0:00 This is my YouTube graph. This is my 0:02 Thread. And this is Instagram. Over the 0:05 past few days, my views and followers 0:07 have gone up like this. I didn't sit 0:09 down every day and upload these one by 0:12 one. I have everything automated— 0:14 writing text, creating images and 0:16 shorts, writing specific captions for 0:18 each platform, and uploading them at 0:20 the scheduled times. All I did was 0:22 press the approve button. I will show 0:25 you how I built this from the very 0:27 beginning today. I will proceed in this 0:29 order. First, I will show you where 0:31 this automation runs. Then, I will show 0:34 you what the AI does and what I click 0:36 on each platform—YouTube, Thread, 0:38 Blog, Shorts, and Instagram. I will 0:42 also explain, one by one, the rules I 0:44 actually implemented to drive traffic. 0:45 These rules are ones you can use 0:47 immediately even if you don't use 0:49 automation. Finally, I will summarize 0:52 how I built this, what I told Claude, 0:54 and provide a step-by-step guide that 0:56 you can follow. If you watch until the 0:59 end, you will have a clear idea of 1:01 which tasks to hand over to the AI 1:03 first and what you need to start with 1:05 this week. Hello. To gather traffic for 1:07 To the J, you ultimately have to 1:09 consistently upload to various places. 1:11 However, as the number of platforms 1:13 increases, so does the workload. You 1:16 have to manage YouTube comments, write 1:18 posts for threads, upload images to 1:19 blog posts, and write different text 1:21 for Shorts for each platform. Each one 1:24 individually isn't difficult, but doing 1:25 it all alone every day is incredibly 1:27 challenging. So, I decided to hand over 1:30 the repetitive daily tasks to the AI 1:32 and focus solely on verifying and 1:33 approving. We started with a single Mac 1:36 Mini. This computer is used exclusively 1:38 for the AI. Filming, checking results, 1:40 and approving are done at the Mac 1:42 Studio next door. So why did I go out 1:44 of my way to separate them? Because if 1:46 the AI opens a browser and transfers 1:48 files on my computer, I can't use that 1:50 computer properly during that time. 1:53 That is why I gave the AI a separate PC 1:56 . There are two types of work performed 1:58 on the Mac Mini: Claude Co-work and 2:00 Chaeji Work. Neither of them interacts 2:03 with questions and answers in a chat 2:04 window; instead, they handle files on 2:06 this computer, use the browser, and 2:08 start working automatically when a set 2:10 time arrives. Claude and Chaeji exist 2:13 as agents. This is called scheduled 2:16 tasks. All my scheduled tasks are set 2:19 up on Co-Work and Work, with over 20 2:21 tasks in Co-Work alone. I use Chrome on 2:23 this Mac Mini as my browser. Co-Work 2:25 directly accesses platforms like 2:27 YouTube, Studio, Thread, and Tistory to 2:29 perform the work. Nowadays, you can 2:31 also launch these tasks using AI 2:33 browsers like Aside. However, I use 2:35 Co-Work and Work for greater freedom. 2:39 This allows me to access not only the 2:41 browser but also the files on this 2:43 computer. The results of the work done 2:45 by Co-Work and Work are collected on a 2:46 screen called the "Operations Room." It 2:48 looks like a regular web page, but you 2:50 should view it not as a publicly 2:52 accessible site, but as a screen 2:53 running inside the Mac Mini. There is 2:56 one Operations Room for each platform; 2:57 while the appearance may differ, the 2:59 structure is the same. When I enter 3:01 something, the AI generates a request, 3:04 which I read and then click approve. 3:06 All tasks are configured to be 3:08 processed through the Operations Room. 3:11 This ensures that the AI does not 3:12 perform arbitrary tasks, allowing it to 3:14 work safely only within the boundaries 3:16 I have authorized. Even today, while I 3:18 am filming, there are this many tasks 3:20 waiting for approval. Now, let's go 3:21 into each platform one by one. First is 3:24 the YouTube Operations Room. Claude 3:26 Co-Work collects unanswered comments 3:28 three times a day and attaches a reply 3:30 tag to each one. The "Like" button is 3:33 checked by default. I read them, 3:35 approve them if they are okay, and ask 3:37 for corrections if they are not. In the 3:39 past, there used to be a lot of backlog 3:41 in comments and replies. I knew I 3:43 should keep track of them, but they 3:44 kept getting pushed back while I was 3:45 making videos. The tip here is to do it 3:47 consistently at a set time. That is why 3:49 I set it to three times a day. If done 3:51 by a human, this is the first thing to 3:53 fall apart. Now that I just have to 3:55 read and click, there is rarely any 3:57 backlog. The Thread Operations Room is 3:59 where posts from my thread account are 4:01 published. Co-Work looks at how popular 4:03 thread posts are written and writes the 4:05 replies in my style. It means that he 4:09 learns from other popular posts as a 4:11 reference and rewrites them in my tone. 4:14 The tip here is timing. So, I don't 4:16 attach the date in advance. Instead, I 4:18 have set a specific time for them to be 4:20 uploaded. On weekdays, it is at 8 AM 4:22 and 9 PM, and on weekends, at 10 AM and 4:25 9 PM. Once I approve it, it is placed 4:27 at the very end of the approved content 4:29 . Approved materials are uploaded at 4:31 the nearest available time. You can 4:33 change the order by dragging and 4:35 dropping. With this setup, the release 4:36 time remains consistent regardless of 4:38 when you write a post. Even if you 4:39 approve ten posts at once, they won't 4:41 all be uploaded in a single day. Also, 4:43 Coke doesn't just run at the exact time 4:45 a post is supposed to be uploaded; 4:46 instead, it schedules a few posts in 4:48 advance. The Blog Management Office 4:50 monitors both Naver Blog and Tistory 4:51 simultaneously. The tip concerns images 4:53 . I insert at least three images into 4:55 each post. I create these images using 4:57 the "Chechi" image feature. I 4:59 instructed Claude Co-worker to use 5:02 Chechi to create images and insert them 5:05 between his text. On Naver, using 5:07 images from other websites is 5:09 detrimental to your blog ranking. 5:10 Therefore, using images you create 5:12 yourself is a better approach. There is 5:15 one more tip: even if the topic is the 5:16 same, I never upload the exact same 5:18 post to both blogs. Uh, since 5:20 publishing the same post would have a 5:22 negative impact on both blogs. Based on 5:25 my experience with Naver, I had the AI 5:27 write by copying others on Tistory. 5:29 Also, before writing, I have the AI 5:31 read about five to ten articles that 5:33 are well-received for that keyword. The 5:36 idea is to organize what is frequently 5:37 covered, missing information, and 5:39 questions people actually ask to make 5:40 the writing more useful. It doesn't 5:42 copy the sentences. Once you write down 5:44 these rules, the AI follows them 5:45 automatically from the beginning. Oh, 5:47 there is also the method of designating 5:48 them as skills to ensure they are 5:49 consistently followed. For Tistory, 5:51 Claude Co-Work even handles scheduled 5:53 pronunciation. However, since there are 5:55 access restrictions for Naver Blog, if 5:56 I set up scheduling tasks with Chaebi 5:58 PT Work, it logs in three times a day 6:00 to schedule and publish approved posts. 6:02 That is why I use both Claude Co-Work 6:04 and Chaebi PT Work. They haven't had 6:06 any conflicts yet. Shorts is divided 6:09 into two rooms: a creation room and an 6:11 upload room. AI and IT news go into the 6:14 Shorts Production Room. I turn that 6:16 into a motion graphic short. I use 6:18 Claude for the motion graphics. When 6:20 creating shorts from long-form footage, 6:21 I request it from the editing studio. 6:23 If I select a video and ask them to 6:24 turn it into a short, it comes out in 6:26 the exact same format as my channel's 6:28 shorts—under 1 minute long and 6:29 delivered for $ 1.25. A single short 6:31 created this way is distributed to four 6:33 platforms: YouTube, TikTok, Instagram, 6:35 and Naver Clip. There are three TV 6:36 channels here. First, the text is 6:38 different for each platform. The title, 6:40 description, and hashtags are all 6:42 different. The text below the video is 6:43 also different. For YouTube, use the 6:45 button in the bottom left to view the 6:46 full version. For TikTok and Instagram, 6:48 please check the profile links. Also, 6:51 for the YouTube Shorts, I link the 6:53 original long-form video as a related 6:55 video. This allows viewers to switch to 6:57 the full version while watching the 6:58 shorts. Second is keywords. I include a 7:01 phrase that people actually search for 7:04 within the 15-character title. Phrases 7:06 like "Claude Code,""5 Coding," or "Side 7:08 Hustle." I include that phrase in the 7:10 first sentence of the description as 7:11 well. For hashtags, I use one or two 7:13 major keywords and about two topic 7:14 keywords. I also include a searchable 7:16 phrase in the title of the top two 7:17 lines of the screen. Since screen text 7:19 immediately catches the eye, this is an 7:21 important phrase. So, how do we track 7:22 what people are searching for? Every 7:24 Sunday, the Co-work scheduled task 7:26 monitors YouTube search autocomplete 7:28 and Google Trends. We capture the exact 7:30 spelling that people actually type. 7:32 However, we do not include words that 7:33 do not have accompanying videos in the 7:35 keywords, as that would be incorrect 7:36 information. Third, there is the timing 7:38 . Only approved shorts are released in 7:40 order at 8 AM and 9 PM. Since Naver 7:43 Clip started late, it had fewer videos 7:45 compared to other platforms. Actually, 7:47 I hadn't even thought about Naver Clip 7:48 until I discovered it a few days ago, 7:50 so I decided to include it as well. 7:52 Therefore, I set it up so that the 7:53 shorts I had already uploaded are 7:55 posted four times a day to catch up in 7:56 quantity. I set this up as a scheduled 7:58 task for Chepy PT Work. Since it is 8:01 Naver, I had the work done through 8:03 Chepy PT Work. When starting a new 8:05 platform, there is no need to create 8:06 something entirely new. You just need 8:07 to make good use of existing content. 8:09 Instagram Card News follows the same 8:11 method. When Co-work gathers material 8:13 and creates cards, only the ones I 8:14 approve are released. I have set a 8:16 specific time here as well: 12:30 PM 8:18 every day. This is to ensure it doesn't 8:20 overlap with the 8 AM and 9 PM times 8:21 when Reels are released. Since they are 8:23 all coming out of the same account, I 8:24 have separated the times. If you look 8:26 at this so far, there is one thing in 8:27 common. Nothing is released 8:29 automatically. The AI creates the 8:31 content, but only the ones I approve 8:33 are released. I could have made 8:36 everything upload automatically, but I 8:38 deliberately didn't. There are three 8:40 reasons for this. First, I have to take 8:42 responsibility for the content that 8:43 goes out under my name. The AI writes 8:45 well, and while it sounds plausible, 8:46 some of the resulting text doesn't 8:48 reflect my writing style. Second, as I 8:50 read through the content while 8:51 approving, I can see where the AI keeps 8:53 making mistakes. If I point those out 8:55 again, it makes fewer mistakes next 8:57 time. The tips I showed you today are 8:59 all rules built up in this way. Even if 9:01 the tool changes, I can carry these 9:03 rules with me. Third, there is no loss 9:06 even if I don't like the draft. I can 9:08 simply discard it. People will likely 9:09 have quite different opinions on this 9:11 point. Many people say that true 9:12 automation should only work if 9:13 everything is uploaded automatically. 9:15 However, I am focused on the approval 9:16 process. Since I don't have to approve 9:19 posts every time, if I do it all at 9:20 once, they handle the tasks 9:22 automatically. I only need to approve 9:25 posts about once a day or every two 9:27 days. It is true that these operations 9:29 rooms were built using coding. However, 9:31 I wrote almost no code myself. It 9:34 starts with telling Claude Co-Work, "I 9:36 need a room like this." When I created 9:39 the operations room, this is what I 9:41 used: "I need an approval waiting list. 9:43 I must be able to change the order." I 9:45 used sentences like this instead of 9:47 code. If you are creating a room like a 9:50 thread operations room from scratch, 9:51 you can say, for example, something 9:53 like this: "Write three or four posts 9:55 for my thread every day and add them to 9:57 the list. Make sure only the posts I 9:59 approve are uploaded during whichever 10:01 time slot is empty between 8 AM and 9 10:03 PM on weekdays." But does simply 10:05 speaking make everything work? Not at 10:07 all. There were three places where I 10:09 got stuck while building it. First, it 10:11 is harder than expected to put into 10:13 words exactly what I want. It’s not 10:16 enough to just let it build it 10:17 automatically. You have to specify 10:19 exactly what goes in, what the AI does, 10:21 and what you press to get the room you 10:24 want. What was most important during 10:26 this development process was the 10:28 ability to explain these features, 10:30 rather than coding skills. Second, it 10:32 doesn't work the first time. I had to 10:33 make multiple adjustments while testing 10:36 it out. There are also many rules to 10:37 teach at the beginning. Third, making 10:39 it operate automatically every day is 10:41 the final step. You need to schedule it 10:43 , keep the computer turned on, and have 10:45 the work done via a browser within that 10:47 PC so that the work continues even when 10:49 you aren't watching. Ultimately, this 10:51 is what VIBE Coding is. It’s not just 10:53 about creating services to sell to 10:55 others; it’s about building tools to 10:58 do your work for you and automating 11:00 repetitive tasks in new ways. If you 11:03 want to try it out, you can follow this 11:05 sequence. First of all, open Claude 11:07 Co-work or Chaji Work. Everything I 11:10 showed you today is done here. It is 11:12 absolutely not just a chat window. You 11:14 must open the Claude app, change the 11:16 setting to Co-work, or switch Chaji PT 11:18 to Work, before you start. Make sure to 11:20 check if "Co-work" or "Work" is 11:22 configured as shown in this screen. 11:24 This is necessary so that they can 11:25 perform their work while you use your 11:27 PC, files, and browser. Second, simply 11:30 have them try out a task you want to 11:32 work on as a test. Initially, you don't 11:33 need a scheduling center or a 11:34 management office. Just choose one of 11:36 the tasks you repeat daily. Whether 11:38 it's scribbling, drawing on threads, 11:40 closing shorts, or creating any content 11:42 , do this for example: "Go into YouTube 11:44 Studio and just draft a reply to 11:46 today's comments. Don't upload it, just 11:47 show it as a list." Or, since your 11:52 thread account is logged in to the 11:54 browser, go to a thread and schedule a 11:56 post in my style from the popular 11:58 threads. You can let them know if there 12:00 are any corrections needed after 12:01 reviewing the results. After repeating 12:03 this a few times, you will be able to 12:05 tell if it is a task worth entrusting. 12:07 Don't try to handle multiple tasks at 12:09 once; start with one, verify it works 12:11 properly, and then check again. Third, 12:14 if it works well, schedule it. This 12:15 allows the task you tested to run 12:17 automatically at a set time every day. 12:20 If you are using Claude Co-work, you 12:21 can create three tasks in the scheduled 12:23 tasks section. You enter the name, task 12:25 , and frequency. For the frequency, you 12:26 can choose between hourly, daily, 12:28 weekly, or weekdays. However, if you 12:31 find these settings difficult, simply 12:33 tell Cowork to create a scheduled task 12:36 and explain what the task is, and 12:38 Cowork will create the task for you. 12:41 You can also create scheduled tasks in 12:43 the same way with Chachi T Work. Again, 12:45 if you find it difficult to create a 12:47 scheduled task yourself, you can 12:49 explain the schedule you want to create 12:51 in GPT Work, and GPT will automatically 12:52 create the schedule for itself. Once 12:57 scheduled, this PC must remain turned 12:59 on. Since you need to be able to 13:01 continuously access files or use the 13:03 browser on your computer, the PC must 13:05 be powered on for it to function. I use 13:08 a Mac Mini, but it doesn't necessarily 13:10 have to be one. If you have an unused 13:12 laptop, you can try starting with that. 13:15 The advantages of the Mac Mini are its 13:17 low power consumption, the fact that it 13:19 slows down less even when left on for a 13:21 long time, and its good performance. 13:23 Also, when using macOS, Claude Co-work 13:26 or GPT utilizes the PC slightly better 13:29 than Windows. In that respect, the Mac 13:31 is advantageous. The Mac Mini also has 13:33 its own strengths. Fourth, it allows 13:35 you to learn the workspace yourself. 13:36 When reservations start rolling, 13:38 requests accumulate. That is when you 13:39 ask them to create a screen for 13:41 approval. Create a screen where these 13:43 requests can be approved. When 13:44 requesting a fix, build a rule so that 13:46 the reason is read first when writing 13:49 the next request. That is the start of 13:50 operational execution. Every time a fix 13:52 is made, a line of the reason is added, 13:54 and from the next execution onwards, 13:55 that rule is read and written. If that 13:57 happens, you will not repeat the same 13:59 errors, and you will achieve the result 14:01 of consistently writing exactly as you 14:03 want. All the tips I showed you today 14:05 were built up in this way. The order is 14:07 important here. It is not about trying 14:09 it out with Co-work or Work first, and 14:11 if it works, setting a reservation and 14:12 then automatically uploading to the 14:13 workspace; rather, trying it out and 14:15 then doing the key-up comes first. To 14:18 summarize, the result of the graph I 14:20 showed you at the beginning was 14:21 generated by a single Mac Minn, the 14:23 rules built across the Claude Co-work 14:24 and Chappy PT T-Work platforms, and the 14:26 approval buttons I click. You can do 14:29 this on your own. You just need to 14:30 follow the four steps I mentioned today 14:32 one by one. However, regarding the 14:35 three steps I mentioned earlier— 14:36 speaking out what you want, building 14:38 rules while making corrections, and 14:39 running things automatically every day 14:41 —there may be parts where you need to 14:43 continuously work with the AI to solve 14:44 problems. These issues are unavoidable. 14:48 Since these problems vary depending on 14:50 the situation, you may need to solve 14:51 them by actually experiencing them. 14:54 This is where it can take a long time 14:55 or even make you give up. If that is 14:58 your case, this course can save you 15:00 time. The content introduced today is 15:02 all material you can learn in the 15:04 regular class on "5 Branches of Revenue 15:06 Pipelines Created with Coding." It will 15:08 all be included in the upcoming 15:10 lectures, covering everything from the 15:11 AI-dedicated PC and operations room to 15:13 AX for sending emails via scheduled 15:15 tasks, as well as blogs and Thread 15:16 Shorts. And what I showed you today 15:18 were just the very basics. In To The J 15:21 Class, we teach you various features 15:23 that expand upon these functions, as 15:25 well as how to actually turn them into 15:27 revenue. The link to the To The Class 15:29 regular course page is in the first 15:31 line of the description below the video 15:33 and in the pinned comment. Replacing AI 15:35 guides isn't a novelty anymore. However 15:38 , there aren't many people yet who 15:40 build and use them themselves to suit 15:41 their needs. Please pick just one of 15:43 the things you saw today and try making 15:44 it this week. It takes the most time to 15:47 make the first one, but you can apply 15:49 that method from then on. If you feel 15:52 like giving it a try, I hope you start 15:54 right now. Thank you for listening to 15:57 this long talk. Please subscribe and 15:59 like. It gives me great strength to 16:01 make the next video. This has been To 16:03 The J. I will see you again with 16:04 another great video. Thank you for 16:06 watching.