YouTube2Text

Ajanlarım Artık Uyuyor: Yeni Ofisim — Transcript

by Selma Kocabıyık · 2,309 words · 327 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Yes, you probably remember my agent
  2. 0:02company video. But I have some bad news
  3. 0:04, or maybe it's better news. I’ve
  4. 0:06shut down that agent company. I froze
  5. 0:08all the agents there and came up with a
  6. 0:10brand-new architecture and design. Now,
  7. 0:12from start to finish, I’ll take you
  8. 0:13on a tour of the office first. After
  9. 0:14that, I’ll talk a little bit about
  10. 0:15the architecture. And then, we have a
  11. 0:17small bonus section at the end of the
  12. 0:19video. I have one or two questions to
  13. 0:20ask you there. I’d be very happy if
  14. 0:22you watched that part and left your
  15. 0:23thoughts in the comments. Let’s get
  16. 0:25started. If you recall, I showed you an
  17. 0:27agent company containing 14 agents. We
  18. 0:30even put timers on all of them, and
  19. 0:31they would wake up and work when their
  20. 0:33time came. It was great on paper. And
  21. 0:35you guys really loved the video. It got
  22. 0:37200,000 views. But to be honest, the
  23. 0:39system definitely had some flaws. We
  24. 0:42faced a few problems. For example, the
  25. 0:44cost issue was a pretty big problem.
  26. 0:46After all, they would wake up and work
  27. 0:48on their own, and there were many
  28. 0:50scheduled agents. Especially on the
  29. 0:52research side, there were agents that
  30. 0:54would go off and research things I
  31. 0:56might not need, or work just because
  32. 0:58the time came, even if I didn't want
  33. 1:00them to. But why waste tokens for
  34. 1:02nothing? So, I completely changed that
  35. 1:05system. But this time, I wanted it to
  36. 1:07be more efficient in terms of tokens,
  37. 1:10and I wanted to create an agent team
  38. 1:12that works only when I run it or when I
  39. 1:14want them to. Now, what was our biggest
  40. 1:17problem? Let’s think about that. My
  41. 1:19biggest problem was having 10 different
  42. 1:21terminals open in the background. But
  43. 1:22it runs manually right now, and nothing
  44. 1:25is right before my eyes. I mean, they
  45. 1:27aren’t within a good interface. They
  46. 1:29are in different chat terminals, and
  47. 1:30it’s very hard to track. Because I
  48. 1:32don't remember which terminal I typed
  49. 1:34what into or what I last talked about.
  50. 1:36I came up with a great architecture
  51. 1:37that will solve this problem at its
  52. 1:39root. Now let me explain the core idea
  53. 1:40of the new system. Because in the
  54. 1:43system I use now, it only starts when I
  55. 1:45want it to. What does that mean? It
  56. 1:47means this. These guys sleep unless I
  57. 1:49run them, and only when I go to the
  58. 1:51terminal and type something—or into
  59. 1:53our chat, I'm saying this for those
  60. 1:56using a desktop. When I give my
  61. 1:58terminal any command, only then does
  62. 2:00this agent wake up and start working,
  63. 2:02and it's very easy to track. Because
  64. 2:04they are all within one interface for
  65. 2:06me. Now, let’s take a general look.
  66. 2:09They look like this right now. They
  67. 2:10look a bit messy right now because they
  68. 2:12are running. Let me show you around the
  69. 2:14office a bit. What’s happening? For
  70. 2:16instance, we have a CEO now. This guy
  71. 2:17is sleeping right now. But he is a cat.
  72. 2:20I made him as a cat. Actually, normally
  73. 2:22the cat memes I use here pop up, but I
  74. 2:24couldn't include them due to YouTube
  75. 2:25copyright issues. Let me show you my
  76. 2:27cat that does research in the back.
  77. 2:29Let’s head over to its office. This
  78. 2:31guy handles my research. He just
  79. 2:33finished, for example, for me. For
  80. 2:35example, when I click here, it shows me
  81. 2:38how many tokens it has spent during the
  82. 2:41day. It shows how many tokens were in
  83. 2:43the input and how many were in the
  84. 2:45output. I can see the tasks I have
  85. 2:47assigned to it from here. I can see the
  86. 2:49model itself. Uh, we can see it working
  87. 2:51when a request comes in. Now, tasks I
  88. 2:52assign from the buttons in the office
  89. 2:54run with Sonnet, but in the terminals I
  90. 2:56open myself, Opus 3.5 runs, and it
  91. 2:58shows both of these separately. So, it
  92. 3:00shows Opus when Opus is running and
  93. 3:02Sonnet when Sonnet is running. I can
  94. 3:04see the task it's doing and everything
  95. 3:05else right here. We also put a small
  96. 3:07design in the background with my own
  97. 3:08renders. This is its room, for instance
  98. 3:10. Uh, I can wake it up from here.
  99. 3:13Production team. The production team is
  100. 3:16sleeping, actually, but when this team
  101. 3:18—this unit—is working, it handles
  102. 3:20all the production I do in the
  103. 3:21background; for example, if I'm doing
  104. 3:23game production, it handles that, or
  105. 3:25website production, or any production
  106. 3:27you can think of. Now, the end-of-day
  107. 3:30agent is the one that looks at how
  108. 3:32efficiently we worked throughout the
  109. 3:34day. Do you know what this measures? It
  110. 3:36measures how many projects I did during
  111. 3:38the day, how hard I worked, and whether
  112. 3:41I wasted tokens or not; this unit
  113. 3:43measures all of that for me. There's
  114. 3:45the Drive and disk section. Let me show
  115. 3:47you this place. You know, all the
  116. 3:49materials I produce for a video, and
  117. 3:51for example, motion—especially the
  118. 3:53Remotion side—takes up a lot of space
  119. 3:54. After the production is finished,
  120. 3:56this 30 GB needs to be moved to the
  121. 3:58Drive, meaning into the cloud. Because
  122. 4:00otherwise, I don't have enough space on
  123. 4:01my computer. That's why there's an
  124. 4:02agent that handles transferring all my
  125. 4:04work into the cloud and deals
  126. 4:05exclusively with this task. Throughout
  127. 4:07the day, for example, if I'm checking
  128. 4:09emails, it's here; if I'm producing
  129. 4:10content, it's here; uh, if I'm editing,
  130. 4:12if I'm creating renders, it's here, and
  131. 4:14the research part is here. This is the
  132. 4:15Apify room directly. By the way, this
  133. 4:18unit collects research and data from
  134. 4:20the web for us. So, how do you think
  135. 4:22the agent goes and fetches the data
  136. 4:24itself? Never. Pay close attention to
  137. 4:25this. It does all the web scraping
  138. 4:27using a tool called Apify. It collects
  139. 4:30all the data. This way, the agent
  140. 4:31understands that this is external data,
  141. 4:34not an instruction. Why do we do this?
  142. 4:36We do this for security. For that
  143. 4:37security system, I also placed a small
  144. 4:39security agent in between. Its job as a
  145. 4:41watchdog is to check every file I
  146. 4:43download before I open it. Let's come
  147. 4:45to the render machine. Let's take a
  148. 4:47general look. Let's go to this one's
  149. 4:48room. Uh, when I click here, I enter
  150. 4:51the render machine. 74.5. I know this
  151. 4:55token amount looks very high, but this
  152. 4:57is, uh, the entire day's tally—from
  153. 4:59midnight today until this hour—the
  154. 5:01amount of tokens this agent used, and
  155. 5:03you see that when it runs under this
  156. 5:05agent, for example, 7 terminals are
  157. 5:06working at once. I want to show that
  158. 5:09example as well. I'm coming to the
  159. 5:11render machine from here. Look, in the
  160. 5:13background, at the bottom, there are 1,
  161. 5:152, 3, 4, 5, 6, exactly 7 terminals.
  162. 5:18Let's continue two of them. For example
  163. 5:21, let's type "finish the job here." And
  164. 5:24let's come to this one and type "finish
  165. 5:26the job here" as well. I am returning
  166. 5:28to my agent dashboard. Look, this agent
  167. 5:30also suddenly started working. The
  168. 5:32Reels machine side is running. Let's
  169. 5:35even go. Token usage has now reached 78
  170. 5:38million. 35.5 is running in the
  171. 5:40background again. I want to show you
  172. 5:42that 7 terminals are working. And 7
  173. 5:43terminals are running. I can even see
  174. 5:45the sub-agent working here in its room.
  175. 5:48Let me also show you the rooms. There
  176. 5:50is an interface here that shows roughly
  177. 5:52what all the rooms are doing. We have a
  178. 5:55memory section. This is where the
  179. 5:57memory is stored. There is the cost
  180. 5:59factor. It allows you to look at the
  181. 6:00daily cost. By the way, I built all of
  182. 6:03these based on the problems you
  183. 6:05mentioned in the comments. Alright, our
  184. 6:08watchman is currently working. Our CEO
  185. 6:11is asleep. For example, in the scenario
  186. 6:13where we want to give work to the CEO.
  187. 6:15Yes, Selma, I'm handling it. Okay, I'm
  188. 6:17explaining my request to the CEO. I'm
  189. 6:19saying research needs to be done for
  190. 6:22new videos. Help out, send it. Okay.
  191. 6:27Now let's look at the overview. What is
  192. 6:29it doing? It's thinking. It will
  193. 6:31definitely run the research side. Let's
  194. 6:33see what it's doing right now. Sonnet
  195. 6:36is connected to it. I assigned two
  196. 6:39agents for the video. It says the topic
  197. 6:41agent should extract candidate topics
  198. 6:43from radar and WALT. Okay. Let's do an
  199. 6:45overview to see if it worked. Yes. The
  200. 6:48topic agent and the research agent have
  201. 6:50both started working. I can see both of
  202. 6:52them from here. The content creator is
  203. 6:55asleep, but it works roughly like this.
  204. 6:58I also see how many tokens it spent. I
  205. 7:00also see when it finished. I can also
  206. 7:03live-track which jobs they are doing
  207. 7:05from here. There are places where they
  208. 7:06ask for input. Remember, I mentioned it
  209. 7:08in the agent company. That there are
  210. 7:10some tasks where they ask for input. I
  211. 7:12can see these on the right side as well
  212. 7:14. Now, I'll explain how you can set up
  213. 7:16such a system roughly. The main
  214. 7:17architecture is like this. We have five
  215. 7:19project files, and they are all located
  216. 7:22under the CEO project file, and you can
  217. 7:24think of each project file as an agent.
  218. 7:27When I start a chat or a terminal
  219. 7:29inside a project file, there is an
  220. 7:31agent that takes care of it, and we
  221. 7:33place that agent inside this map.
  222. 7:35Actually, that's the whole point. So
  223. 7:37how did I do this? I did it by
  224. 7:38prompting. And for example, the
  225. 7:40important thing here is this. For
  226. 7:41instance, inside the Reels machine,
  227. 7:43there are my style and the GitHub repos
  228. 7:45I use. Especially, you know, recently
  229. 7:48with motion design, blender usage,
  230. 7:50reels production, and I didn't sit down
  231. 7:52and edit a single video inside my
  232. 7:53Instagram. I never touch any edits
  233. 7:56either. Claude does all the rest by
  234. 7:57itself, for example. That's why at that
  235. 7:59point, the inside of the Reels machine
  236. 8:01is now full of excellent skill files.
  237. 8:03In the background, there are great code
  238. 8:05snippets that create these 3D motions,
  239. 8:07for example. Especially Python codes.
  240. 8:09But as I said, this didn't happen
  241. 8:11overnight. I sat down and worked on
  242. 8:13this for 2 weeks. Then, for example,
  243. 8:16behind the research team, there is a
  244. 8:18rule file explaining how to research, a
  245. 8:20skill file, additional instructions,
  246. 8:22and lessons learned from our previous
  247. 8:24researches. Or the production part, for
  248. 8:27instance, it produces games, but I
  249. 8:28provide the instructions on what kind
  250. 8:30of game and with which library
  251. 8:31structures it should be built. So, at
  252. 8:32this point, yes, we can follow a nice
  253. 8:34interface where we watch the agents.
  254. 8:36It’s great, but the main point is
  255. 8:37that it looks at how much we’ve
  256. 8:39developed each project file, or rather,
  257. 8:40each agent, in the background. But as I
  258. 8:42said, you need to design your agents
  259. 8:44yourselves. Because I will only put the
  260. 8:45interface in the GitHub repo. Meaning
  261. 8:47the agent files you put in there will
  262. 8:48belong entirely to you. If this also
  263. 8:49gains interest, I can make a second
  264. 8:51part, open up the architecture in
  265. 8:53detail, and explain it. We are coming
  266. 8:55to the bonus part of our video. That is
  267. 8:57, I am developing a game, and as you
  268. 8:58know, my follower count has started to
  269. 9:00increase slowly. By the way, I’m
  270. 9:01truly grateful for this. I’m almost
  271. 9:03at 20,000. The Selman of a few years
  272. 9:04ago wouldn't even have expected
  273. 9:05something like this. So, thank you,
  274. 9:07really. And I don’t have a community.
  275. 9:10I wanted to come up with a more
  276. 9:11original idea, so I decided to build my
  277. 9:13community inside a game. It’s almost
  278. 9:16finished, by the way, there are just
  279. 9:17one or two security parts left. Also, I
  280. 9:19want to ask you a couple of things
  281. 9:20about the design. I don’t want to
  282. 9:21give spoilers, but I’m thinking of
  283. 9:23releasing it in 2-3 days anyway. It has
  284. 9:25a capacity of about 2,000 people. Of
  285. 9:27course, since you are the ones who will
  286. 9:28play the game, I wanted to get your
  287. 9:30opinion on this. We have two design
  288. 9:31types. The first is a huge world
  289. 9:33consisting only of pixels. The other is
  290. 9:36a bit more realistic, a world where you
  291. 9:38can move your character around. Now,
  292. 9:40the pixelated part looks like this.
  293. 9:41There will be a lot of players here.
  294. 9:43You can already see the islands. We
  295. 9:45have flying dragons. There is a royal
  296. 9:46steward here. Everyone will be able to
  297. 9:48talk to everyone. There will be areas
  298. 9:49like bars, by the way. Where you can
  299. 9:51hang out. There will be places where
  300. 9:52you can specifically buy magic or
  301. 9:54skills, and you will earn money by
  302. 9:56doing projects. For example, you can
  303. 9:58see the characters here. I actually
  304. 9:59wanted to show the travel portal a bit
  305. 10:01more. But I won’t give any spoilers.
  306. 10:04Okay. This is how it is. Now let me
  307. 10:06show you the other one. It looks a bit
  308. 10:08more like this, but this one will be a
  309. 10:10bit more realistic. Here, too, you will
  310. 10:13be able to enter locations and play. So
  311. 10:15, the interface is exactly a bit
  312. 10:16different. Therefore, which one would
  313. 10:18you like to play in? In the pixelated
  314. 10:20one or the one that is more realistic?
  315. 10:22I’m very curious. Well then, we have
  316. 10:24come to the end of our video. You can
  317. 10:26find all the details about the agent
  318. 10:27company in the GitHub repo in the
  319. 10:28description section. If you have any
  320. 10:29questions, or if you have extra
  321. 10:31suggestions about the game, or if you
  322. 10:33want to add something, for example. If
  323. 10:35there are places where you say "It
  324. 10:36would be great if this were here," you
  325. 10:37can state them in the comments section.
  326. 10:39I would be very pleased. Thank you. See
  327. 10:41you in the next video.

About this transcript

This page contains the full transcript of Ajanlarım Artık Uyuyor: Yeni Ofisim by Selma Kocabıyık, generated from the public captions YouTube serves with the video. The transcript has 2,309 words across 327 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.