YouTube2Text

Claude и Claude Code: полный гайд от установки до сложных задач — Transcript

by Web3nity · 7,818 words · 1,258 segments · language en · Watch on YouTube

Full transcript

  1. 0:00This video will be a comprehensive guide
  2. 0:03on setting up Claude and Claude Code.
  3. 0:06In this video, guys, there will be
  4. 0:08everything you need to know to use these
  5. 0:11tools as simply and effectively as
  6. 0:13possible.
  7. 0:14And this video will consist of two parts
  8. 0:17. The first is for absolute beginners:
  9. 0:19where to download it, how the interface
  10. 0:21is structured, the difference between
  11. 0:23various operating modes, what everything
  12. 0:25is for, and so on.
  13. 0:27The second part will be about how to set
  14. 0:29up Claude and Claude Code for advanced
  15. 0:31work.
  16. 0:33We will create a working folder, connect
  17. 0:35the necessary services, and add a
  18. 0:37selection of skills that I recommend and
  19. 0:39use daily.
  20. 0:40And I will show you three ways to work
  21. 0:42with Claude that will unlock its
  22. 0:44potential.
  23. 0:46There will be timestamps in the video
  24. 0:48description, so you can jump straight to
  25. 0:50the section you need or find your way
  26. 0:52back later to review something.
  27. 0:54And additionally, guys, I have also
  28. 0:57prepared a guide for you. It will be
  29. 0:59available on my Telegram channel. I will
  30. 1:01explain what is in this guide and how to
  31. 1:03use it throughout the video. Guys, this
  32. 1:05will be an incredibly useful video. You
  33. 1:07simply won't need another video. Let's
  34. 1:09go. So, let's start with the very basics
  35. 1:11. Where to use it and how to download it
  36. 1:13. Claude has a web version that you can
  37. 1:16use right in your browser, and there is
  38. 1:19a desktop application. Spoiler alert: it
  39. 1:22is the desktop application that unlocks
  40. 1:25all the possibilities of this tool. You
  41. 1:27can download it from the official
  42. 1:29website. From this one right here. I
  43. 1:31will leave the link in the description.
  44. 1:32We select our operating system, download
  45. 1:35the file, install it, and log into the
  46. 1:37account. You can also authorize using
  47. 1:40Google. Okay. Now, we will need the
  48. 1:42downloaded application a little later,
  49. 1:44but we will start our acquaintance with
  50. 1:46the web version so you can understand
  51. 1:48the difference. And at the same time, I
  52. 1:50will tell you about Claude's models and
  53. 1:51the differences in subscriptions. Okay.
  54. 1:54We open our Claude in the browser, and
  55. 1:56this is what the start screen looks like
  56. 1:57.
  57. 1:59In the center is the field where we
  58. 2:01write, and if you click the plus sign,
  59. 2:03you can attach files, attach documents,
  60. 2:05or select a connected tool.
  61. 2:07Next to it, we also have a microphone to
  62. 2:09dictate tasks by voice. This is an
  63. 2:12absolute must-have when working with
  64. 2:13agents, because naturally, it is easier
  65. 2:15to speak a lot more details than to
  66. 2:17write a prompt. Speaking for myself, I
  67. 2:20would train myself to write more using
  68. 2:22text. Why? Because when using our voice
  69. 2:25, we tend to say many, many things with
  70. 2:27very little substance.
  71. 2:29With text, however, we approach the task
  72. 2:31more thoughtfully.
  73. 2:32Therefore, if we are talking about
  74. 2:34architecture, if we are talking about
  75. 2:37planning your project or your results in
  76. 2:39general, I would practice writing it out
  77. 2:41in text and only use voice for minor
  78. 2:44additions.
  79. 2:45That is exactly how I structure my work
  80. 2:47, and it is exactly what I recommend you
  81. 2:48do. Because with the arrival of AI, the
  82. 2:51most valuable skill is defining the task
  83. 2:53correctly and communicating exactly what
  84. 2:56you want to do. Let's move on. Click on
  85. 2:58the model name, and a list drops down.
  86. 3:01The menu contains four families. Let me
  87. 3:04explain. There is Haiku, Sonnet, Opus,
  88. 3:06and Fable.
  89. 3:08They differ in speed, capabilities, and
  90. 3:10, naturally, token limit consumption.
  91. 3:13After all, tokens are the currency of
  92. 3:14the future. I have a separate video on
  93. 3:16that, you can watch it later. Briefly
  94. 3:18about the models. Haiku is our fastest
  95. 3:21and most economical model. We choose it
  96. 3:24for short answers, for summarizing text
  97. 3:26, or for extracting, for example,
  98. 3:27specific data from a document. In other
  99. 3:30words, even the simplest model handles
  100. 3:32simple tasks well and with high quality
  101. 3:34. That does not mean it is bad. No.
  102. 3:37Sonnet is our universal model, more for
  103. 3:39everyday work.
  104. 3:41For texts, code, analysis, and
  105. 3:42multi-step tasks, Sonnet is a good fit.
  106. 3:44Use it, don't be afraid, it is good. Now
  107. 3:47for Opus.
  108. 3:48Opus is suitable for complex analysis
  109. 3:50and tasks that require deep thought.
  110. 3:52For example, comparing several studies,
  111. 3:55checking conclusions, fact-checking data
  112. 3:58, and synthesizing it all. We use Opus.
  113. 4:01And the fourth model is Fable. A model
  114. 4:03for the most complex and long-term
  115. 4:05projects with many steps. This model
  116. 4:08takes longer to think and, naturally,
  117. 4:10consumes the most limits.
  118. 4:12Therefore, for example, I use it only
  119. 4:14when I need to think through a truly
  120. 4:16architectural solution or when I need to
  121. 4:19compare and see the big picture.
  122. 4:21An example regarding my health.
  123. 4:23I have test results, doctor's
  124. 4:24recommendations, I need to compare
  125. 4:26studies I found, and synthesize all of
  126. 4:28this into a complete picture.
  127. 4:30I can use Fable for that.
  128. 4:32Or I want to create a custom system for
  129. 4:34myself, a dashboard with a lot of data.
  130. 4:37I need it to be architecturally
  131. 4:39well-thought-out. I can use Fable for
  132. 4:41that as well. You understand, right, for
  133. 4:42what kind of tasks this is needed.
  134. 4:43Therefore, for a fast task we choose a
  135. 4:45fast model, for a complex one we assign
  136. 4:47a more complex model. Personally, I use
  137. 4:49Opus most often, but I use Opus with
  138. 4:52sub-agents.
  139. 4:53Meaning, I have a main Opus, and then it
  140. 4:55has sub-agents that already use Sonnet
  141. 4:56and Haiku.
  142. 4:58We will talk about all this, and you
  143. 4:59will understand. And in the second part
  144. 5:01, I will show you how to combine all of
  145. 5:03this in one task. What I just mentioned
  146. 5:05. For example, how a more expensive
  147. 5:07model directs cheaper ones. This is
  148. 5:09called orchestration. Because right now
  149. 5:11, it is absolutely essential to know how
  150. 5:13to build these kinds of links so you
  151. 5:15don't burn through all your limits in a
  152. 5:17day.
  153. 5:18Tokens are the currency of the future,
  154. 5:19so we treat them exactly the same way as
  155. 5:21your money. We approach it wisely. Okay
  156. 5:23, let's get back to the interface. Which
  157. 5:25models are available? Your menu will
  158. 5:27depend on your subscription. Let me
  159. 5:29explain. On the free plan, if you happen
  160. 5:31to use it, you only have access to Haiku
  161. 5:33and Sonnet, so you won't see other
  162. 5:35models. On Pro, Opus is added to them.
  163. 5:38And Fable can be used for an additional
  164. 5:40fee within Opus.
  165. 5:42On Max and Ultra plans, all four models
  166. 5:44are available, and Fable is already
  167. 5:46included in the subscription.
  168. 5:48But remember, guys, you can spend up to
  169. 5:50half of your weekly limit on it. Now,
  170. 5:53next to the model selector, we have
  171. 5:56another toggle. And this toggle is where
  172. 5:59you can set permissions. Meaning when
  173. 6:01Claude can act on its own and when it
  174. 6:04needs to ask you. For example, in the
  175. 6:06future, you might want it to check your
  176. 6:08email. Naturally, you can grant it
  177. 6:10permission here. Now let's look at the
  178. 6:12menu on the left. Let's break down what
  179. 6:14each section is for. We have chats.
  180. 6:16Chats, I think it's clear, are our
  181. 6:17conversation history in Claude. Here you
  182. 6:20can find a past dialogue you need,
  183. 6:21continue where you left off, or simply
  184. 6:23open the dialogues you need that are
  185. 6:25tailored to specific topics.
  186. 6:27We shouldn't have everything mixed up in
  187. 6:29one chat. A separate chat for a separate
  188. 6:31topic. Now the next one is Projects.
  189. 6:33Projects are separate workspaces for us
  190. 6:35. Within projects, we can group multiple
  191. 6:38chats, collect documents, and general
  192. 6:40instructions on one specific project
  193. 6:42topic. And Claude will use these
  194. 6:45materials as context. So, to start
  195. 6:48working, it won't need to be reminded or
  196. 6:49have things uploaded again; it will have
  197. 6:51all of this by default. For example, you
  198. 6:53could create a project where you write
  199. 6:55posts for your social media. Accordingly
  200. 6:57, it would already contain examples of
  201. 6:59your posts, rules, and the topics you
  202. 7:01write about.
  203. 7:03And when you start working within that
  204. 7:05project, you won't need to re-explain
  205. 7:07what you're doing or how you write. I
  206. 7:09think that's clear. But we will analyze
  207. 7:11projects in detail in the second part of
  208. 7:12the video. Next, we have the Artifacts
  209. 7:14tab.
  210. 7:16Artifacts will store the materials you
  211. 7:18created in Claude, so you can easily
  212. 7:20find them instead of digging through
  213. 7:22chats to look for them.
  214. 7:25For instance, if you created HTML pages
  215. 7:27in some chat, or built small apps,
  216. 7:28visualizations, or documents.
  217. 7:31Basically, all of that will be there,
  218. 7:32and you'll be able to find it easily.
  219. 7:33The next tab is Scheduled Tasks. These
  220. 7:36are our scheduled tasks.
  221. 7:39For example, you can set your agent to
  222. 7:42send you news or a summary once a week
  223. 7:44or every day at a specific time.
  224. 7:48If you forget or want to check what
  225. 7:50tasks are scheduled, you can go to this
  226. 7:52tab and view the schedule to keep track
  227. 7:54of these runs.
  228. 7:56The next tab we have is Design.
  229. 7:58This is a section for visual work.
  230. 8:01Here you can start a presentation,
  231. 8:04design a page, create an interface, or
  232. 8:06make an interactive prototype.
  233. 8:09It’s a dedicated tab with a UI/UX design
  234. 8:11environment where you can change
  235. 8:12absolutely everything. It is amazing
  236. 8:14what they have done.
  237. 8:16And again, what you create here will
  238. 8:18then be stored in artifacts.
  239. 8:20You can easily return to them as well.
  240. 8:21And another very important tab is
  241. 8:23Personalization. This is where our
  242. 8:25skills are located. Skills are
  243. 8:27instructions for specific step-by-step
  244. 8:29tasks. Connectors are also found here.
  245. 8:32Connectors are how Claude links to your
  246. 8:34other services so you can work within
  247. 8:37them.
  248. 8:38For instance, if you work with your CRM
  249. 8:40, Telegram, or an application like
  250. 8:42Eleven Labs for voice generation.
  251. 8:46This is where it will all be stored, so
  252. 8:48you can easily go in, see your
  253. 8:49connectors, and link new apps to your
  254. 8:51Claude so it can interact with them.
  255. 8:54This will be located here.
  256. 8:55Again, in the second part of the video,
  257. 8:57we will return here, install the
  258. 8:58recommended skills, and connect the
  259. 9:00necessary services, so we will work with
  260. 9:02it more.
  261. 9:03Next, we move on to the important part:
  262. 9:04Chat and Cowork. Another thing important
  263. 9:06for me to tell you regarding the
  264. 9:07interface.
  265. 9:08We already released a guide on Claude,
  266. 9:10and back then there was a distinction
  267. 9:12between three modes: normal chat, Cowork
  268. 9:15mode for longer office tasks, and Code
  269. 9:17mode.
  270. 9:18Well, it's also called Claude Code,
  271. 9:20which we need for web coding to build
  272. 9:21tools, websites, and various complex
  273. 9:23artifacts.
  274. 9:24So, in the updated version, since they
  275. 9:26are constantly updating the app, the
  276. 9:28first two modes have been merged: the
  277. 9:29normal chat and Cowork mode.
  278. 9:31Now, in one conversation, you can
  279. 9:34discuss an idea and then ask it to find
  280. 9:36materials, compare them, and assemble an
  281. 9:39artifact. This means Claude now
  282. 9:40understands for itself when to just
  283. 9:42discuss things and when to activate
  284. 9:44Cowork mode. This now happens
  285. 9:45automatically under the hood. They
  286. 9:47simplified it. But the Code mode has
  287. 9:49remained separate.
  288. 9:51And here we have a toggle, and that is
  289. 9:53where many possibilities are hidden. And
  290. 9:56, by the way, the word "code" is
  291. 9:58important. There is no need to be scared
  292. 10:00. This mode, it seems to me, actually
  293. 10:02provides even more opportunities for
  294. 10:04those who don’t code at all, but who
  295. 10:07have their own project ideas, product
  296. 10:09ideas, and creativity. Meaning it helps
  297. 10:12by coding for you. The main thing is for
  298. 10:15you to describe it well.
  299. 10:17And in the browser, the code works with
  300. 10:19cloud projects, but next I will show you
  301. 10:21another option—Cloud Code with a folder
  302. 10:23on your computer.
  303. 10:24Accordingly, we are now moving from the
  304. 10:26browser to the installed Cloud Code
  305. 10:28application on the computer. What is
  306. 10:30important to say? Everything I have just
  307. 10:32told you about the interface up to this
  308. 10:34point is absolutely applicable to the
  309. 10:36computer application as well.
  310. 10:39There are all the same tabs, all the
  311. 10:41same buttons for you to understand.
  312. 10:43Then you might ask, what is the
  313. 10:45difference?
  314. 10:46The main difference is that when Cloud
  315. 10:48Code runs specifically on your computer
  316. 10:50, it gains access to your computer's
  317. 10:52file system.
  318. 10:55It can read documents, create new ones
  319. 10:57on your computer, edit them, and
  320. 10:58organize everything into folders.
  321. 11:02And the files and all the results will
  322. 11:04be stored on your disk, and you can
  323. 11:05return to them in future tasks.
  324. 11:08Furthermore, it can use the capabilities
  325. 11:11of your computer, you see? It can write
  326. 11:13and execute code on your computer, it
  327. 11:16can run installed programs and tools,
  328. 11:19and even use the browser itself.
  329. 11:21And for all this, your computer's
  330. 11:23processor, RAM, and, for suitable tasks
  331. 11:25, the graphics card are utilized.
  332. 11:28For example, for audio transcription, it
  333. 11:31can download and run a separate local
  334. 11:33model directly on your computer to use
  335. 11:36it.
  336. 11:37And then the entire recording will be
  337. 11:39processed on your device, and the
  338. 11:40finished text will be saved to a file,
  339. 11:42you see? Naturally, you will need to
  340. 11:44grant permission for everything it does
  341. 11:45on your computer, so be attentive to
  342. 11:47this. It is incredibly convenient.
  343. 11:49I have many agents working, and they
  344. 11:51periodically open my calendar on their
  345. 11:53own, add various tasks, and sort
  346. 11:55everything out themselves.
  347. 11:57The only thing is, of course, that you
  348. 11:58need to monitor this periodically.
  349. 12:00Meaning, do not trust AI completely;
  350. 12:02periodically check these folders, verify
  351. 12:04all the permissions it has, and see what
  352. 12:06it changes or adds. We are looking out
  353. 12:09for security. Trust, but verify. And an
  354. 12:11important note—Cloud itself continues to
  355. 12:14run in the cloud, but the commands and
  356. 12:16programs it manages are executed on our
  357. 12:19computer. So, in essence, you are giving
  358. 12:22Claude hands inside your computer,
  359. 12:24access to files and tools it can use to
  360. 12:26perform its work. Basically, that is the
  361. 12:29value of this mode. Next, I will explain
  362. 12:31in detail how to customize this workflow
  363. 12:32for yourself. But for now, let's move on
  364. 12:34to the initial settings. We have covered
  365. 12:36the basics, right? Now we will configure
  366. 12:38Claude itself.
  367. 12:40Open the Claude app and, in the
  368. 12:42environment selection where you will be
  369. 12:44working with Claude, specify this
  370. 12:45computer.
  371. 12:47In other words, by doing this, we
  372. 12:49confirm that Claude is allowed to run on
  373. 12:50this computer. We are pointing it out to
  374. 12:53him. Now, let's give Claude a working
  375. 12:55folder.
  376. 12:56This will be the place where he creates
  377. 12:59files, stores all instructions, and
  378. 13:01keeps everything updated.
  379. 13:03In short, this will be his workspace.
  380. 13:06Create a separate folder on your desktop
  381. 13:08called "Claude Projects". You can do
  382. 13:11this yourself via Finder or File
  383. 13:13Explorer, or you can even ask Claude to
  384. 13:15create the folder.
  385. 13:17Just write to him: "Create a Claude
  386. 13:18Projects folder on the desktop." That's
  387. 13:20it, he will create it for you. He will
  388. 13:22ask for permission, you grant it, and he
  389. 13:23will create it.
  390. 13:24Once the folder is ready, click "Select
  391. 13:26Folder" and open the specific folder we
  392. 13:29just created.
  393. 13:30This is how we give the agent access to
  394. 13:32our workspace.
  395. 13:34And from now on, this will be the main
  396. 13:36folder for working with Claude.
  397. 13:39Next, ask him directly in the chat to
  398. 13:41save all materials inside it and create
  399. 13:43a separate subfolder for each project.
  400. 13:47This way, documents, spreadsheets, code
  401. 13:50, and finished results will always be
  402. 13:52organized, and you can easily find them
  403. 13:55.
  404. 13:55Next, we need to set up your
  405. 13:57personalization. I will share a little
  406. 14:00life hack with you. This comes from me.
  407. 14:03You can do it on your own if you know
  408. 14:04how, but I will give you a different
  409. 14:06method that is simpler and even more
  410. 14:07effective.
  411. 14:09To set up personalization and configure
  412. 14:10Claude for your needs, we won't rack our
  413. 14:12brains thinking about what to tell him;
  414. 14:14we will ask him to conduct a short
  415. 14:16interview with us. Just turn on
  416. 14:18dictation and say: "I just installed
  417. 14:20Claude.
  418. 14:21Help me configure it for my tasks. Ask
  419. 14:25me what I do, what projects I have, what
  420. 14:27services I work with, and in what format
  421. 14:29I want to receive the results." And now
  422. 14:31he will ask you questions, and you will
  423. 14:34answer them one by one. You can tell him
  424. 14:36that you have several projects so that
  425. 14:38he creates different subfolders for you
  426. 14:39right away. Here you can share the full
  427. 14:42picture of your work, and he will take
  428. 14:44it into account. For instance, I explain
  429. 14:47that I have several projects and in one
  430. 14:48of them, I prepare videos and texts. I
  431. 14:50need scripts. In research, for example,
  432. 14:52primary sources are important to me. If
  433. 14:54there isn't enough information, please
  434. 14:56ask me to clarify.
  435. 14:58Basically, by answering these questions
  436. 15:00, you're syncing up with the AI agent so
  437. 15:02it understands exactly what you want and
  438. 15:04how you intend to work with it.
  439. 15:07It’s just like getting to know a new
  440. 15:09person. Now, here is a very important
  441. 15:11question: where will it store this
  442. 15:13information?
  443. 15:14Well, in case you missed it, I mentioned
  444. 15:16earlier that it will store it in
  445. 15:18specific files locally on your computer
  446. 15:20.
  447. 15:21So, we’re going to walk through that
  448. 15:23right now. This is extremely, extremely
  449. 15:25important. So, memory and instructions.
  450. 15:28After the interview, we ask Claude: "
  451. 15:30Save my general rules to the user
  452. 15:32Claude.md file and show me where it is.
  453. 15:34"
  454. 15:36To be clear, this will be a very
  455. 15:37important place that you—guys—need to
  456. 15:39visit and periodically review to ensure
  457. 15:42everything is correct, and see what it
  458. 15:44changes or adds. We open this file.
  459. 15:47Here is recorded what Claude should know
  460. 15:49about us and how we want to work with it
  461. 15:51.
  462. 15:52Basically, this will be a standard text
  463. 15:55file with permanent instructions.
  464. 15:58You can open it, read it, and edit it if
  465. 16:01the AI misunderstood something.
  466. 16:04You can add important details here
  467. 16:05yourself so you don't have to keep
  468. 16:07telling it to redo or add things; just
  469. 16:08go in and add them manually. It will be
  470. 16:10much easier that way. And this Claude.md
  471. 16:12is simultaneously your agent's identity
  472. 16:15, its knowledge about you, and the rules
  473. 16:17for working with you. Once again.
  474. 16:20Your agent will maintain this main user
  475. 16:23file by default.
  476. 16:26I’m just saying that you can go in there
  477. 16:27and change things yourself. Next to this
  478. 16:29file, there may be auxiliary files, for
  479. 16:33example, called Memory. Memory is a more
  480. 16:36detailed record about us. In most cases
  481. 16:38, we don't need to mess with that. This
  482. 16:41file will simply sit in your program's
  483. 16:43folder and get accessed by the script
  484. 16:45often.
  485. 16:46That is, the agent itself will fill it,
  486. 16:48update it, and record details it deems
  487. 16:50important for working with you.
  488. 16:53But this is not the only instruction
  489. 16:57file, Claude.md.
  490. 16:59A very important detail is that these
  491. 17:02kinds of instructions can be stored at
  492. 17:04three different levels simultaneously
  493. 17:06and used for different tasks.
  494. 17:09So, I’m going to break everything down
  495. 17:11for you in detail right now.
  496. 17:14Your agent can have several files that
  497. 17:16are all named Claude.md.
  498. 17:19And only the location of the file
  499. 17:21determines which tasks its rules are
  500. 17:23meant for. Let me explain. The first
  501. 17:25level is the general rules used for
  502. 17:27working with Cloud in general.
  503. 17:30This is the custom cloud.md in the
  504. 17:33hidden Cloud folder located inside your
  505. 17:35user folder on the computer.
  506. 17:38That’s exactly the one we just looked at
  507. 17:40.
  508. 17:41Here, I repeat, the language of
  509. 17:42communication, your preferences for
  510. 17:44responses, and general work requirements
  511. 17:46are stored.
  512. 17:48The second level, where I recommend
  513. 17:50setting instructions and creating an
  514. 17:52additional cloud.md, will be our main
  515. 17:54Projects Cloud folder that we created on
  516. 17:56the desktop.
  517. 17:58In other words, we ask Cloud to create
  518. 18:01another cloud.md inside it.
  519. 18:04And here we save more specific and
  520. 18:06detailed knowledge about what we do,
  521. 18:09what our work areas are, such as, in my
  522. 18:12case, separate tracks for online
  523. 18:15education and content production.
  524. 18:19This is where we outline the materials
  525. 18:21and where to save the results.
  526. 18:24These rules will be general for all
  527. 18:26projects within this folder.
  528. 18:28And the third level is the Cloud.md
  529. 18:31located in a specific project folder.
  530. 18:34Here, there will be narrower rules,
  531. 18:36materials, and requirements related to a
  532. 18:39specific task.
  533. 18:40You see, right? You may have many
  534. 18:42different tasks that you do with agents
  535. 18:44. You likely won't have just one agent.
  536. 18:46Different departments. We understand. I
  537. 18:48will show you with an example now.
  538. 18:49Content project. For example, we've done
  539. 18:52the user instructions and the general
  540. 18:54project rules.
  541. 18:55Now, inside the Projects Cloud folder,
  542. 18:57let's create a Content folder. So, this
  543. 18:59will be a separate project for us. In
  544. 19:01this folder, we will prepare posts and
  545. 19:03materials about artificial intelligence
  546. 19:04, so let's write it down. Well, you,
  547. 19:06accordingly, adapt all of this to
  548. 19:08yourself.
  549. 19:09Inside the Projects Cloud folder, create
  550. 19:11a Content folder.
  551. 19:12Posts about AI will be written here.
  552. 19:15Create a cloud.md inside and write
  553. 19:17instructions for this specific project
  554. 19:19in it. Topics, material formats, text
  555. 19:22style, and the fact-checking process.
  556. 19:25Save the finished results in a "Results
  557. 19:28" subfolder.
  558. 19:30And ask me everything needed to create
  559. 19:33this project.
  560. 19:34We understand.
  561. 19:36He won't just make it all up from his
  562. 19:37head; he will ask us now, and we must
  563. 19:39answer every single question.
  564. 19:41And then, after he analyzes them, he
  565. 19:43will create everything properly. There,
  566. 19:45we’ve stated our requirements, for
  567. 19:47example. Now we open this Content folder
  568. 19:49and see the cloud.md inside. MD. And
  569. 19:52here, as we can see, the work on content
  570. 19:54is already described: who we are writing
  571. 19:55for, what texts we are preparing, what
  572. 19:57we check, and where we put the finished
  573. 19:59materials. So, this is a special
  574. 20:01instruction specifically for this
  575. 20:03project. I think you’ve got it. Now, how
  576. 20:05will all these three levels work
  577. 20:07together?
  578. 20:08We create a new chat and ask it to start
  579. 20:11working on a new Telegram post.
  580. 20:13What is happening inside Claude?
  581. 20:16It is currently reading and considering
  582. 20:18the general user rules from the
  583. 20:20top-level file.
  584. 20:22You see, right? Where, for example, it
  585. 20:24says to answer in Russian, to keep
  586. 20:25answers brief if possible—basically,
  587. 20:27general instructions for all my projects
  588. 20:29.
  589. 20:30Then, it reads the instructions from the
  590. 20:33main "Claude Projects" folder and
  591. 20:35understands that I have an online
  592. 20:37education stream and a content stream,
  593. 20:40and that they are connected.
  594. 20:42So, it understands my work context and
  595. 20:44realizes it needs to go into the content
  596. 20:46folder.
  597. 20:48And after that, it reads the rules of
  598. 20:49the content project itself, which
  599. 20:51details exactly how I write posts and
  600. 20:53where the result should be saved.
  601. 20:55You see, right? So, all these files are
  602. 20:57three levels, and they complement each
  603. 21:00other. This reading of the instructions
  604. 21:02takes mere fractions of a second.
  605. 21:05We don't notice it, but when you have
  606. 21:07all these levels, the agent has enough
  607. 21:10context to perform tasks for us as
  608. 21:12efficiently and personalized as possible
  609. 21:15.
  610. 21:16For instance, it won't emphasize points
  611. 21:18in posts that would contradict the
  612. 21:19principles we promote in our online
  613. 21:21school, for example.
  614. 21:23Even though that's a completely
  615. 21:24different project, right? Online
  616. 21:25education is a completely different
  617. 21:27project. Right now, we are writing a
  618. 21:28project for posts. But it takes all of
  619. 21:29that into account because everything is
  620. 21:31interconnected. My life is
  621. 21:33interconnected with this.
  622. 21:35I also have separate projects for my
  623. 21:36health, and I have separate projects for
  624. 21:38my other ventures.
  625. 21:40These are all important details because
  626. 21:42, despite them being different
  627. 21:43businesses and different directions,
  628. 21:45they are all linked to me, which means
  629. 21:47they all affect me.
  630. 21:49And overall, my approach to each project
  631. 21:51defines me as a person, and it will
  632. 21:52define you too.
  633. 21:54But there is another scenario.
  634. 21:56Another scenario could be that you
  635. 21:57create a separate Claude MD folder that
  636. 21:59won't take your work context into
  637. 22:01account at all, but will be focused, for
  638. 22:03example, on analyzing your test results
  639. 22:05, if you don't want it to be linked.
  640. 22:08And this project will work in isolation
  641. 22:10. How do you make it work in isolation?
  642. 22:13You need to start a chat from within
  643. 22:15that specific new "lab results analysis
  644. 22:17" folder.
  645. 22:18It is done like this.
  646. 22:20When we create a chat and start a
  647. 22:21dialogue within this specific narrow
  648. 22:23project, Claude will only consider the
  649. 22:25general user instructions and the
  650. 22:26instructions of that specific project,
  651. 22:28separate from everything else.
  652. 22:31And you can have as many such divisions
  653. 22:33and sub-systems as you like. Again,
  654. 22:35don't worry if something is unclear
  655. 22:37right now. In the process of working,
  656. 22:39you will build all the necessary levels
  657. 22:41and directions for yourself. Okay, let's
  658. 22:43move on.
  659. 22:45Now we need to give our Claude more
  660. 22:47hands and more tools so that it can be
  661. 22:49more useful to you.
  662. 22:51I will tell you how and which services
  663. 22:54you should connect first. First and
  664. 22:56foremost, of course, you need to connect
  665. 22:59Google, specifically Google Drive, for
  666. 23:01example, Gmail, and Google Calendar.
  667. 23:04Through these, Claude will gain access
  668. 23:06to our documents, emails, and meetings—
  669. 23:09essentially our plans—and will be able
  670. 23:11to add events to them and remind us
  671. 23:13about them.
  672. 23:15To do this, we go to personalization,
  673. 23:18select connectors, and find the service
  674. 23:20we need.
  675. 23:21For example, here is Google Drive. Click
  676. 23:24connect. Then we log into our account,
  677. 23:26and after authorization, the service
  678. 23:28appears among those connected. That's it
  679. 23:30; now it can access it, and they can
  680. 23:32exchange information. And we add mail
  681. 23:34and calendar using the same principle.
  682. 23:37If, for example, you also want to create
  683. 23:39documents, edit them, or make
  684. 23:41presentations, you can additionally
  685. 23:43connect Google Docs, Sheets, and Slides
  686. 23:45.
  687. 23:45Convenient.
  688. 23:47And now, through these connectors, your
  689. 23:49agent will be able to create and edit
  690. 23:50your documents, spreadsheets, and slides
  691. 23:52.
  692. 23:53In short, everything you need. Schedule
  693. 23:55meetings for you and set reminders in
  694. 23:57the calendar.
  695. 23:58And, of course, besides Google, you
  696. 24:00understand there are dozens of other
  697. 24:01platforms and services you use, and you
  698. 24:03can connect them here using the same
  699. 24:05principle.
  700. 24:06For example, if you use Dropbox, connect
  701. 24:08it. If you use Figma, connect it. If you
  702. 24:11use Notion, connect it. There is a huge
  703. 24:14list here. How do you work with these
  704. 24:15connectors? Let me explain. Just write
  705. 24:17in the chat.
  706. 24:19For example, if you want to work with
  707. 24:20the calendar, write: "Find the next work
  708. 24:22call in the calendar, look at the
  709. 24:24created document, and the latest emails
  710. 24:25on this topic."
  711. 24:27"Gather a short summary of what to
  712. 24:28discuss, what questions are open, and
  713. 24:30what needs to be solved." And we see
  714. 24:32right in the chat how Claude begins to
  715. 24:34use different tools one by one.
  716. 24:36Everything is ready; it checks the
  717. 24:38calendar and completes the task for us.
  718. 24:39Everything is clear. Now, an important
  719. 24:41question that might cross your mind.
  720. 24:44What if I didn't find the service I want
  721. 24:46my agent to work with in the list?
  722. 24:49For this, you will now be introduced to
  723. 24:51a new concept: MCP.
  724. 24:53MCP is a special adapter that helps an
  725. 24:56agent work with almost any service.
  726. 24:59I have a separate video about MCP, so I
  727. 25:02won't go into technical details; I'll
  728. 25:04just show you how to implement it. If
  729. 25:06you find you lack the knowledge, just
  730. 25:07watch that video and you will understand
  731. 25:09everything. For example, let's connect
  732. 25:11Telegram via MCP. We'll use it, for
  733. 25:13example, to retrieve agreements,
  734. 25:15commands, messages, and files needed for
  735. 25:17our tasks. So, we just type exactly that
  736. 25:19into the chat. Help me set up an MCP
  737. 25:21connection to Telegram. I want to read
  738. 25:24messages in my work chat, download files
  739. 25:27, and track agreements. Basically, we
  740. 25:29write down everything we need from this
  741. 25:31service. Do you understand? This is
  742. 25:32important. It’s not just about
  743. 25:33connecting Telegram and letting the
  744. 25:35agent set everything up for you. Yes, it
  745. 25:37really is just that simple. The only
  746. 25:39thing, guys, is that I must mention,
  747. 25:41since we have a large media presence, we
  748. 25:43must naturally keep security and privacy
  749. 25:46in mind.
  750. 25:47That’s why I always create a separate
  751. 25:49email, a separate Google account, and a
  752. 25:51separate Telegram for Claude, not my
  753. 25:53main Telegram, not my main Google email
  754. 25:56, and so on.
  755. 25:57We need to be mindful of this. You know
  756. 25:59what I mean? There is also an option to
  757. 26:01use a Telegram bot, which you would need
  758. 26:04to add to all the groups you want to
  759. 26:06read messages from later. But there are
  760. 26:08limitations; the bot won't be able to
  761. 26:10download files larger than 20 MB.
  762. 26:13For instance, that’s critical for me,
  763. 26:15because our team often sends video files
  764. 26:17, so my preferred option is to connect a
  765. 26:20real account.
  766. 26:21But I will use a separate work number, a
  767. 26:23separate work account, not my personal
  768. 26:25one.
  769. 26:26Now, to connect to a Telegram account,
  770. 26:28you will need Telegram API data.
  771. 26:31The agent will go to my.telegram.org
  772. 26:33itself, open API Development Tools, and
  773. 26:36create an application. Two parameters
  774. 26:39will appear there: API ID and API Hash.
  775. 26:42The agent will take them and save them
  776. 26:44to the local connection settings.
  777. 26:46Alternatively, it might ask you to visit
  778. 26:48those pages yourself to get these
  779. 26:50parameters and pass them over securely.
  780. 26:53And naturally, we confirm the Telegram
  781. 26:56login ourselves.
  782. 26:58And guys, this is also very important—
  783. 27:00never paste any APIs, passwords, or
  784. 27:02authorization codes into your chat with
  785. 27:05the model.
  786. 27:06In principle, they have already
  787. 27:08addressed this, and the security system
  788. 27:10is set up so that it will guide you on
  789. 27:12where and how to save them.
  790. 27:15If it doesn't ask, then you ask it: "
  791. 27:17Create a separate file where I should
  792. 27:19load or paste these API passwords, and
  793. 27:21walk me through it—tell me where to put
  794. 27:24what."
  795. 27:25It will create instructions for you. We
  796. 27:27never send anything important like that
  797. 27:28into chats or the cloud. Keep that in
  798. 27:30mind.
  799. 27:31Also, I’ve left a special file in the
  800. 27:33Telegram channel that contains
  801. 27:34everything needed to connect to Telegram
  802. 27:36.
  803. 27:37So, if anything, just give it to your
  804. 27:39agent, and it will set everything up for
  805. 27:41you. So, that’s basically it—nothing
  806. 27:43complicated.
  807. 27:44After setup, we open the list of MCP
  808. 27:47connections here in this tab and check
  809. 27:49the status of Telegram.
  810. 27:52Then we type: read the messages in the
  811. 27:54work chat from today and tell me what
  812. 27:56decisions we made and what questions
  813. 27:58remain open.
  814. 27:59This is a check. After all connections
  815. 28:01are made, we must verify our agent to
  816. 28:03see if it understood everything and set
  817. 28:05it up correctly. Important. Accordingly
  818. 28:07, our agent goes to Telegram, reads the
  819. 28:09messages, and brings us the answers.
  820. 28:12And again, in exactly the same way, we
  821. 28:14can set up almost any other service that
  822. 28:16doesn't have an official connector in
  823. 28:18the app.
  824. 28:20What else can be said here?
  825. 28:23If you don't know, or you want some tips
  826. 28:25on what else you can connect, I'll just
  827. 28:27show you my basic list of what needs to
  828. 28:30be connected.
  829. 28:32It is very simple.
  830. 28:33I would recommend connecting all Google
  831. 28:35services if you work in Google, Telegram
  832. 28:37or the messenger you use for
  833. 28:39correspondence, and my favorite search
  834. 28:40engine.
  835. 28:42The search engine enhances your agent's
  836. 28:44capabilities. It is a special search
  837. 28:46engine for agents. It is called Exa.
  838. 28:50In short, it is a special tool so that
  839. 28:52agents can search the internet
  840. 28:53semantically, that is, by meaning.
  841. 28:56Understood. Let's move on. Now, skills.
  842. 28:59I will tell you everything you need to
  843. 29:01know about skills. What are they,
  844. 29:02generally speaking? A skill is a saved
  845. 29:05instruction for a specific task. What
  846. 29:07does it look like?
  847. 29:09It looks like a regular text file, but
  848. 29:11it contains all the details, meaning a
  849. 29:13specific workflow of yours, step by step
  850. 29:15.
  851. 29:16For example, how to prepare a brief, how
  852. 29:18to format a page correctly, how to edit
  853. 29:21text.
  854. 29:22Inside a skill, there can be rules,
  855. 29:24examples, links to supporting files, and
  856. 29:26references.
  857. 29:27By the way, I have a separate detailed
  858. 29:29video specifically about how skills are
  859. 29:31structured internally, so I won't go
  860. 29:33into technical details now; you can
  861. 29:34watch that. I will only say the main
  862. 29:37thing: a skill is essentially digitized
  863. 29:39expertise. And right now, 1,000 skills
  864. 29:41are publicly available. Designers, for
  865. 29:43example, digitize their expertise on how
  866. 29:46pages should look, post their expertise
  867. 29:49publicly, and you can simply take this
  868. 29:51expertise and use it in your project.
  869. 29:54There are paid skills, there are free
  870. 29:56skills, and there are huge marketplaces
  871. 29:57for these skills. Where are they located
  872. 29:59? We have skills within Claude itself.
  873. 30:02We also have a separate website. How do
  874. 30:04you find the skills that are in Claude?
  875. 30:05They are located in the personalization
  876. 30:07section.
  877. 30:08We open personalization, then Skills,
  878. 30:10and here we see all the pre-installed
  879. 30:12skills that Claude has already set up
  880. 30:14for us.
  881. 30:15We can also add skills ourselves, that
  882. 30:17is, upload them to our agent. Skills can
  883. 30:19be uploaded here directly as a file; you
  884. 30:21can add a text file here, one you
  885. 30:23downloaded somewhere or someone gave you
  886. 30:25, and upload it here. Claude will read
  887. 30:27it, Claude will add it, and now it can
  888. 30:29use it for a specific task. But there is
  889. 30:31a simpler way to install a skill.
  890. 30:34You can actually just give Claude a link
  891. 30:36to a skill you find on the internet.
  892. 30:38Usually our skills, or rather most of
  893. 30:40them, the most capabilities are found on
  894. 30:43GitHub. It is a kind of social network
  895. 30:45for programmers, but it is used by
  896. 30:47non-programmers now too.
  897. 30:49And we can just provide a link to the
  898. 30:51skill we need, the one we like, and ask
  899. 30:53your agent to install it by itself. How
  900. 30:56do you search for these skills on GitHub
  901. 30:57? There are entire collections of skills
  902. 30:59. That is, these are separate
  903. 31:01repositories with individual skills.
  904. 31:03And if you want, you don't even have to
  905. 31:05go to GitHub yourself and search for
  906. 31:06anything there.
  907. 31:08We can simply ask Claude itself to find
  908. 31:10the necessary skills, for example.
  909. 31:13For instance, you want to create a
  910. 31:14specific website right now.
  911. 31:15You just ask it: find suitable skills on
  912. 31:18GitHub and explain what each one does.
  913. 31:20Even if you don't understand anything at
  914. 31:22all, this is the simplest way, right?
  915. 31:24That's it, it opens, it searches, it
  916. 31:26suggests, it sends you a list, and you
  917. 31:28read and study it.
  918. 31:30If you like a certain skill, you must
  919. 31:32check it for safety. That means you ask
  920. 31:35it to check for safety, to see if
  921. 31:37everything is okay, if there are any
  922. 31:39pitfalls, or if there are any hidden
  923. 31:40instructions. Actually, tell me, tell me
  924. 31:42, because it is very important.
  925. 31:44Absolutely anyone can post any skill on
  926. 31:46GitHub. And there are skills that are
  927. 31:48not verified, and there are skills that
  928. 31:50could potentially hack you. Therefore,
  929. 31:52even if someone doesn't mention it, we
  930. 31:54always check everything.
  931. 31:56And if it has checked the skill and
  932. 31:58everything is great and it is safe, only
  933. 32:00then do we ask it to install it.
  934. 32:02And immediately, as soon as you have
  935. 32:04installed a specific skill, you ask it
  936. 32:06to work with it. Because it is one thing
  937. 32:07that someone created it for themselves,
  938. 32:09but it is another thing how you can use
  939. 32:10someone else's skill for your own tasks
  940. 32:11. If you like it, you keep it in your
  941. 32:13arsenal. If you don't like it, you just
  942. 32:15delete the skill saying, I didn't like
  943. 32:16it. Or, if you liked it but something
  944. 32:19about it doesn't suit you, you also have
  945. 32:21the option to redo, supplement, or
  946. 32:23improve it. For example, you like
  947. 32:25everything about it, but you don't like
  948. 32:26certain details. You just write: listen
  949. 32:28, I don't like these specific details,
  950. 32:29change them to fit our needs. That's it
  951. 32:31, it will take a ready-made skill; it
  952. 32:33won't be doing half the work from
  953. 32:34scratch, just adapting and improving it
  954. 32:36for you. And you will have your own
  955. 32:38improved, your very own skill.
  956. 32:40If you ask me now which skills I would
  957. 32:42recommend, I will tell you about four
  958. 32:45skills that I suggest starting with if
  959. 32:47you are starting from scratch. Each of
  960. 32:49them is useful for its own specific task
  961. 32:51.
  962. 32:52The first skill is one I would call a
  963. 32:54meta-skill.
  964. 32:56It helps us think through the task
  965. 32:57itself from different angles and how we
  966. 32:59are going to solve it.
  967. 33:01The workflow is defined within this
  968. 33:03skill.
  969. 33:05That is, first Claude studies the
  970. 33:06context, then it asks clarifying
  971. 33:08questions, then proposes several
  972. 33:09approaches and discusses with us which
  973. 33:11one to choose.
  974. 33:13The key thing here is that you might
  975. 33:15simply not think about a certain side of
  976. 33:18your issue, and this skill helps you
  977. 33:20look at it and suggests things you
  978. 33:22aren't noticing.
  979. 33:23For example, you want to figure out who
  980. 33:25your audience is, who your main client
  981. 33:27is.
  982. 33:27You export everything you need from your
  983. 33:30side, from your Telegram channel, and
  984. 33:32ask it to analyze it.
  985. 33:34And with the help of this skill, the
  986. 33:36agent can suggest that you also look at
  987. 33:38, for example, your region's
  988. 33:39demographics to get a broader view of
  989. 33:41your issue.
  990. 33:42You see, right? So it looks for a few
  991. 33:45more angles on your task. It is
  992. 33:47incredibly useful. I personally start
  993. 33:49every new task with a discussion like
  994. 33:51this.
  995. 33:52I answer various questions, and we sort
  996. 33:54of brainstorm the idea together, compare
  997. 33:56options, and agree on an approach.
  998. 33:58And only after that do we move on to
  999. 34:00execution.
  1000. 34:01The second skill I recommend is one I
  1001. 34:04suggest for web coding, when you'll be
  1002. 34:06creating something with an agent,
  1003. 34:08meaning where you need a visual
  1004. 34:10interface, like making websites or apps
  1005. 34:12, and you want them to look good.
  1006. 34:16This skill contains a collection of
  1007. 34:18visual styles, palettes, and font
  1008. 34:20pairings, and Claude will rely on them
  1009. 34:22to, for example, choose the design for
  1010. 34:25your projects, headings, colors, spacing
  1011. 34:27, buttons, and other interface elements
  1012. 34:29.
  1013. 34:30For example, we ask Claude to build a
  1014. 34:32landing page that describes our channel
  1015. 34:34and our principles.
  1016. 34:36And we immediately add: use and select
  1017. 34:38the style, colors, and fonts for this
  1018. 34:40project, and make a selection for me so
  1019. 34:42I can choose.
  1020. 34:43That’s it, we look at the various
  1021. 34:45options it immediately suggests and ask
  1022. 34:47it to apply one to the page. If you
  1023. 34:49managed to get something. Today we ended
  1024. 34:50up with a page like this. I also use
  1025. 34:52this skill myself. I already posted a
  1026. 34:54selection of visual skills on Telegram,
  1027. 34:56and this one is there, by the way. And
  1028. 34:59there are other important skills and
  1029. 35:01tools for this field too, so if you are
  1030. 35:03interested in visuals, go to Telegram
  1031. 35:05and look for that very useful post.
  1032. 35:08Next, our third skill is called "It
  1033. 35:11scans the text for things that
  1034. 35:13immediately give away an AI."
  1035. 35:16These are repetitive introductory
  1036. 35:18phrases, unnecessary contrasts,
  1037. 35:20antitheses, pompous wording, too many
  1038. 35:22dashes, and artificially choppy
  1039. 35:24sentences.
  1040. 35:25Well, I think everything is clear here.
  1041. 35:27This is just for anyone writing with AI
  1042. 35:29.
  1043. 35:30And the fourth skill is called "It
  1044. 35:32contains both a video creation tool and
  1045. 35:34a skill that explains to the agent how
  1046. 35:36to work with it."
  1047. 35:38The agent, for example, writes the code
  1048. 35:40, and it then turns it into a video.
  1049. 35:43I think there are a lot of tasks related
  1050. 35:45to video content right now, but even if
  1051. 35:48you don't post anything, you can just
  1052. 35:50ask it to explain a complex topic with a
  1053. 35:52short, captioned, animated video, right
  1054. 35:54?
  1055. 35:55It’s not necessarily for creating videos
  1056. 35:56for social media. It can just be a video
  1057. 35:58created for your colleagues, for your
  1058. 36:00child, or for yourself. It's incredibly
  1059. 36:02interesting. This way of working with AI
  1060. 36:04is trending right now. That is, the most
  1061. 36:06advanced users are using it for such
  1062. 36:08tasks.
  1063. 36:09And, of course, you can also make Reels
  1064. 36:11, subtitles, intros, and animated
  1065. 36:13graphics with it.
  1066. 36:15Well, guys, up to this point, we have
  1067. 36:17prepared for work.
  1068. 36:19We created folders, saved instructions,
  1069. 36:22connected services, and added skills.
  1070. 36:25Now let's figure out how exactly to work
  1071. 36:27with it. So everything before this was
  1072. 36:30just the setup. These are the tools, the
  1073. 36:33instructions, everything needed for the
  1074. 36:35work. But how exactly to work is
  1075. 36:37different. So I will show you three ways
  1076. 36:40to work effectively, three AI modes. In
  1077. 36:42Claude, there are essentially three ways
  1078. 36:45of working. We have the plan mode, the
  1079. 36:47goal mode, and the orchestration mode.
  1080. 36:49One by one. You need the plan mode when
  1081. 36:52the specific result you want to get is
  1082. 36:54not yet clear.
  1083. 36:56First, we build and agree on the steps
  1084. 36:58together with Claude, and only then,
  1085. 37:00after you have agreed, does it proceed
  1086. 37:02to execute the plan. And you see what
  1087. 37:05result you arrived at. I'll show you an
  1088. 37:07example. Here is a table with our
  1089. 37:08channel's analytics. There are 57
  1090. 37:10episodes, views, subscriptions, and
  1091. 37:12other metrics for different periods. And
  1092. 37:15our task is to choose the next four
  1093. 37:17episodes, based on the data. For example
  1094. 37:20, I don't really understand what the
  1095. 37:21episodes could be, what specific topics
  1096. 37:23should be in this plan.
  1097. 37:25But I do know that I need to check which
  1098. 37:27videos are getting views, which are
  1099. 37:29bringing in subscribers, and where
  1100. 37:30success is repeated across several
  1101. 37:32episodes.
  1102. 37:33To do this, we turn on plan mode and
  1103. 37:35write. Look at the channel analytics.
  1104. 37:38We need to choose four topics that will
  1105. 37:40help get views and new subscribers.
  1106. 37:42Let's leave one episode for an
  1107. 37:43experiment. First, propose an analysis
  1108. 37:45plan and clear it with me.
  1109. 37:47Then, accordingly, in this mode, Claude
  1110. 37:50now suggests steps to complete this task
  1111. 37:52.
  1112. 37:53We check, answer questions, and correct
  1113. 37:55the approach if we need to adjust it.
  1114. 37:58When you are satisfied with everything,
  1115. 38:00we say the plan works, go ahead. That's
  1116. 38:02it. And now Claude will carry out
  1117. 38:04everything specifically, following the
  1118. 38:06steps you agreed upon. We understand why
  1119. 38:09we need it. Accordingly, we open the
  1120. 38:11result, four topics and a justification
  1121. 38:13for each. That is what plan mode is for
  1122. 38:15. Now we have goal mode. In goal mode,
  1123. 38:18you know what you want to get in the end
  1124. 38:20. You already know the result, of course
  1125. 38:23, but in this case, you just don't know
  1126. 38:25which steps and tools to use to get
  1127. 38:27there. Accordingly, here we set the goal
  1128. 38:29and the criteria for completion. But the
  1129. 38:32path to this goal, to the result, Claude
  1130. 38:34will choose for himself. For example,
  1131. 38:36let's take the same spreadsheet. But
  1132. 38:37here we will set a different task.
  1133. 38:39Choose an episode and a specific clip
  1134. 38:41that is worth turning into a short
  1135. 38:43vertical video and making a finished
  1136. 38:45Reel.
  1137. 38:46To do this, we select goal mode and
  1138. 38:48write: Look at the channel analytics,
  1139. 38:50choose an episode for a short clip, find
  1140. 38:53the necessary tools and skills yourself
  1141. 38:55; I need a finished, edited Reel. And
  1142. 38:57it's better to clarify with him right
  1143. 39:00away what "finished" means. And if
  1144. 39:02necessary, you can add to this chat, for
  1145. 39:04example, a vertical video of about a
  1146. 39:06minute with a clear message, readable
  1147. 39:08subtitles, and good sound. Well, so that
  1148. 39:10he understands what "finished" means to
  1149. 39:12you. What, accordingly, happens next in
  1150. 39:14this mode? Claude studies the analytics
  1151. 39:16, he chooses the most promising video in
  1152. 39:19his opinion, and then he goes into our
  1153. 39:21archive, takes the files from that
  1154. 39:23episode, connects a transcriber, matches
  1155. 39:26frames with audio, then cuts out a clip
  1156. 39:28, adds graphics, and sends you the
  1157. 39:30finished video. That is, the agent opens
  1158. 39:33the file himself, checks himself if it
  1159. 39:34meets the "finished video" requirements
  1160. 39:36, if the subtitles are readable, if the
  1161. 39:38voice is clearly audible, if the thought
  1162. 39:40is understandable to someone who hasn't
  1163. 39:41seen the full episode. And if something
  1164. 39:44needs to be corrected, the agent himself
  1165. 39:46goes and redoes it. And it will keep
  1166. 39:48doing this until the goal is achieved.
  1167. 39:51So, in your goal, be sure to explicitly
  1168. 39:53write down what is important to you and
  1169. 39:55what the exact final outcome should be,
  1170. 39:56you understand? Then it will keep
  1171. 39:58redoing the task until it reaches that
  1172. 40:00point. That is, the goal itself helps
  1173. 40:02your agent continue working until the
  1174. 40:05result meets the criteria you set.
  1175. 40:09If it lacks a source or tool access for
  1176. 40:11the next step, Claude will first try to
  1177. 40:13solve the problem itself and only then
  1178. 40:15ask you for help.
  1179. 40:18And now, the third mode is the most
  1180. 40:20complex, but also the most advanced
  1181. 40:23workflow for working with Claude.
  1182. 40:26This is orchestration and sub-agents.
  1183. 40:29Let me explain.
  1184. 40:30Orchestration is a way to divide the
  1185. 40:32work within a single task between
  1186. 40:34different models.
  1187. 40:36Because the main Claude, using an
  1188. 40:38expensive and smart model, coordinates
  1189. 40:41the process, while separate, cheaper
  1190. 40:43assistants receive their parts of the
  1191. 40:45work. For example, let's return to our
  1192. 40:48table and again pick the topic of four
  1193. 40:50future episodes.
  1194. 40:52But now we instruct Claude to involve
  1195. 40:53three sub-agents and assign them
  1196. 40:55different roles.
  1197. 40:56The first, for instance, will search for
  1198. 40:58topics with steady reach.
  1199. 41:00The second will look at view depth,
  1200. 41:02subscriptions, or shares.
  1201. 41:04The third will check their calculations
  1202. 41:06and conclusions. That is, it will look
  1203. 41:08for errors, comparisons, and
  1204. 41:09counterexamples. We run the first two in
  1205. 41:11parallel, while the checker first
  1206. 41:13studies the data and then their results
  1207. 41:15.
  1208. 41:16And we ask the main Claude to now
  1209. 41:18compare these arguments and make the
  1210. 41:20final editorial decision.
  1211. 41:23So, in this way, we distribute the load
  1212. 41:25and, consequently, the token costs. This
  1213. 41:28is insanely important, because in the
  1214. 41:30future, tokens are literally your money
  1215. 41:32, and orchestration helps distribute
  1216. 41:35them wisely. That is, orchestration
  1217. 41:37helps allocate tokens more rationally.
  1218. 41:39Complex, high-volume tasks that don't
  1219. 41:41require much brainpower use fewer of
  1220. 41:43your limits.
  1221. 41:44Complex, high-volume tasks that don't
  1222. 41:46require much brainpower use fewer of
  1223. 41:48your limits because these tasks are
  1224. 41:50performed by cheaper models. And the
  1225. 41:53smartest, most expensive model analyzes
  1226. 41:55the results and provides you with
  1227. 41:57ready-made solutions. If you want to
  1228. 41:59know my recommendation, I personally use
  1229. 42:02Opus as my default main model. And for
  1230. 42:04super architectural tasks, I use Haiku,
  1231. 42:06but that is extremely rare. But Sonnet,
  1232. 42:08for example, I use as a sub-agent, an
  1233. 42:10assistant that performs all the routine
  1234. 42:11work.
  1235. 42:13Phew, I think I've told you absolutely
  1236. 42:15everything in this video that you need
  1237. 42:17to get started.
  1238. 42:19And that is the guide we’ve put together
  1239. 42:21. I think everything I’ve shared will be
  1240. 42:23enough for you. If you find you’re
  1241. 42:25missing something, you can watch my
  1242. 42:26additional videos.
  1243. 42:27And don't worry if there's something you
  1244. 42:29don't quite understand while watching
  1245. 42:30the video. Everything comes with
  1246. 42:32practice.
  1247. 42:33Start with this guide, start working,
  1248. 42:36and you'll figure everything out as you
  1249. 42:38go.
  1250. 42:39I've left all the instructions, settings
  1251. 42:40, prompts, and links to skills and tools
  1252. 42:42in my Telegram channel; go there, grab
  1253. 42:44them, try them out, and experiment.
  1254. 42:46Please leave a like, subscribe, comment
  1255. 42:47, and share with your friends, because
  1256. 42:49this was an incredibly useful video.
  1257. 42:50That's it, see you in the next video.
  1258. 42:52Bye.

About this transcript

This page contains the full transcript of Claude и Claude Code: полный гайд от установки до сложных задач by Web3nity, generated from the public captions YouTube serves with the video. The transcript has 7,818 words across 1,258 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.