YouTube2Text

Дай мне 1 час и ты получишь 1000 часов знаний о Codex (2026) — Transcript

by Матвей Шульга · 11,257 words · 1,643 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Today we are going to break down 24 key
  2. 0:02concepts of Codex. This is exactly what
  3. 0:05you need to know to master Codex by 95%
  4. 0:08and start using it today, without
  5. 0:10spending weeks studying it. And it
  6. 0:12doesn't matter if you know how to
  7. 0:14program or not. By the end of this
  8. 0:17video, you will understand how Codex
  9. 0:19works, and you will have a fully
  10. 0:20functional system that will work even
  11. 0:22while you sleep. Let's get started.
  12. 0:25Before the first part, let's first take
  13. 0:27care of the installation and purchasing
  14. 0:28a subscription. To download the Codex
  15. 0:31application to your computer, type "
  16. 0:33Codex download" into your browser and
  17. 0:35go to the first link from openai.com.
  18. 0:38As soon as you enter the site, there
  19. 0:39will be a download button on the main
  20. 0:41block. Click on this button and
  21. 0:43download it. There will be no
  22. 0:44difficulty with the installation of the
  23. 0:46application itself. The first problem
  24. 0:48that may arise is registering a new
  25. 0:50account. If you are registering in
  26. 0:52Codex for the first time, you will be
  27. 0:54asked for a phone number. And this
  28. 0:56phone number, naturally, must be a
  29. 0:59foreign one. If you don't have a
  30. 1:00foreign phone number, there are two
  31. 1:02options. The first is to find a service
  32. 1:05for renting foreign numbers. And the
  33. 1:07second option, the most reasonable one,
  34. 1:09is to find an acquaintance who has a
  35. 1:11foreign phone number. Well, another
  36. 1:13option is to buy an account on Avito,
  37. 1:15for example, but that's probably not a
  38. 1:17very reliable thing to do. I wouldn't
  39. 1:18recommend doing that. That's it, you
  40. 1:20verify it once, and as a rule, it
  41. 1:22doesn't ban you and doesn't ask for the
  42. 1:24number again. At least, as of the time
  43. 1:26of recording this video. After you
  44. 1:28install Codex on your computer, you
  45. 1:30will need to choose a paid plan. If you
  46. 1:33need Codex, then only a Plus
  47. 1:34subscription or higher will work. The
  48. 1:37Plus plan has all the most powerful
  49. 1:39models, has Codex, and you can already
  50. 1:41fully develop something with it. The
  51. 1:44only limitation of this plan is its
  52. 1:46limits. If you are going to use some
  53. 1:49powerful models or handle some large
  54. 1:51tasks, these limits will not be enough
  55. 1:52for you. Therefore, for those who want
  56. 1:56to work fully and powerfully with Codex
  57. 1:58, you need to buy the Pro plan for $
  58. 2:00100 or $ 200. On the $ 100 Pro plan,
  59. 2:02you will have five times more limits
  60. 2:04than on the Plus plan. Previously, the
  61. 2:07$ 200 Pro plan had 20 times more limits
  62. 2:09than Plus, but now it has 10 times more
  63. 2:11limits than Plus. And recently, they
  64. 2:14also rolled out a $ 500-a-month plan. I
  65. 2:17don't know what kind of rich people
  66. 2:19this plan is for, but if you are
  67. 2:21working with some large tasks, really
  68. 2:23very large ones, then you might need it
  69. 2:25. On this plan, you have 25 times more
  70. 2:28limits than on Plus. And there's also a
  71. 2:31fast mode added, so you spend less time
  72. 2:33on a task. Meaning the model will work
  73. 2:37up to eight times faster, but you'll
  74. 2:38hit those limits much more often. We
  75. 2:41will talk about tokens, limits, and
  76. 2:43context a bit later. In short, if you
  77. 2:45are just starting to learn AI coding
  78. 2:47and trying to develop something, then
  79. 2:48get the Plus plan first. Now for the
  80. 2:51next problem. How do we actually buy
  81. 2:53this plan? To do this, you need a
  82. 2:56foreign bank card. To get one, you can
  83. 2:59either ask a friend abroad to make the
  84. 3:01payment or get yourself a virtual
  85. 3:03foreign card. I pay for my
  86. 3:05subscriptions using a Telegram bot, the
  87. 3:07link to which will be in the
  88. 3:08description. There are, of course, fees
  89. 3:10, but thanks to this bot you will still
  90. 3:12be able to pay for subscriptions. And
  91. 3:15one more alternative way to buy a
  92. 3:17subscription is to use the GGcell
  93. 3:19service or any other equivalent. With
  94. 3:21this service, you can buy a
  95. 3:22subscription using rubles. That's all.
  96. 3:25After we have registered a new account,
  97. 3:27installed the desktop application,
  98. 3:29selected a plan, and purchased it, we
  99. 3:32can finally start using our Cursor. So,
  100. 3:36concept number one is that Cursor and
  101. 3:38ChatGPT are different things. You can
  102. 3:41notice that we have ChatGPT, which has
  103. 3:44two modes inside: Chat and Work. And
  104. 3:47right here in the left panel, you can
  105. 3:49switch from ChatGPT to Cursor. In total
  106. 3:52, the application has three modes. The
  107. 3:53first one is Chat. This is our usual
  108. 3:56ChatGPT. Meaning we can ask it
  109. 3:58something here, find something, ask it
  110. 4:00to draft texts, and so on. Then there
  111. 4:03is the Work mode. This is a tool you
  112. 4:06can use to work with your computer,
  113. 4:07perform certain tasks, make documents,
  114. 4:09presentations, and a bunch of other
  115. 4:11things. We will talk about this in more
  116. 4:13detail later. And, accordingly, for
  117. 4:15working with code, for creating
  118. 4:18projects, automations, services, and
  119. 4:20websites, we have Cursor. Just so you
  120. 4:23know, the Work mode uses up your Cursor
  121. 4:25limits, but the Chat mode does not.
  122. 4:27Meaning you can talk to it for as long
  123. 4:29as you want, and your Cursor limits
  124. 4:31won't be used up. Concept number two.
  125. 4:33Cursor doesn't work in thin air; it
  126. 4:36works with a folder on your computer.
  127. 4:39The first thing we must do before
  128. 4:40working in Cursor is to create a folder
  129. 4:42on the desktop. So, I created a folder,
  130. 4:45named it Cursor, and after that, I go
  131. 4:48into Cursor, select the necessary tab,
  132. 4:50and click New Chat. Now we are in a new
  133. 4:53chat. And here we click Select Project.
  134. 4:56Next, click New Project, Add Folder
  135. 4:58from this computer, and select the
  136. 5:00corresponding folder. Click Open. Next,
  137. 5:03click create project. After that, it
  138. 5:05will ask if we trust this folder. We
  139. 5:07trust it. Codex will see everything
  140. 5:10inside this folder. All your files, all
  141. 5:13your spreadsheets, all your photos,
  142. 5:15anything at all. Concept number three
  143. 5:19is that we choose a specific model for
  144. 5:21a specific task. Choosing a model
  145. 5:24directly depends on how you spend your
  146. 5:26tokens and your limits. Let me explain
  147. 5:29what tokens and limits are, so you
  148. 5:30understand. A token is roughly a piece
  149. 5:33of a word. It can be a whole word, a
  150. 5:35part of a word, or some preposition.
  151. 5:37Our subscription consists of these
  152. 5:39tokens. Our subscription gives a token
  153. 5:42limit for a week or for 5 hours. The
  154. 5:45Pro plan does not have this five-hour
  155. 5:47window. It only has a weekly limit.
  156. 5:49That is, you are given a certain amount
  157. 5:51of limits for the week. And if you go
  158. 5:54over these limits, you won't be able to
  159. 5:55use Codex for the rest of the week.
  160. 5:58There is such a concept as input tokens
  161. 6:00. These are the tokens that you write
  162. 6:01to it. That is, all the text that you
  163. 6:03send to it. And not just text. Images
  164. 6:06also consume tokens. Any working files
  165. 6:08also consume tokens. In general,
  166. 6:11everything you send in your request is
  167. 6:13input tokens. And there are also output
  168. 6:16tokens. This is the model's response.
  169. 6:18Moreover, tokens are spent not only on
  170. 6:20the model's output that it writes back
  171. 6:23to you, but also on its reasoning. That
  172. 6:26is, while it is answering your question
  173. 6:28, it goes online, researches something,
  174. 6:30studies, looks through other files on
  175. 6:32your computer if they are needed for
  176. 6:34the task, checks some documentation,
  177. 6:36and so on. All of this also consumes
  178. 6:39your tokens. And we can influence all
  179. 6:42of this by choosing a specific model
  180. 6:43and a specific level of reasoning.
  181. 6:45Let's see what models are there and
  182. 6:47what they are used for. If we take the
  183. 6:50latest generation of models, that is
  184. 6:52the GPT6 generation, and 6.1, we have
  185. 6:55several models. These are the Luna
  186. 6:57model, the Soul model, and the Astra
  187. 6:59model. GPT6 Luna is accordingly the
  188. 7:02cheapest and the fastest. It is needed
  189. 7:05to work with simple, specific tasks.
  190. 7:08Reformatting something, renaming,
  191. 7:10adding a line of code, changing a
  192. 7:12button color, and so on. So here we are
  193. 7:15not dealing with any complex, large
  194. 7:17tasks that require deep reasoning. You
  195. 7:19can assign it any routine task that can
  196. 7:22be done easily according to a pattern,
  197. 7:24without applying any mental effort. For
  198. 7:26example, sorting emails, analyzing
  199. 7:29comments. Well, in general, these are
  200. 7:31not very complex, specific tasks. The
  201. 7:34GPT6 Astra model is the most powerful
  202. 7:37and expensive model that currently
  203. 7:39exists from OpenAI. It is needed for
  204. 7:41complex tasks where a high level of
  205. 7:43reasoning is required. For example,
  206. 7:46developing a project idea, analyzing
  207. 7:48project files, or finding a tricky bug
  208. 7:51that has crept into the code. In short,
  209. 7:53it is needed for complex tasks that
  210. 7:55require a lot of thought. And there is
  211. 7:58also GPT 6.1 Sou. It is almost as smart
  212. 8:02as Astra, but five times cheaper. This
  213. 8:04is what you will work on every day. And
  214. 8:06honestly, I use it for 99%of all my
  215. 8:09tasks. The main beginner mistake in
  216. 8:12this matter is choosing the coolest,
  217. 8:14most powerful model and using it to
  218. 8:15solve all your tasks. Why is this bad?
  219. 8:18Firstly, because when performing a task
  220. 8:20with a specific outcome, the Sou model
  221. 8:22will handle it just as effectively as
  222. 8:24the Astra model. But Astra will spend
  223. 8:27more time on it, firstly because it
  224. 8:29simply needs to think, and secondly, it
  225. 8:31will spend a lot of money, or rather,
  226. 8:33your tokens. In addition to this model
  227. 8:35selection, there is also a choice of
  228. 8:37reasoning level here. You see, it
  229. 8:39changes: light, medium, high, very high
  230. 8:42, and ultra. The higher the reasoning
  231. 8:45level you set, the more the model will
  232. 8:47think before answering you. A light
  233. 8:49reasoning level is when the model
  234. 8:51barely thinks about your task. That is,
  235. 8:55we will use it for specific tasks, like
  236. 8:57changing a button color, replacing a
  237. 9:00string with this one, making a specific
  238. 9:02edit, and so on. At a light reasoning
  239. 9:06level, it will perform such specific
  240. 9:08tasks quickly and cheaply. The medium
  241. 9:10level is accordingly higher, meaning
  242. 9:12the model starts thinking before it
  243. 9:14answers you. You can already use it for
  244. 9:17tasks where there is no specific answer
  245. 9:19and some reflection is needed. And,
  246. 9:22accordingly, the high and very high
  247. 9:24reasoning levels are for when you are
  248. 9:26analyzing code, finding bugs, or, for
  249. 9:28example, when you want to analyze a
  250. 9:30huge number of sources, do some
  251. 9:32research, and so on. This means the
  252. 9:35model will work longer and spend more
  253. 9:37of your tokens, but it will thoroughly
  254. 9:39think through your tasks. And there is
  255. 9:41also an ultra mode. This is a
  256. 9:43completely different level of model
  257. 9:44operation, and we will talk about it a
  258. 9:46little later. Concept number five is
  259. 9:48dedicated to permissions. Code-base can
  260. 9:50do various things on your computer.
  261. 9:52Create and delete files, rename files,
  262. 9:55change their location, run programs,
  263. 9:58access the internet, and so on. In
  264. 10:00general, it has a complete carte
  265. 10:02blanche when it comes to working with
  266. 10:03your data. Plus, if you connect a few
  267. 10:06more external services to it, it will
  268. 10:07also have access to them. And, of
  269. 10:10course, blindly trusting Codecks with
  270. 10:12your data is, as you can understand,
  271. 10:14not entirely safe. That is why Codecks
  272. 10:17has a special setting that is
  273. 10:18responsible for all of this. It is
  274. 10:20called permissions. It is located right
  275. 10:22here to the right of the plus sign. And
  276. 10:24there are three options here. The first
  277. 10:26is to request permission. In this case,
  278. 10:29it will ask for your permission for any
  279. 10:31changes to external files or before
  280. 10:33using the internet, that is, if it
  281. 10:35performs any operations online. Then
  282. 10:38there is the "confirm for me"
  283. 10:39permission level, where it will only
  284. 10:41request permission for potentially
  285. 10:43unsafe actions. For example, if you are
  286. 10:46about to delete an important file or
  287. 10:48make a change to the code that could
  288. 10:51affect a huge number of functions, it
  289. 10:53will ask for permission before all such
  290. 10:55edits. And another permission level
  291. 10:58that is here is full access. As you
  292. 11:01understand, it won't ask for any
  293. 11:03permissions at all here. Deleting files
  294. 11:06, moving files, removing or adding
  295. 11:08lines of code. It will do all of this
  296. 11:11without your permission. By default, it
  297. 11:13will most likely not be set to this. To
  298. 11:16activate it, you need to go to settings
  299. 11:18in the bottom left corner, find
  300. 11:20permissions here, and they will be open
  301. 11:21right away. And right here there will
  302. 11:24be a toggle switch, full access. You
  303. 11:26click it here and press confirm. And
  304. 11:28that is it, now you will have full
  305. 11:30access activated. As soon as we try to
  306. 11:32turn it on, it warns us that it is
  307. 11:34dangerous, that something might be
  308. 11:36deleted, or that a terminal command
  309. 11:38might be unsafe, but we click confirm.
  310. 11:41And honestly, I use this mode more
  311. 11:43often than the others. To be more
  312. 11:45precise, I only use this mode. Why am I
  313. 11:47not afraid of this? Modern models have
  314. 11:49already become quite powerful. And
  315. 11:51overall, I have never had a situation
  316. 11:53where Codecks for some reason deleted
  317. 11:55something important from my computer.
  318. 11:58But it is important to craft the
  319. 11:59request correctly here. If you just
  320. 12:02write to it, for example, "clean up my
  321. 12:04desktop," it might delete both
  322. 12:05unnecessary files and important folders
  323. 12:07you need. Or, for example, if the task
  324. 12:10is to clean up your computer to free up
  325. 12:12more memory. Here, too, it might delete
  326. 12:14both junk and various duplicates, as
  327. 12:16well as important, useful applications.
  328. 12:19In short, it is important to provide a
  329. 12:20correct, specific request without any
  330. 12:22double meanings, and then you will not
  331. 12:24have any problems with this level of
  332. 12:26permissions. Why don't I like the first
  333. 12:28two modes? Because it will constantly
  334. 12:30ask you for these permissions. That is,
  335. 12:33if you are working on a big task where
  336. 12:34it needs to think a lot and perform
  337. 12:36many actions, you might give it about
  338. 12:3820 permissions in a single request.
  339. 12:40Honestly, that gets tiring. It slows
  340. 12:42down work significantly, and you
  341. 12:44essentially can't switch to other tasks
  342. 12:46while it's running because you always
  343. 12:49have to click "allow." In short, when
  344. 12:51using full access, it is important to
  345. 12:53assign tasks correctly. Concept number
  346. 12:55six: the limit is a resource. It's time
  347. 12:58to explain how the context window works
  348. 13:00and how your limits are spent. We
  349. 13:02already know that the limit is consumed
  350. 13:04by the chosen model and the depth of
  351. 13:06its reasoning. Now let's find out where
  352. 13:09to see it. This very limit, and we'll
  353. 13:11learn how not to burn through it. I'll
  354. 13:13just type a slash into the input field.
  355. 13:16After entering the slash, a huge number
  356. 13:18of commands pop up, but let's find "/
  357. 13:20status." Here it is. We click on it.
  358. 13:23And you can notice that, again, there
  359. 13:24are no five-hour limits here. There are
  360. 13:26only weekly ones. And I've only used up
  361. 13:293%of this limit. If for some reason you
  362. 13:32hit the limit, OpenAI sometimes gives
  363. 13:35you the option to reset these limits.
  364. 13:38This is done in the bottom left corner,
  365. 13:40right here in the "Usage" tab. Let's go
  366. 13:42there. And right here, as you can see,
  367. 13:44I have two resets available. One
  368. 13:46expires on the 22nd, the other on
  369. 13:48October 29th. If for some reason I hit
  370. 13:51the weekly limit, I can easily restore
  371. 13:53these limits using this button. OpenAI
  372. 13:56provides this as a bonus. You can't get
  373. 13:59these resets just like that. How can we
  374. 14:01save these limits? The first rule you
  375. 14:05can apply is to brainstorm in ChatGPT
  376. 14:08and then work directly in Codex. As I
  377. 14:11said, regular ChatGPT does not affect
  378. 14:13the Codex limit. It has its own
  379. 14:15restrictions, but that's not the point.
  380. 14:16Again, this doesn't apply to Codex at
  381. 14:18all. So, if you want to discuss an idea
  382. 14:21, outline a structure, or ask questions
  383. 14:24before starting a task, you can do all
  384. 14:26of that using ChatGPT. After you've
  385. 14:29discussed the task with ChatGPT and
  386. 14:31figured out how to solve it, you can
  387. 14:33turn to Codex. Next, there is another
  388. 14:36rule: one chat, one task. That is, for
  389. 14:39example, you work out the design in one
  390. 14:41chat, you work out something else in
  391. 14:43another, integrations in a third, and
  392. 14:45so on. Why is this important? To
  393. 14:48understand that, we need to look at how
  394. 14:50context works. Let's say we have a chat
  395. 14:53. We give it a prompt, and it replies.
  396. 14:56That is, to answer us, the neural
  397. 14:58network processed our request. Then,
  398. 15:00after that, we input a second prompt.
  399. 15:02And to process it, the neural network
  400. 15:04must process our previous request and
  401. 15:07the answer to that previous request. So
  402. 15:10, in this way, to answer the second
  403. 15:11prompt, it is already processing more
  404. 15:13information than when it answered the
  405. 15:14first. And, accordingly, when you
  406. 15:18provide so many edits, reason with it,
  407. 15:20or give it new tasks, not only does it
  408. 15:23spend a lot of tokens thinking, but it
  409. 15:26also processes previous answers and
  410. 15:28requests to address the current one.
  411. 15:32And all of this accumulates. And, of
  412. 15:34course, when answering, say, the
  413. 15:36fiftieth question, it will spend a lot
  414. 15:38of tokens just processing previous
  415. 15:39answers. But it's not all that strict.
  416. 15:42We have a thing called caching. Here is
  417. 15:45the point. When the model has read the
  418. 15:47beginning of a chat, OpenAI remembers
  419. 15:48what it understood from it. For the
  420. 15:50next message, the beginning hasn't
  421. 15:52changed, so it doesn't need to be read
  422. 15:54again. It simply pulls the information
  423. 15:56from memory. But this caching has
  424. 15:58conditions. The first is time. The
  425. 16:00memory lasts for 30 minutes after your
  426. 16:02last message. Each new message extends
  427. 16:04it, accordingly, for another half hour.
  428. 16:08If you, for example, go away for two
  429. 16:09hours and then want to continue, there
  430. 16:11will be no caching anymore. The second
  431. 16:13point is switching models. If you
  432. 16:15change the model in the middle of a
  433. 16:17chat, that also resets the cache.
  434. 16:20Therefore, in long chats, it is ideally
  435. 16:21better not to change the model. In
  436. 16:24short, caching is a useful thing, but
  437. 16:25it still doesn't solve the problem of
  438. 16:27the context window. By the way, let's
  439. 16:29move on to the next concept. This is
  440. 16:31exactly the context window. That is,
  441. 16:33the model does not have memory in the
  442. 16:35usual sense. It has a context window,
  443. 16:38which is a limited volume that holds
  444. 16:40everything it sees right now. These are
  445. 16:43the instructions I give it, my files,
  446. 16:46chat history, command results, and so
  447. 16:48on. All of this together fills the
  448. 16:51context window. If you overload your
  449. 16:54context window, for example, over 70%,
  450. 16:56there is a very high probability that
  451. 16:58the neural network will miss something
  452. 17:00—some important fact you threw in or
  453. 17:03information you provided—and because
  454. 17:05of this, it will start to act slow.
  455. 17:08This is called context degradation. To
  456. 17:11somehow shrink the volume of the
  457. 17:13context window you have used, there is
  458. 17:15a command called compact. This is what
  459. 17:18it looks like. I click here, and it
  460. 17:20compresses the context of this chat.
  461. 17:22Meaning, it creates a separate summary
  462. 17:25from all of our previous chat. Thus,
  463. 17:27the context window size decreases,
  464. 17:29meaning we can work with it fully, but
  465. 17:31it might accidentally remove an
  466. 17:33important fact, and your performance
  467. 17:35will still be much worse. Therefore, it
  468. 17:38is ideal to assign one task per chat.
  469. 17:40Actually, there are many life hacks for
  470. 17:42saving tokens. I have detailed them in
  471. 17:44this video. Feel free to check it out.
  472. 17:48You might ask me: "How do I constantly
  473. 17:50work in a new chat when I keep dumping
  474. 17:52a lot of context, and then it
  475. 17:54disappears when I start a new one?""Do
  476. 17:57I really have to feed it a huge amount
  477. 17:58of information every single time before
  478. 18:00I start working?" My answer: no. And
  479. 18:03now, in the following concepts, we will
  480. 18:05figure out how to work with all of this
  481. 18:07. Concept number nine. It concerns the
  482. 18:10most important file in the codebase. We
  483. 18:13just talked about how every new chat is
  484. 18:15a blank slate. Meaning, the codebase
  485. 18:18doesn't know who you are, what the
  486. 18:19project is, what the project rules are,
  487. 18:21what its structure is, and so on. And
  488. 18:23you explain everything from scratch
  489. 18:24every single time. To avoid explaining
  490. 18:27context a thousand times, there is a
  491. 18:29file called .agents.md. It is a regular
  492. 18:33text file in Markdown format that GPT
  493. 18:35reads at the start of every session, at
  494. 18:37the start of every chat. So that we
  495. 18:39don't have to create it ourselves from
  496. 18:41scratch, there is a command called "
  497. 18:43init". I type it in like this, and "
  498. 18:46initialize" pops up in the suggestions.
  499. 18:48This .agents.md file can be global or
  500. 18:51project-specific. The global .agents.md
  501. 18:53affects every single chat throughout
  502. 18:55the entire codebase. So, regardless of
  503. 18:57which chat or which folder you enter,
  504. 18:59these rules will apply. And there is a
  505. 19:01project-specific .agents.md—this is a
  506. 19:04file that only acts within the folder
  507. 19:06or the project where it is located. So,
  508. 19:09while you can put global settings like
  509. 19:11"answer in simple language,""write in
  510. 19:14Russian," or anything else in the
  511. 19:15global one, in the project .agents.md
  512. 19:18you write specific rules for your
  513. 19:19project. What is important to
  514. 19:22understand is that this file must be
  515. 19:24short. Perhaps this doesn't seem to fit
  516. 19:26with the idea of needing to load a huge
  517. 19:28amount of company information there,
  518. 19:30but I will explain how to handle it
  519. 19:32best. If it is included in every chat
  520. 19:34and every session, even unnecessary
  521. 19:37information will clutter the chat. For
  522. 19:40example, you might be working on design
  523. 19:42, but information about the structure,
  524. 19:44backend, or some functionality is also
  525. 19:46there. All of that also burdens the
  526. 19:49context window and wastes tokens. To
  527. 19:51avoid this, we create several files
  528. 19:53with different information. For example
  529. 19:56, one for the design system, one for
  530. 19:58the backend, one for the structure, and
  531. 20:00so on. And we use the Agent MD as a
  532. 20:02navigator, as a coordinator. That is,
  533. 20:06for example, in the project's Agent MD,
  534. 20:07you can write lines like: for the
  535. 20:09design system, go to files/design.md.
  536. 20:12For the structure of my pages, go to
  537. 20:14files/structure.md, for example. So, in
  538. 20:16the project's Agent MD, you can put a
  539. 20:19few words about what we are working
  540. 20:21with and the paths to the files where
  541. 20:23specific information is located, and
  542. 20:25our chat will load the information as
  543. 20:28needed. This way, tokens are saved, and
  544. 20:30we adequately immerse our chat into the
  545. 20:33context of our project. This is a very
  546. 20:35important rule that needs to be
  547. 20:36understood. Concept number 10 is memory
  548. 20:38. Cursor has a feature called Memory.
  549. 20:41By default, it is turned off. And to
  550. 20:44turn it on, you need to go into
  551. 20:45settings. So, we go into Settings, then
  552. 20:47Personalization. And right here, there
  553. 20:49is Cursor Memory. What is it and how
  554. 20:51does it work? You don't specifically
  555. 20:53tell it: "Remember this or that
  556. 20:54information." Cursor automatically
  557. 20:57accesses a chat you haven't touched for
  558. 20:59several hours and extracts useful
  559. 21:00things from it, like your preferences,
  560. 21:02agreements, how the project is
  561. 21:04structured, and so on. It puts all of
  562. 21:07this into a Memories folder inside the
  563. 21:09Cursor folder. I, to be honest, don't
  564. 21:11really like Memories, because it can
  565. 21:12write any kind of nonsense in there.
  566. 21:15Meaning, you have to keep track of what
  567. 21:16it writes and then fix all of it. Why
  568. 21:19do we need this extra hassle when we
  569. 21:21have Agent MD, when we have all these
  570. 21:23other project MD files where we record
  571. 21:25all the necessary context? In short, I
  572. 21:28don't really like this feature, so I
  573. 21:30turn it off. Concept number 11 is
  574. 21:31dedicated to something just as
  575. 21:33important as Agent MD. That is Git.
  576. 21:36What are we dealing with? Cursor is
  577. 21:38constantly changing files, and
  578. 21:39sometimes not very successfully. For
  579. 21:42instance, you are working on a project
  580. 21:44and decide to implement some new
  581. 21:46function, a new feature. And at the
  582. 21:49moment you try to implement this new
  583. 21:50feature into your project, for some
  584. 21:52reason, everything stops working. That
  585. 21:55is, for example, this feature affected
  586. 21:57some important functionality, some
  587. 21:58important logic, and because of this,
  588. 22:00everything in your project went haywire
  589. 22:02. So, if you don't save successful
  590. 22:04versions of your project, you can very
  591. 22:06easily lose what you have been working
  592. 22:08on for so long. To save our versions,
  593. 22:12there is Git, a program that saves
  594. 22:14versions of a folder. Each saved
  595. 22:17version is called a commit. Just a
  596. 22:19fancy term. In fact, it is just a save.
  597. 22:22How does it work? For example, you have
  598. 22:24a project open where you are developing
  599. 22:25your service, website, or anything else
  600. 22:27. You just write a prompt for it. "
  601. 22:30Connect Git to this folder and make the
  602. 22:32first commit," and you hit Enter. There
  603. 22:34is nothing here, of course, but an
  604. 22:36empty folder is also a commit. Git is
  605. 22:38connected to the folder, the main
  606. 22:39branch is created, and the first commit
  607. 22:41is made. The commit is empty because
  608. 22:42there are no files in the folder yet.
  609. 22:44That’s it, now our first empty
  610. 22:46version is saved. Let’s say we
  611. 22:48implemented some important feature, and
  612. 22:50we succeeded. After we succeed, we
  613. 22:52write directly to Codeix to make a
  614. 22:54commit. A commit saves the state of the
  615. 22:57folder, the state of the files, and
  616. 22:59creates a description. That is, it
  617. 23:01describes what has changed in this
  618. 23:03folder. Thus, if you want to roll back,
  619. 23:05that is, return to a previous version
  620. 23:07of your project, you can do so thanks
  621. 23:09to the description, and ideally, you
  622. 23:11should make commits after every working
  623. 23:13piece. That is, you implemented a
  624. 23:16feature, made a commit, finally set up
  625. 23:18an integration, and made a commit. This
  626. 23:21way, you won’t lose your project
  627. 23:23during any unsuccessful changes. But
  628. 23:25there is one important flaw here. All
  629. 23:27these saves are located locally on your
  630. 23:29computer. That is, if for some reason
  631. 23:32your computer breaks or drowns, I
  632. 23:34don’t know, whatever happens to it,
  633. 23:36you will never recover this project
  634. 23:38because all its files are on your
  635. 23:40computer. To avoid such a potential
  636. 23:43problem, there is a service called
  637. 23:45GitHub. This is what this site looks
  638. 23:47like. Again, you just register on it,
  639. 23:49there is no problem at all. To create a
  640. 23:51repository, you press this plus sign on
  641. 23:53the top panel and click "create new
  642. 23:56repository." Here you come up with a
  643. 23:57name for it. You can write anything
  644. 23:59here in Latin characters. Next, you
  645. 24:01choose a public or private project. A
  646. 24:03public one will be available to
  647. 24:05everyone who can access this repository
  648. 24:06. Let’s choose private. We don't
  649. 24:08change anything here. We click "create
  650. 24:10repository." Done. Now we have a
  651. 24:12repository where our project files will
  652. 24:14be stored. We click the "Copy" button
  653. 24:16right here. We go into Codeix. Let’s
  654. 24:18ask it to create some simple page. "Can
  655. 24:21you create a page with two blocks?"
  656. 24:24Really, any page on any topic. Just not
  657. 24:27related to artificial intelligence and
  658. 24:29technology. Otherwise, it usually does
  659. 24:31the same thing. Let’s see what it
  660. 24:32does this time. We click Enter.
  661. 24:34That’s it, it made some thing about
  662. 24:36the sea. Well, it doesn't really matter
  663. 24:37what it made. This is how it looks. A
  664. 24:39three out of five, let's say. But
  665. 24:41that’s not important to us right now.
  666. 24:42Paste the link to our repository that
  667. 24:44we copied, and type the following: "
  668. 24:47Push all changes to the GitHub
  669. 24:48repository." I hit Enter. By the way, "
  670. 24:52push" just means publish. That's
  671. 24:54another term for you. Now, since you
  672. 24:56haven't connected your GitHub to Cursor
  673. 24:58yet, it will likely provide
  674. 24:59instructions on how to do so, and it
  675. 25:01won't be complicated. Once you do that,
  676. 25:04you'll be able to easily push all your
  677. 25:06project files to the GitHub repository.
  678. 25:09This way, you'll save your versions not
  679. 25:11just locally, but also on the internet.
  680. 25:13There you go, Matvey, all changes have
  681. 25:15been sent to the GitHub repository.
  682. 25:16Let's move on. Once again, paste this
  683. 25:18link into your browser. And now, the
  684. 25:20index HTML is located here. I'll
  685. 25:22quickly show you one more thing; it's
  686. 25:24also a very important story. How to
  687. 25:25easily and quickly deploy your projects
  688. 25:28to the internet. "Deploy" is another
  689. 25:29term, which also means to publish.
  690. 25:31Except "pushing" is publishing changes
  691. 25:33to a repository, while "deploying" is
  692. 25:35publishing to the internet. For this,
  693. 25:38there is a service called Vercel. It's
  694. 25:40also quite easy to sign up for it. Just
  695. 25:42sign in via Google, ideally via GitHub.
  696. 25:45You'll find out why in a moment. Click
  697. 25:47"Add New Project" here. You need to
  698. 25:50link it with GitHub here so it pulls
  699. 25:52your entire list of repositories. You
  700. 25:55select the one you need. Click "Import.
  701. 25:58" Next, consult with Cursor regarding
  702. 26:00the settings on how to configure it for
  703. 26:02publishing. Well, my project is simple,
  704. 26:04so I don't need to configure anything.
  705. 26:06I click the "Deploy" button, and that's
  706. 26:07it. Then, click "Continue to Project."
  707. 26:09Click this link here. And now our
  708. 26:12project is available on the internet.
  709. 26:13This way, you can quickly share your
  710. 26:16projects with clients, colleagues, and
  711. 26:18so on. This, of course, doesn't replace
  712. 26:21a full-scale publication on a server or
  713. 26:23hosting, but for a quick publication to
  714. 26:24show someone a project, why not? The
  715. 26:27next batch of concepts will be
  716. 26:29dedicated to how to write prompts
  717. 26:31correctly and how to set tasks properly
  718. 26:33. The main mistake most people make is
  719. 26:36writing vague prompts. For example, "
  720. 26:39make me a beautiful website." The AI
  721. 26:41doesn't know what "beautiful" means.
  722. 26:43And since it lacks specific references
  723. 26:46or examples from you, it will use what
  724. 26:49it knows. And it knows those typical,
  725. 26:51sloppy AI-generated designs. You can
  726. 26:53also make it easier for the AI to
  727. 26:55understand your task by giving it the
  728. 26:57path to the file it needs to reference.
  729. 26:59For example, don't just say "study the
  730. 27:01design system," but "the design system
  731. 27:03is located here." Type an "@" symbol
  732. 27:04and tag the markdown file containing
  733. 27:06the design system. This way, it will
  734. 27:08specifically understand where to look
  735. 27:10and what to do. Plus, besides text and
  736. 27:13a specific prompt, you can add images,
  737. 27:16that is, references or HTML code. For
  738. 27:19example, you’ve visited a website
  739. 27:20that you find visually appealing. Well,
  740. 27:22let’s say this one. Instead of just
  741. 27:24taking screenshots, you can do
  742. 27:26something even better. Let’s copy the
  743. 27:28link. Paste it into a service called
  744. 27:30View Page Source. The link will be in
  745. 27:32the description. Again, you get the
  746. 27:34HTML code, click the download button,
  747. 27:36and drop this code file into your
  748. 27:38project folder. Now let’s go back to
  749. 27:42our Cursor and type the following.
  750. 27:44That’s what it was called. I press
  751. 27:46Tab and write: "Let’s replicate the
  752. 27:49main block on this site exactly." And
  753. 27:52let’s also take a screenshot just to
  754. 27:54be sure, so it understands exactly what
  755. 27:55it looks like. So, after 5 minutes, it
  756. 27:58finished and did it almost perfectly,
  757. 28:01exactly the same. Let’s take a look.
  758. 28:04There is a slight issue with this black
  759. 28:06bar overlapping the orange arrow shape,
  760. 28:08but otherwise, everything looks almost
  761. 28:10perfect. And, as you can see, because
  762. 28:13we provided the HTML, it didn't just
  763. 28:15copy the image, it actually pulled the
  764. 28:17video from the source. The twelfth
  765. 28:19concept will be dedicated to plan mode.
  766. 28:21That is, when the task is defined or
  767. 28:23the task is small, you can just write
  768. 28:25about it to the neural network. And the
  769. 28:27neural network will do it. But when the
  770. 28:29task is large or there are too many
  771. 28:31files, for example, and it affects a
  772. 28:33large number of files, jumping straight
  773. 28:34to writing code can be a mistake. It
  774. 28:37might interpret the task in its own way
  775. 28:39, do it not the way you need, and
  776. 28:41you’ll have to revert to the previous
  777. 28:43version. To think through the task
  778. 28:46before executing it, there is a
  779. 28:47planning mode. To turn it on, click
  780. 28:51slash and find planning mode right here
  781. 28:54. I click here like this. And now this
  782. 28:56little badge has appeared here,
  783. 28:58indicating that plan mode is enabled.
  784. 29:00In this mode, Cursor will not change
  785. 29:02your project files in any way. It reads
  786. 29:05the files, sees what you have, asks
  787. 29:07some questions, and so on. After you
  788. 29:09answer its questions, it will show you
  789. 29:12a plan. If you are satisfied with the
  790. 29:14plan, you approve it, and it begins the
  791. 29:16work. The next concept will be familiar
  792. 29:18to those who have been working with
  793. 29:20neural networks for a long time. When
  794. 29:21you are working on a task, Cursor works
  795. 29:23step-by-step. It completes a task,
  796. 29:26writes you a report, and you tell it "
  797. 29:28okay, continue." And it starts the next
  798. 29:30stage of the task. So, while working,
  799. 29:33you have to click "continue" a certain
  800. 29:35number of times, which is cool on one
  801. 29:36hand because you can check every
  802. 29:38intermediate step, but it's not great
  803. 29:40in terms of slowing down your work,
  804. 29:41constantly getting distracted, and
  805. 29:43typing the next actions for it. But
  806. 29:46what if the task is large, doesn't
  807. 29:48require your intermediate checks, and
  808. 29:49you just want to set it to run
  809. 29:51overnight and go to sleep? For this
  810. 29:53purpose, they created the "GO" command,
  811. 29:55which is the goal mode. Let me type it
  812. 29:58in. There, we’ve clicked it. Now, a "
  813. 29:59goal" has appeared in the bottom panel.
  814. 30:01What is this mode? You give it not a
  815. 30:03task, but a goal. And Cursor will work
  816. 30:06until it achieves it. It won't stop
  817. 30:08halfway; it will go all the way to the
  818. 30:11end. Well, at least until the point it
  819. 30:13considers to be the finish. A goal is a
  820. 30:15final result that can be verified.
  821. 30:17Therefore, it must contain three things
  822. 30:19: what should be working in the end,
  823. 30:21and how can we tell that it works? And
  824. 30:24what are the boundaries? Meaning, what
  825. 30:25should not be touched. For example, if
  826. 30:27we take the development of an admin
  827. 30:29panel—a tool that allows you to make
  828. 30:31changes to a project without using an
  829. 30:33AI—it would be a bad goal to tell it:
  830. 30:35"Make a cool admin panel." Again, it
  831. 30:37doesn't know what "cool" means. It has
  832. 30:40no metric for "coolness," and it will
  833. 30:42keep improving it forever. So, this is
  834. 30:45an unlimited, non-stop process. Here is
  835. 30:47an example of a normal goal. Create an
  836. 30:48"Orders" section in the admin panel.
  837. 30:50Include a list of all orders with a
  838. 30:52status filter, an individual order page
  839. 30:53, and a button to change the status.
  840. 30:55Work until the verification passes. I
  841. 30:58place a test order on the site and see
  842. 30:59it in the list. I change the status to
  843. 31:02"completed," refresh the page, and the
  844. 31:04status is saved. Do not touch
  845. 31:05authorization or other sections of the
  846. 31:07admin panel. So, it has a verification
  847. 31:09scenario. It will place a test order
  848. 31:11itself, open the admin panel itself,
  849. 31:13and refresh the page itself. If it
  850. 31:15doesn't work, it fixes it and checks
  851. 31:17again. If everything works, then,
  852. 31:19accordingly, the goal is achieved, and
  853. 31:21Cursor stops. One of the downsides of
  854. 31:24this approach, naturally, is that when
  855. 31:26performing a long task, it will be
  856. 31:27working the entire time, meaning it
  857. 31:29will spend our tokens. Accordingly, in
  858. 31:32"goal" mode, it can very easily hit the
  859. 31:35limits. That’s not great, but if
  860. 31:37you're setting a task overnight or just
  861. 31:39setting a big task and going out for a
  862. 31:41walk, then why not? Next concept. Trust
  863. 31:45, but verify. Cursor might say that it
  864. 31:48has done everything you asked. And you
  865. 31:50can take this "done" status with about
  866. 31:5260%confidence. Meaning, it definitely
  867. 31:55messed up somewhere, overlooked
  868. 31:57something, or didn't account for a
  869. 31:58certain scenario, so ideally, you
  870. 32:00should check everything manually. To
  871. 32:02verify what our neural network did, we
  872. 32:05can either use some skills or plugins (
  873. 32:07we'll talk about those soon), or, again
  874. 32:09, just manually go through all the
  875. 32:12scenarios, testing and checking
  876. 32:14everything—every single scenario,
  877. 32:16both successful and unsuccessful. And
  878. 32:19I’m sure that by the end of your
  879. 32:20check, you’ll have a huge list of
  880. 32:21comments to send back to it. And
  881. 32:24unfortunately, if you are working on
  882. 32:26large projects or a big task, you will
  883. 32:29have to perform these tests and checks
  884. 32:31yourself. You can, of course, dump it
  885. 32:34all on the neural network, but you’ll
  886. 32:35highly likely end up with errors. In
  887. 32:38reality, AI coding is one thing, but
  888. 32:40when you work on large projects or
  889. 32:42commercial orders, you can run into
  890. 32:44major problems without understanding
  891. 32:46how it all works under the hood. You
  892. 32:50could ruin a project for a client, cost
  893. 32:52them money, overlook some legal aspect
  894. 32:54that causes them trouble, and so on.
  895. 32:58Even though creating projects has
  896. 32:59become easier with the arrival of
  897. 33:01neural networks, without understanding
  898. 33:02the stages and the final outcome, it
  899. 33:04will be very hard for you to work and
  900. 33:06monetize this skill. Actually, on this
  901. 33:08very topic, I recently finished my
  902. 33:10first training cohort. Somewhere around
  903. 33:12here there will be testimonials for
  904. 33:14this cohort, and a link in the
  905. 33:15description to the results of this
  906. 33:17first group. There, I was teaching the
  907. 33:19guys how to create AI-coded projects
  908. 33:21from scratch, from writing the tech
  909. 33:22specs to deploying to a server, where
  910. 33:24we worked through every stage of
  911. 33:26project creation. That means design,
  912. 33:29integrations, payment systems, delivery
  913. 33:31services, security audits, SEO audits,
  914. 33:34refactoring, debugging, deployment,
  915. 33:36post-deployment maintenance, the legal
  916. 33:39side, and sending notifications to
  917. 33:41Roskomnadzor. Plus, we studied client
  918. 33:44work in great detail, meaning how to
  919. 33:46properly sign a contract, how to sign
  920. 33:48an acceptance act, how to draft a
  921. 33:49commercial proposal, and what a brief
  922. 33:51is. In short, it’s a very intense
  923. 33:53course. I still haven’t fully
  924. 33:54recovered from it. And right now, I am
  925. 33:55accepting applications for the second
  926. 33:57cohort. If you want to be part of the
  927. 34:00second cohort, check the description or
  928. 34:01use the QR code on the screen, go into
  929. 34:03the bot, answer three questions, after
  930. 34:05which you'll be added to the waitlist
  931. 34:07and will be the first to hear news
  932. 34:08about the new cohort, as well as
  933. 34:10receive a discount. Anyone who isn't on
  934. 34:13the waitlist won't get the discount.
  935. 34:14The next concept will cover the "Work"
  936. 34:16mode. This is the same mode found in
  937. 34:19the ChatGPT tab. It’s not Claude, but
  938. 34:22it’s a convenient tool that you can
  939. 34:24use even on your phone. So, we open a
  940. 34:27new chat and select "Work" mode. What
  941. 34:30is it? Essentially, it’s the same
  942. 34:32agent as Claude. The same engine, the
  943. 34:35same models, the same limits. With it,
  944. 34:38you can do more than just chat; you can
  945. 34:40work with folders on your desktop,
  946. 34:42create files, documents, presentations,
  947. 34:44and so on. Unlike Claude, this thing is
  948. 34:47also available in the web version.
  949. 34:49Let’s give it a small task here. For
  950. 34:52example, putting together a
  951. 34:53presentation. Why is GPT cool compared
  952. 34:56to Claude? Because Claude doesn't know
  953. 34:58how to generate images. Accordingly, it
  954. 35:00cannot create illustrations for a
  955. 35:02presentation on its own. GPT, however,
  956. 35:05has no problems in this regard, and it
  957. 35:07can create much more interesting and
  958. 35:09engaging presentations for users.
  959. 35:11Let’s enter a prompt like this, for
  960. 35:13example. So, I’ve given it access to
  961. 35:15a folder, and the files it needs are
  962. 35:17inside that folder. Analyze the files
  963. 35:19in the product.mstyle.mdstructure
  964. 35:21folder. Create an eight-slide
  965. 35:24presentation in the style of style.md.
  966. 35:27There are no facts about the model in
  967. 35:29the folder. Find them on the internet.
  968. 35:30I asked it to make a presentation for
  969. 35:32Apple Watch. And these files were also
  970. 35:34generated by the neural network. Well,
  971. 35:36it’s just for testing purposes. List
  972. 35:37the sources on the last slide. For each
  973. 35:40slide, generate one minimalist
  974. 35:42illustration in the colors of style.md
  975. 35:44without people and without the Apple
  976. 35:46logo. Why without the Apple logo?
  977. 35:48Let’s remove that. Before finishing,
  978. 35:50check that every number on the slides
  979. 35:52is in the sources and the titles are no
  980. 35:54longer than five words. Do not modify
  981. 35:56the files in the folder; save the
  982. 35:57presentation alongside them. In fact,
  983. 35:59thanks to connectors, we can do more
  984. 36:00than just save the presentation locally
  985. 36:02; for example, we can have it sent to
  986. 36:04me via Google Mail. Let’s write that:
  987. 36:06"Duplicate the file of this
  988. 36:08presentation to my Google Mail." I will
  989. 36:10explain how to connect Google Mail and
  990. 36:12other connectors a bit later. Actually,
  991. 36:14why wait? Go to plugins. And right here
  992. 36:17is the popular Google Mail. Go to the
  993. 36:20plugin and install it; I already have
  994. 36:21it installed, so there’s no "install"
  995. 36:23button here. And that's it. Now let’s
  996. 36:25see what it comes up with. Okay, it's
  997. 36:27all finished. It took 10 minutes and 48
  998. 36:30seconds. Matvey, the presentation is
  999. 36:32ready. Eight slides, eight
  1000. 36:33illustrations, verified numbers and
  1001. 36:35sources. On the last slide, the
  1002. 36:36original files remain unchanged. A copy
  1003. 36:39has been sent as an attachment to my
  1004. 36:41email. And overall, it turned out quite
  1005. 36:43well. Obviously, it’s not a design
  1006. 36:45masterpiece. Well, because there wasn't
  1007. 36:47even a proper task description here.
  1008. 36:49But it generated the images more or
  1009. 36:50less okay. Although, these two slides
  1010. 36:52are super weird. What even is that? It
  1011. 36:54included some links here, which you can
  1012. 36:57actually click on. And overall, if I
  1013. 36:59had given it a proper task and at least
  1014. 37:01shown what the watch looks like, it
  1015. 37:03would have been pretty good. And let's
  1016. 37:05also open my Google mail. Here is a
  1017. 37:08message I received from myself. And
  1018. 37:10there you go, Matvey, the finished
  1019. 37:12presentation based on your files is in
  1020. 37:14the attachment. In short, with this
  1021. 37:16tool, you can work quite effectively
  1022. 37:18with presentations, documents, files on
  1023. 37:20your computer, and so on. You can, for
  1024. 37:22example, ask it to take and sort your
  1025. 37:24downloads folder. You just select work,
  1026. 37:27select downloads here, click new
  1027. 37:30project, add folder, downloads folder,
  1028. 37:32and ask it to sort, for instance, some
  1029. 37:35files. Please sort all the duplicates
  1030. 37:38that are in the downloads folder and
  1031. 37:40delete them. That’s it. This way, you
  1032. 37:42can easily clean up your downloads
  1033. 37:44folder. But since you have Cursor, you
  1034. 37:46don't really need this GPT Work at all.
  1035. 37:48That is, you can comfortably use Cursor
  1036. 37:50and perform all the same tasks. Concept
  1037. 37:53number 16. Turn a process into a skill.
  1038. 37:56What are skills? For example, we just
  1039. 37:59made a presentation together. It turned
  1040. 38:01out quite well, but still not perfect.
  1041. 38:04Somewhere images were small, somewhere
  1042. 38:06we didn't like the font, and so on.
  1043. 38:09Ideally, we should have clarified the
  1044. 38:11style, asked for a certain image size,
  1045. 38:13requested a specific aesthetic, and so
  1046. 38:15on. And let's say in a week we need the
  1047. 38:18same kind of presentation, well, for
  1048. 38:20another product. What should we do? We
  1049. 38:22have two options. Either we describe
  1050. 38:25the whole process to it all over again,
  1051. 38:27what kind of images it should make,
  1052. 38:28what fonts, and all that. Or you can
  1053. 38:31create a skill, that is, an instruction
  1054. 38:34on how to approach creating
  1055. 38:36presentations. Essentially, it's a
  1056. 38:38regular markdown file that just states
  1057. 38:40in text what the neural network needs
  1058. 38:41to do. A skill has its own name. So,
  1059. 38:45for example, we type a slash and the
  1060. 38:47name of this skill. And that is how we
  1061. 38:49use the skill in our work. Then the
  1062. 38:52skill has a description. For example,
  1063. 38:55we created this skill and labeled it as
  1064. 38:57a skill for bringing our presentations
  1065. 38:59up to brand standards. And we will also
  1066. 39:01add that we should use it when we are
  1067. 39:03dealing with presentations. And the
  1068. 39:06moment we ask it to make a presentation
  1069. 39:08, it will read the skill's description,
  1070. 39:10understand that it needs to be used,
  1071. 39:12and will use it. So, we don't
  1072. 39:14necessarily have to specify this skill
  1073. 39:16with a slash, but the description must
  1074. 39:17be such that the skill is definitely
  1075. 39:19pulled in. Therefore, in this regard,
  1076. 39:21the description needs to be specific.
  1077. 39:23Let's not create a new skill right now;
  1078. 39:25I'll just click slash and show you
  1079. 39:26where they can be found. We scroll down
  1080. 39:29like this and find those very skills
  1081. 39:31here. There is a huge variety of them
  1082. 39:33here. Some skills were in the codebase
  1083. 39:35initially, others I added myself. The
  1084. 39:38set of skills, plugins, and MCPs that I
  1085. 39:40use in my work, I will leave in my
  1086. 39:42Telegram channel. So you can go there
  1087. 39:44via the link in the description or via
  1088. 39:46the QR code on the screen and get all
  1089. 39:48the additional materials. You can
  1090. 39:50create these skills yourself. To find
  1091. 39:53these skills, you can either go to the
  1092. 39:56skills tab, this outlet icon, and
  1093. 39:58select everything you need. You can
  1094. 40:01create a skill or record a skill. That
  1095. 40:03is, you can perform an action right on
  1096. 40:05a screen recording, show how it works,
  1097. 40:08and later the neural network will rely
  1098. 40:10on this and, accordingly, do what you
  1099. 40:12need. This is a bit of a hassle, but in
  1100. 40:15certain situations, it can be useful.
  1101. 40:17And there is simply a button to create
  1102. 40:18a skill. You can click on it. The skill
  1103. 40:20creator is pulled up here. This is a
  1104. 40:22skill for creating a skill. And here we
  1105. 40:24write, for example, I want to create
  1106. 40:26such and such a skill, so that it
  1107. 40:28performs such and such functions and
  1108. 40:30uses such and such tools. That is, you
  1109. 40:33can give a skill not just an
  1110. 40:35instruction, but also a set of tools it
  1111. 40:38will use. For example, the Google Mail
  1112. 40:41MCP, to write messages to email or any
  1113. 40:44other tool. There are a huge number of
  1114. 40:47ready-made skills on the internet that
  1115. 40:49you can use in your work. For example,
  1116. 40:51there is a site called skills. Here you
  1117. 40:54can find skills for every taste. One of
  1118. 40:57the important skills is "find skills,"
  1119. 40:59with which you can install skills from
  1120. 41:00this service. So you just click on the
  1121. 41:03desired skill, select the prompt here,
  1122. 41:06copy it, and paste it into the codebase
  1123. 41:08. That's it, click enter. And this
  1124. 41:10skill is installed for you. The next
  1125. 41:12concept concerns plugins and MCP. We
  1126. 41:15have, of course, already tried to
  1127. 41:16connect Google Mail, so you already
  1128. 41:18know what that is. But let's go over it
  1129. 41:20one more time. MCP is a way to connect
  1130. 41:23your codebase with another external
  1131. 41:25tool, for example, Google Calendar,
  1132. 41:27Google Mail, Teams, BRIX, anything at
  1133. 41:30all. Well, anything that has an MCP;
  1134. 41:33for example, let's go back to the
  1135. 41:35plugins tab right here and find
  1136. 41:37something, let's say, for example,
  1137. 41:39something like Vers. This is the same
  1138. 41:42service we used for our deployment. It
  1139. 41:44redirects us to this page. We give our
  1140. 41:47authorization to connect our Codex to
  1141. 41:49Vercel. Then we go into the application
  1142. 41:52, and that's it, it's loading.
  1143. 41:53Everything is ready, the VS plugin is
  1144. 41:55installed. Now let's go to the project
  1145. 41:58we were working on. To select a
  1146. 42:00specific plugin, we press the@sign and
  1147. 42:02type its name. I clicked Tab to mark it
  1148. 42:06. Deploy this project, which is in my
  1149. 42:08folder, and give me the link right in
  1150. 42:10the chat. In the link, use a name like
  1151. 42:14"test project," but in English. Let's
  1152. 42:17go. So, it told us that the project is
  1153. 42:19deployed and verified. Let's open it
  1154. 42:21and see that it's working. Open in an
  1155. 42:23external browser. The beginning of the
  1156. 42:25link has what I asked for. Test project
  1157. 42:27. And, accordingly, the site itself
  1158. 42:30opens up. We can, for example, use
  1159. 42:32several MCPs at once. That is, ask it
  1160. 42:34to deploy the project we are developing
  1161. 42:36to Vercel and, for example, send it
  1162. 42:38somewhere via email. In general, with
  1163. 42:41these connectors, you can create great
  1164. 42:43integrations between Codex and various
  1165. 42:45tools. And besides MCPs, which are
  1166. 42:48these integrations with applications,
  1167. 42:50there are also just regular plugins. A
  1168. 42:52plugin is essentially a set of skills
  1169. 42:54and tools that solve a specific task.
  1170. 42:57For example, Codex Security. There it
  1171. 43:00is. This is a plugin for checking our
  1172. 43:03code for security issues. It has
  1173. 43:05several built-in skills. As you can see
  1174. 43:07, there are 15 of them. You can choose
  1175. 43:09which ones to use and which ones not to
  1176. 43:11use. And if we click, we can read about
  1177. 43:13each skill. Concept number 18 is
  1178. 43:15dedicated to working in the browser.
  1179. 43:17Many services actually don't have MCP
  1180. 43:20or AI, meaning the neural network can't
  1181. 43:22interact with them at all. Previously,
  1182. 43:25this was a dead end, but now Codex can
  1183. 43:27work in the browser. This means it can
  1184. 43:29visit a site itself and click on things
  1185. 43:31. To use Codex in Chrome, for example,
  1186. 43:35we go to the computer use settings. And
  1187. 43:38right here, there is an extension for
  1188. 43:40Chrome. Click "Install." Click "Add to
  1189. 43:43Chrome" and install it. Next, you log
  1190. 43:46into your account and follow the short
  1191. 43:48instructions here. That is, open the
  1192. 43:51extension in the sidebar, pin it if you
  1193. 43:54need to, and try GPT with real work
  1194. 43:56tasks. Then go to GPT. And you can
  1195. 43:59notice that Chrome is installed. Now we
  1196. 44:02can ask it to do the following. I don't
  1197. 44:04know, I haven't come up with any task.
  1198. 44:05Let's make it something like: you can
  1199. 44:08go into Claude via the Chrome browser
  1200. 44:10and write something in the chat, like:
  1201. 44:13"Haha, Matvey's computer has been taken
  1202. 44:15over by ChatGPT." So, as you can see,
  1203. 44:18this glowing cursor has appeared here.
  1204. 44:21That is our GPT. I mean, you see, it
  1205. 44:23clicked the "Send" button all by itself
  1206. 44:24. This way, you can connect GPT to
  1207. 44:27various tools in the browser. This way,
  1208. 44:31by the way, you can test your
  1209. 44:32application, meaning you can directly
  1210. 44:34ask it to go through all the tabs,
  1211. 44:35click all the buttons, test everything,
  1212. 44:37and make sure everything is actually
  1213. 44:39working. Or if something isn't working,
  1214. 44:42it can provide feedback, create a
  1215. 44:44report, and then you can go through it
  1216. 44:45and make corrections. But, honestly,
  1217. 44:48sometimes it's very slow. I mean, there
  1218. 44:52are certain tasks that are easier to do
  1219. 44:53yourself than to wait for the neural
  1220. 44:55network to figure things out and do
  1221. 44:57what's needed. Plus, it can consume a
  1222. 44:59fairly large amount of tokens because,
  1223. 45:01most likely, it will be taking
  1224. 45:03screenshots along the way, comparing
  1225. 45:05things, and all of that will take a
  1226. 45:06very long time. And it can happen that
  1227. 45:08with some incorrect prompts, the neural
  1228. 45:10network might wander where you didn't
  1229. 45:12ask it to go. Therefore, you need to be
  1230. 45:14careful here as well. Concept number 19
  1231. 45:16. Distribute work to copies, that is,
  1232. 45:18sub-agents. Up to this point, the code
  1233. 45:20was working alone, meaning essentially
  1234. 45:22it had one chat, and it performed all
  1235. 45:24tasks sequentially. Sub-agents are when
  1236. 45:27a chat launches several copies of
  1237. 45:28itself; each takes its own piece, and
  1238. 45:30they all work simultaneously. This way,
  1239. 45:33you can parallelize several tasks at
  1240. 45:35once. I mean, if previously three
  1241. 45:37ten-minute tasks took you half an hour,
  1242. 45:39that is 10 minutes, 10 minutes, 10
  1243. 45:41minutes, now it's all distributed, for
  1244. 45:43example, among three sub-agents and
  1245. 45:45completed in 10 minutes. You can launch
  1246. 45:48them simply via a prompt. For example,
  1247. 45:50check our site for SEO and security.
  1248. 45:53Break the task down into two sub-agents
  1249. 45:54. Let one check the site for SEO and
  1250. 45:57the second for security. Well, let's
  1251. 45:59send our site to it. There's nothing to
  1252. 46:01check there, of course, but I just want
  1253. 46:03to show how it works. You see, it says
  1254. 46:05that it is launching two sub-agents.
  1255. 46:07One is called SEO Audit, the other is
  1256. 46:08Security Scan. And here our sub-agent
  1257. 46:11has appeared with an icon. And this
  1258. 46:14pink one is the Security Audit. You can
  1259. 46:16open any agent and see what it's doing.
  1260. 46:18For example, the SEO Audit. It's doing
  1261. 46:20something there, keeping busy. What
  1262. 46:22else is the advantage? Besides speed,
  1263. 46:25each sub-agent has its own separate
  1264. 46:27context window. That is, its work does
  1265. 46:30not burden our shared context window.
  1266. 46:32It works in a separate space, and only
  1267. 46:34the result of the sub-agent's work
  1268. 46:36enters the main context window. Claude
  1269. 46:39has a feature in its skills called "
  1270. 46:41context fork," which means the task is
  1271. 46:43performed in a separate context window.
  1272. 46:45For some reason, Codeium doesn't have
  1273. 46:47this, so only sub-agents handle it here
  1274. 46:48. But one must understand that when we
  1275. 46:51break a task into multiple sub-agents
  1276. 46:53—and there are modes where there can
  1277. 46:56be 50, 100, or 150 sub-agents—tokens
  1278. 46:58will naturally be consumed at an insane
  1279. 47:01rate. And, by the way, the "Ultra"
  1280. 47:03reasoning level we discussed is exactly
  1281. 47:05what forces the neural network to
  1282. 47:07create these sub-agents. In short,
  1283. 47:10thanks to these sub-agents, we can
  1284. 47:12parallelize a task across multiple
  1285. 47:14executors. This way, we will complete
  1286. 47:16our work much faster. The next concept
  1287. 47:19is dedicated to automations. Everything
  1288. 47:21we have done so far was triggered
  1289. 47:22personally by you. That is, you write
  1290. 47:25to Codeium, and Codeium does it. But
  1291. 47:28Codeium has a special feature that
  1292. 47:30allows you to run automations, i.e.,
  1293. 47:32scheduled tasks. It is called "
  1294. 47:35Scheduled," or "
  1295. 47:35запланированные" in
  1296. 47:37Russian. This is how it looks in the
  1297. 47:39left panel. Those little clocks there.
  1298. 47:41Click here. New task. Let's make it so
  1299. 47:44that every day at 9:00 AM, for example,
  1300. 47:47Moscow time, it sends a post to my
  1301. 47:49Telegram with an image and a text
  1302. 47:51description of the latest news in the
  1303. 47:53world of artificial intelligence for
  1304. 47:55that day. To do this, I need to get the
  1305. 47:59bot's API key and my user ID. Let's do
  1306. 48:02it. Go into Telegram and type "
  1307. 48:04BotFather" here. It's a bot; the link
  1308. 48:07to it will be in the description. Click
  1309. 48:08"Open." Let's create a new bot. Let's
  1310. 48:11name it "newsmaker." Doesn't matter
  1311. 48:14here. And here: news_news_matvey. And
  1312. 48:18it must end with the word "bot." So I
  1313. 48:21type "bot" and click "create bot." And
  1314. 48:24here is its API token. I copy it. To
  1315. 48:29get my user ID so that this bot sends
  1316. 48:31messages specifically to me, I need to
  1317. 48:33go to a bot called "userinfobot." The
  1318. 48:36link to it will also be in the
  1319. 48:37description. Send it any message, and
  1320. 48:40it will give me my user ID. Just before
  1321. 48:43creating the automation, you need to
  1322. 48:44click the "Start" button in the bot,
  1323. 48:46because without that, it won't be able
  1324. 48:48to send messages to my personal account
  1325. 48:50. That's it, now we can insert the task
  1326. 48:52for it. Copy it and paste it. Find the
  1327. 48:55top five news items of the week from
  1328. 48:56the world of artificial intelligence
  1329. 48:57and AI coding on the internet. New
  1330. 49:00models: OpenAI, Anthropic, Google
  1331. 49:01releases, and Codeium/Cursor updates.
  1332. 49:04Choose the most significant one from
  1333. 49:06them, and generate an image. A cozy
  1334. 49:08kitten sitting in an armchair and
  1335. 49:09reading a newspaper. The newspaper is
  1336. 49:11facing the viewer. On the front page, a
  1337. 49:13large headline of one to three words
  1338. 49:15about the biggest news. For example,
  1339. 49:17GPT7 is out. The headline must be
  1340. 49:19legible and fit perfectly into the
  1341. 49:21newspaper. The style should be a warm
  1342. 49:23illustration without any company logos.
  1343. 49:25Send it to Telegram via a bot in a
  1344. 49:26single message. An image and a summary
  1345. 49:29in the photo caption. The caption
  1346. 49:30should be up to 1,000 characters. Five
  1347. 49:32news items, each one to two lines long,
  1348. 49:34with a link, in simple language,
  1349. 49:35without long dashes. That's it. Next,
  1350. 49:37we send the bot token and the chat ID.
  1351. 49:40In general, it is better not to expose
  1352. 49:42your keys like this and put them in a
  1353. 49:44separate file, for example. Let's
  1354. 49:47choose GPT6 Soul as the model for this.
  1355. 49:49Let's make it 61. There is an important
  1356. 49:52setting here: "run on this computer."
  1357. 49:55If it is active, this automation will
  1358. 49:57only work when our computer is turned
  1359. 49:59on and the Codeks app is running as
  1360. 50:01well. That is not very convenient.
  1361. 50:04However, in this situation, Codeks has
  1362. 50:06access to your folders and files. If we
  1363. 50:09run it not on our own computer, then
  1364. 50:11the work will be done in the cloud. It
  1365. 50:14will run using our subscription without
  1366. 50:16any additional costs. And we won't need
  1367. 50:18to turn on the computer. That is, we
  1368. 50:20can turn it off, throw it away, or go
  1369. 50:23somewhere. And this automation will
  1370. 50:25still be executed. The downside is that
  1371. 50:27, in this case, it won't have access to
  1372. 50:29your files on the computer. But here we
  1373. 50:31don't need any files on the computer,
  1374. 50:33so we will naturally do this in the
  1375. 50:35cloud. You can choose to repeat hourly,
  1376. 50:37daily, or set it up to your preference.
  1377. 50:39Time and end of repetition. That is,
  1378. 50:42choose the moment when this automation
  1379. 50:44will finish. Start each run in a new
  1380. 50:46chat. Let's select this so our chat
  1381. 50:47doesn't get cluttered. "Effort: medium"
  1382. 50:49should be enough, I think. Everything
  1383. 50:51seems to be set up now. Let's click "
  1384. 50:53create." All done, your task is
  1385. 50:54scheduled. The next run is on October
  1386. 50:577th at 9:00 AM. Now we can click on the
  1387. 51:00three dots here and click "run now" to
  1388. 51:02check how it works. So, everything is
  1389. 51:05ready. It sent me this post. It
  1390. 51:08considered the release of Geminii 4
  1391. 51:10Argon to be the main news of the week,
  1392. 51:11included all the links, wrote
  1393. 51:13everything out, and compiled a top list
  1394. 51:14. And this thing will work
  1395. 51:17automatically every day at 9:00 AM,
  1396. 51:19regardless of whether my computer is on
  1397. 51:22or not. This is a very convenient story
  1398. 51:24. It is especially cool that GPT can
  1399. 51:26generate images and create such
  1400. 51:28interesting stories. In short, with
  1401. 51:32this feature, you can automate a huge
  1402. 51:34amount of your routine work, collect
  1403. 51:37reports, analyze data for a specific
  1404. 51:39period, and so on. Concept number 21 is
  1405. 51:43a pet. To find the command to make this
  1406. 51:46pet appear, we click in the bottom left
  1407. 51:48corner. And right here there is a
  1408. 51:50button to show mini. For me, it's
  1409. 51:52option-space. Let me press it. And here
  1410. 51:55on the right side, this little pet
  1411. 51:57appears. You can talk to it by voice,
  1412. 52:00give it tasks, or type something to it.
  1413. 52:03This thing stays on top of all windows.
  1414. 52:05I mean, see, I went into Claude, and
  1415. 52:07it’s still here. It hasn't gone
  1416. 52:09anywhere. To be honest, I don't really
  1417. 52:10understand why one would use it.
  1418. 52:12Probably just to avoid opening the
  1419. 52:14Codex unnecessarily. But it at least
  1420. 52:16shows the status of the chat. Create
  1421. 52:18the simplest page in the world. Just a
  1422. 52:20white background, black text, in one
  1423. 52:22block. We hit Enter and watch what this
  1424. 52:25pet does. I switched to another window,
  1425. 52:27and you could see it was sitting at its
  1426. 52:30laptop typing something. Let's open
  1427. 52:32this up. And when it finishes the task,
  1428. 52:34it will notify us. See, this green
  1429. 52:37badge has appeared here. And you can
  1430. 52:39see that the task was completed. You
  1431. 52:42can also talk to it by voice or
  1432. 52:43exchange messages and give it a prompt.
  1433. 52:46Open the page we just made. See, I gave
  1434. 52:49it a task, and it's doing something
  1435. 52:51over there. Well, we've opened that
  1436. 52:53page. In short, it can notify us about
  1437. 52:55what is happening in the Codex. At the
  1438. 52:59same time, we can also interact with it
  1439. 53:01, give it certain plugins, skills, and
  1440. 53:03so on; we can talk to it by voice and
  1441. 53:05give it text prompts. In short, it's an
  1442. 53:08interesting, interactive gimmick, but I
  1443. 53:10don't know who exactly it's for.
  1444. 53:12Honestly, I don't really like it, so I
  1445. 53:14usually remove it. By the way, you can
  1446. 53:16go into settings and change its
  1447. 53:18appearance. For example, choose this
  1448. 53:20owl. I think it's cool. Well, anyway,
  1449. 53:22that’s the story. Concept number 22.
  1450. 53:26Working from a phone. Codex runs on our
  1451. 53:28computer, but we don't have to sit at
  1452. 53:30the computer all the time. In the GPT
  1453. 53:32app on your phone, there is a Codex tab
  1454. 53:34, so all chats from the computer are
  1455. 53:36identical. You can see what it's doing,
  1456. 53:38see the progress, see the questions it
  1457. 53:40asks, and so on. To enable this, go to
  1458. 53:43settings, connection. And right here
  1459. 53:45you can add a device. Add a device for
  1460. 53:48remote control of this Mac. Click add.
  1461. 53:51Connect a device to this Mac. Continue
  1462. 53:53from where you left off. Stay informed,
  1463. 53:55start something new. Click start. To
  1464. 53:58enable this feature, you will need to
  1465. 54:00turn on multi-factor authentication. To
  1466. 54:02do this, you need to go to the web
  1467. 54:04version. Let's go into settings here
  1468. 54:06and find security and login. Let me
  1469. 54:08zoom in a little bit. Here we select
  1470. 54:10multi-factor authentication. We turn it
  1471. 54:13on. Here we confirm our email again. We
  1472. 54:16receive a code. Next, we scan the QR
  1473. 54:18code in any authentication app. There
  1474. 54:20will be examples of these apps right
  1475. 54:22here. And we enter the six-digit code
  1476. 54:24that it provides. That’s it. After
  1477. 54:26that, we have enabled authentication,
  1478. 54:28and now we can press add again, click
  1479. 54:30start, and let's click allow. Then you
  1480. 54:33scan the QR code and that's all. Now
  1481. 54:36you can use Kodeks on your mobile
  1482. 54:39device. Concepts 23 and 24. These two
  1483. 54:42concepts will be dedicated to the new
  1484. 54:44features released on September 29th at
  1485. 54:46the event called DevDay. The first one
  1486. 54:49is Spaces. A new icon has appeared on
  1487. 54:51the left here that wasn't there before.
  1488. 54:54What is it and what is it for? Let's
  1489. 54:56click continue. That was a cool
  1490. 54:58animation. Everything that ChatGPT
  1491. 55:00creates for me lives in different chats
  1492. 55:01. Plans, tables, summaries. In a week,
  1493. 55:05all these files can be lost because
  1494. 55:07they are in different places. Well,
  1495. 55:09Spaces fixes all that. Meaning, if we
  1496. 55:12open the Spaces tab, there is this huge
  1497. 55:15thing with a bunch of stuff, like pages
  1498. 55:17, some sites, images, some texts, and
  1499. 55:20so on. This whole thing works not with
  1500. 55:23Kodeks, but with the cloud, that is,
  1501. 55:26with ChatGPT and GPT Work. So, if you
  1502. 55:28go into images, you won't find the
  1503. 55:30images you generated, for example, in
  1504. 55:33some folder in Kodeks. Or, for example,
  1505. 55:35the website files we made together,
  1506. 55:37they won't be here either. But if, for
  1507. 55:39example, you just went into ChatGPT and
  1508. 55:42used it to make a document, file, or
  1509. 55:45generated an image, all of that will be
  1510. 55:47here. And thanks to this collection of
  1511. 55:50all the files you created in the cloud,
  1512. 55:52you can manage them more easily.
  1513. 55:54Examples are shown here of what can be
  1514. 55:57done in a to-do list. A project tracker
  1515. 55:59that can be filled out quite flexibly,
  1516. 56:02a weekly summary. Here, for example,
  1517. 56:05you can click the plus sign and insert
  1518. 56:07some image, file, or anything at all.
  1519. 56:09We can generate it or attach it. You
  1520. 56:12can upload it from your computer, or
  1521. 56:14you can export it from this Spaces
  1522. 56:17media library. You can also click the
  1523. 56:20slash here and insert either headers,
  1524. 56:22some lists, quotes, a piece of code,
  1525. 56:24and so on. In short, a very flexible
  1526. 56:27setup that works almost like Notion,
  1527. 56:29but unlike Notion, it has these kinds
  1528. 56:31of features. For example, creating a
  1529. 56:33task. And here we write something like,
  1530. 56:36hey, write me five topics for an AI
  1531. 56:38video that I can film based on the new
  1532. 56:41releases from the past week. There is
  1533. 56:44such an interesting feature here. You
  1534. 56:46can rearrange all of this. You can
  1535. 56:48simply insert instructions for the
  1536. 56:50agent like this. You see, I wrote to
  1537. 56:52him, and he answered me in this format.
  1538. 56:55If you're interested, again, I can make
  1539. 56:57a separate video where I analyze
  1540. 56:59several use cases for this space. But
  1541. 57:02in short, it’s just a space that
  1542. 57:04holds everything you’ve been working
  1543. 57:07on in the cloud. And you can mix it all
  1544. 57:10up, share it, process it, and so on. Or
  1545. 57:13, to put it briefly, it's Notion with
  1546. 57:15ChatGPT built into it. And the last
  1547. 57:18thing I’d like to discuss, also a
  1548. 57:20novelty released at DevDay on September
  1549. 57:2229th, is Dots. Again, this thing
  1550. 57:25deserves a separate video, but let me
  1551. 57:26tell you about it concisely and briefly
  1552. 57:28. Everything we did before this, we
  1553. 57:30launched personally. Even the schedule
  1554. 57:32—we wrote it ourselves and told it to
  1555. 57:34do this every Monday. Dot works
  1556. 57:37differently. It is a single persistent
  1557. 57:39agent running on the GPT-4o Astra model
  1558. 57:42, which has its own computer in the
  1559. 57:44cloud, a browser, files, and a terminal
  1560. 57:46. Meaning, if you close your laptop, it
  1561. 57:48will still keep working. It is
  1562. 57:50connected to your email, your calendar,
  1563. 57:52your codebase, and it doesn’t wait
  1564. 57:53for your command. It periodically
  1565. 57:56checks on what is happening with you on
  1566. 57:57its own. And it writes to you first.
  1567. 57:59For example, it might write to you: "
  1568. 58:01Hey, I see an advertiser hasn't replied
  1569. 58:03to you in 2 weeks. Should I remind them
  1570. 58:05that they need to respond?" Or, for
  1571. 58:07instance: "A nighttime task in the
  1572. 58:09project failed; I’ve looked into it."
  1573. 58:11"Here is the reason why it failed." So,
  1574. 58:13you closed your laptop, but it’s
  1575. 58:15still working. To open it here, we
  1576. 58:17click on "your dot" right here and go
  1577. 58:19in. Click continue. And we can
  1578. 58:22immediately connect it to several
  1579. 58:23services. Click continue here. So, hi,
  1580. 58:26I'm your Dot. Write or call me anytime.
  1581. 58:29I will help with tasks even when we
  1582. 58:31aren't talking, and I'll come back with
  1583. 58:33news or questions. Do you want to give
  1584. 58:35me a name? Let's give one. Let it be
  1585. 58:37Anatoly. See, he left a heart as a
  1586. 58:39reaction all by himself. Well, that’s
  1587. 58:41already a bit of a strange story. Saved
  1588. 58:43. Now I am Anatoly. I will start
  1589. 58:45looking for ways I can help. Is there
  1590. 58:47anything you want to tackle first right
  1591. 58:49now? Actually, see, I haven't even had
  1592. 58:51a proper conversation with him yet.
  1593. 58:52He’s already found something here. He
  1594. 58:54found two items in my connected email
  1595. 58:56that he can help with. Checking the BAS
  1596. 58:58soup protection and sorting out the
  1597. 58:59late-night check-in in Hong Kong. Let's
  1598. 59:01set some boundaries and areas of
  1599. 59:03responsibility right away. Anatoly,
  1600. 59:05your zone is my tasks in Coda and the
  1601. 59:07daily news summary that comes to my
  1602. 59:09Telegram on schedule. If the summary
  1603. 59:12doesn't arrive in the morning or a task
  1604. 59:14drops into Coda, find out why and
  1605. 59:16message me. Don't send me anything
  1606. 59:17about the email, I'm tracking that
  1607. 59:18myself. Don't send anything on my
  1608. 59:20behalf without my confirmation. There,
  1609. 59:22you see, he has accepted his area of
  1610. 59:24responsibility and confirmed the custom
  1611. 59:26rules. I don't do anything further with
  1612. 59:28him. That's the whole point. He will
  1613. 59:30look into my projects and the summary
  1614. 59:32himself and only message me when
  1615. 59:34there's a reason. The message will come
  1616. 59:36here, into this chat. This feature is
  1617. 59:37not needed for one-off tasks. For
  1618. 59:39example, making a presentation,
  1619. 59:41creating a file, or setting up a page.
  1620. 59:44All of that will be faster and more
  1621. 59:45cost-effective to do directly in Coda.
  1622. 59:47This feature is not available on the
  1623. 59:49Plus plan. It is available on the Pro
  1624. 59:51plan starting at $ 100. Conversations
  1625. 59:53with him don't use up limits, but the
  1626. 59:55tasks he performs in Coda do use limits
  1627. 59:58. You can have one Dot per account, but
  1628. 1:00:01OpenAI promises teams of these Dots in
  1629. 1:00:03the future. Well, I think I've said
  1630. 1:00:06everything I wanted to say. If I forgot
  1631. 1:00:07something, I'll write about it in my
  1632. 1:00:09Telegram channel. Plus, if you still
  1633. 1:00:12have any questions about using Coda,
  1634. 1:00:13again, write them in the YouTube
  1635. 1:00:15comments, I will answer them. And you
  1636. 1:00:18can also share your own experience
  1637. 1:00:20using that space, this space and this
  1638. 1:00:23Dot. It will also be interesting to
  1639. 1:00:25read. Thanks everyone for watching. I
  1640. 1:00:27hope this video was useful for you.
  1641. 1:00:29Check out other videos on my channel
  1642. 1:00:31too. For example, this pretty good
  1643. 1:00:33video. Bye everyone. Yeah.

About this transcript

This page contains the full transcript of Дай мне 1 час и ты получишь 1000 часов знаний о Codex (2026) by Матвей Шульга, generated from the public captions YouTube serves with the video. The transcript has 11,257 words across 1,643 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.