7 Free GitHub Repos That Make Claude So Good It Feels Illegal — Transcript

Full transcript

1. 0:00I feel like I've been using Claude wrong
2. 0:01for the last six months. How did I not
3. 0:03find out about these GitHub repos
4. 0:05sooner? Because ever since I started
5. 0:06using these, I feel like my Claude has
6. 0:09become orders of magnitude more
7. 0:10powerful. I went through all the major
8. 0:12topics on GitHub. Claude Code, agent
9. 0:14skills, Claude skills, MCP servers,
10. 0:16Claude AI, AI agents, and I found the
11. 0:18best GitHub repos to inject into your
12. 0:21Claude. These are repositories that you
13. 0:23can simply download, connect to your
14. 0:24Claude, and turn Claude into a monster.
15. 0:26One of these repos can help you scale up
16. 0:28your Claude into an entire 30 agent team
17. 0:31that can run your business for you.
18. 0:32Another one of these repos is the exact
19. 0:35same system that the CEO of Y Combinator
20. 0:38uses to run their multi-billion dollar
21. 0:40business. Another one was just launched
22. 0:41and was created by Jack Dorsey and is
23. 0:44the future of Agentic Work. And another
24. 0:46one is a repo that I've been using for
25. 0:48the last week to scale and optimize my
26. 0:51ads. So, I've been putting these to the
27. 0:52test across my real workflows. I'm going
28. 0:54to show you the top seven which have
29. 0:55almost a million stars on GitHub between
30. 0:57them. So, they're very highly rated and
31. 0:59I'm going to show you how to turn Claude
32. 1:00into an absolute beast. All of the repos
33. 1:03that I select today are going to be
34. 1:05available for free in the description
35. 1:06below. If you join my free school
36. 1:08community under the resource section,
37. 1:09just like every video, I'm going to have
38. 1:11the full guide from this video. It's
39. 1:12going to contain the seven repos and
40. 1:14installation guide so you can just chuck
41. 1:15them straight into Claude Code and
42. 1:17they're going to start working pretty
43. 1:19much straight away. The install was
44. 1:20actually really, really easy. So the
45. 1:22first one is last 30 days. This is very
46. 1:25cool because when most people are
47. 1:26researching, they just use Claude and
48. 1:28Claude just researches the internet. So
49. 1:29when they're researching a topic, they
50. 1:31basically just get an aggregation of
51. 1:33news articles. But what last 30 days
52. 1:34does is it also scans the social side.
53. 1:37So it'll scan Reddit X, YouTube, Hacker
54. 1:39News, Poly Market, and the web all at
55. 1:41once to give you real opinions, not just
56. 1:43from news articles, but also from real
57. 1:45people. So you can basically ask it
58. 1:46about any topic and it's going to come
59. 1:48back to you with what's rising in
60. 1:49popularity, what's overrated, what
61. 1:51people are saying with examples. And
62. 1:53this becomes very valuable in many ways.
63. 1:55I'll give you some example prompts. If
64. 1:56you run a business, for example, you can
65. 1:58stress test the market's opinions about
66. 2:00a particular app you want to onboard or
67. 2:02a particular funnel that you're thinking
68. 2:03of running. If you are wanting to use a
69. 2:05product, you can stress test whether the
70. 2:06product is good or not. If I'm
71. 2:08researching for a YouTube video, I can
72. 2:09get real research based on what people
73. 2:11are saying. So, I don't just need to ask
74. 2:13Claude and rely on Claude for
75. 2:15researching because I know I'm covering
76. 2:16all of the major platforms and social
77. 2:18opinion. So, if you do anything research
78. 2:20related, which in reality is all of us
79. 2:22because if you run a business, you're
80. 2:23constantly researching. If you make
81. 2:24content, you're constantly researching.
82. 2:26And just in general, if you want to
83. 2:27learn anything, you're probably
84. 2:28researching. Last 30 days is an amazing
85. 2:31tool. So, let me show you. I just asked
86. 2:33I'm running a prop in the background to
87. 2:34get the public sentiment and the
88. 2:36opinions and the current vibe check on
89. 2:39Claude versus Chad versus Gemini. what
90. 2:41people are using and what people are
91. 2:42switching to and a complaint nobody at
92. 2:44the AI companies wants to hear. Last 30
93. 2:47days basically describes themselves as
94. 2:48an AIEL search engine scored by up
95. 2:50votes, likes, and real money, not
96. 2:52editors. So, you can see what's actually
97. 2:54moving the needle in terms of social
98. 2:55interest and engagement, which is very
99. 2:57important. It's great for me because in
100. 2:58content creation, that's all you really
101. 3:00care about. You care about social
102. 3:01sentiment more than news rankings cuz at
103. 3:03the end of the day, you want real people
104. 3:05to watch your videos. So, you can see
105. 3:06here the engine has done the research.
106. 3:09It's broken down the community
107. 3:10sentiment. Polarized on Claude. Loyal on
108. 3:13output quality. Furious on limits.
109. 3:15Strength sticks your existing code base
110. 3:17better than Astra. Highest ambition
111. 3:19ceiling. Co-workers landing with the
112. 3:21non-coders. Weaknesses. September 14th.
113. 3:24Limit cut. Public cancellations and
114. 3:25lawsuit with Chad GPT. Enthusiastic on
115. 3:28GPT Astra. Uneasy on privacy and ads.
116. 3:31Strengths what people love. Weaknesses
117. 3:32common complaints. Gemini love price
118. 3:35reliability and volume not for being the
119. 3:37smartest. And then it summarizes all of
120. 3:39the information here with what the
121. 3:41agents have reported back. So that's
122. 3:42just an example of a broader prompt. And
123. 3:44you can see that the research outputs
124. 3:45here in my opinion are stronger than
125. 3:46just asking Claude, especially for
126. 3:48social stuff. But here's where it gets
127. 3:49interesting. If you run a business and
128. 3:51you want data about a potential product
129. 3:54or how to launch a product, for example,
130. 3:55I have a school community and I might
131. 3:56want to know what are people saying
132. 3:58about the current school communities.
133. 3:59What do people complain about? What
134. 4:01would make one potentially worth money
135. 4:02if I were to do a paid offering? Mine's
136. 4:04free, but if I were, what would make it
137. 4:06worth it? So you can ask questions like
138. 4:08this. You could ask questions about what
139. 4:09ads people are responding to, what
140. 4:10products people are responding to, and
141. 4:11use a tool like this to do feasibility
142. 4:13studies on a product or an offer because
143. 4:15you're getting opinions from real
144. 4:16people. So let's say you were running a
145. 4:18consulting business like a, you know,
146. 4:20white glove AI concierge type business,
147. 4:22you could ask what the average person is
148. 4:24looking for. And because it's crawling
149. 4:26through X, Reddit, YouTube, getting real
150. 4:27people's thoughts, it's going to
151. 4:29aggregate that information much better.
152. 4:30So I'm sure you guys, you know, could
153. 4:32start to brainstorm and think of the
154. 4:33power of a tool like this. And because
155. 4:35it has the agents preset and pre-trained
156. 4:38to sweep in a certain way and scan in a
157. 4:40certain way, the output's going to be
158. 4:41better than just asking Claude, "Hey,
159. 4:43get a social score from XR Reddit or
160. 4:45YouTube." So, based on this prompt, I'm
161. 4:46already starting to get some insights.
162. 4:47Premium buyers aren't paying for more
163. 4:49information. They're paying you to help
164. 4:51them get something done 100%. Low ticket
165. 4:53trap is a math problem, not a marketing
166. 4:55problem. When you're charging, what
167. 4:56separates the ones people stay in is
168. 4:58call frequency and feedback on real
169. 5:00work, key patterns from the research.
170. 5:02And then you've got the source material
171. 5:03down below. So the next MCP that I've
172. 5:05been finding very very useful in my
173. 5:07workflow especially for content creation
174. 5:09is Playright MCP. So this essentially
175. 5:12gives Claude browser automation
176. 5:14capability. So it enables an LLM like
177. 5:16Claude or GPT to interact with web pages
178. 5:19through structured accessibility
179. 5:20snapshots which bypasses the need for
180. 5:23screenshots or visually tuned models
181. 5:25which means that it can be a lot faster
182. 5:26to interact with the browser and do
183. 5:28research than computer use. I don't know
184. 5:29if any of you have been using Grockbot
185. 5:31recently. Although it's very capable, a
186. 5:33lot of the time it's extremely slow.
187. 5:34Whereas this bypasses everything. This
188. 5:36is literally Microsoft's official MCP
189. 5:38server for Playright. So it's the tool
190. 5:40that real developers use to automate
191. 5:42their agentic workflows in their
192. 5:44browser. So it means that Claude can
193. 5:45essentially interact with a real
194. 5:46browser. So clicking, typing,
195. 5:48researching, scanning, filling forms,
196. 5:50reading pages, and coming back with
197. 5:52information, which essentially gives
198. 5:53Claude hands, right? So Claude on its
199. 5:55own can't do anything on the browser. It
200. 5:57can just read it. If you put claude in a
201. 5:59Chrome extension or you use computer use
202. 6:00or if you use its own browser
203. 6:02technically it can take a screenshot it
204. 6:03guesses where to click and then it can
205. 6:05click on pixels but it's very slow and
206. 6:07it lives in the browser whereas
207. 6:08playright MCP opens its own browser
208. 6:10reads the page structure clicks the
209. 6:12exact button needed to interact and it
210. 6:14can run without you. So it's essentially
211. 6:16headless so it can stay logged in which
212. 6:18means you can chain it onto other
213. 6:19workflows more efficiently. Let me give
214. 6:21you an example of something it can do.
215. 6:22So, I'm going to ask it to open a
216. 6:24browser and go to my channel, AI Edge.
217. 6:26And then I want it to open the videos
218. 6:28tab, read my most 10 recent uploads, log
219. 6:31all of the data, then do the same for
220. 6:32one of my competitors. I'm going to use
221. 6:34Nate Herk as one of the biggest AI
222. 6:36channels for this example. Put all 20 in
223. 6:38one table sorted by views per day since
224. 6:40upload. Tell me in three lines what
225. 6:41their titles are doing that mine aren't.
226. 6:44Now, I can see here that it's actually
227. 6:46loaded up a browser. You can see it's
228. 6:48going on to both of our channels. It's
229. 6:49taking the data. You can see here it's
230. 6:51already taken the titles, it's loaded
231. 6:54them, it's looked at the competitor
232. 6:55page. So, it's basically just
233. 6:56interacting with the browser. This is a
234. 6:58tool that many professional developers
235. 7:01use. So, it's Microsoft's browser
236. 7:03automation framework. So, it's become
237. 7:05the standard for endto-end testing and
238. 7:07web development. So, if you're a
239. 7:08front-end team, if you're building
240. 7:09landing pages, UIs, it's very common to
241. 7:11use Playright MCP for tweaks. So if an
242. 7:14agent writes a UI change, it can open
243. 7:16the page in playright, check that it
244. 7:18rendered instead of the developer like
245. 7:20alt tabbing to look and essentially you
246. 7:22end up with a scenario where the agent
247. 7:23can now see the result of its own code.
248. 7:25So it can use it for autonomous checking
249. 7:27of its own work. The key distinction to
250. 7:29understand is that it actuallyes the
251. 7:31real browser as opposed to cla or gbt
252. 7:34which takes screenshots. It slows the
253. 7:36whole process down and it misses
254. 7:37information and it's less accurate and
255. 7:39it's less autonomous. So you can see
256. 7:40here it's given me the output. It's
257. 7:42given me the views, the titles, the day
258. 7:44since upload, and it gives me some of
259. 7:46the reasons why he's getting more views
260. 7:47than me. This is a very quick test, but
261. 7:49it's something that's very useful when
262. 7:50it comes to YouTube research. Like for
263. 7:52example, I could say research all videos
264. 7:53on GitHub repos. You know, pull all of
265. 7:55those videos, go in, pull the
266. 7:57transcripts, come back to me with the
267. 7:58transcripts, and summarize the reason
268. 8:00why the most viral videos went viral. Or
269. 8:02I could say, search Claude, pick the
270. 8:03most viral outlier topics and help me
271. 8:06ideate 10 content ideas based on that.
272. 8:07So there are many ways that you can use
273. 8:09Playright. It's just a bit more seamless
274. 8:11than you know using the clawed browser.
275. 8:13So if you've ever run into issues with
276. 8:14the browser, if you do a lot of browser
277. 8:15work, I think this is a great MCP for
278. 8:17you. All right, so for the next two
279. 8:19repos, repos three and four, we're going
280. 8:21to focus specifically on jobs. So we're
281. 8:23going to focus on ads. This is something
282. 8:25that I do a lot in my business. If you
283. 8:26run a marketing company as well, this is
284. 8:28going to be super useful and something
285. 8:30that I think all of you can use as
286. 8:32number four, which is a skill that
287. 8:33enables you to build diagrams and
288. 8:36workflows from scratch. So you can
289. 8:37literally describe a system, a workflow,
290. 8:39and it's going to make an amazing visual
291. 8:41demonstration of that workflow, which
292. 8:42you can then give to an agent. It's
293. 8:44super cool. Then we're going to move on
294. 8:45to company use cases in repo 5 and six.
295. 8:48I'm going to show you how you can run a
296. 8:4930 employee company with Claude. All
297. 8:51thanks to a repo from Gary Tan, the CEO
298. 8:53of Y Combinator, which literally makes
299. 8:56Claude work like a Y Combinator company.
300. 8:58And then number six and seven is just
301. 9:00going to take any startup idea, any
302. 9:02company, pretty much if you're running
303. 9:03any business on AI to the next level.
304. 9:05All right, so the next repo is Claude
305. 9:07Ads. This is so sick if you're running
306. 9:09paid ads. Now, although a lot of
307. 9:11creators or agencies or companies, they
308. 9:13focus on organic content, one thing that
309. 9:16is really effective, especially right
310. 9:17now, is meta ads. And this is a way that
311. 9:20you can track all of your ads from Meta,
312. 9:22from YouTube, pretty much any platform,
313. 9:23as you can see here, all in one place.
314. 9:25So, you can optimize your ads. You can
315. 9:27find out why they're working, why they
316. 9:28aren't working. You can track versus
317. 9:30competitors. This is pretty much a way
318. 9:31to have an entire ads agency within
319. 9:33Claude. And I really wanted to speak
320. 9:35about it today because no matter if you
321. 9:36are working in the agency model, if
322. 9:38you're working for clients, if you're
323. 9:39doing AI concentric consulting, if
324. 9:40you're running your own business, you
325. 9:41probably need a repo like this. So, it
326. 9:44has 9.3K stars. I honestly think it's
327. 9:46underrated cuz I think it's great. So,
328. 9:48this is essentially a claude first
329. 9:49portable paid media operations for
330. 9:51agencies, consultants, and in-house
331. 9:52performance teams. Don't you think it's
332. 9:54just crazy now? And especially as we get
333. 9:55to the last repost, I think it'll become
334. 9:57obvious that we have software tools that
335. 9:59can literally put a professional
336. 10:01marketing team in your pocket. I was
337. 10:02testing out some of the hedge fund stuff
338. 10:03before and I find it mind-blowing
339. 10:05because you can literally have an in
340. 10:06professional hedge fund using proquant
341. 10:08strategies in your pocket now on your
342. 10:10claude account. Like it's just crazy
343. 10:11what you can do these days. But
344. 10:13essentially the way claude ads works is
345. 10:14you run one command, the agents fan out.
346. 10:16So they have 11 specialist agents which
347. 10:18audit every platform in parallel. It
348. 10:19will give you a health score of all your
349. 10:21ads and then it will give you an action
350. 10:22plan. So it will rank the potential
351. 10:24fixes by impact and what you can ship
352. 10:26right now. So you can audit all of your
353. 10:29ads at once. Not only does it do audits,
354. 10:31it can also plan campaigns, budget, you
355. 10:33can strategize, it can tell you when you
356. 10:34should up ad spend in a certain area,
357. 10:36when you should lower ad spend in a
358. 10:37certain area, what your ROI is if you're
359. 10:39missing something. And if you combine it
360. 10:41with the first repo, last 30 days that I
361. 10:43shared with you as well, where you can
362. 10:44get new ad creative ideas based on what
363. 10:46people are talking about, it's just
364. 10:47going to put rocket fuel on top of this.
365. 10:49And that is the real power of these
366. 10:50repos and MCPs. It's really stacking.
367. 10:52For example, you could do the research
368. 10:54on last 30 days. You could use this ads
369. 10:57agent to do the grunt work. You could
370. 10:59use the Higsfield MCP to generate the
371. 11:01media cuz it has access to the latest
372. 11:03video and photo models. You could use
373. 11:05Playright MCP to actually go into a
374. 11:07browser and and do something like you
375. 11:09can stack all of these on top of each
376. 11:10other. And the last ones I'm going to
377. 11:11talk about, you can actually sync all of
378. 11:13this with your other AI agent employees.
379. 11:15So that is how you have to think of AI.
380. 11:17Now it's tool stacking, using each tool
381. 11:19for what it's good at and building a
382. 11:20genuine company. You know, you want to
383. 11:21have your ads manager, you want to have
384. 11:23your financial officer, you want to have
385. 11:24every agent with a specific trained
386. 11:27role. And that's why these repos are
387. 11:29good because you don't need to go
388. 11:30through the effort of training it.
389. 11:31Someone has already done it and not just
390. 11:32someone. You have in this case four
391. 11:34contributors and these are top
392. 11:35contributors you know literally Claude
393. 11:36and Codex. Sometimes you have crazy
394. 11:38amounts of contributors. Like let's go
395. 11:40on to like last 30 days for example has
396. 11:42140 contributors. So there are a lot of
397. 11:44amazing developers who are putting you
398. 11:45know time behind this. So as you can see
399. 11:47in the demo here you have a bunch of
400. 11:49commands. You can choose what command
401. 11:50you want to run. For example the audit
402. 11:51command. Then it will spin up an
403. 11:53orchestrator. It will run parallel audit
404. 11:54agents and then it will provide an
405. 11:56output. And you can see the example
406. 11:57output here. So that's an absolute
407. 11:58banger one. Let's get into Archify,
408. 12:01which is a skill to create beautiful
409. 12:03verifiable architecture, workflow,
410. 12:05sequence, data flow, and lifestyle
411. 12:07diagrams, which are their own HTMLs,
412. 12:09which you can put into notion. You can
413. 12:10export them. I'll show you exactly how
414. 12:12to do it. And this one has 64.4K stars.
415. 12:15So essentially, it's a way to take plain
416. 12:17English and build it into architecture.
417. 12:19So this is especially amazing if you're
418. 12:21doing client work because you can just
419. 12:23describe the client's problem. Let's say
420. 12:24you get on a discovery call with a
421. 12:26client. They tell you their problems.
422. 12:27You can then determine a workflow in
423. 12:29plain English with Claude and then you
424. 12:31can use it to create an amazing diagram
425. 12:33which then you can automate or you can
426. 12:35give to an agent to automate. And I find
427. 12:38being able to graph engineer and
428. 12:39visualize processes not only helps your
429. 12:42agents but helps you understand
430. 12:43processes better which in turn helps you
431. 12:45get the results that you need out of AI.
432. 12:47And this was actually one of the top
433. 12:49trending repositories. People have been
434. 12:51loving it. As you can see it creates
435. 12:52these diagrams. You can switch between
436. 12:53dark mode and light mode for different
437. 12:55flows that you would like to create. And
438. 12:57what I actually recommend people do is
439. 12:59they build out SOPs for their real
440. 13:01business. Like you've really just got to
441. 13:02be writing everything down in your
442. 13:04business. Have a database like notion
443. 13:06for example. Have every single SOP. So
444. 13:08how you go about creating a video, how
445. 13:10you go about planning for an ad
446. 13:13campaign, how you go about onboarding a
447. 13:15client, how you go about doing your
448. 13:16financials, whatever you do, if you're
449. 13:18working a 9 to5, how you do your
450. 13:19specific task. Write down each
451. 13:21component. draw it up into an
452. 13:23architecture and a workflow using
453. 13:25Archify and then from there go and find
454. 13:29each bit that you can automate because
455. 13:30once you've mapped it out and you've
456. 13:32graph engineered and you have a visual
457. 13:33exemplification of what you do, you can
458. 13:35actually go and start picking individual
459. 13:36components that you want to outsource.
460. 13:38And this is really important because
461. 13:40before you set an agent to do something,
462. 13:42you really want to understand exactly
463. 13:44what it's doing. And I feel like this
464. 13:45also helps you get a much better
465. 13:47understanding of AI. It genuinely makes
466. 13:49you better at using AI because you
467. 13:52understand why things are happening and
468. 13:53how things are happening. And you're
469. 13:54going to get much more consistent
470. 13:56outputs with this because it's already
471. 13:57been trained. It's already been stress
472. 13:59tested and it already has top developers
473. 14:02versus trying to graph engineer
474. 14:03yourself, which to be honest, I was
475. 14:04doing for a long time. And as I said in
476. 14:06the beginning of the video before
477. 14:07finding some of these repositories, I
478. 14:08genuinely think I was using Claude and
479. 14:10AI wrong in many ways. Not wrong, but I
480. 14:12was just putting too much work into
481. 14:14things when people have already done the
482. 14:16grunt work for me. Like the reality is
483. 14:18developers all around the world are
484. 14:20having the same issues that you have in
485. 14:21terms of memory in terms of AI not
486. 14:23sounding human enough in terms of it
487. 14:25being too inconsistent or too hard to
488. 14:27graph engineer in terms of the web or
489. 14:29browser capabilities not being correct
490. 14:31and that's why developers have made
491. 14:32workarounds and that's why these highly
492. 14:34rated GitHubs exist because genuinely
493. 14:36they're better for many of these things
494. 14:37than just using the stock and you know
495. 14:39plugandplay chatgbt or claude. Now we're
496. 14:42moving on to number five and number six.
497. 14:44These are use cases for your company or
498. 14:47to create a company. And this is one of
499. 14:49the heavy hitters of this video. This is
500. 14:50Gary Tan, the CEO of Y Combinator's
501. 14:53exact Claude Code setup. If you don't
502. 14:55know what Y Combinator is, they're
503. 14:56basically venture capital. So they
504. 14:57invest in startups. They incubate a lot
505. 14:59of startups. They manage an estimated
506. 15:01$12.2 billion in assets and they are at
507. 15:04the front foot of AI. They're investing
508. 15:06and backing in a lot of the up and
509. 15:08cominging exciting tech. So you best
510. 15:10believe these guys are ahead of the
511. 15:12curve when it comes to AI. And Gary Tan
512. 15:14released his claude code setup with 23
513. 15:16opinionated tools that serve. This is
514. 15:18crazy. As a CEO, a designer, an
515. 15:20engineer, a manager, a release manager.
516. 15:22You literally have an entire company in
517. 15:24your pocket. This has 133,000
518. 15:27stars, but I still don't hear many
519. 15:30people talking about it. And even I was
520. 15:31onto this a little bit late to be
521. 15:33honest. When I started testing this, I
522. 15:34realized how crazy it actually is. So to
523. 15:36install it, you need to follow these
524. 15:37steps. If you are in the school, I'm
525. 15:39going to link all of these anyway with a
526. 15:41direct link to each GitHub and have a
527. 15:43oneshot prompt so you can install all of
528. 15:45these at once. Cuz if you go into claude
529. 15:46code and you just drop that MD file and
530. 15:49install everything at once, then you can
531. 15:50just open any chat and you can just
532. 15:52start using this stuff cuz it's already
533. 15:53installed on your device or you can just
534. 15:55go through and install manually.
535. 15:56Completely up to you if you just want to
536. 15:57install this one specifically, for
537. 15:58example. But my school community is
538. 16:00free. So you can see that GStack has a
539. 16:02collection of skills. So each skill
540. 16:05feeds into the next. For example, office
541. 16:07hours writes a design document that the
542. 16:09plan CO review skill the CEO needs to
543. 16:12actually think about the problem. Then
544. 16:14it can hand off to the engineering
545. 16:15manager who revites a test that the QA
546. 16:18picks up. Review catches bugs which
547. 16:20means that nothing falls through the
548. 16:22cracks because every step knows what
549. 16:24came before it. So what I recommend is
550. 16:26using for/off hours to kick off a chain.
551. 16:29For example, you know, let's say you
552. 16:30want to build a daily briefing app for
553. 16:32your calendar. you go forward/off hours.
554. 16:34Then claude because it's tapped into
555. 16:36GStack would start asking questions. You
556. 16:38give it a response. It's going to
557. 16:40challenge it. It's going to give you a
558. 16:41recommendation. Then you can kick off
559. 16:43the next skill and then it's going to go
560. 16:45through the system until you've
561. 16:47actually, you know, shipped the product.
562. 16:49So this is a real flow that is created
563. 16:51by a real venture capital and incubation
564. 16:54company. But it doesn't just stop there
565. 16:56because they also have power tools. So
566. 16:57you can use for/codeex for a second
567. 16:59opinion. So an independent code review
568. 17:01from OpenAI. So let's say you're using
569. 17:04Claude to code. You can get a second
570. 17:06opinion from Codeex. Claude code the
571. 17:08other way around if you are using
572. 17:09Codeex. You've got careful which states
573. 17:12the safety guard rails. You've got a few
574. 17:14really cool ones like iOS QA as well to
575. 17:16connect to your iPhone. So basically
576. 17:18every time you use it, you want to start
577. 17:19with plan CO review to kick off a
578. 17:21process. And you can also spin up
579. 17:23multiple sub agents and really run this
580. 17:25as its own autonomous team using the
581. 17:27training from GStack. The way to really
582. 17:29understand GStack is from the lens of
583. 17:31Gary Tan because this is essentially the
584. 17:34same skill stack that he runs in Claude.
585. 17:36This is how he at YC would interrogate
586. 17:38an idea. So if you are trying to ship
587. 17:40something new for the company, if you
588. 17:42are trying to launch a new product, if
589. 17:44you are trying to make a tweak or a
590. 17:46change, you can just run it through a co
591. 17:48review. For example, if I want to launch
592. 17:50a new product, which is the prompt I use
593. 17:51here, I'm going to run it through CO
594. 17:52review as long as it's tapped into my
595. 17:54memory. So if my folder has a MG file
596. 17:56and a instructions file with my actual
597. 17:58memory and data or it's connected to my
598. 18:00notion, for example, which has all the
599. 18:01information on my company, it'll be able
600. 18:03to extract off that information as well,
601. 18:05which I think is really important for
602. 18:06this to, you know, actually give
603. 18:07applicable advice to my business. It's
604. 18:10essentially like you've hired YC to work
605. 18:12in your company and it's constantly
606. 18:14updated. I think there was literally an
607. 18:15update today. So they're always adding
608. 18:17new stuff, which is good because some
609. 18:19repos get stale and aren't that active,
610. 18:21but this one's super active. All right,
611. 18:22while that's running the analysis, let's
612. 18:24get into the onetwo punch here, which is
613. 18:26so powerful if you want to start a fully
614. 18:28AI business. Like, if you are hellbent
615. 18:29on starting a one person AI business, or
616. 18:31even if you have a few employees and you
617. 18:33just want to scale up your business with
618. 18:35agents, but you don't want to have to
619. 18:36code agents yourself from scratch, this
620. 18:39is a plug-and-play solution. Probably
621. 18:40the best plug-and-play solution I've
622. 18:42found, and not just me, it's literally
623. 18:44got 152,000 stars and 124 contributors,
624. 18:47so the market seems to think so as well.
625. 18:48that is going to basically give you an
626. 18:51entire AI agency or AI company from
627. 18:53scratch. So the Y cominator Gary Tan
628. 18:56stack that I just showed you is going to
629. 18:58interrogate a product launch and it's
630. 19:00going to run through start to finish.
631. 19:02This is going to be the operations side
632. 19:04that runs your company on a day-to-day
633. 19:06basis. So you've got entire divisions to
634. 19:08tap into here. And remember each agent
635. 19:10is actually trained on a specialtity. So
636. 19:12you've got an engineering division with
637. 19:14a front-end developer, backend
638. 19:15architect, mobile app builder, AI
639. 19:17engineer. You can plug this into any
640. 19:18model you want. Code reviewer. The list
641. 19:20goes on and on and on. I won't read them
642. 19:22all out or you'll probably get bored.
643. 19:23You've got a design division, a UI
644. 19:24designer, a UX researcher, a UX
645. 19:26architect, a brand guardian, a visual
646. 19:28storyteller. You've got a paid media
647. 19:30division. And once again, you can
648. 19:32combine this with some of the other
649. 19:33tools that I spoke about earlier. And
650. 19:35you can train the agents to orchestrate
651. 19:37and delegate work to those other MCPS as
652. 19:40well. You've got an ad creative
653. 19:41strategist. You got a paid social
654. 19:42strategist. You've got a sales division,
655. 19:44an outbound strategist, a discovery
656. 19:45coach, a deal strategist. You've got a
657. 19:47marketing division, a growth hacker, a
658. 19:48content creator. And these are trained
659. 19:50by real people. You got to remember this
660. 19:51isn't just slop that someone's just spun
661. 19:53up in 5 seconds. This is contributed by
662. 19:55124 people that have built all of these
663. 19:58skills to act and behave and, you know,
664. 20:01work like these employees really would
665. 20:03in the real world. You've got a product
666. 20:04division. It just goes on and on. I
667. 20:06literally can't even scroll fast enough.
668. 20:08I'll go through quickly. security
669. 20:09division, support division, special
670. 20:10computing division, specialized
671. 20:12division. Look, depends what company you
672. 20:13have. But what they've done is they've
673. 20:14gone out and essentially built something
674. 20:16for every single company. So, you can
675. 20:18literally install this, and I'll show
676. 20:20you how to do that in a second. And
677. 20:21based on whatever business you're
678. 20:22starting, you can just pick the agents
679. 20:24that you want. Oh, by the way, the Gary
680. 20:26Tan skill is doing an amazing breakdown
681. 20:28in the background of how I should
682. 20:30actually structure my launch. So, you
683. 20:32can see much better than a normal cord
684. 20:33output. Just the way that the skill has
685. 20:35been refined to critically think. That
686. 20:36makes sense cuz it's literally Gary's
687. 20:38workflow. So, let me just think of a
688. 20:39prompt off the top of my head for agency
689. 20:41agents. I want to utilize agency agents
690. 20:44to start a company. I am a marketing
691. 20:47company focused on marketing AI
692. 20:50products. I need a design specialist for
693. 20:52my ad creatives. I need a marketing
694. 20:54manager and I need an accountant because
695. 20:56I'm always losing track of my finances.
696. 20:58I want to start with a small team to
697. 21:00begin with. By the way, I use Whisper
698. 21:01Flow for all of my voice transcriptions.
699. 21:03Highly, highly recommend. All right. So,
700. 21:04it's saying it has what it needs from
701. 21:06the roster. It's going to set up the
702. 21:08company's own workspace with a three
703. 21:09agent team. Remember, there's literally
704. 21:11hundreds of agents to select from. So,
705. 21:13you can just scale it up. It has 279
706. 21:15agents installed. Like, this is crazy.
707. 21:18We just installed this for free. I have
708. 21:20almost 300 agents just sitting in Claude
709. 21:22now that have already been trained.
710. 21:24These are actual Claude skills, right?
711. 21:25So, it's not like me just asking around
712. 21:26Claude chat, oh, be a design strategist.
713. 21:28It's like this has already been trained.
714. 21:30Now, maybe it's not perfectly trained to
715. 21:31how you want it. That's where you still
716. 21:32need to make tweaks, but it's going to
717. 21:34be a hell of a lot better than just
718. 21:35doing it completely from scratch because
719. 21:37there's real people, real contributors,
720. 21:39real developers, as I showed you, 124
721. 21:41people that have contributed towards
722. 21:42this. And you can actually just scroll
723. 21:43through if you want because it's, you
724. 21:45know, a repository at the end of the day
725. 21:46and just look at the individual agents
726. 21:49or skills under each single category.
727. 21:51Here's finance, here's healthcare,
728. 21:53here's spatial computing, here is
729. 21:55scripts. There's so much here. So, you
730. 21:57can see that I'm now speaking to the
731. 21:59designer. I'm about to launch a new
732. 22:01product. What tools are you going to use
733. 22:03to design my web page for something
734. 22:05super professional? You can see that
735. 22:07it's routing to the agent and then the
736. 22:09agent is going to respond to me now. So
737. 22:11I'm speaking to So this is really smart.
738. 22:14It wants to use claw design canvas. Then
739. 22:15it's going to use G stack design
740. 22:18consultation. Then it's going to use
741. 22:19Higsfield and then it's going to use
742. 22:21Versal plugin for hosting playright and
743. 22:23the inapp browser for screenshots and QA
744. 22:25on the live page which we installed
745. 22:27before. This is really smart. This is
746. 22:28crazy. I can literally have a whole
747. 22:30company running here. However, something
748. 22:31that's obvious is that running a company
749. 22:33like this is not very efficient. I mean,
750. 22:35you're literally in a claw terminal.
751. 22:36It's going to get messy. It's just not
752. 22:37the right UI for things. So, what is the
753. 22:39right UI for things? Well, this is where
754. 22:41number seven comes in. It's called Buzz.
755. 22:43It's by Jack Dorsey. This is very recent
756. 22:45on GitHub, but obviously they recently
757. 22:47launched their desktop application. And
758. 22:48this is the place like Slack, but for
759. 22:51agents to have all of your agents
760. 22:53conversing with each other. So all of
761. 22:56those agents, the ad specialist, the
762. 22:58designer, the finance accountant, the
763. 23:01CEO, all of those skills we went
764. 23:03through, they can now all be in Slack
765. 23:05channels. So I can set up the company
766. 23:06like a real company. So I can create a
767. 23:08new channel. I can make the channel
768. 23:09YouTube strategy. And then I can click
769. 23:12create channel. And then I can drop my
770. 23:14YouTube strategist in here. And I can
771. 23:16connect it to Claude Code through the
772. 23:17CLI. So I can actually tap into the
773. 23:19agent capabilities from the agency
774. 23:21agents repo that I just showed you. So,
775. 23:24this is how you can actually set up a
776. 23:26functional agentic company where you can
777. 23:28properly visualize things but still tap
778. 23:30into all of their skills through Claude.
779. 23:32I'm going to probably have to do a
780. 23:33separate video on this. It deserves its
781. 23:35own video. So, if you want that, let me
782. 23:37know in the comments below. But number
783. 23:39seven is Buzz. This is how you tie
784. 23:40everything together and all you need to
785. 23:42do is install it from the GitHub. Once
786. 23:43again, I'll leave my free school
787. 23:45community below. But you could start
788. 23:46developing a whole army of agents
789. 23:49working for you as you can see here. And
790. 23:51you can have employees working with
791. 23:52agents. I think that's the big diff.
792. 23:54Obviously, you could do it on Slack,
793. 23:55too. Don't get me wrong. But Buzz,
794. 23:56because of the CLI integration, is very
795. 23:58AI forward, and that was why Jack Dorsey
796. 24:00decided to do it cuz he wanted an AI
797. 24:02forward solution. So, you can have real
798. 24:03employees in Buzz working alongside
799. 24:06agents and agents completing work when
800. 24:08employees aren't working. So, this is
801. 24:09the future of work in my opinion. It's
802. 24:11not just human employees. It's not just
803. 24:13agent employees. It's agents and humans
804. 24:16working together. So, let me do a video
805. 24:18to show you how to stack number six and
806. 24:19seven on top of each other cuz I think
807. 24:20it'll be crazy. But that's all I have
808. 24:22for today. Make sure to subscribe to the
809. 24:24channel for more content like this so I
810. 24:26can help you get the edge in AI and my
811. 24:28free school communities below if you
812. 24:29want all of the GitHubs that I spoke
813. 24:31about today. And you can just drag and
814. 24:33drop that MD file that I'm going to give
815. 24:34you into claude and it will install
816. 24:36everything. So you can go ahead and use
817. 24:38this stuff yourself. This is how you 10,
818. 24:4220, 30, 40, 50x the capabilities of the
819. 24:45stock standard claude. I'll see you in
820. 24:47the next one. Have a lovely rest of your
821. 24:48day. Peace out.