Transcription
Hello legends. In this video, I'm going to give you an overview of the entire Claude ecosystem. This is a great video if you're either brand new to Claude or you've just been using Claude chat and then you want to improve your skills so you can get better at using AI.
We'll start by looking at the three different products that Claude offers, which is Claude Chat, which is great for everyday conversations and simpler questions. We'll look at Claude Co-work, which is great at working at tasks and bigger projects. And then we'll look at Claude Code, which is actually running on the same engine as Claude Co-work, but this version has been optimized to write code, software, and create automations for you.
We'll then go through and understand what level of access each of these products has on our computers. So if they can access local files, can they access the terminal, or they only accept drag and drop files? We'll look at more advanced concepts, things like skills, connectors, and plugins. We'll understand what these are, how to add them to our workspace, and then how to use them for our workflows.
And then we'll wrap up by looking at some new places that you can use Claude. So outside of the desktop app or outside of the browser, one of my new favorite places to use Claude is inside Excel. This version of Claude knows how to work with pivot tables, with graphs. It knows how to create sheets, and it can also read and write Excel formulas. So whenever it works with your data, it never hallucinates and it's always 100% accurate.
So for this demo, I'm actually using the Claude desktop app. The desktop app gives me access to all three versions of Claude into the one space. But really, in order for us to access Claude Co-work, we have to do that in a desktop app.
We'll start by looking at Claude Chat. This is the best for those simpler tasks, stuff that you can solve within about a five or ten-minute conversation with Claude. Stuff like doing research, helping you with your emails, trying to brainstorm an idea, the chat mode is actually the best place to do that.
Now, as with any AI tool, it's all about context. So you'll notice that all the different functions and features for the chat mode, as well as Co-work and Code, it's all about the information that you bring into the context. So the chat mode of Claude has access to files that you can upload, and you can upload files into here. You can upload PDF contracts and get them summarized. You can upload financial data in an Excel sheet and then get Claude to work with that.
A really common feature for the Claude Chat is to use this projects feature. A project is like creating a separate room inside Claude where you can upload a private set of instructions and a private set of files. Private in the sense that these are only applicable to the chats that you have within this project. So as long as I create a new conversation within this project, I'll be able to access these custom instructions and custom files. But if I was to go back into my main chat window, all this context would not apply.
Now, over time, as you have more and more conversations, Claude will then also begin to create a separate memory file. It'll pick and choose the most important stuff from each conversation, and then every time you create a new conversation, it'll actually dip into the memory file and just get smarter and better over time.
Now, you may have noticed this quick link button where it says "Start a new task in Co-work." That's actually the second product for Claude. And the reason that you have this button is because typically when you go across to a project, you're actually working on a slightly bigger piece of work. This is normally stuff that you cannot solve in that five or ten-minute session. It's usually a multi-touch task where you have to go and come back multiple times and kind of chunk through the work bit by bit, but we'll go across to Co-work a little bit later on.
Another interesting feature of Claude Chat is artifacts. Artifacts are little visual apps that live in the Claude Chat that you're having. They can be as useful or as pointless as you like. For example, an interesting artifact can be an email simulator where you can upload your thoughts, you can write your tone, and then you can get an email generated on the output. So, as you can see in this case, it's a pretty useful tool that you might use for work.
Now, for the chat mode, the final thing I want to explore is this "Customize" tab. And here you will see some of those advanced concepts like skills, connectors, and plugins. So, this settings panel is actually a global settings panel. We could come into here from the Co-work tab or the Code tab. It's just important to note that not every one of these concepts is available in every single product.
So, we were just taking a look at Claude Chat, and before I go into Co-work and Code, I actually want to cut away now and start looking at some of these advanced capabilities, things like skills, connectors, and plugins. So, I've just pasted the logo of the product here so it makes it easier for us to see. But the main thing is that the different products actually has access to different capabilities.
So the Claude Chat has access to connectors and to a default version of skills. If you go across to Claude Co-work, you have a wider range of capabilities. You have the skills, connectors, and plugins, the happy trio that everyone talks about. And you have this new capability called scheduled tasks. And then finally, Claude Code has a slightly different set of capabilities. So we've got skills, connectors, and plugins. In this case, the connector is a raw MCP. I'll get more on that later on. And then we also have these things called hooks and remote control.
So the first learning is that just because you can see skills, connectors, and plugins, it doesn't mean that every tool can access each of these features.
Now let's start by looking at the three most common capabilities. In this case, we have a skill, which is a way for Claude to automate a repeatable task. We have a connector, which is a way for Claude to interact with a third-party tool that we're using, something like Gmail or ClickUp or Slack. And then we have plugins, which are typically connections or groups of skills and connectors where when you put them together, you give Claude the ability to start to automate a role.
So, let's dive a little bit deeper into each of these. Starting with skills, let's actually open up the "Skill Creator" skill. This skill is actually a default skill built into Claude, which gives it the ability to create more skills. And over here, we can see the anatomy of a skill, where the first part of a skill is actually a lightweight description of what that skill is and what it does. And the second part of a skill is the bundled set of resources that are needed to actually make that skill run.
These resources can be things like simple markdown instructions, like plain English instructions that say, "Step one, do this. Step two, do that." A skill can also include things like code, like a Python script that will run and give you a deterministic output. And it can also include things like templates and icons and faults. So you can imagine that this folder in a skill is actually a collection of specialized resources that give Claude the ability to complete a specialized task. And that's one thing we can see over here.
If we actually just start exploring this "Skill Creator" skill, we can see in the "agents" folder, we have a bunch of MD files as instructions for things to complete. We have an "assets" folder with an HTML page. We have an "evaluations" folder with a Python script.
Now, as you can see, this skill actually has a bunch of different folders and files and scripts. So, if your Claude had ten different skills, you might think that's a lot of context to constantly bring into conversations. But the cool thing is that Claude doesn't bring that entire context into a conversation unless it needs it. The only thing that's always in the context is a short description of that skill. That lets Claude know how and when to use it. And then only when it needs it does it actually pull up the entire skill. So, imagine you have a filing cabinet and you're looking for Bart's file. All you need to do is brush through all those files at the very top. You find the name with Bart, and then when you pull it out, you can open it up and see all the different papers inside.
Skills are super easy to create or to upload. You can either go online and find different skills, and you can find online resources like this collection of different skills that you can just take and import into your Claude. Or if this is your first time really playing around with skills, just create one yourself. Just click the "Create with Claude" button, and then you're prompted to create a skill.
Now, a quick note is that since we're in the chat, we currently have a restricted version of skills that we can use in Claude. If you create a skill in that specific chat session, you can call it again, but as soon as you start a new conversation with Claude, that skill doesn't persist. The best way to use skills is either in Co-work or Code. As soon as you add a skill to your workspace, it's persistent across all different chat sessions, and it can be accessed by either of these products.
Another great way to create a skill is to just have a conversation with Claude. Get it to complete a series of steps for you. And then once you're happy with the output, you can just ask Claude to create a skill based on the steps that it took in your conversation. For example, let's say every single Monday, you want to upload your Excel sheet from your last week's financial sales. You want to get Claude to summarize the results, create a PowerPoint presentation, and then upload it to ClickUp. You can just complete those sets of actions in one Claude chat and then ask it to package it into a skill, and then you've got a reusable process at your fingertips.
Our next capability is a connector. Now, like we spoke about before, a connector is a way to give Claude access to a different tool that you're working with. So, let's dive into the connectors over here. And here we have the connectors view.
Now, under the hood, a connector is just access to an MCP, which means typically when you add a new connection to a different tool, you actually get a full suite of different actions that you can take in that tool. It's normally a mixture of read actions and write actions. So a read action is when you want to pull information out of the tool, and a write action is when you want to take information and add it into your tool. It's so awesome because you have a granular level of access. For example, over here I can do "always allow," "needs approval," or "blocked." So when you add a new tool into your workspace, you can actually control exactly what it can do.
Let's now check out the different connectors that we have available. As you can see, I've plugged in my Gmail and my Google Calendar. There's so many connectors available with all the most common tools that you want to use. But say, for example, you don't see one of the tools here. You can typically go to that tool's documentation and just try and find an MCP connection. But sometimes there might not be a native MCP connection. But over here, I found a GitHub repo that gives me Zendesk MCP access into my Claude. You could just follow these instructions and add them into your Claude, and then you'll be able to use Zendesk.
Authenticating is super easy. So let's give it a go. We just get a new window to authenticate our connection. Let's click "Authorize with Claude." And now we've just connected into our Superbase. Let's find our Superbase tool over here. And like we saw before, we have a bunch of read tools and write tools. It looks like we can do a lot of actions in Superbase. And we can go through each of these and one by one change the approval status.
And finally, we go across to our plugins. Now, as we just saw, a skill is just a predefined process that teaches Claude how to complete a certain task. And a connector gives Claude the ability to actually do something in a separate tool. So, if you join these two things together, you actually let Claude complete tasks in different tools. So that's why a plugin is just a collection of skills and connectors because it lets it operate in different domains and complete different tasks.
Let's actually explore one of these built-in connectors. This is an Anthropic connector that, if I open up, we can see the anatomy over here. We have a bunch of commands, which I'll go through in just a second, but we have a collection of connectors that are predefined for this specific plugin. And we can go ahead and install all these separate connectors. And we have a bunch of predefined skills. So for this example, this is a sales plugin. So, this is going to be a sales agent that I can run inside Claude. And each of these skills is a separate process that the sales agent can complete.
If we flick through this "Account Research" skill, we've got a bunch of rules and policies that we must adhere to, as well as an explanation of different processes, output formats, and tools that we can connect to. The slash commands are actually the most interesting because this is a way that you can specifically call on a skill from Claude. So, over here in our "Skills" plugin, we have this "/call summary."
Now, let's open up a brand new Co-work chat and let's paste in "/". So, if we just go "/" drop down to "Plugins," go across to our "Sales" plugin, and go across to our "Call Summary" skill, let's hit enter. And now we automatically get an initial questionnaire. So, this questionnaire is from the instructions that we set in the skill. In this case, it's a call summary skill where I can just upload the transcript of my call summary. I can hit enter, follow the questionnaire, and then this skill will actually transcribe this and follow my next sales steps.
Now, the most useful thing about Co-work is its ability to actually help you progress with tasks. So, the first thing that you want to do is actually define a folder to work within. And the second thing you can do, once you're actually inside a task and working on it, you can actually schedule follow-up tasks or reminders for you to actually keep moving things forward.
Now, these folders are actually folders on your actual computer. So, the best way to do this is to actually figure out what tasks you need to work on, what kind of projects you have, and create a separate folder for each project. Don't just create one overall folder like a "Co-work" folder and do all your different types of work in that one space because all the files are added into that one folder and things get very messy. The better way to do is actually structure it one project per folder, and then work from that folder each and every session.
So in this case, I just created a new folder called "Fitness." Let's "Always allow." Now, Co-work is geared to be as helpful as possible, and it typically wants to take things and turn them into tasks. It'll then take those tasks and pitch them up here into this one single session. And then as you work through it in the session, it'll basically check everything off.
So I've just shared some additional details for Claude based on my fitness goals. And as you can see, we've created a plan for this session. We've already checked off the first step of the plan. Claude is then going to be updating instructions into a "claude.md" file. Now, this "claude.md" file is very important because as soon as you close your computer and turn it off for the night, this session just ends. It doesn't actually persist. So the next time you open up your Co-work app, it's a brand new session. So all these instructional files that go into here, and as you have a longer and longer conversation, Code will actually add more files and resources into this space over here, into the context area, as well. And this is why it's important to actually have a separate folder for each main project or task that you're working on.
Now the best way to think about Co-work is that you're not going to complete your entire fitness goal, for example, in one session. If I want to lose ten kilos across three months, I'm going to be coming back into this goal multiple times across the next few weeks or months or whatever until I reach my fitness goal. So, that means every time you spool up a session, you want to have a very concise set of steps that you can do in that specific session. Maybe it's a two-week check-in. You check the progress of your weight loss. Maybe you have to readjust your workout plan so that for your next two weeks, you can actually start optimizing and doing better. The core project lives in the overarching folder structure on your computer. And then each session is a minimal check-in and basically doing smaller chunks of work. And this is what it means.
Looks like Claude's actually helped us set this initial stage of our fitness goal. We can scroll down to here. We got a bunch of different things we can look at. And if we open up our "Fitness Tracker," we actually have this Excel sheet in our folder saved on our computer. But we can look through here. We have all the information that we need to do. We have different sheets and different tabs that we can actually go through, log our results. So then if we come back into here two weeks later after filling this out every single day, we can get Claude to review it and help us optimize the next two weeks.
Another really awesome feature that you should look into is actually using this scheduled feature. So I can actually keep having a conversation with Claude and say, "Hey, I'm going to need reminders to actually fill this stuff out." So every day at 9:00 AM, can you please send me a notification to my desktop app to remind me to actually fill out this information?
Now, you can also go into here and just create a brand new schedule for whatever work you want to do. It can be totally unrelated to an existing project. It could just be something that you want done every single day. For example, a really good default one is to do a daily briefing where you give your agent access to your Gmail inbox and you get all your unread email summarized and anything urgent highlighted. For this, just make sure that you have your Gmail or your Outlook connected up because you need that tool access. You can also set a specific frequency, which I like for the email one to actually just do weekdays instead of every single day. And then for the model, I would just choose the Haiku 4.5. And then create a custom folder for this reminder.
The final product for us to look at is Claude Code. When it comes to Claude Code, just like Claude Co-work, we actually have access to the files on our computer. We can create a separate folder and have Claude Code work directly from that folder. And Claude Code also by default has access to our terminal. So a really cool thing is that you can create software on Claude Code and then have it be installed directly onto your computer like little apps and widgets. Those apps and widgets, unlike the artifacts from the Claude Chat, actually live on your computer. They're persistent on your computer, and you can build really cool tools for yourself.
We also have the same set of skills and plugins, just like before. Actually, online you can find a lot more skills and plugins for Claude Code than you can for just a regular Co-work. Then we have these two custom capabilities here as well. Remote control came out very recently. This gives you the ability to actually control a Claude Code session from your mobile phone. You can just download the Claude app on your phone and then access the session that's on your computer. So you can be out going for a walk and your computer can be at home coding a project for you, and you can quickly check on a status and help unblock any issues.
And then finally, we have hooks. So a hook is a deterministic action. This is a hard-coded script that executes at certain stages of the coding life cycle. So some of the pre-built stages are: as soon as you start a new session, execute a specific task. As soon as you, or before you use a tool, please send me a notification. After you use a tool, please give me the update. As soon as the coding session stops, please notify me.
One nuance when it comes to hooks is that Claude Code must actually follow the instructions of a hook. But when it comes to some of these other things like skills, for example, it's still an LLM that's reading the skill, and LLMs by nature are probabilistic. 99 times out of 100, it's going to follow the skill to a T. And when it comes to hooks, it's going to follow this 100 out of 100 times.
Now, when it comes to Claude Code, the best way to get started is just to give a very simple instruction. And actually, I found the best performance by using something like Sonnet or Opus 4.6. I typically like to use Haiku 4.5 when I'm doing a lot of work with Claude, mainly because it conserves my tokens and I don't hit my limit as quickly. But when it comes to coding tasks, yeah, I have found that Sonnet 4.6 is really good, and then Opus 4.6 if you need something better. The cool thing about Claude Code is that you can actually change the model during the session, unlike the other like Chat or Co-work. If you want to change the model, you have to start a new session.
Now we also have access to all the different skills, connectors, and plugins, just like before. So, if you go online and you find anything super useful that you want your Claude Code to have access to, be it a tool, be it a specialized skill, like being able to design more beautiful websites or being able to work with different languages and frameworks. All you need to do is install it into your Claude instance, and then you can access it in your coding session. And then finally, make sure that you use a custom folder for this app that you're creating. Then hit enter on your request. Say that you trust this workspace if this is the first time that you're using this folder. And then your code will go away and do a bunch of activity for you.
Now, by default, your permission setting is going to cause Claude to actually always come to you and ask if you can run certain commands, if it can do things like web research or read your files, whatever it needs to do. One of the best ways when you're first getting started is to actually stay close to the system and understand exactly what it's doing, what it's reading. But over time, it is a bit more risky. You can use the "bypass permission" setting. And by enabling this, your Claude will literally run any command that it wants to. So, this is a lot riskier because it could delete files that you don't want deleted. But it's also a lot faster and simpler to actually code in a session. We don't have to be here all the time.
Now, if your Claude is creating an app for you that has a visual interface, you can press the "Preview" button, and then you'll have a demo browser that you can actually view the app that it's building. Now, after this coding session is finished, and let's say the app is ready to run, you can actually get it to install it on your computer, and it can read and understand what kind of computer you're using as well. So, if you're using a Mac app, you can actually create an app for you that works on your Mac. You can make it install into your dock bar, into your top navbar, pretty much anything that you want to do.
Now, there are some different places that you can use Claude. For example, if you use it on your mobile phone, it can access files and resources on your phone. So, you can actually go away and access things like your health data and then analyze it or build different apps from that data to help you get more healthy. You can use Claude Code inside different places as well. So, right now we're using it on the desktop app. That's not the best experience. A lot of people use it in a separate coding tool or directly in a terminal. For example, this is Cursor, a popular AI coding tool. You can still use the visual interface for Claude Code. The benefit here is that you get better control over the files and your app that you're building. And then the more you use Claude Code, you actually start defaulting to using it in a terminal. The terminal is a little bit scarier to use because it's a lot less visual, but it's the exact same thing. You just have a plain text conversation, and then your Claude can go away and build out your app.
And now, as of more recently, there's a couple of different places that you can use Claude. For example, if you just download the Chrome extension, and then this version of Claude can actually access the browser and automate tasks. For example, I'm just asking to find the Anthropic newsroom. Let's hit enter. And as we can see, Claude is navigating across to the Anthropic newsroom.
Another really interesting feature is this little button over here. It says "Teach Claude." If you go into "Teach Claude," you can actually run through a process on the browser, and Claude will learn your process and then commit it to memory. Kind of like creating a custom skill but directly on a browser.
Now, as of recently, one of my favorite ways to use Claude is actually in Microsoft Excel. You can upload really complicated data sets, and Claude is able to actually review everything. And if you use Claude Opus 4.6, and I'm pretty sure also Sonnet 4.6, inside Excel, it's got access to a range of new spreadsheet editing tools. Things like working with pivot tables, chart editing, conditional formatting, and it can even use equations directly in Excel. So, it doesn't actually hallucinate and work with your data. It can create new tabs and new pivot tables and modify your data by using actual expressions and equations instead of just by making up numbers by itself. And this version of Claude also knows how to work with very common financial forecasting.
All right, thank you very much for watching this video. If you want to learn more, make sure you subscribe to my channel because I'm going to be releasing deeper dives into all the different products and capabilities.