletter use my Ben's newsletter writer skill. So lastly what you want to

writer skill. So lastly what you want to do before using this skill and automation is to test it once make sure it actually works. Now cloud now has a built-in feature that can automatically test. So how do how would we do it? We

test. So how do how would we do it? We

would say something like cloud please um just test this scale now. Now generally

when you want to test you want to also define what you want to optimize this test for right what criteria does it has to have to test. So in this case I would say I want to optim um optimize this test for making sure the newsletter

aderes to my uh tone of voice and style as closely as possible. And then you can also define the criteria to test upon.

So in this case I would say something like the criteria for the test are first of all is the scale functional? Second

is the word count similar to my newsletter examples? Is the sentence

newsletter examples? Is the sentence structure similar? Is the tone of voice

structure similar? Is the tone of voice similar? And then we can also define how

similar? And then we can also define how the test should be done. So we can say something like um please run three test variations based on this YouTube link.

I'll paste that same link and now run a test. So it now finished the test. It

test. So it now finished the test. It

basically span up three agents to do test in parallel. And you can see that criteria one it passed all three. It's

functional. The word count one of them failed and criteria three uh the sentence structure two of them actually failed. So we can see a full report of

failed. So we can see a full report of the test results here. in the outputs.

Um, but it already gives me suggestions on how to improve this scale based on my criteria. So, it recommends adding two

criteria. So, it recommends adding two scale rule updates. So, I can just tell Claude now, please uh update the scale according to the the the rules you suggested.

And this is basically a really easy way to um let Claude update and improve your skills really fast because we can um test and optimize based on any criteria.

So we can go back again a few times back and forth to really get your skill to the next level. So now it updated the skill with the rolls. I can just click on copy to your skills and upload and replace and it will make a new version

of the skill. And now I can do this exact process again by just going to a slash command and uh typing in the skill or I can just tell clot please use my uh Ben's newsletter writer skill um to

write a newsletter.

Now let's say now I share it an article.

It'll follow the exact same process. You

can see it used ampify to scrape the link and now it follows the process. Let

me read the reference files before suggesting angles. So it gives me the

suggesting angles. So it gives me the five angles exactly like the original process. Let's say I go four. Now again

process. Let's say I go four. Now again

this is a human in the loop process but of course you can make this into a non-human in the loop process also.

Again gives me three outline options.

Let's say a now gave me the full newsletter. When I first heard about AI

newsletter. When I first heard about AI automation tools I thought it was for developers. I was the CMO of a 25 person

developers. I was the CMO of a 25 person SAS company. So used personal stories

SAS company. So used personal stories from my reference files. It span it up into a version that's relevant for my audience. Of course, I'd probably go

audience. Of course, I'd probably go back and forth a few times or change some things manually, but this can really help you get to good copy a lot faster. And the amazing thing with these

faster. And the amazing thing with these skills is I can now give team members in my team access to my copywriting skills.

And even though they have no marketing background or understanding around me, they can come up with very good drafts of newsletters, LinkedIn posts, etc. according to my tone of voice and style.

And I've built skills like this around all my copyrightiting tasks. For

example, here I have a LinkedIn writer with a very similar process. So you can see I invoked the LinkedIn writer with a YouTube link. In this case, it gave me

YouTube link. In this case, it gave me 10 angles for the LinkedIn post, then gave me suggestions for the writing frameworks, then suggestions for the hooks, and then a LinkedIn post in my

style. And you can see that my LinkedIn

style. And you can see that my LinkedIn skill has very similar reference docs.

my profile background, my ICP examples, but in this case for LinkedIn voice personality, what we do, and in this case, one more, which is a hook template document. But yes, it is a little bit of

document. But yes, it is a little bit of work creating these context documents, but it's going to save you a lot of time in the long run. I also have a copyrightiting skill for YouTube intros.

For example, let's write an intro for YouTube. Please load the skill, which

YouTube. Please load the skill, which goes through the process of writing a good intro, and it gives me a YouTube intro according to my process. Now most

of my copywriting skills have human in the loop because that's how I like to work with AI. But you can also of course make this completely autonomous and get good draft outputs which you can iterate on. And if you do that you can also

on. And if you do that you can also start scheduling your tasks because the next thing you can start doing with your skills is to schedule them automatically. So you could say for

automatically. So you could say for example to schedule my newsletter writing skill every morning at 8 a.m.

check if there's new YouTube videos on my channel and automatically repurpose them into a newsletter. So when I wake up in the morning, I already have a draft email newsletter prepared. And

I'll give some examples in the ideation part on how to use this scheduling feature. We can also start bundling

feature. We can also start bundling different skills after each other to automate even more complex workflows.

For example, in this chat, I bundled the LinkedIn post writer, a gift creator, and a newsletter writer in the same chat. So you can see from one YouTube

chat. So you can see from one YouTube link, we get a LinkedIn post, a GIF, and a newsletter all in the same chat.

And this was of course impossible with custom projects etc. Now I'll show you later how to bundle multiple of these skills into commands to automate even more complex tasks. But first let me show you how you do brand aligned

visuals. Now we can also use co-work and

visuals. Now we can also use co-work and build skills around any brand aligned visuals we need. For example around infographics uh ads uh uh brand aligned

dashboards gifs uh presentations anything you can imagine. For example,

here I created a skill around building brand aligned case studies from uh a client transcript. So you can see this

client transcript. So you can see this is uh aligned to my um colors of my brand uh to my style etc. So if you look at my website it's the same same style

and here I created a skill to create brand aligned infographics for LinkedIn.

So how do you go about building these brand aligned visual assets? I'll show

you by going through an example of how I build my uh infographic skill. But

before that, uh, very quickly, if you want access to all of the skills we're building out, more than, uh, 40 we have right now, you can also check out my AI community where we, uh, list them all out for you to customize and maybe help

you speed up building some of these skills a little bit faster. Also, we uh, do weekly AI workshops where we dive a lot deeper into these tools and have unlimited one-on-one live tech help available to help you set up things like

this. So, if that's interesting, you can

this. So, if that's interesting, you can check it out in the link in the description below. So you can see the

description below. So you can see the approach I took for the LinkedIn infographic uh creator skill is very similar to the one I showed for the copywriting. Right? So I just went and

copywriting. Right? So I just went and did the process once. For example, here I said okay I want to turn this LinkedIn post into an infographic image for LinkedIn. I threw in the LinkedIn post

LinkedIn. I threw in the LinkedIn post and then I told it first give me five suggestions on which part of this post would make the strongest visual. So it

gave me five suggestions. I chose option two and then I asked it now give me five specific ways this could be visualized best inside of an infographic so we can choose a good way to visualize this concept and again I do this because I

don't want one-off outputs or an immediate infographic because in general in my experience with a few of these decision points or human in the loop points the output gets a lot better a lot faster and eventually when your skills getting better and better maybe

you can start taking out some of these human in the loop uh steps to make it more and more autonomous. So it gave me five different ways to visualize it. I

chose the first one suggest me three different ways the copies can be structured. It gave me options for copy.

structured. It gave me options for copy.

Then I picked one and added one more uh detail and then I asked it to generate the infographic. Now cloud is not

the infographic. Now cloud is not capable of actually um generating PGs.

So I used Nanobanana. And Nano Banana is Google's image model and in general is by far the best still. So highly

recommend if you're trying to create images or maybe ads or infographics like this is to use the Google nanobanana mcp. Now in this case I told it okay now

mcp. Now in this case I told it okay now let's generate the prompt for nano banana and use nano banana mcp to generate the infographic because of course it has to prompt nano banana to

generate the infographic. I want you to use the brand guidelines of my brand here below to know how to prompt nano banana to generate the infographic that we defined above in my style. Now, this

is the key document you want to have for anything you want to do with visual assets is you want to have a brand guideline to make sure that any visual you create will follow your brand. Now,

I've experimented a lot with this too. I

found a good sort of template that you can fill out again it will be available in the link in the description below and adapt just in a chat together with claude. Uh you can also give it your

claude. Uh you can also give it your website for example or any asset you have so it can sort of start filling out uh itself. But once you have that, it's

uh itself. But once you have that, it's going to be a gamecher because anything you do, right, generating presentations, dashboard, etc., all you have to do is add the brand guidelines into the skill

or give it access to the brand guideline and it will do it in your uh brand style. So, so I added the brand

style. So, so I added the brand guidelines in the prompt and I added in the MCP instructions for using the nano banana Google image generator uh which I'll make sure to add in the the free

resources too. So you can just paste

resources too. So you can just paste that in so Claude knows how to use uh the Google Nana Banana MCP. It then

asked me for my Gemini API key which of course we need to generate images from uh Google Nana Banana. Highly recommend

to do it if you're going to plan to create any images um together with Clockco work. So you can go into the

Clockco work. So you can go into the Google AI studio and just click here on get an API key. Uh you have some free credits but then you have to add in your um billing information. Again, it's not very expensive to generate an image and

it's definitely worth it if you're going to do this. And then it created a prompt for another banana as you can see here.

And then it generated uh the infographic using the tool. And you can see it follows my brand pretty well. It's not

perfect yet. Probably have to make some changes here on the style. But this was a good first version. So I said now build a skill out of this exact process that we followed. It should be called this and triggered when I say this. But

before this process, I want you to add two more steps to the skill. First I

want uh it should ask the user for the content or idea it should generate these infographics on and then it should ask the user for which platform this infographic is going to be generated to

apply the right format for the image to it. Then I asked it that to make a

it. Then I asked it that to make a reference file out of the brand style guide and a guide on how to prompt the nanobanana mcp in order to get a good output. Now this is key when we give it

output. Now this is key when we give it a lot of context for example on the nanobanana MCP or a brand guideline without a docs then we have to ask claude to create a separate reference

file for it. Why is that is because we want to keep the skill MD as clean as possible. So all the extraformational

possible. So all the extraformational context should be stored as reference files in a skill and the skill MD should mostly focus on the step-by-step process it has to follow. That's how your skills are going to perform best. Yeah, I

doubled down on that rule that I usually uh use. Make sure to clearly make it

uh use. Make sure to clearly make it obligatory steps in the scale to read those reference files in order to ensure good outputs. Again, add in a

good outputs. Again, add in a progressive update rule that makes the scale automatically update itself when a user specifies something not to do anymore. So again, it created the scale

anymore. So again, it created the scale as you can see here. Add two different um I copied that to the scale. Then I

kept updating the scale probably around 5 to 10 times until got so good that almost every time uh a new infographic is generated, it's uh good enough for me to use. And as you can see, my last

to use. And as you can see, my last version here, the infographic v2 has a lot more reference files because I've started giving it more and more context around doing it a better doing a better job and therefore became more and more

complex and more and more information.

And that's really the key with these skills, right? They're never really

skills, right? They're never really finished. The more you work with it, the

finished. The more you work with it, the better they get and the more extensive they get. So you can start with

they get. So you can start with something simple, but it's really about just using it. I've also set up skills for branded excellraw presentations. As

you can see here, these basically generate slides uh automatically from YouTube transcripts or outlines. And as

you can see again, it uses my colors of my brand, the shadows of my brand, etc. Now, for most visual assets we want to create, we don't actually need any external connectors because Cloud can now generate presentations in PPT and

Google Slide format. Without these

connectors, it will be able to generate HTML in your style. So, any dashboard you want to visualize. I'll show you an example of that in a second. Excell you

can do it right from within cloud because it's based on JSON. So you can just copy the JSON into Excel draw and have your slides. The only thing Cloud is not capable of is generating PNG. So

that's where you want to set up the Google NanoBanana MCP connector. So

that's the one that I set up here in the connectors. Again, in the free

connectors. Again, in the free resources, I'll give you a guide on how to set up this Nanobanana MCP. And

again, if you want to copy all of the skills that I'm showing here and all of the other skills we have, you can also check out my AI community if that's interesting to you. Now before moving on to how you use cloud for ideation and

research very quickly let me go over commands. You can see that here inside

commands. You can see that here inside of customization we have different plugins for marketing sales operations etc. And these plugins you can basically just see as ways to organize and

distribute a bundle of skills connectors and command. For example, I can share

and command. For example, I can share this creative plugin with one of my teammates and when he imports it, he will already have the nano banana MCP,

all the skills inside of this plugin and my commands. Now, what are these

my commands. Now, what are these commands? If I look here in the

commands? If I look here in the marketing one, commands basically allow us um to make even more complex automations because inside of a command, we can lay out the process of using

multiple skills in sequence. For

example, here in the marketing plug-in here, I have uh one command set up for repurposing. And this one combines the

repurposing. And this one combines the LinkedIn writer, the newsletter writer, the excellraw, the gift creator, and the infographic creator into one workflow.

So, if I trigger this command, it'll be able to automatically repurpose for my LinkedIn, for my newsletter, and use the visual creators to generate visuals for them right away. And all this basically

is is one text file that maps out the process with multiple skills. can invoke

a command exactly the same way as I would do with a skill through the slash command uh with repurpose or by just telling clot please repurpose uh my content. I can build commands the exact

content. I can build commands the exact same way as we build skills. You can

just open a chat and say hey please build me a command and then you can uh instruct the process to follow in this command and which skills to use and then we can even start scheduling these commands and skills automatically through scheduled task which I'll show

you an example of uh in ideation but generally I recommend building these commands only when you have built some skills already you want to get to the next level of complexity but you can see that you can really start to do more and

more complex automations right inside cloud co-work and by the way the same is the case for these plugins we can just tell cloud to build you a plugin in. So

you can just define all the skills, the commands and connectors you want to include and cloud can then generate you the plugin which you can then later share to other people or people on your team. I have a full video that also

team. I have a full video that also covers plugins entirely. So if you want to learn more about that um you can check it in the link in the description below too. Now let's move to ideation

below too. Now let's move to ideation and research which of course are big parts of marketing and what cloud co-work can really start automating right now. For example, I've been using

right now. For example, I've been using this for ideation for my newsletter and for YouTube by tracking news updates, competitors, um, but can really be used for any ideation or research uh, task that you might have. Now, let me show

you an example of how I've built a skill out of my newsletter ideation process and how I then scheduled this into a scheduled task to do on a daily basis automatically. So, in this use case, I

automatically. So, in this use case, I said I'm trying to ideate for my newsletter. Inside the newsletter

newsletter. Inside the newsletter ideation folder, uh, you will find our ideal customer profile and our newsletter strategy. So this is another

newsletter strategy. So this is another thing you can do in cloud cowork is instead of um passing it through the documents that it needs to read each time you can also just save those in a specific folder and just give clot

access to that specific folder. For

example, I have a folder here with Beni context copy. I just give it that and

context copy. I just give it that and then it already has access to all those files. So that's another way to do it.

files. So that's another way to do it.

So I gave it access to that folder in this chat. Then I said I'm going to

this chat. Then I said I'm going to paste in some resources that you can explore to extract information.

Basically, the idea is that you need to navigate through and identify the latest pieces of content that have been published there in the last 24 hours and qualify them as relevant or irrelevant for our ICP and my newsletter strategy.

So, you can see here at the bottom, I gave it five resources that I uh normally read or use and give me inspiration for potential newsletters or YouTube videos. Now, again, here's where

YouTube videos. Now, again, here's where the Appify MCP can come in really handy because of course, we can also start uh tracking competitor posts on LinkedIn, YouTube, etc. with the Apify MCP. Now,

in this case, the sources I gave it were websites like A16Z and Anthropic. And

basically, I asked it to based on my newsletter strategy, identify if there are potentially interesting uh new updates on these websites that I could um make a newsletter about. I gave it a

little bit more context and then I said, what I want as an output is a well structured HTML report that I can open in my Chrome to see what all the content ideas uh I can create newsletters around

from these sources. It then started going through the websites and all the new content pieces they published and it found seven potential newsletter ideas and added it in an HTML report or dashboard which I can open in my

browser. As you can see, I get a nice

browser. As you can see, I get a nice dashboard and of course you can make this branded too if you want by giving it the brand guidelines. So, as you can see, it gives me an overview of all the new content published on these sources

if they're relevant potentially for a newsletter or not. here gives me full newsletter ideas and potentially interesting stats to share in my newsletter and potentially interesting infographics that I could add in this

newsletter. And then because I was going

newsletter. And then because I was going to plan to make this a schedule task and I don't want to have duplicates, um we actually need a database so it can track and make sure that it doesn't make uh duplicates all the time. What I asked it

to do is I plan on running this every 48 hours. And to ensure that we do not go

hours. And to ensure that we do not go over the same content over and over again, we need a database of all the sources that we have used in the past.

So please create a CSV that consists of all the qualified sources and unqualified sources with the date. What

I would like to see in the database is a source name, the source URL, the date, the execution time, whether it was qualified or unqualified, and then whenever this workflow runs again, it

first checks the database to know what it's already looked at and can then skip those sources already. And this

basically how we can build databases inside of cloud co-work is by just telling it to make a CSV or an Excel file. So you can see it made a

file. So you can see it made a newsletter source database with all the different columns that I instructed. And

now I created a skill out of this exact process just like I showed you before.

It then created the skill and added in the database as a reference file. And a

really powerful thing with cloud co-work we can now do is schedule these tasks or skills automatically on a daily basis.

For example, I can either just tell Cloud here, please schedule this skill every day at uh 8:00 a.m. to run

automatically, or I can go into the schedule section here and set up a schedule task. So, here I can say, for

schedule task. So, here I can say, for example, newsletter ideation ideates on my newsletter. And then in the prompt, I

my newsletter. And then in the prompt, I just tell it which skill to use. Use the

newsletter ideation skill um to ideate for my newsletter. And this basically schedules a prompt every day at 8 a.m.

The skill will already know to go through the database to make sure it doesn't do duplicates. And every morning I can open my cowork and already get ideas for my content. I can define the frequency, the folder it can access. So

in this case I could give it access to that context folder. So I don't even need to save it inside of the skill. And

I can define the model opus 4 or sonnet.

So, I could let's say daily at 8 a.m.

I could save this and now it's scheduled every day at 8 8 a.m. And this is sort of how you can make cloud co-work become proactive. You can open up your cloud

proactive. You can open up your cloud desktop in the morning and go through your uh scheduled tasks, you can imagine if you start bundling the newsletter ideation with the newsletter writing and

the visual creator skill into a command and run it on autopilot, you can start to automate more and more autonomously together with Glock work. You can see that the scheduled task will be on a separate tab here in scheduled. It

already loaded the skill here to know exactly how to do this process and the progress bar here. Now with clockwork, we can also completely automate our data analysis and reporting of analytics. for

example, for SEO reporting, uh meta ad reporting, in my case, YouTube uh studio data analysis, but any marketing analysis you can do because we have built-in skills now from Entropic on uh

being able to read Excel files, being able to create these brand aligned presentations or dashboards. We can

build skills out of them, schedule them on a monthly or weekly basis and completely automate our data analysis workflows. For example, here I created a

workflows. For example, here I created a skill around my YouTube analytics. I

gave it a CSV file of my uh YouTube studio analytics and it gave me suggestions on what data analytics could be useful for me. I then asked it to create a dashboard again in that HTML

structure using the front-end design skill which is a built-in skill from Entropic. And because it also has access

Entropic. And because it also has access to my YouTube scraper, it can also analyze my transcripts and give me conclusions on what I do in my intros for example for my best performing

videos. It then again created a full

videos. It then again created a full dashboard with a complete overview of my total views, top 10 videos by views, engagement, the top videos with the top

like rate and the top comment rate and share rate, which videos have the highest subscriber conversions and CTR, monthly upload frequency, and by

comparing the data with my transcripts, it could also identify which video length is optimal, my best opening hooks, for example, they include numbers and statistics generally, in the best

performing type of videos, how-to tutorials. Then again, I can build a

tutorials. Then again, I can build a skill out of this that runs every month to analyze my past month's YouTube videos and give me good insights. So for

any repetitive analytics or reporting tasks you might have around marketing, again, you can build skills out of them and just schedule them on a weekly or monthly basis and automate these tasks entirely. And lastly, if you do SEO, we

entirely. And lastly, if you do SEO, we can now automate large parts of our SEO workflows. We've built skills around

workflows. We've built skills around programmatic SEO, SEO audits, and SEO optimizing. Now, some of these like

optimizing. Now, some of these like programmatic SEO and SEO optimizing are pretty comprehensive skills that actually work best when used in clot code. I'll soon do a full video on using

code. I'll soon do a full video on using clot code work and clock code for SEO.

But let me show you a quick example here. For example, I run our SEO audit

here. For example, I run our SEO audit skill that does a full audit around legal compliance, performance, EAT, content, and geo readiness. It then

gives me a full report uh with action items on how to improve geo and SEO uh for my website or any other website you would want to run this on. It gives an overview of the categories and what

needs to be improved. Uh critical issues that needs to be fixed immediately, the medium priorities and high priorities and quick wins that you can implement fast. Now, this skill combined with our

fast. Now, this skill combined with our SEO optimizer and the web flow MCP where my website is hosted can almost autonomously keep our site uh geo and SEO optimized. Again, this is a little

SEO optimized. Again, this is a little bit more technical and SEO specific. So,

I'll do a full video on SEO soon. But if

you're a marketing or SEO agency, onboarding your team on cloud code or co-work and starting to become uh good at building skills out of the common workflows and services you provide uh should be one of your top priorities

right now. So now that you've seen some

right now. So now that you've seen some marketing use cases, let me show you another few use cases for uh sales cloud co-work can really entirely transform the way you do sales. So in this next section, I'll cover plenty of use cases

and skills to automate uh your sales processes. Let's start with using

processes. Let's start with using co-work for prospecting, lead qualification, and research. Now, as

most people have unique prospecting workflows, I'll just show you an example of a prospecting skill we've built, how we've built it, and the connectors that are most useful for prospecting because I think it will give you a good idea of

how to do this for your own workflows.

For example, here I built a skill out of a workflow that can scrape LinkedIn post engagers, qualify them against my ICP, and enrich the qualified leads and even do personalized outreach on LinkedIn.

So, you see here, I called the skill by just using a slash command here with LinkedIn post engagers. That's how we can use skills by just using the slash and using the skill here. Then I added someone else's LinkedIn post to scrape

the engagers from. And I added the qualification criteria. And because the

qualification criteria. And because the skill already has the workflow on how to do this embedded, it already went ahead and found the last LinkedIn post of Chris Long, it then scraped all the

engagers from that post using Ampify and identified 127 unique engagers. 17 were

qualified according to my criteria. It

also enriched the data with a company size, a company description, and the LinkedIn post they engaged with. And it

gave me the information of those leads with the enriched data in a CSV sheet as you can see here. So, we have the LinkedIn public URL, the about uh the company size, the company website, company description, and the post they

engaged with. Now, this skill stops

engaged with. Now, this skill stops here, but we can now run another skill to actually make personalized outreach messages and actually do the outreach on LinkedIn. Now the point of this is you

LinkedIn. Now the point of this is you can build skills like this for any repetitive prospecting qualification or research workflow you might have through building these skills. And I'll show you in a second how I built this one. But

whether you get prospects or data from LinkedIn, Google maps, Apollo, your CRM or uh any other social media, we do need the right connectors to actually let claude access the data on these

platforms. So the key connectors you want to have installed in clockwork for any prospecting, research or outreach task is first of all appy. If you don't know Appify, Appify is basically a

marketplace with scrapers that can scrape any social media like LinkedIn profiles, LinkedIn posts, LinkedIn post engagers, X, Instagram or any other social media. They can even scrape some

social media. They can even scrape some uh lead databases like Apollo. So this

is really one of the most important to install for any um lead prospecting or lead enrichment. Uh you can start for

lead enrichment. Uh you can start for free with Ampify and they also have a native integration with Cloud Coork. You

can just set up new connections by going here to the plus sign and go to browse connections and then I can choose ampify and you can do the setup. You can also connect it with for example Apollo or

any other lead database you might use.

You can now also connect it uh with clay as they have a native integration if you use clay. And another one that's really

use clay. And another one that's really interesting to explore if you're looking for lead data is vi prospecting which is an API purely made to give to AI agents to provide you with more data on leads.

And if you do outreach, you can connect it to your email outreach platform like instantly. Or if you do LinkedIn

instantly. Or if you do LinkedIn outreach, I recommend using the Uniow connection, which can automate outreach on LinkedIn, WhatsApp, Facebook, and X for you. Of course, for some of the

for you. Of course, for some of the other sales skills that I'm going to show you, a connection with your CRM, in my case, AIO, is of course very important. And what I highly recommend

important. And what I highly recommend is to have a meeting transcriber, in my case, Fireflies. And lastly, of course,

case, Fireflies. And lastly, of course, you want to have it connected to probably your calendar and your email.

So, how did I build that skill that I showed you? I basically went through the

showed you? I basically went through the workflow once together with Clav and then asked it to build a skill out of it. So, already knows and went through

it. So, already knows and went through the process and knows what to save as the workflow. So, here I said, I want

the workflow. So, here I said, I want you to scrape the most uh recent 10 posts of Ben's profile, extract the post commenters and engagers data and present them in a CSV along with the link

profile URL, name, headline, and the post that they engaged with. In this

case, I specified the specific Apify scraper that it should use, but it can figure this out itself, too. In this

case, I already knew, so I specified it to make the process a bit quicker. I

then gave it my LinkedIn profile to scrape the post from. It then went ahead and used the appy scraper to scrape the post engagers and it found 154 unique

people who engage with this post. Then,

of course, I want to qualify these according to my uh ICP criteria. I then

noticed they didn't actually use the LinkedIn scraper to base the qualification on. So I specified to use

qualification on. So I specified to use an appy scraper to scrape the LinkedIn and get specific data points like the about section, company name, and website. And that's the point of giving

website. And that's the point of giving these instructions. If you do it once,

these instructions. If you do it once, then build a skill, you will already know how to do it next time, so you don't have to correct it. It then

actually scraped the LinkedIn profiles and created an enriched CSV with potential leads for my business. And

then I asked it to build a skill out of this process and I gave it a little bit of extra context on how the skill should be built. Claude then built the skill as

be built. Claude then built the skill as you can see here which you can add to your library. And as you can see here in

your library. And as you can see here in the skill itself which is just an MD file in this case it's just a step-by-step process on how to do this and which scrapers from aify to use for

example. So you can see step one scrape

example. So you can see step one scrape post with engagement data use this appy actor. Step two extract and dduplicate

actor. Step two extract and dduplicate engagers. Step three, ICP qualification.

engagers. Step three, ICP qualification.

And step four, full LinkedIn profile scrape. And step five, company LinkedIn

scrape. And step five, company LinkedIn profile scrape. And lastly, the final

profile scrape. And lastly, the final output, which is in a CSV. You can, of course, add steps here where you actually generate personalized outreach messages, but we have these separated into different skills. Now, the second

use case we've been using it for is lead nurturing and follow-ups. Often, some of the lowhanging fruit are just leads that were lost or not properly followed up with before. So, we built a skill that's

with before. So, we built a skill that's called uh prospect minor and it basically goes through uh the last lead column of our CRM. It researches all of these leads, scrapes their LinkedIn and

prioritizes them based on our ICP criteria and checks all the email communications we had with this lead to understand how we could follow up. So,

here I said run the CRM prospect mining scale for my CRM. It then asked me for my criteria to find potential opportunities. I gave it the criteria.

opportunities. I gave it the criteria.

It then found 160 records in that column. It then researched these leads

column. It then researched these leads and found their LinkedIn profile and started scraping their LinkedIn profiles with Apify again. And then it went through the email communications to identify the last communications with

each of these clients. And it gave me a CSV sheet um prioritizing the loss lease based on company size and my ICP criteria and the comm's communication summary. Again, based on this, of

summary. Again, based on this, of course, we can start drafting potential follow-up messages and actually send them out through the Gmail connector. So

for this one specifically, we used the Gmail app and our CRM connector which is Atoio in our case. So how do we build this one? Again, same thing. We just did

this one? Again, same thing. We just did the process once together with cloud and asked it to create a scale around it. So

I've just told it here to go through my CRM. It's wrongly spelled, but it

CRM. It's wrongly spelled, but it understood. Told it what to do and I

understood. Told it what to do and I also told it to launch sub agents for this and give each sub agent 15 leads to analyze. Now this is important when

analyze. Now this is important when you're running bulk tasks. You can now run sub aents also in cloud co-work and the big advantage of this is these sub aents can run in parallel and this means

that the context window here doesn't get overloaded because they a lot of the research is done by these sub agents and because they run in parallel it can be done far faster and this is a big advantage we have in co-work now too we

can actually run bulk tasks on lead list of 100 to 200 which was previously hard to do with the normal cloud chat but you do sometimes have to specifically mention to use sub agents when running

bulb tasks and then again I asked it here to I need to create a skill around this process with some extra context and it gave me the CRM prospect mining skill which I can just copy to my uh library.

Now the third use case that I think is very useful especially with the new feature in coowwork of scheduled task which I'll show you in a second is a call preparation skill which is uh one

we customized from enthropic skills. So

you can see here I run the call prep for my meeting with a client on Monday. The

skill then used my CRM to find the lead and then uh started researching the email communications with this lead uh any past Firefly transcripts and researching the lead on the web. And

then based on the research it did, it gave me a call prep brief in HTML format which is a nice sort of dashboard uh here with the company overview, an account snapshot uh what they do, the

interaction history, the suggested agenda and discovery question suggestions. Now, of course, we can

suggestions. Now, of course, we can customize this skill from Entropic by adding more data on how you want your call preps to be structured, what data points he should look for in the research, etc. Now, you can customize these entropic built-in skills really

easily by just telling Claude. For

example, you can say, I want to customize the entropic uh call prep skill according to my how my company does um call preps. You can then give it all the context uh you need and what you

want to change and CL can then instantly create a new skill that's adapted and customized to your specific way of working. Now these kinds of use cases

working. Now these kinds of use cases become especially powerful now that we have scheduled tasks here in the sidebar in uh co-work and here we can basically schedule our skills or task to run at a

specific time interval. For example,

every morning at 7 a.m. So for this one I could say something like um every morning at 7 a.m. check my CRM and go through all of the meetings I have today

and use the call prep scale for each of the meetings I have today. present all

the call prep briefs in a nice HTML dashboard for me to go through every morning. I can then decide the frequency

morning. I can then decide the frequency daily at 7 a.m. And as soon as you will open your co-work in the morning, it'll start running this skill and preparing all your calls for today. Then for sales

analytics, cloud co-work can become really powerful too. For example, we've used it for win loss analysis and sales rep performance analysis. You can see here I run it by calling the win loss

analysis scale. And this scale basically

analysis scale. And this scale basically goes through all of our leads in our win column and in our loss column in our CRM. It researches both of these types

CRM. It researches both of these types of leads, but also the meeting transcripts from all of these leads and the email communication. And based on that, it makes an extremely comprehensive win loss analysis report

in a Google Docs that gives us an overview of the win rate, the key findings of our best and worst performing customer profiles, common objections, patterns on actions we tend

to take with converted leads like engagement depth, predicts outcomes. It

shows us deals won and lost by company type and by pricing. Gives us case studies on one leads and why they were won. It does a full analysis on lost

won. It does a full analysis on lost deals and you can see that we have to get far better at follow-ups which co-work will hopefully help us with.

Then the key differentiators between one versus lost deals and red flags and early disqualification criteria for us.

Honestly, this was extremely insightful and pretty impressive to read this report. And of course, we can now run

report. And of course, we can now run this through scheduled task on a monthly basis to analyze our wins and losses over the last month. And again, this one was built by just doing the process once manually with cloud and then saving it

as a skill at the end. And one more sales analysis skill we built is for a sales rep performance analytics by going through past transcripts, CRM data, and their email communication. So here I ran that skill on one of our sales reps and

it gave me a full Google Docs with a full uh performance report. So it gives an overall grade, the top three strengths, the top three improvement areas, and a full performance scorecard

per client. It also grades across

per client. It also grades across different stages in the sales pipeline like discovery, demo, objection handling, pricing, follow-ups, etc. And

the top five recommendations for this sales rep to start focusing on improving. Then lastly, I think another

improving. Then lastly, I think another very useful one for most sales reps is the review sales pipeline skill, which we also customized from Entropics built-in skills. Now, this can be very

built-in skills. Now, this can be very useful to run on a daily basis too with scheduled tasks because you can basically check your pipeline on a day-to-day basis and give you a brief on all the important leads to focus on

today by analyzing your sales pipeline, email communication, and past transcripts. So, you see why these

transcripts. So, you see why these connectors become so important because you give God all the context it needs to actually come up with really strategic advice for you. So here I ran it with the pipeline review command and then it

asked me questions on which pipeline it should review, what stage I want to focus on and where to send the pipeline summary. In this case, I told it to DM

summary. In this case, I told it to DM me on Slack. It then send me a full report to Slack and also presented here in the chat and it basically gives me an overview of the leads I have to focus on today because they're hot or uh we might

risk losing them. You can also adapt this to even include uh pipeline hygiene. Uh so it can automatically

hygiene. Uh so it can automatically start updating your pipeline which is of course a big pain point for anyone who's in sales because it can now analyze your uh email communications, your transcripts and the pipeline itself. It

can make your pipeline automatically update itself when changes happen. But I

think you get the point. You can really start automating any repetitive tasks you have around sales. Uh I really highly recommend starting to experiment more and more with co-work and building skills because it's really going to make a big impact on your productivity. Now,

lastly, there's one more thing I want to cover that ties everything together. Um,

throughout this course, I've talked a lot about the importance of giving Clot the right context. And in this last section, I'll take that up a notch and show you how to connect Obsidian to Cloud Co-work to essentially give your

AI agent a second brain. And this really allows co-work to become your main operating system or interface for doing work. This setup will essentially give

work. This setup will essentially give Cloud unlimited memory, persistent context across every conversation. And

it's especially interesting for businesses because it can be scaled across teams. So, in this last part, I'll walk you through the main uh advantages of this setup and show you an easy way um to get started with this

today. Now, before showing you how this

today. Now, before showing you how this works, how to set it up, and why I recommend starting with this today, let me quickly go over the five big advantages with some examples of having a second brain. Now, the first one is the most obvious one, and it's

persistent context. Right now, most

persistent context. Right now, most people use AI in isolated conversations.

You have to reexlain everything in each chat about your situation, your project, your workflows, etc. And with a second brain, your AI agent has persistent access to all of this context. And not

just a few facts, but detailed context around everything. You can see in my

around everything. You can see in my Obsidian, I have context saved around everything. My business, my strategy, my

everything. My business, my strategy, my projects, my brand, my workflows, my team, uh about myself, my meetings, literally everything. And here in this

literally everything. And here in this graph view, we can see the relationships between all of these documents. For

example, my AI accelerator here, and here's me. Now, I'll show you later how

here's me. Now, I'll show you later how this actually works and how to set this up. But because of this, I can now open

up. But because of this, I can now open a new chat session with, for example, Cloud Co-work, and ask something like, "What should I focus on today?" It's

already connected to my knowledge, Obsidian Vault, how they call it. And

you can see Claude now pulls my context to give me an answer. And because it pulled this context, it now knows that my main priority should be landing page copy changes, recording the Obsidian video, and organizing the Spain offsite

in April. I could even ask it things

in April. I could even ask it things like, "Write me a LinkedIn post based on AI topics we discussed in our team meetings this week. Use the LinkedIn instant uh skill." It will now go

through our team meetings, see what AI topics we discussed, and use a LinkedIn skill to actually write it in my tone of voice, then pulls the context from my second brain, uses the skill, and then output a LinkedIn post according to the

topic we discussed, which is of course this second brain topic. This week, my team and I built something different. We

created a second brain for our entire business. Now, this is just an example,

business. Now, this is just an example, but you can see the power of this. And

with the new scheduled task feature, this becomes even more powerful.

Secondly, besides co-work now always being able to pull up to date and complete context around me and my business in any chat, it can also directly update the context in my second brain. So, any decision, any rule, any

brain. So, any decision, any rule, any project update I make in an AI chat, it can log it directly back into my second brain. For example, if I see something

brain. For example, if I see something in this LinkedIn skill that I don't like, for example, never use um m dashes when writing content for me, I can say something like remember this in my

second brain or in my AI operating system. You can see it now saved and

system. You can see it now saved and updated this in my second brain as a role in the writing preferences. Now,

this is huge because it means the more you and your team use AI to do tasks, the more context is built, the more guidelines it has and the better your AI becomes for yourself and your entire team. Thirdly, there's also a big

team. Thirdly, there's also a big advantage of having this second brain when building and using skills. Now, if

you don't know what skills are yet, skills are basically saved instructions for your AI agent on cloud code or cloud co-work on how to do a specific process or task. For example, in this LinkedIn

or task. For example, in this LinkedIn skill, it goes through the step-by-step process of how to write a LinkedIn post in my style and essentially allows these agents to automate workflows just like I showed you in the example. Now, I have a

full video covering skills if you're still unfamiliar with it, which I'll make sure to link in the description below, too. But what changes with this

below, too. But what changes with this second brain? Skills usually have

second brain? Skills usually have reference files and context inside the skill folder. For example, in this uh

skill folder. For example, in this uh LinkedIn skill, I have a pen profile background document, a hook template document, an ICP document, a LinkedIn example document, and a voice personality document. Now, we use these

personality document. Now, we use these reference files in skills of course to get to better outputs, but it usually takes a long time to provide all the context every new skill needs to get to a good output. And if you have all of

the relevant context around your ICP, your tone of voice, your business already in your second brain, it means you can build good skills far faster.

This means you only need to lay out the process and point it to the right context in your second brain. And you

don't have to give it the exact same context over and over again for each new skill you're building. For example, I have dozens of skills that share reference files around my ICP, their pain points, what my business does, etc.

As you can see, my newsletter writer skill shares a lot of the same context files as my LinkedIn one. So the new way I'm building out my skills with this second brain setup is not by adding this

extra context or reference files in the skill itself but by just pointing my skill to the right context in my second brain. For example, this is my new

brain. For example, this is my new LinkedIn skill the LinkedIn instant and in this case you can see that I only have the skill MD only the process instructions and there I direct it to

where I can find the specific context files in my second brain. And this means that any update I make to a reference file in my second brain, for example, in the ICP document, all of the skills that

use this ICP document are instantly updated instead of me manually updating dozens of skills myself. It also means that this rule, for example, that I added, uh, never use mashes when writing

content for me, is now updated as a rule in my writing preferences, which is one of the documents many of my content writing skills point towards. So, my

newsletter skill is automatically updated with this same rule. Now,

another huge advantage of this setup, of course, is that it works across any AI you use. So, your second brain in

you use. So, your second brain in Obsidian is really just a folder of markdown files, which I'll explain in more detail in a second. But that's all it is. It's just a folder that I can now

it is. It's just a folder that I can now give Clot Co-work access to. I can give Clot Code access to. I can give Codeex, Anti-gravity, or any other AI agent provider access to. For example, I can

go to the code tab or just use Cloud Code in the terminal. give it access to the same folder of my second brain, the Ben iOS. And I can ask the same

Ben iOS. And I can ask the same question. What should I focus on today?

question. What should I focus on today?

And as you can see, clot code has access to the same context. As you can see, landing page copy, Obsidian video, and Spain offsite. And this even works

Spain offsite. And this even works across different AI providers. Here, I

gave Codex access to the same folder, as the same question. As you can see, landing page copy, YouTube production, and Spain offsite. And then lastly, which is huge if you're a business, is that this is not only for yourself, but

can actually scale across your entire team and business. Me and my team now share this same second brain with my business strategy, ICP understanding, to voice references, company goals, etc. So

instantly, my entire team's AI agents have access to this context that make them far more powerful and productive for my business. For example, my entire team will always have access to up-to-date strategy documents, ICP

documents that their agents can instantly use. And with the context and

instantly use. And with the context and skills that I've built, I can now let an engineer write my LinkedIn post with an onpoint tone of voice. In Obsidian, I can sync these updates in my context

across the entire team here. And you can imagine that this setup could completely change the way you run a business, which I'll get to later in this video. But let

me first explain how this actually works because it might look a little bit overwhelming and complicated, but it really is not. So, what is Obsidian? All

Obsidian really is is just a visual overlay of a folder and its files on your computer. As you can see here, all

your computer. As you can see here, all these folders that I'm seeing in Obsidian contacts daily departments intelligence, onboarding, I also have available here in the folder that Obsidian is connected to. And all we

really do is we point cloud co-work, cloud code or codeex or any other AI agent we use to the same folder on your computer. So they can directly read and

computer. So they can directly read and write to the same files you see in Obsidian. And that's why Obsidian is a

Obsidian. And that's why Obsidian is a great app to do this on because we don't have to sync anything, use an API or an MCP or a cloud-based software. It's all

local. So in coowwork for example, all I do is give it access to that folder. But

if we have thousands of these context files, how does our AI agent on clockwork or clock code actually know what context to use and how does it update it? Now again, this is pretty

update it? Now again, this is pretty simple. The way it updates and retrieves

simple. The way it updates and retrieves the right data from your Obsidian folder or vault is through the cloud MD file.

Now the cloud.md file is basically just an instruction you give to your AI agent on cloud co-work or cloud code on how to navigate the second brain or the folder.

So if I ask coworker a question like what did we talk about in our team meeting yesterday? My AI agent first of

meeting yesterday? My AI agent first of all knows that it needs more context to answer this question. It then reads the cloud MD file to understand where in my Obsidian vault or folder it will find

more information around this. Then it

reads those specific documents like for example yesterday's Firefly transcript to answer the question back to the user.

So you can basically see that cloud.md

file as sort of like a system prompt or an instruction layer that tells your AI agent how the vault or your folder is structured and where to retrieve and save data. So as you can see in this

save data. So as you can see in this session with cloud co-work where we wrote the LinkedIn post, it has access to one file which is the instructions or the cloud MD and this is basically just instructions for the AI agent on where

to find specific information in the second brain and where to save it. So

you can see how this system works, the file structure, knowledge routing, and this same clot.md file or the instructions you also see here in the Obsidian vault. Now, don't worry, I'll

Obsidian vault. Now, don't worry, I'll show you later in this video exactly how to get to this cloud MD and how to set up these instructions easily. Now, you

might be asking, how's this actually different from CL's built-in memory, and why would we actually need Obsidian if it's just a folder? Now, first of all, CLA's built-in memory is very limited and is basically designed to remember

the most essential facts about you. It's

generally stored in one document. So the

difference really is the scope of the context. My Obsidian Vault, as you can

context. My Obsidian Vault, as you can see, has thousands of pieces of context.

And secondly, if Obsidian is just a folder, why would we need Obsidian? The

short answer is you don't need it. Uh

you can set this up in a folder yourself, too. But honestly, it's just a

yourself, too. But honestly, it's just a nice way for you to visualize, organize, navigate, uh search, and link your notes and files together. my context and knowledge sources over the last weeks have been growing really fast and

honestly without Obsidian I wouldn't be able to organize it the way uh that I have right now through the graph view we can also see the relationships between all of these context files it automatically makes these connections

between different documents or context files uh for example in the brand identity document you can see that we have the voice positioning we have links to the ICP for our ideal customer profile and the pain points of our

customer this is also what your AI agent is able to navigate so for example if it reads the brand guidelines and feels like it needs more context around my ICP, you can see this link or wiki link, what they call it, and actually look up

this document to find more information.

And the nice thing about Obsidian is I can really easily sync the uh updated context across my team if you're going to use this in a team setting. It's also

entirely free to use and download. So, I

just recommend using it. Now, before

showing you how to set it up, let me zoom out for one second because I think this setup has much bigger implications than just some extra productivity. I

think it could entirely change the way people work and businesses are run. So

there are multiple of these big developments coming together right now in AI. Uh everyone can see that these

in AI. Uh everyone can see that these L&Ms are becoming better at reasoning.

MCPs are getting better and now allow them to efficiently navigate softwares and the internet and skills plugins, schedule tasks, etc. now allow you to automate uh repetitive tasks fast and

easily. But the missing layer for those

easily. But the missing layer for those AI agents was really context. And with a setup like this, I think it will slowly allow people and businesses to start adopting an AI interface like cloud

co-work or cloud code as the main interface to do their work instead of hopping between 15 different softwares all the time. But maybe more importantly, I think this is the development that will slowly allow AI

agents to start doing work autonomously without our involvement. Personally, for

example, since I've really started using co-work on a daily basis, I've been less and less in my Gmail inbox. I've been

less on Google doing research. I'm

barely in my CRM anymore. And now with these combination of MCPs, connectors, and scheduled skills, I can now automate end-to-end processes like email follow-ups without my involvement. And

that's why I really believe you need to start building this today because the value of this setup isn't in the setup itself. It's in the context that builds

itself. It's in the context that builds over time. Every decision that gets

over time. Every decision that gets logged, every correction or role that gets saved, every project that gets documented, and every skill that gets made, it all compounds. So the AI agent

you and your entire team have after 6 months of using this is far more powerful than the one you start with on day one. And if your competitor for

day one. And if your competitor for example starts 6 months after you, they're not just behind on the tool, they're behind on 6 month of intelligence that makes the tool actually perform far better for you. And

even when better models come out, and they will, the same context just becomes more powerful. So the context I think

more powerful. So the context I think will be your actual mode in the upcoming months and years. So how do you actually set this up for yourself or your business? Now, the key thing to keep in

business? Now, the key thing to keep in mind when you get started is that you don't want to overoptimize. It might

look very overwhelming what I just showed you in my own setup, but this setup I started with just probably five files a few weeks ago. This context will grow very naturally the more you use AI.

You just want to sort of start very simple and let the system evolve naturally. The same is the case for the

naturally. The same is the case for the file structure. It is important, but you

file structure. It is important, but you do want to start simple and let it evolve naturally. There is really no

evolve naturally. There is really no perfect file structure because it's going to be highly context dependent. It

will depend on your context, your business, your goals, and your projects.

Um, now that being said, there are two file structures that I recommend and I've seen work well as a starting file structure. One for if you're running a

structure. One for if you're running a business uh and you want to use this across your team and one if you uh want to use this for yourself as a professional or as a soloreneur. Now, go

over the file structure quickly so you understand what's in each because it is important. But we've also built a

important. But we've also built a plug-in you can use in your AI agent on cloud co-work or cloud code to help you set up and get to these starting structures fast, which I'll show you in a second how to use. And second, what you want to keep in mind when I'm going

through this is that many of these files are and can be created by your AI agent.

So don't get overwhelmed. You'll get

there naturally. So I'll cover the file structure that I use for my business setup and then I'll show you quickly the personal setup which is basically the same but with less files. So first we have the context folder and this is where you store general context around

who you are, your business, your strategy, your team, your brand and it's basically everything your AI agent needs to understand about you and your situation always. For example, in

situation always. For example, in context, I have information about my team strategy stakeholders pain points, organization, operator, the ICP, and the brand. Second, we have daily.

And this is basically where your AI agent logs everything that happened each day across your sessions, maybe across your meetings. And this is probably the

your meetings. And this is probably the most important one because it gives your ai agent that continuity between conversations. Then third we have

conversations. Then third we have departments. Now this is if you run a

departments. Now this is if you run a business uh you will of course have different departments. For example in my

different departments. For example in my case community, content engineering partnerships operations etc. And then in the community folder for example we can have SOPs around work that needs to be done in my community. For example here

YouTube to community repurposing. The

fourth one is intelligence. And this is a bit like the first one context but much more detailed. And this is the place where uh things like meeting transcripts, decisions, uh competitor

research, market insights get stored over time. Then we have onboarding. Um

over time. Then we have onboarding. Um

here you can have SOPs around uh onboarding new team members or even clients. Then we have projects here. Now

clients. Then we have projects here. Now

projects will highly depend on your context. For me, projects can be for

context. For me, projects can be for example different YouTube videos I'm working on. So I can ideate and work on

working on. So I can ideate and work on scripting uh on one video between different chats. Uh if you run an

different chats. Uh if you run an agency, this can be a project for each client you're managing. But this will be highly context dependent. Then fifth, we have resources. And resources is

have resources. And resources is basically anything reusable. So you can imagine it like a library of prompts, templates, frameworks, maybe content output examples, good examples, things

like this. Then we have the skill

like this. Then we have the skill folder. Uh an important one uh where the

folder. Uh an important one uh where the reference material of your skills live.

for example, your strategy docs, your voice guides, your ICP descriptions, uh, basically additional information that your skills point to. So, you can see I have all my skills here, uh, with the reference files laid out. By the way, if

you want access to all of the skills that me and my team are building out and using, uh, you can also check out my AI accelerator in the link in the description. Then, lastly, we have here

description. Then, lastly, we have here tasks. And tasks can basically be uh,

tasks. And tasks can basically be uh, to-do lists. And then we have teams with

to-do lists. And then we have teams with more context uh, around each team member's role and responsibilities in your business. So your agent always has

your business. So your agent always has context around anyone in your team. And

that's really it. And then at the root here, you have the clot.md file, the brain file, which is the instruction layer that tells your AI agent how this whole file system here works and how to

navigate it. And this will also appear

navigate it. And this will also appear in the co-work section in their folder instruction that I showed you before.

Now, if you're setting this up for yourself uh as a solopreneur or maybe as a professional, you can basically have the same uh structure but a little bit simpler. Uh so here I have an example of

simpler. Uh so here I have an example of the personal OS. It's basically the same file structure uh without the department, without the team section, and without the onboarding. So, same

file structure, just a bit simpler. Now,

again, might look overwhelming, but I started this a few weeks ago with just five files, and this sort of naturally grew, and a lot of this context has been created by my AI agent. And the plug-in we built is going to make this process a

lot easier and faster to do. So, how do we set it up? You can just go to Obsidian and download Obsidian for free.

Once you've done that, you'll land on a screen like this. So you can just go here to create new vault. For example, I call it Benai test and then I have to choose a folder. So I just create a new

folder and this is the folder that you're going to point cowork or cloth code or any AI agent you use towards to access that same vault that Obsidian visualizes for you. So we open this one

and click create. So we now have an empty folder. Now, if you're going to

empty folder. Now, if you're going to set this up in cloud code, you can also use Obsidian uh CLI, which I'll make sure to put in the description below.

That can help you get to a basic generic setup a bit quicker. But as said, we've built our own plug-in, which is available in my AI accelerator together with all our other plugins and skills to get you to that file structure that I

showed you before a lot quicker. It

helps you also populate uh your essential context a lot faster. And this

plug-in will work across cloud co-work or cloud code, wherever you want to use it. So, if that's interesting to you,

it. So, if that's interesting to you, you can check it out in my AI accelerator in the link in description below. We also have unlimited one-on-one

below. We also have unlimited one-on-one live tech help available if you want uh some help setting up these things. We

also do AI workshops where we dive a lot deeper into these setups and tools. So,

if that's interesting, definitely check it out. Also, if you're a business and

it out. Also, if you're a business and you want me and my team to help you in a more personalized way to set up a business AI OS for your company, we're now opening a few limited spots to help businesses set this up. So, if you want

more information, you can also check the link in the description below. In there

you can find a link uh with the marketplace of plugins and skills of my accelerator. And from there if you do it

accelerator. And from there if you do it in co-work you can just go to customize.

Click here on the plus and you click add marketplace and you add the link that you find in accelerator. Once you've

done that and you go to browse plugins you'll find a tab here at personal that says Benai skills and in there at the bottom you'll find our plugin Ben Obsidian plugin. You can install that.

Obsidian plugin. You can install that.

And now you'll see that in your plugins, this one will appear. And there we have the skills that help you get to that setup in Obsidian a lot faster. And all

we do then is we point to the same folder that we just set up in Obsidian.

So in this case, Beni test. We click

always allow. And now we uh use uh the plug-in and the scale setup.

Right? We run this and it will walk you through uh the setup on getting to this file structure a lot faster. and I'll

start asking you some question to get the essential context set up. So first

it asked me what kind of vault do I want a business setup or a solopreneur setup.

So in this case I'll do solopreneur just to show you as an example. So now it's already created the initial folder structure and the cloud MD with the instructions on how to navigate this type of folder structure. As you can see

now in obsidian we already have this folder structure. Now most of these are

folder structure. Now most of these are still empty of course because we haven't given cloud any context. So the next question is to really start giving it context, right? And that's what this

context, right? And that's what this plug-in does, right? It it's going to ask you some questions to populate your initial context data set. Now want to spend probably half an hour to an hour here to get your your initial setup,

have the initial context data set and from there it will naturally expand and I can tell you if you start using AI more and more in a couple of weeks you'll have a very expanded data set of context that really makes your AI far

more powerful. So a couple of important

more powerful. So a couple of important things to keep in mind once you get to that initial data set is every new task you start in cloud code cloud co-work you always want to point towards that same uh folder. Second when there are

things that you want your AI or your second brain to remember clearly tell cloud co-work to remember this in your second brain or whatever you call your folder. If you can ideally even point it

folder. If you can ideally even point it to the specific file it has to save that role to. Thirdly, if you see it have

role to. Thirdly, if you see it have issues navigating the folder structure, tell it to update the claude MD, which you can also do yourself because remember this is the bridge to point it

to the right direction. So I can add in rules all the way at the end here too.

You can see I've already added some rules on how it should navigate the folder structure. And then lastly, if

folder structure. And then lastly, if you're going to build skills, I highly recommend to take a new approach now and instead of embedding reference files into the skill here in Cloud Co-work or

in Cloud Code, save the reference files in your second brain and let the skill point towards the right folders. You can

also adapt your old skills by just telling claude. For example, here I

telling claude. For example, here I said, can you adapt my LinkedIn skill and create a new skill and gave it here specific instruction instead of having the reference files in the skill point towards the files in the Ben AIOS to get

the additional info instead of having them saved in the skill. And again, the earlier you start with this, the more powerful your AI agent is going to get over the long term. So, highly recommend to start soon with this. Now, that's it for this video. If you're still

watching, first of all, congrats. And

second of all, thanks for sticking with me for so long. Again, if you want access to some of the skills and plugins uh that I've shown in this course, of course, you can uh check out my AI accelerator in the link in the description below where we list them all

out. We also do multiple AI workshops

out. We also do multiple AI workshops and Q&As every week to dive a lot deeper into these tools and setups. We have

unlimited one-on-one live tech help available to help you resolve any issues uh you you might have when setting this up for yourself and a great community of professionals, business owners, and department leaders. So, if that's

department leaders. So, if that's interesting to you, you can check it out in the link in the description below.

Also, if you got any value out of this video, I really highly appreciate a like and a subscribe. It really does help me.

Um, thanks again for watching. And if

you want to learn more about AI and claude, you can check out the video here above.