Transcription
Guys, in this video, I suggest we look at a tool that Claude recently released. It's a Chrome extension that allows you to automate any tasks, any of your actions in the Chrome browser, and use, in fact, Claude's agent for this, which can help you simplify your actions, simplify your tasks, and take on all the routine daily actions that you sometimes feel too lazy to do, or take on more complex, yes, tasks, not just routine ones. Let's break this down and look at it today. So, first of all, what kind of extension is this for Claude? So, an extension is added to our browser, yes, an extension, which opens in a browser window and which allows interaction with the browser's content, yes, and solving some tasks here, writing some prompts to control your browser. Yes, I've said this word a million times. Let's figure it out step by step, where we can read information about the extension and how to install it. So, we go to the official website clod.com/hrome. And here we can see the description that Anтроopic himself has prepared for us. Here we can read about use cases. One of the tasks they give here as an example is, for example, to go through your calendar, go through the meetings that are in this calendar, check the availability of rooms, uh, for rooms, I understand they mean some online room, like Google Meets or something like that, and, like, book this room. Next, the next task they offer us here is, like, background task. I'm looking for a three-bedroom house, such-and-such a price, blah-blah-blah. Can you search and show me the top five options? In general, here they suggest we give a task to search for information on the internet for, like, booking houses and something like that. And the next task they offer us for consideration, schedule workflow, that is, a schedule that is executed on a planned schedule. Go through my sent items. And here they give a task, yes, as an example. Go through my sent emails from the past month. Find messages that haven't received a reply but should have been replied to. And write a follow-up message, like, write an additional message. These are the kinds of tasks they offer us, well, these are the kinds of cases, rather, that Anтропик offers us today. Well, some of these might be applicable to you, like, I don't know, booking and searching for houses, booking and searching for plane tickets. This is just a classic that all neural networks give us, yes, when they describe their tasks for agents. This is not a super frequent occurrence, so it's unlikely to be very useful to you. But the story with scheduling or the story with daily tasks, this is already a more applicable topic that can help you in your daily routine. And today I will show you three cases that seem most interesting to me, most optimal, which differ slightly from what Anтропик himself offers us here. And they also offer several more options for daily tasks below, which you can assign to Claude, or rather, to Claude's extension for Chrome. These are, for example, taking metrics and analyzing dashboards. This is, by the way, a cool story when we can go into any services and get some information from there. This is a quite applicable topic and, I think, has, uh, a place to be popular in terms of usage. Organizing files in Google Drive, preparing presentations, and planning your calendar. Then comparing products on websites. This is also a cool thing when you can analyze some data on different websites automatically. This is also, I think, an interesting case to consider. Logging sales calls, uh, in general, performing some actions in CRM, cleaning promotional emails, that is, cleaning up emails in your inbox when it's very cluttered. Each of these cases can be applied to yourself, slightly modified for your own tasks. Also an interesting thing.
So, now let's look at how to install it. Installation is as simple as possible. So, on this same site, you click "Add to Chrome," you are redirected to the Chrome Web Store, and here you simply install the extension. After you have installed it, it appears here in your extensions. If it hasn't appeared, then click on this extension button and check the box so that it is pinned to, uh, the screen, yes, so that it is not in the list of extensions, but so that you see it here immediately. In principle, that's all. After that, you simply click, and this extension window opens. And here at the bottom in the input field, you can already write your commands that you want to be executed on the site that is open, on the tab that is open. Well, and not necessarily open.
Now let's look at the use cases that seem most relevant to me and that deserve such interesting attention from our side and that might be useful to you in your work. I suggest we consider three cases today. These are YouTube video analysis, Telegram channel and chat analysis, and the skill of working in any program. So, I have arranged them in order of their application interest. YouTube video analysis is a cool thing, but it's not something super new, yes, so I gave it one star. Telegram channel and chat analysis is a more interesting, cool thing. So, this case gives us the opportunity to look into private channels, yes, your own, yes, which you are a member of, and extract some data, some information from there. This is already a more interesting topic that is not so super accessible to existing agents. Here I have already given this case two stars. And what deserves truly super special attention and interest. And I think this is a super powerful cool thing that will really blow up the internet, yes, and start to develop further. It is currently in a nascent stage, as it is a beta version, but I am sure it will grow into something super cool. This is the ability to train an agent to work in any program, yes, within the browser. That is, within CRM, within any constructor, within, like, any of your apps that you use, and so on. Here I have given it a full three stars out of three, yes, possible. My evaluation criterion is three out of three. And let's see how this whole thing works. I will show you how to train an agent to work. And let's look at an example. Let's move on to the first one. This is YouTube video analysis. What do we have here? First of all, how does this whole thing look visually? Look, when you open a tab in the browser, for example, Yandex Market, we can open it, it opens here in the standard format. If you, like any other tab, click on the extension, then, you see, this tab is marked with an orange stripe. And next to it, we start to see the inscription Claude. This means that this page now has access to our Claude, yes, our agent, our extension. This inscription in the browser also symbolizes this. The tool has launched debugging for this browser, so that you understand that the tab is now available for the agent to work with. And also, here below the address bar, below the URL, you see tab groups. Here I have a YouTube video tab open. We can also quickly click on it. And a window opens, which I already have open. Here, you see, under this tab, two windows are open. If we click on the red, yes, name at the top in the tabs, which the agent has come up with for this tab, then they collapse for convenience. The same thing you can do, yes, within one, let's call it a micro-workspace, you can open many, many tabs, and, accordingly, the agent will have access to each of these tabs, and in each of them he can perform certain actions, yes, these group actions that you have programmed into him. Therefore, it is quite convenient. You can also delete these groups simply by closing the tabs. Well, and in principle, by right-clicking, you can change the name here, you can change the colors here. That is, manage it as you see fit. You can ungroup them. Accordingly, in this case, I am removing, yes, the agent's control over this tab. Well, let's move on to YouTube videos, let's look at the use case. I open a tab. And for what case do I suggest using the agent? In this case, I want to analyze videos from channels that I will specify to our agent so that he watches them, extracts key ideas from them, and writes these ideas down for me. Accordingly, I write him a task: "Analyze the latest videos from YouTube channels." I give him, as links, the link to the Antropic channel, the Open AI channel. And here you can give him a whole list, yes, of channels that you want to follow. For video analysis, I ask him to use transcription. I'll show you what that is now. That is, he cannot watch the video itself, but he can look at the text of this video and analyze it based on that. And for each video, give me a conclusion, the publication date, three top ideas from the video, and send me this video. And, accordingly, I launch this task. And that's it, he's off to complete it. Or rather, he first writes us a plan according to which he will act. And then we approve the plan. And he proceeds to the execution process. So, in this way, you can analyze, since there is a lot of content, a lot of videos on YouTube. You can analyze the channels you want. You can watch, like, multi-hour videos, make summaries of them with the help of Claude's agent, using this extension, using a similar prompt that I just wrote. We now see, yes, that he is opening the video, he is starting to scroll through it. That is, in principle, we can see the history of the actions he is starting to perform here. And he will describe the entire history here on the right in the menu. I'll warn you right away that a task from Claude will take, its solution will take quite a long time. Therefore, in principle, it is not necessary to wait for all of this. You can give him a task, go to another tab, and continue to perform any actions you want there. This, by the way, is one of the advantages of the extension. That is, we can, we will now move on to the next task, I will give it to Claude, and he will also perform it in the background, and you don't have to be on this tab. Only if you are interested and want to observe what he is doing here, I do not recommend clicking anything, pressing anything, scrolling, etc., here at the moment when he is working on this tab. That is, you set the task and leave here so as not to interfere with his work.
Next, let's set him the next task in parallel. Telegram channel and chat analysis. And let's also put the task in the background for him to perform. Here I also open pre-prepared tabs. I have the web version of Telegram open. Accordingly, by opening the web version of Telegram, you give Claude access to all chats, channels, open, closed, in general, all those that you are a member of. And you can extract any information from here. This is already a cooler story. What do I want to do here? Or rather, I have already done it. I will just show you now, and I will not launch this task, yes, in the background, so as not to consume resources. I will just show you the possibilities that can be implemented here. What was my task? I wanted to take a specific channel. In this case, I chose Mikhail Dokovinin and asked Claude to take the latest posts, or rather, the posts from December, take all the reactions, and write the key ideas for each of these posts, and then put all of this into a Google Sheet. And here's what he did for me in the end. That is, he collected the posts, wrote the date when the post was published, wrote the topic of the post, wrote the main idea of the post, collected the number of reactions, collected the number of views. That is, you see, the agent has access to all information, yes, to internal information, to reactions, to views, and to the ability to read these posts. And moreover, he can transfer all this information somewhere else. That is, he can act not only within the current, yes, window, he can, for example, save the information to your Google Sheets, which you can then use for your own purposes. This is already a more interesting, cool, great case for application. So, what prompt did I write for him to complete this task? I used the following prompt here. Analyze posts from the Telegram channel of such-and-such and I specified the link to the channel for December. And write conclusions for each post in Google Sheets and inserted a link to my Google Sheet. Then I described what columns I have in Google Sheets. Date, that is, the date of the post publication, Topic. Determine the topic. Post 1 word. Main idea. Write the key ideas of the post, Number of reactions. Calculate the total number and number of views. Number of post views. And I launched this task. It also starts to be executed. That is, we see, it starts to be executed, yes, in parallel. If we go back to the previous tab, where we analyzed YouTube, then he continues to perform this task in parallel in the background. I will not wait, yes, within the video, for the Telegram channel task to be completed. I will stop it, as I have already done it in the background. And, accordingly, as a result, he made me a summary table in which he wrote everything down. This is already a pretty cool mechanism that you can use to analyze all your Telegram channels, chats, which, I am sure, many of you, naturally, do not visit daily. And keeping track of content and always being aware of events is quite difficult now, as there is a huge amount of content, yes, accumulating. And this is one of the ways, in principle, to be able to track this content and extract some key ideas from it. In channels, again, I repeat, in chats, in all your private communities that you are a member of, in any of them, if you open the web version of Telegram, then Claude's agent will be able to access it and extract the summary data you need from there. This is an awesome story. What else is cool here? You can, this story, that is, you can launch this task on a permanent basis, that is, on a specific schedule. How can this be done? Look, let's go back to the prompt input field. Here we type slash, select shortcuts. And, for example, we can click the plus sign to create a shortcut, yes, a quick command. And here you come up with a name. For example, "Check Telegram." Here you set the prompt, the same one that I showed you above. You can put it here. Then "start from," that is, the page in Telegram from which everything should start. And here there is a "schedule" slider. And here you set the schedule according to which this task should be launched. "Once," yes, once, "daily," that is, daily, for example, daily. Then you choose the time and choose the Claude model that should work. You can leave Opus 4.5 here. What's here? Weekly, monthly, annually, that is, daily, weekly, monthly, and annually. Click, select the schedule you need, click "Create Shortcut," and your task is created. If, accordingly, you type slash again, then here in shortcuts, it appears, "Check Telegram," and you can return to it and edit it a little. That is, you can also, for example, it will be executed tomorrow, you will check if there are any flaws or not. You can edit the prompt, or you can change the schedule, or you can delete it altogether. This is also a cool thing when you can collect analytics on a daily basis, not only, again, I repeat, from your Telegram channels. You can analyze, for example, some websites, competitor websites, how they change prices, what they add to their websites, like offers, promotions, I don't know, other things. Put all this into daily execution. This is also a cool business task that can be solved with the help of its agent. Also an awesome thing.
Let's go back to our first task on YouTube. You see, it's still running. Again, I repeat that it can take quite a long time to complete. This is a minus for now. I don't know if Claude will fix this in the future or not with subsequent versions of this project, but we see that it is reading the transcription of this video. And here, by the way, yes, I didn't tell you where you can see the transcription text. If you scroll down the description, here there is "Show video text." If you click this button, then on the right, here, the text of any YouTube video will open. And, accordingly, you can analyze based on this text. That is, for a neural network, it is not necessary to watch the video itself, naturally, to extract some key ideas from it. This is how, as one of the outputs, it can simply analyze based on the text. So, here I will not wait either. That is, I will stop the task so as not to consume resources. Ultimately, after some time, when Claude finishes working, you will see a summary here of the information you requested. That is, you will see the date, you will see the ideas for each video, and you will see, well, what else you will see. Ah, nothing. The date and three top ideas for each of the videos. Plus, it might write some of its own recommendations additionally. Therefore, in this way, using this task, you can now consume a huge amount of content. This could have been done before with the help of other neural networks, but now it is also available in Claude and within its browser extension.
And now let's get to the main thing, to the most awesome tool that, I think, has appeared. It is currently in its initial stage. It doesn't handle things perfectly yet, but I think when Claude perfects this, it will be a real game-changer. This is the skill of working in any program. How does this whole thing work and what is it and why did I give this thing three stars? Look, I examined it using the platform called JTB. This is a chatbot constructor, that is, a no-code platform for creating chatbots. I go to the platform, I get to my personal account, I open a bot, where, accordingly, this bot is created, by creating blocks, yes, like in any code editor. And my task is to train an agent to create a bot within the platform on its own, so that I only write him a prompt, and he does everything. The platform, I note, is currently running in English, because Claude's training is currently only in English. But I think they will also improve this in the future. After that, I launch the trained agent, and he starts to work some magic. That is, in this format, actions begin to occur on the platform. That is, I am currently not touching anything. You see, yes, that the agent is working on the right, there are some thought processes happening, and on the platform itself, some actions are also happening. That is, he adds blocks, configures these blocks, establishes connections between them, deletes these connections, adds text, buttons, connects blocks to each other, and so on. That is, the agent is building the bot that I asked him to build. And he does this based on the pre-training that I conducted earlier. It looks quite chaotic. That is, he scatters blocks, you see, around the platform. Well, they overlap each other on the workspace, but nevertheless, the process is happening, and it looks cool, awesome, and great. Now, how to train an agent to do this kind of thing? We do the same, yes, we open our extension here. And then we have the option, we type slashcut and "record workflow," that is, record a workflow, a working flow. And here we proceed to training. That is, after we click "Start Recording," the microphone will start, and you will speak directly to Claude and tell him how he should act. That is, move the block here. To create a new block, click here, open its settings. In the settings, click this button, then move it here, right-click, add the next block, and so on, and so on. And you tell him all this and speak it, that is, you teach him naturally, in simple human language. You tell him how the platform works, on which you want to train him. At this moment, he makes his notes. Here on the screen, he takes screenshots, transcribes your text, which you give him. He records all this in his memory. And then, after you have trained him, you will also have a shortcut, as we did when I showed you about scheduling. And here, for example, this shortcut will be called "Create Bot." Here it is, yes, I already have it. I recorded it. And, accordingly, here the system automatically prepares the prompt, also sets the starting page. Here the prompt is, naturally, larger with a description of the work. And then you can refer to it and say: "Create a chatbot." Then, blah-blah-blah-blah-blah, with such and such parameters. And he starts creating. So, how to train, how to speak the text. I cannot do it within this video now, because otherwise my microphone will be transmitted, yes, to, well, to the training, and most likely the recording will be interrupted. Therefore, all this process, so that my current video recording is not disrupted, I have left it all in my Telegram channel. So, the entire process of training an agent to work with a third-party application in the browser, I have left in my Telegram channel. To get there, scan the QR code you see on the screen, or it will also be there. The link to it is in the video description. As usual, there will be an "Open" button in the pinned post of this channel. Don't forget to subscribe to my Telegram channel. Three topics that I like, that I am interested in. These are no-code, low-code, and AI coding, development, that is, using AI. My Telegram channel is generally about all of this, and this YouTube video is also about all of this. So, come over there, subscribe, there is also additional interesting content. And in the pinned post, I repeat, there is an "open" button. Click on it, you will get to the bot, and in the bot, you will open a mini-app. You can open it. I have a separate mini-app that I created, in which I create various tutorials on these three topics: no-code, low-code, and AI coding. Some content is available for free, some content is available for a fee. So, in no-code, everything is free. In low-code, part is paid, part is free. In AI coding, it's the same. There is a monthly subscription, at the moment it costs very little. So, depending on when you are watching this video, you can get in for a couple of thousand rubles a month and watch content for super cheap money, which, I think, is not available on the market now. for the content that is provided there. And in the AI coding section, there is a tutorial called "How to train an agent to work in any program." So, here, in this video, in this tutorial, I have left the training on how to train an agent to work in Jetbot, that is, I recorded the entire process there. Therefore, click on this button, go there and watch the videos. I repeat, all of this is currently available for free, so come and watch. That is, in principle, you can train it to work, yes, in any program. Jetbot is just an example that I told you about, but you can use it in any CRM, I don't know, in any email client, or in any other application where you work, train your Claude code, your agent, and let it perform tasks there. Of course, there is a caveat, that is, it currently works far from perfectly, far from perfectly, but I think, since it is a beta version now, Claude clearly plans to develop all of this, and something tells me that it will definitely develop into something interesting. So, test it, jump in at the early stages and see what's possible.
Well, and let's conclude the video by summarizing all the features that I mentioned today, let's go over them concisely, compactly, so that we can see the pros and cons of the Chrome extension. So, let's talk about the pros. You can run tasks in parallel. This is a really cool thing. Since tasks currently take a long time to complete, so that you don't have to sit at the monitor, waiting for the task to complete. You launch a couple of tasks, go do your other tasks, yes, away from the computer or also at the computer. It solves them in parallel. The next cool feature. You can set execution on a schedule. Again, I showed this with the example of Telegram. You set all this for daily execution, and let it churn out the information you need for you. Well, and of course, there is, yes, room for growth. Clearly, since it is a beta version now, just released. Clearly, they will refine it, improve it, add some cool things to it, and make it, well, probably a separate service, if everything goes well for them. Well, and speaking of pros, it is naturally worth mentioning the cons. As for the cons, I would highlight now. It works slowly and not perfectly. Whether Antropic will fix this or not, we will see over time. But now we have what we have. Slowness in this, I don't actually see any significant problem. As for not being perfect, this rather relates to training, working in third-party applications, yes, for example, the same Jetbot and the training example. If you watch the one I told you about, you will see nuances of its work there. That is, it cannot perform some super complex tasks now. Again, with the caveat that maybe I trained it poorly, and perhaps I gave it flawed prompts. That is, you need to play around with it more here. But nevertheless, the potential for growth, I am sure, is cool. The potential for growth, I am sure, is huge. The function is cool, interesting, deserves its own separate attention. Therefore, use it in your work, automate what can be automated, and spend your time not on routine tasks, but spend the resources of neural networks, which now allow you to do this.
That's all, guys, I'm saying goodbye. Don't forget to subscribe to my YouTube channel, give likes so that YouTube pushes me in the algorithms. I will be extremely grateful for that. And also come to Telegram, subscribe to it, there is additional content and also announcements of videos that I release. That's all, bye.