Transcription
Claude Cowork is a new feature from Claude that appeared at the beginning of this year and instantly went viral on social media, captivating many people. Essentially, Claude Cowork is your AI assistant that will help you complete all routine tasks on your computer. All you have to do is monitor its work and make adjustments if necessary. In this video, I will show you how to get this Claude Cowork. I will demonstrate several really cool application scenarios for this technology, and also briefly touch upon its security. I think this video will definitely be useful and informative. Let's not waste time, let's go.
Let me briefly explain again what Claude Cowork is. Essentially, Claude Cowork is the same Claude, but with executor functions. This means you give it, for example, access to a folder on your desktop, assign it a task, and it can do anything with that folder, from deleting files to creating text documents, presentations, and various other files. And if you also connect the Chrome extension, you can do anything, even on the internet. Essentially, it's like hiring an intern. You give them a task, go have coffee, come back, and the task is already done.
So, what's the deal? Come on, jump in. To get Claude Cowork, you need to download the desktop application for Claude. To download Claude Cowork, you need to type "Claude," go to the very first link, and click download for either MacOS or Windows. By the way, until recently, before February 2026, Cowork was only available on MacOS, but now it's also available on Windows.
Let me show you with a simple example how it works. Here we select the folder we want to work with. Let's choose Downloads. Click "Allow" to let it work with this folder. And let's enter the following request. Create a folder on my desktop, name it "Photos from Downloads," and put all the photos from the Downloads folder into it.
So, we have entered the request. Let's see if it can do it. Click "Let's go." Let's wait a bit. Sometimes Cowork works quite slowly, but overall, if you want to give it a task and calmly go have coffee, there's no problem with that.
So, since I didn't give it access to the desktop, but only to the Downloads folder, it suggests creating the folder within Downloads. In general, that suits me. Let's click "Create in Downloads."
So, as you can see in the right panel, look here, there's progress. It's creating a plan to complete this task and crossing off completed items. It has already counted the photos in the Downloads folder, and is now creating the folder and moving the photos. Wow. I thought I had cleaned up my downloads, but a lot has accumulated again. And it seems like everything is ready. Now we'll check if it all worked out.
The "Photos from Downloads" folder has been created directly in your Downloads folder, and 4,611 photos have been moved into it. The folder can be found in Downloads under the name "Photos from Downloads." Let's check if it's really there.
So, I've opened the Downloads folder. And as you can see, there are indeed 4,611 photos here. Let's open it. And yes, there's a lot of junk here that has accumulated in my downloads, and that I would never have cleaned up myself. In short, excellent. This is the simplest example of its use. Now let's move on to more complex cases.
Next case. I want it to create a PowerPoint presentation for me, but not just any abstract one, but based on my technical specification, which is on my desktop. There will be two files there. The first file is the technical specification. It will contain all the text, slide by slide. You see, here are two files for the presentation's technical specification. Let's open and look, and here it is, all laid out slide by slide. And the second file. Here are just three references and a note about why I liked them. I didn't bother much with the references, so it turned out as it turned out.
Let's try it and see if it can create a decent presentation. Select our "Presentation" folder on the desktop and click "Open." Grant access and enter the following prompt here. Please study the files in the presentation folder. There are two files there. The first is the technical specification, and the second contains links to references. I ask you to study these references, understand why I like them. I've noted why I like them, and create a PowerPoint presentation for me according to my technical specification and according to the references. Click "Let's go." Oh, cool name for the button.
So, as you can see, it has created a plan for completing this task. It is studying references from Dribble. So, you see, it's asking me if it can allow Claude to use the browser on dribble.com. Click, allow. Let's see what it's doing there. By the way, these are my hands here. It's doing this itself. I found three references on Dribble. I didn't delve too deeply into what those references were, so let it study and do what I asked.
As we can see from the chat, it is already studying our references. Let's give it some time and see what the result will be. By the way, notice that it has loaded the PowerPoint skill on its own, without my participation. Essentially, Claude Cowork can find the necessary skill itself to perform a given task. This is also one of the advantages of this feature. By the way, in general, if I hadn't written down which specific references to look at, I could have just given it a prompt to go to Dribble itself, find presentations with a suitable style, and use them as references. In general, you can work like that too, but to use it in Google, you need to connect the extension. I didn't show you, by the way, but I'll show you after this case. Without it, unfortunately, Claude cannot work adequately in the browser.
So, Claude Cowork has finished its work. While it was doing this, I managed to have lunch. So, it made an orange accent, a warm, stylish off-white. Dark slides, rounded cards with shadows, block structure. Well, according to the description, everything should be good. Let's see what the result is. We open it with Knee, because I don't have PowerPoint on Mac, and let's look at the result, how artificial intelligence transforms recruiting.
Listen, it turned out quite well, by the way. Quite good blocks. As far as I'm concerned, this is not the worst presentation a neural network could create. I would diversify it, perhaps with some images, but again, that would have needed to be specified in the prompt. In short, in this simple way, using Claude Cowork, you can create presentations.
Let's move on to the next case. In the next case, I want to show you how Claude Cowork interacts with the browser. First, let me show you how to install the Claude Chrome extension. To find it, just type "Claude extension" in the browser bar and open it. It will say that this product is unavailable, but we don't believe it. We know it's available. Find Claude here, press Enter. And in the first line, we see an extension called Claude with a rating of 2.7. Don't be scared by this rating, this is it. Why such a low rating? There are certain peculiarities with the Russian language, which I will explain a bit later, but still, thanks to this extension, Claude can work excellently in the browser.
As soon as you install it, it will appear here in your extensions. Or you'll need to click on the puzzle piece and click on this button to make it appear here. Next, no matter what browser page we are on, we can click the button here, and Claude will appear right in the browser.
So, what case do I want to show you? I would like to demonstrate how Claude Cowork will interact with the browser for image generation. I will have two files on my desktop. The first file is a CSV table. It will list the themes and descriptions of the images. There will be five or six of them. And the second file will contain a script for creating images in a unified style. That is, I will specify that it will be, for example, a drawn style, a resolution of 1920x1080, and so on.
Accordingly, what do I want to do? I want Claude Cowork to analyze my script, analyze the CSV table, understand what images it needs to generate. Then, I want it to go to a website called LM Arena. By the way, LMA is a service with which you can generate images absolutely for free. I want it to generate images using LM Arena, save them to my desktop, and then create a folder on my desktop where it would put all the images, and name them according to the image theme. I hope you understood me. Let's get started.
Don't forget to select "Desktop" here so it has the ability to work with files on your desktop. And after that, click.
So, as you can see, in the right panel, it has created a plan for completing this task. Let's see what files are presented. Here's a small CSV table with a theme and description. And also a script file where the image format, image style, color palette, and so on are specified.
So, as you can see, it's asking for our confirmation that we allow it to work on the LM Arena website. We naturally allow it. So, let's go to Chrome and see how it works there. Right now, by the way, these are my hands here. Around LM Arena, there's this orange glow. This means Claude is working with it. It has composed the prompt itself. As you can see, it's trying to adhere to our script. Here's the 1920x1080 resolution that I specified in the script. And everything else is written in English, I don't want to read it. The only thing that could go wrong is that GPT Image 1.5 will generate images poorly. But I decided not to use Nanobanu for this video, because it would be too long. "Let me try clicking the image to save it directly. Let's go, man. Let's go."
By the way, as you can see, in the right window, or rather in the right panel, we have context, and here it says the connectors that are connected to this task. By the way, connectors allow you to connect Cowork with specific services. This can be a browser, or, for example, directly Google Mail, Google Calendar, and so on. It has quite a few connectors. I'll even show you where. Here, if you click the plus sign, you can select connectors. And here there is a whole library of connectors. If you click "Manage Connectors." And here you can connect to GitHub, Google Calendar, Gmail, Google Drive, and also click "Browse Connectors," and here there will be a huge, countless number of various integrations.
So, not much time has passed, by the way. I thought it would take longer. As a result, it generated all the images, but the problem is that it downloaded them to the Downloads folder for some reason. You see, it worked with all the images here and wrote the following: "All images have been successfully generated and downloaded in the browser. Now I need access to the Downloads folder on your computer to move the files to the Photos folder on your desktop." Let's do that. Click "Choose folder" and select Downloads here. And our Cowork has finished its work. Let's now see if everything really worked out.
As you can see, it named the images exactly as written in the CSV table. Let's open the folder and see. And yes, indeed, all five images have been generated and named as required. And by the way, even though ChatGPT messed up the 1920x1080 resolution, everything looks really in a unified warm drawn style. The only problem Claude had was with saving the images.
If, by chance, during the execution of some tasks, you encounter difficulties with Claude, you can combine the work of Claude Cowork with the Chrome extension. Now I'll show you what I mean. Let's open Chrome and open LM Arena again. For example, you gave Claude Cowork the task of saving images, and for some reason, it can't do it. We can delegate this task to the Chrome extension. But why on earth should the Chrome extension know how to save images correctly? Good question, and there's a solution. We will use shortcuts. Shortcuts are a Claude function in the Chrome extension so that it can learn from you how to perform certain actions.
Let me show you how it works in practice. Here in the right panel, this button is called "Teach Claude." As soon as we click it, it starts remembering all our movements, everything we do on the site. And then, when performing a similar task, it won't be slow like before. Let's, for example, show it the full process of creating and saving an image. Click this button, and it suggests clicking the "Start Recording" button. What is this all about? Claude conceived this function as follows. We perform certain actions on the site, Claude records them, and simultaneously as we perform these actions, we also comment on them. I don't know about others, but I encountered a problem: when I started dictating to Claude what I needed to do, it recorded my words, first, not in Russian, but in English. And second, it didn't just translate my words into English, but simply took similar-sounding English words and put them next to each other. In short, descriptions in Russian don't work for him, so if you know English well, it's better to explain to him in English. But I think this is just a beta version, and in the future, the developers will simply fix this bug.
Let me click "Start Recording." I click this button so that my voice is not recorded, because it was useless. And look, I click anywhere on the page, and it highlights it with a blue circle. You see, in the right panel. Let's write here: "Generate a corgi." Press the arrow. Okay, GPT is selected. Everything is okay. Okay, my limit is reached. Five images. Let's choose another model. So, as we can see, the image has been generated. Now click this button. Then click here, "Save image," and save the image. It has recorded all of this. Then we click the "Done" button and finish the Shortcut. All done.
As you can see, it has created a description of our prompt, so that in the future we don't have to write this whole huge block of tasks for him. We name it whatever we want. Let's change "corgi" here and that's it. Then click "Create Shortcut." Then, to use it, we need to type "/" here, select "Shortcuts," and choose the one we need. Here it is. "Generate image corgi." And after we type the shortcut here, we will need to write the following prompt. That is, something like: "Generate an image according to my shortcut and save it to my desktop, for example." If you hear noises, it's my cat eating dry food.
And that's it. Thanks to the fact that you have taught Claude how to save images correctly, it should no longer be slow with this in the future. But again, there can be any glitches, it's still a beta version, so we are waiting for the most high-quality and robust version of Claude to use this imba. Let's move on.
And one more thing I would like to show you is plugins. Plugins are only available in Cowork. They are not in the regular Claude chat. Essentially, you install a plugin, and Claude starts to understand a specific topic. Inside each plugin, there are already instructions, connections to services, commands, and so on. This means you don't have to configure anything for a long time, just install the plugin and everything works.
Let me show you where to find them. Here, open the left panel and select the "Plugins" tab. The tab opens, and here are 3x3=9 + 1=10 plugins. These are plugins that are already built into Cowork, and we can use any of them. For example, let's try this plugin called "data." Click "Add Plugin." We clicked "Add Plugin," and a banner appeared at the top saying "Data is installed and ready to use," which means it's ready for use.
Let me show you how to work with the plugin in practice. I have this CSV table. Let's open it now. Here's a regular data table. A neural network compiled this for me. Here, you see, Apple technology, category, region, and various numerical data. We click "/". We select "Plugins" here, we select "Data" here. This is our plugin. And here there are a bunch of commands with descriptions and skills. Let's use the "build dashboard" command and write something like this. "Create a dashboard from the file sales data 2025. Revenue by month, top products, comparison of regions." And let's write "Let's go."
"Let's go." In general, Claude Cowork could process this CSV table without this plugin, but honestly, with it, it will do it much better. As you can see, it says: "Dashboard ready." Here's what's inside. KPI cards, revenue by month, top products, well, a lot of stuff. Let's open it. And we got this interesting dashboard. Dashboard. Let's open it in Yandex. Here, as you can see, there's a lot: pie charts, other charts, various graphs, and so on. Here, by the way, you can even select by category, look at Kazan-Moscow too, which, in my opinion, is cool. And specifically by technology. This is a very convenient thing.
Instead of writing a long text to Claude Cowork, you can just enter a command from a plugin, and it will understand everything and do it as required. I want to mention a few points that are quite important if you plan to work with Claude Cowork. For complex tasks, Cowork consumes significantly more resources than Claude Code or just Claude. All file reading, website navigation, working with images, files, and so on. All of this consumes a lot of tokens. Therefore, even if you have a Max subscription, you can quickly run out of limits. Accordingly, I would not delegate all your work processes to Claude Cowork.
I would also like to briefly discuss security, because, essentially, the neural network has access to all your pages, to all your files, and, accordingly, it can make mistakes and do something with them. There are five simple rules on how to protect yourself.
First, again, do not give Claude access to all the resources you have. Accordingly, we specify a specific folder, allow it to work only in a specific folder or on a specific website. Carefully observe what it wants to do. If it offers you to click the "Allow" button, then we definitely look at what we are allowing it to do. Just in case, you can write in the prompt that it should not delete anything, even if for some reason it wants to do so. And if it plans to delete something, let it ask for our confirmation. Although, as a rule, it always asks for permission before deleting, but it doesn't hurt to be extra cautious, so to speak.
Third. If, by chance, you want to work with some important files directly in these files, then I recommend copying them somewhere else, because who knows what might happen, what might come into Claude's head, it might just take and corrupt your file.
Fourth. Do not give Claude access to your financial resources and other important accounts, again for security reasons.
And fifth. Always monitor what Claude is doing. If at some point it starts acting weird and doing something incomprehensible, it's better to stop the process and restart it. Always read what it's doing, track the process it outlines in the right panel, and never trust it blindly like that.
Phew. This is such an interesting tool from Claude, which allows non-developers to do a lot of interesting things. Again, I might have missed something or not fully disclosed the topic, so if you have anything to add, be sure to write about it in the comments. Thank you all for watching. I hope this video was useful for you. Goodbye.