YouTube2Text

My Favourite AI Workflow Nobody Talks About — Transcript

by Tina Huang · 4,847 words · 693 segments · language en · Watch on YouTube

Full transcript

  1. 0:00This is my favorite AI workflow right
  2. 0:01now. It is so powerful, and you don't
  3. 0:03even need to know how to code, although
  4. 0:04it does help. And it is essentially
  5. 0:05free, or at least much, much cheaper
  6. 0:07than paying for most software
  7. 0:09subscriptions. Let me explain by just
  8. 0:10showing you guys first some use cases,
  9. 0:12and then I will break it down
  10. 0:13step-by-step with a five-step framework
  11. 0:15on how to do it this AI workflow. Sounds
  12. 0:17good? A portion of this video is
  13. 0:19sponsored by Bolt.
  14. 0:22Use case number one, which was the bane
  15. 0:24of my existence, accounting. If you
  16. 0:26yourself are a small business owner, or
  17. 0:28know somebody that is a small business
  18. 0:30owner, you may understand what I mean
  19. 0:31when I say that accounting is the bane
  20. 0:33of my existence. Stuff like bookkeeping,
  21. 0:35filing taxes, payroll, filling out a
  22. 0:37million forms about a million things. It
  23. 0:39causes me physical pain just thinking
  24. 0:41about it.
  25. 0:43Especially because I do a lot of
  26. 0:45cross-border work, meaning I hire people
  27. 0:47from all over the world, and we also
  28. 0:48have clients from all over the world.
  29. 0:50So, I have this very specific kind of
  30. 0:51setup that is not optimal for most
  31. 0:53commercial accounting software. Except
  32. 0:55now, because we actually built our own
  33. 0:57hyper-specific AI app to do accounting
  34. 1:00for us. It is so custom to this company.
  35. 1:02It's able to deal with specific
  36. 1:03regulations from specific countries. It
  37. 1:05can interpret different kinds of
  38. 1:06currencies, and is able to do
  39. 1:08bookkeeping and forms and things like
  40. 1:09that that abide by the laws of multiple
  41. 1:12different countries as well. We're also
  42. 1:13using open-source AI and hosted locally,
  43. 1:15so it's not leaking things to the
  44. 1:17internet, and it's free, no matter how
  45. 1:19many documents, receipts, and invoices,
  46. 1:21and whatever that I want to feed through
  47. 1:22it. Commercial software, by the way,
  48. 1:23usually costs around like $15 to $30 per
  49. 1:25month. Use case number two, making
  50. 1:27custom slide decks. If you've been to
  51. 1:29any of my live streams, or have done any
  52. 1:31of our courses or workshops, you would
  53. 1:32have seen this very specific style of
  54. 1:34slides that we use. While previously it
  55. 1:36would have taken ages to make slide
  56. 1:38decks for every single presentation, now
  57. 1:40we built this super custom slide deck
  58. 1:42maker where we can feed in the
  59. 1:43information and have it just come out
  60. 1:45with these very beautiful slides, so
  61. 1:47streamlined and custom to exactly what
  62. 1:48we want, and also free. Use case number
  63. 1:51three, another example. Couple months
  64. 1:53back, for whatever reason, I I that I
  65. 1:55wanted to make a manga. But the problem
  66. 1:57is that I don't know how to draw. Like
  67. 1:59this is my drawing level. Seeing as I
  68. 2:01did not have the patience,
  69. 2:02unfortunately, to spend months and
  70. 2:03months drawing, which by the way,
  71. 2:05totally cool if that is what you want to
  72. 2:07do. But for me, it was kind of like a
  73. 2:08bucket list thing where I just wanted to
  74. 2:10make a manga. So, I made, you guessed
  75. 2:12it, a hyper-specific custom AI app that
  76. 2:14was able to make an entire manga just
  77. 2:16based upon text prompts and my terrible
  78. 2:19drawings. I have so many more examples.
  79. 2:21Let's see, what else have I made? Here's
  80. 2:23another custom email AI agent that's
  81. 2:24able to screen my emails and pre-draft
  82. 2:26them for me. And about a year ago, I
  83. 2:28also made another one that is able to
  84. 2:29handle like my seven calendars and tell
  85. 2:31me what I'm supposed to do every day.
  86. 2:32Yes, I have so many other examples, but
  87. 2:35I think you get my point. So, these are
  88. 2:37examples of my favorite AI workflow
  89. 2:39right now, which I don't really know
  90. 2:41what the name is for this. So, I shall
  91. 2:43call it hyper-specific apps. Definition
  92. 2:46of hyper-specific app is when you use
  93. 2:48AI-assisted coding to build software for
  94. 2:51very specific personal or business use
  95. 2:53cases that would never exist as
  96. 2:55commercial software. Here are the pros.
  97. 2:57I think the biggest pro of building a
  98. 2:59hyper-specific app is that you're able
  99. 3:00to build custom features that are so
  100. 3:03specific to your use case that it's
  101. 3:04impossible for you to actually find from
  102. 3:06commercial software. Another very
  103. 3:08attractive pro is that it is essentially
  104. 3:10free, minus the building and setup
  105. 3:12costs, especially if you use open-source
  106. 3:13models. Like on average, these apps cost
  107. 3:15me around $10 to $20 to build. And
  108. 3:17depending on where I'm hosting these
  109. 3:18apps, like where these apps live, you
  110. 3:20can run them again and again without
  111. 3:22incurring additional costs, which I will
  112. 3:24actually talk about like hosting and
  113. 3:25stuff like that a little bit later in
  114. 3:26the video. But yes, for now, just
  115. 3:28realize that at the very minimum, it's
  116. 3:29going to be a lot cheaper than buying
  117. 3:31like a monthly subscription for most
  118. 3:32software. With building your own apps,
  119. 3:34you also do have full control over your
  120. 3:36software. So, you don't need to be
  121. 3:37worried about like where it's sending
  122. 3:38its data, what's happening to it, things
  123. 3:40like that. And finally, you don't need
  124. 3:42to write code yourself to build
  125. 3:44something like this. Don't get me wrong,
  126. 3:45I think you're interested in learning
  127. 3:46how to code, you absolutely should still
  128. 3:48do it because there is a ceiling to what
  129. 3:50you can actually build without knowing
  130. 3:51how to code, and overall, it would also
  131. 3:53save you a lot more money if you knew
  132. 3:54how to code. But you don't need to know
  133. 3:55how to code when you're starting out
  134. 3:57building your hyper-specific apps, which
  135. 3:58is pretty cool. I'll explain the process
  136. 4:00in a bit. But first, I do want to
  137. 4:01mention some cons as well, some caveats.
  138. 4:04The first one is that it does take time.
  139. 4:06On average, it takes me about 2 to 3
  140. 4:07hours up to a day to build the first
  141. 4:09version of these hyper-specific apps.
  142. 4:12But there is a little bit of a learning
  143. 4:13curve. So when I was just starting out,
  144. 4:15or if you're someone who has no idea how
  145. 4:16to build anything, zero engineering
  146. 4:17background, no coding skills, it might
  147. 4:19take you a little bit longer, like maybe
  148. 4:20a few days to a week to get started
  149. 4:22properly. So some time and effort, yes.
  150. 4:24Also, it could require a little bit of
  151. 4:25setup and building cost. Specifically,
  152. 4:28if you are going to be using AI-assisted
  153. 4:29coding, you're probably going to need to
  154. 4:30pay for a subscription to use a coding
  155. 4:33agent of your choosing. And if you want
  156. 4:34your app to be able to run continuously,
  157. 4:36you might have to pay for some type of
  158. 4:38cloud service or a local machine that
  159. 4:41will continuously run. Although, if
  160. 4:42you're okay using like open-source
  161. 4:44coding models that are not like as
  162. 4:45optimized for coding, um or you're okay
  163. 4:48with things like maybe not running all
  164. 4:49the time, you can technically also do
  165. 4:51this for free. All right, great. Let me
  166. 4:52now explain the five-step process for
  167. 4:55how to do this AI workflow to build a
  168. 4:57hyper-specific app. Step number one is
  169. 5:00to identify which workflow you want to
  170. 5:02build a hyper-specific app for. There's
  171. 5:05like three categories of things that my
  172. 5:06hyper-specific apps usually fall under.
  173. 5:08So hopefully, this is also helpful for
  174. 5:10you to come up with what you want to
  175. 5:11build. The first one are things that are
  176. 5:12the bane of your existence, but
  177. 5:14unfortunately, to exist, you must
  178. 5:17actually do it. Biggest example for me
  179. 5:19was accounting. I will put on screen now
  180. 5:21some other examples in this category.
  181. 5:23The second is things that you've been
  182. 5:24procrastinating on because it's like you
  183. 5:26know it's important, but you just don't
  184. 5:29want to do it cuz it's a lot of work and
  185. 5:31you just don't feel like doing it. But
  186. 5:32you know, if you did it, it would
  187. 5:33actually be much better for you. An
  188. 5:35example for me is tracking my exercise,
  189. 5:37my food, and my sleep. Things that I
  190. 5:39know that I should do, but it's just
  191. 5:41like effort, you know? I will put on
  192. 5:43screen some other examples in this
  193. 5:44category. And the third category is
  194. 5:46something that you want to do, but you
  195. 5:48can't or haven't because you lack the
  196. 5:50skill or you don't have the time and you
  197. 5:52just like maybe don't have the resources
  198. 5:53or energy to to dedicate to doing it.
  199. 5:56This could be like a fun hobby project.
  200. 5:57For example, for me it was making a
  201. 5:59manga, except it was impossible because
  202. 6:01I'm so bad at drawing. And I know for
  203. 6:02others like many people want to build
  204. 6:04like a software or a game, but they just
  205. 6:06like don't have the skills and time to
  206. 6:08do that. I'm going to put on screen out
  207. 6:09some other examples in this category.
  208. 6:11And I do think this is like a really
  209. 6:12interesting category because it
  210. 6:14literally enables you to do something
  211. 6:16that you simply could not do before.
  212. 6:21>> [screaming]
  213. 6:23>> That's pretty cool. All right, I hope
  214. 6:24this helps you decide what you may want
  215. 6:26to build a hyper-specific app on. And do
  216. 6:28put in the comments what you want to
  217. 6:29build. I genuinely really want to know
  218. 6:31what you guys are interested in
  219. 6:32building. Step number two is to map out
  220. 6:35the process workflow. As in like map out
  221. 6:37the current existing workflow. This step
  222. 6:40can take a lot of time or very little
  223. 6:41time depending on how well you
  224. 6:43understand the process. Let me explain.
  225. 6:45For example, for the accounting stuff
  226. 6:46that I wanted to build this
  227. 6:47hyper-specific app for, a core feature
  228. 6:49would be bookkeeping. And this is
  229. 6:50something that I already understand the
  230. 6:52full workflow of because I've been doing
  231. 6:53it for like 5 years. Every month I
  232. 6:55needed to one, collect all of the
  233. 6:57receipts and invoices. Two, match them
  234. 6:59to all of my corporate credit cards and
  235. 7:01also my bank statements. And then three,
  236. 7:03have a final deliverable as a
  237. 7:04spreadsheet with all of the entries as
  238. 7:06transactions and the supporting
  239. 7:08documents for each transaction. And that
  240. 7:10would be the mapped out process. Didn't
  241. 7:11take me a lot of time to figure this out
  242. 7:12because I had already done it so many
  243. 7:14times. But for mapping out the workflow
  244. 7:15for generating for example, I literally
  245. 7:18had no idea the process of making manga.
  246. 7:20So in order to map out this workflow, I
  247. 7:22actually have to go take an entire
  248. 7:23course from an expert to map out the
  249. 7:25process of what it takes to make a
  250. 7:27manga, which in case you're curious is
  251. 7:29this. So after you map out your
  252. 7:30workflow, we can then move on to step
  253. 7:32three, which is identifying where your
  254. 7:34hyper-specific app would fit in in that
  255. 7:36workflow. As amazing as technology and
  256. 7:38software and AI is these days, it's
  257. 7:40still not possible to just like take
  258. 7:42your entire workflow that you just
  259. 7:43mapped out and stick it into AI and just
  260. 7:45tell it to like build the entire thing,
  261. 7:46unfortunately. And also, maybe you don't
  262. 7:48actually want that, right? Like part of
  263. 7:50the process, you might want it to remain
  264. 7:51manual. Or there's no need for you to do
  265. 7:53that. So, that's why we do want to
  266. 7:54explicitly go in there and identify
  267. 7:56which parts of the workflow we want to
  268. 7:57be part of the hyper-specific app. Let's
  269. 7:59use the manga example here. So, here's
  270. 8:00the workflow for making manga, which I
  271. 8:02learned from taking that course from a
  272. 8:04professional mangaka. I learned that
  273. 8:05there were nine steps to making a manga.
  274. 8:07One was to pick a topic, including
  275. 8:08central values and central themes that I
  276. 8:10wanted to address. Two was to design the
  277. 8:12characters. Three was to have a major
  278. 8:14plot synopsis, like an explanation for
  279. 8:16what the manga should be about. Four is
  280. 8:18to write out the entire story and split
  281. 8:20them into chapters and scenes. Five is
  282. 8:22then to transform these scenes and
  283. 8:23chapters into panel descriptions, like
  284. 8:26what each little panel drawing should be
  285. 8:28portraying. Six is to actually sketch
  286. 8:30out these scenes and draw them and then
  287. 8:32design the panels as well, like the how
  288. 8:34the panels are all supposed to fit
  289. 8:35together and how the pages are exposed
  290. 8:37to be divided. I was really surprised by
  291. 8:38how much I thought and detail and
  292. 8:40process went into just constructing how
  293. 8:42the panels fit together. And seven is to
  294. 8:44fill in the details of these sketches,
  295. 8:45like actually draw them nicely. Eight is
  296. 8:47to then finally bind them all together
  297. 8:49into one whole manga entity. And nine is
  298. 8:52to go publish it. So, that is the entire
  299. 8:54workflow, but I don't actually want to
  300. 8:55include this entire workflow in the
  301. 8:58hyper-specific app itself. Because steps
  302. 9:00one to five actually involve a lot of
  303. 9:02writing and I actually really like
  304. 9:03writing and I think it's really fun.
  305. 9:05It's also where a lot of creativity
  306. 9:06comes in. So, I don't actually want this
  307. 9:08to be like auto-generated for me. I
  308. 9:10might like use AI to help me, you know,
  309. 9:12flesh out some stuff or like bounce
  310. 9:14ideas and stuff like that, but I do want
  311. 9:15to keep this process manual. Now, I do
  312. 9:17need a lot of help with and what I do
  313. 9:19want this hyper-specific app to be able
  314. 9:21to do is steps six, seven, and eight.
  315. 9:24Cuz that's like the visual parts, like
  316. 9:25sketching out the scenes, designing the
  317. 9:27panels, filling the details, and then
  318. 9:29binding them all together. This is like
  319. 9:30the blocker for me. It is impossible for
  320. 9:32me to do this because I don't know how
  321. 9:33to draw. So, that's why I want my
  322. 9:34hyper-specific app to be able to take
  323. 9:36care of these steps for me. And finally,
  324. 9:38step number nine, which is the publish,
  325. 9:40I don't actually care that much about
  326. 9:41that cuz I'm just doing it for my own
  327. 9:43like bucket list enjoyment. Amazing,
  328. 9:45great. You now know what you want your
  329. 9:47hyper-specific app to do. We can now
  330. 9:49move on to the next step, which is
  331. 9:50actually building your hyper-specific
  332. 9:52app using good AI coding practices. I've
  333. 9:55actually made an entire video on how to
  334. 9:57use AI coding in order to build apps
  335. 9:58both using code and no-code tools. So, I
  336. 10:01do recommend checking that out. I'll
  337. 10:02link it over here. I also made a videos
  338. 10:03where I show you guys me building these
  339. 10:05apps and I'll link them in the
  340. 10:05description as well. So, do check those
  341. 10:07out when you get serious on building
  342. 10:08your app, but I do want to give you a
  343. 10:09very fast quick start guide in this
  344. 10:11portion of the video, too. So,
  345. 10:12basically, three steps. You want to take
  346. 10:14what you want your hyper-specific app to
  347. 10:15do, and then change that into what is
  348. 10:17called a product requirement document.
  349. 10:19This is a standard document for defining
  350. 10:22the purpose, features, functionalities,
  351. 10:23and the behaviors of a product. This is
  352. 10:25what you're going to be feeding into
  353. 10:26your AI coding tool, so it knows what
  354. 10:29you're supposed to be building. I'm
  355. 10:29going to actually link a prompt in the
  356. 10:31description, which you can just copy
  357. 10:32paste into like any chatbot of your
  358. 10:34choosing, whether that be like ChatGPT,
  359. 10:36Claude, Grok, Gemini, whatever you want.
  360. 10:38And it would actually like ask you
  361. 10:39questions to help you come up with this
  362. 10:42PRD. See, I got you. All right, so after
  363. 10:44you have your PRD, you then want to pick
  364. 10:46your AI coding tool to build your
  365. 10:47hyper-specific app. In the AI coding
  366. 10:49fundamentals video, I literally like
  367. 10:51have an entire section dedicated to
  368. 10:53explaining tools, so please go to that
  369. 10:55video for a lot more details. But
  370. 10:56basically, there are range of different
  371. 10:58AI coding tools that range from
  372. 11:00completely no-code to for pro
  373. 11:02developers, and those are good for
  374. 11:03specific types of use cases as well. For
  375. 11:05me personally, because this is literally
  376. 11:07my job, I have tested pretty much the
  377. 11:10full spectrum, like all of the different
  378. 11:12AI coding tools out there. And I would
  379. 11:13use different AI coding tools for
  380. 11:15different purposes. Say for example, if
  381. 11:16I know that this hyper-specific app I'm
  382. 11:18trying to build is going to be like a
  383. 11:19big, complex project with lots of
  384. 11:21technical details, maybe even like
  385. 11:22pretty obscure technical things included
  386. 11:24in it, and at some point I might even
  387. 11:26want to bring in other developers to
  388. 11:27help me with it, then I would go
  389. 11:29directly with a code-based pro developer
  390. 11:31tool. Like the accounting software, for
  391. 11:32example, I primarily use Cloud Code for
  392. 11:34that. And when the project became bigger
  393. 11:36and I wanted to do more parallelization,
  394. 11:38like having multiple coding agents
  395. 11:40working at the same time and also like
  396. 11:41working with other developers, I
  397. 11:42switched over to Warp, which is really
  398. 11:44good for that kind of stuff. But, for
  399. 11:45the hyper-specific apps I'm making
  400. 11:47slides and also for the generator,
  401. 11:49these are web-based well-scoped projects
  402. 11:51that have pretty standard integration
  403. 11:53needs. So, I actually built these using
  404. 11:54Bolt, who I'm actually really excited to
  405. 11:56say is the sponsor of this portion of
  406. 11:57the video. I really like their product
  407. 11:59as a web-based option. using them for
  408. 12:01over a year and a half now. So, thank
  409. 12:04you so much for reaching out to sponsor
  410. 12:05this portion of the video, Bolt. Bolt is
  411. 12:07my go-to for creating really fast
  412. 12:08AI-powered software and fully functional
  413. 12:10web apps. I built this generator
  414. 12:12and the slide generator using Bolt. It
  415. 12:14lets you go from an idea to a fully
  416. 12:16deployed production-ready app just by
  417. 12:17describing what you want in plain
  418. 12:19English. It's crazy cuz it's not just
  419. 12:21the prototype. It's like an actual
  420. 12:22working app with a real back-end and
  421. 12:24running entirely on your browser. They
  422. 12:26also dropped two new features that I
  423. 12:27think will be really relevant for you if
  424. 12:29you want to build hyper-specific apps.
  425. 12:30The first one is image generation right
  426. 12:33inside Bolt. So, if you need a
  427. 12:34thumbnail, a mock-up, or any visual
  428. 12:36assets for whatever it is that you're
  429. 12:37building, you don't have to jump between
  430. 12:38tools anymore. You can just generate it
  431. 12:40directly within the platform itself. The
  432. 12:41second is connectors, and this is what
  433. 12:44I'm most excited about. Bolt now
  434. 12:45directly plugs into platforms like
  435. 12:47GitHub, Notion, Linear, Jira, and many
  436. 12:49more. So, your app is able to pull from
  437. 12:50real data, not just placeholder content.
  438. 12:53Now, that is what it takes from a
  439. 12:54prototype to something that you actually
  440. 12:56can ship. Over 7 million people are
  441. 12:58using Bolt now, including teams from
  442. 13:00Fortune 500 companies. It is free to
  443. 13:02start and no credit card required. Link
  444. 13:04is in the description. Thank you so
  445. 13:05much, Bolt, for sponsoring this portion
  446. 13:06of the video. Now, back to the video.
  447. 13:08Finally, to round out my brief guide to
  448. 13:09AI coding tools, I want to say that
  449. 13:11there are free options as well if cost
  450. 13:13is a really limiting factor, too. You
  451. 13:15can use open-source coding models. Just
  452. 13:17know that the coding experience itself
  453. 13:18is not going to be as optimized. And you
  454. 13:20need to be more careful about things
  455. 13:22like version control and and privacy and
  456. 13:24security things. So, I would only
  457. 13:25recommend this option to people who do
  458. 13:27have a little bit of coding experience
  459. 13:29already and a lot more patience. I'll
  460. 13:31put on screen some tools and models to
  461. 13:32check out as well here. Okay, great. So,
  462. 13:34now you've picked your AI coding tool.
  463. 13:35You want to take your PRD, remember your
  464. 13:37PRD from earlier, and stick that into
  465. 13:39your AI coding tool to get started on
  466. 13:42the process of building your
  467. 13:43hyper-specific app. Sorry for sounding
  468. 13:45like a broken record here, but in the AI
  469. 13:47coding fundamentals video, I go into a
  470. 13:49lot more detail about how to go through
  471. 13:51this building process. So, please do
  472. 13:53check out that video. The AI coding
  473. 13:54principles that I presented in that
  474. 13:56video, I have dubbed the tiny ferrets
  475. 13:58carry dangerous code framework, which
  476. 14:01stands for thinking, as in always be
  477. 14:03thinking and rethinking deeply about the
  478. 14:05features that you're building. You must
  479. 14:06not throw your PRD at your AI coding
  480. 14:08tool and just run away. Second is the
  481. 14:10focus on existing frameworks. You don't
  482. 14:12need to reinvent the wheel as you're
  483. 14:13building your app. There are most likely
  484. 14:16existing frameworks that other
  485. 14:17developers have already built. So,
  486. 14:19instead of trying to do something from
  487. 14:20scratch, help your AI agent provide
  488. 14:22relevant documentation and frameworks to
  489. 14:25your AI tool, so we can draw on the
  490. 14:26knowledge of past work. C stands for
  491. 14:28checkpoints. Create checkpoints as
  492. 14:30you're building your app, thinking about
  493. 14:31version control using Git GitHub. D
  494. 14:34stands for debugging. Debug with care.
  495. 14:37Debug as you build. And C stands for
  496. 14:39context. Provide as much context as
  497. 14:42possible to your AI coding tool in the
  498. 14:44form of screenshots, examples, more
  499. 14:47detailed descriptions. The more context
  500. 14:49and details that you can provide your
  501. 14:51AI, the better able it would be able to
  502. 14:52build what you have in mind. As a bonus
  503. 14:54tip, if the hyper-specific AI app that
  504. 14:57you're trying to build, you also want to
  505. 14:58include an AI component to it, I know,
  506. 15:01right? Very meta here. You're using AI
  507. 15:03to build an app that has AI in it. Think
  508. 15:05carefully about what kind of model you
  509. 15:07want to use. I'm going to link in the
  510. 15:08description a couple of videos that I've
  511. 15:10done that may be able to help you decide
  512. 15:12which models to use. Wow, great. If you
  513. 15:14follow these steps, by now you should
  514. 15:15have the first version of your
  515. 15:17hyper-specific app. The first time that
  516. 15:19you do this probably going to take you a
  517. 15:20little bit longer, especially if you've
  518. 15:22never done AI coding or built apps
  519. 15:24before. But, as you build more and more
  520. 15:26of these, I promise you it's going to
  521. 15:27become so much easier and faster. These
  522. 15:29days, depending on how complex the app
  523. 15:31I'm trying to build is, I can usually
  524. 15:32get the first version out in maybe like
  525. 15:352 to 3 hours up to like a day. Just
  526. 15:37saying, prior to AI coding tools, the
  527. 15:40development of really any type of
  528. 15:41application would have taken you like
  529. 15:43months to do. And there's no way you
  530. 15:46could do this if you didn't know how to
  531. 15:47code, obviously. But, now you can, and
  532. 15:49you can do it in like a few hours. That
  533. 15:51is pretty magical. All right, time for
  534. 15:53the last step of this framework in
  535. 15:55developing your hyper-specific app. And
  536. 15:57that is hosting. Hosting is providing
  537. 16:00the infrastructure of where your app
  538. 16:02lives and runs, so you and other people
  539. 16:05can access it. Roughly speaking, there
  540. 16:06are three different categories of
  541. 16:08hosting options. The first one is cloud
  542. 16:10options. This is when you can rent
  543. 16:12compute, and you can dump this code like
  544. 16:14the application that you build and have
  545. 16:15it run on these cloud services, and like
  546. 16:18get other people to run it for you. You
  547. 16:20pay for how much compute, like how much
  548. 16:21running of this application that you
  549. 16:23use. It can scale better, so you can
  550. 16:25have one user using it or 10,000 people
  551. 16:27using it. And there's also other tons of
  552. 16:28managed services that you can use as
  553. 16:30well, like databases or queries. As an
  554. 16:32analogy, sort of like renting a hotel
  555. 16:34room. You're paying per night or hourly,
  556. 16:36sometimes. There's like managed services
  557. 16:38like a concierge, there's like a mini
  558. 16:40fridge, and you can like invite lots of
  559. 16:42other people over, and you can like
  560. 16:43leave anytime you want as well. You
  561. 16:45don't have to like clean up after
  562. 16:46yourself. This might be the best option
  563. 16:48for you if your hyper-specific app that
  564. 16:49you're building you can foresee that
  565. 16:51there's going to be unpredictable
  566. 16:52traffic. There might be like a lot of
  567. 16:53people using it at the same time, and
  568. 16:55then suddenly like nobody using it for a
  569. 16:57while. Or they involve like other team
  570. 16:58members coming in and also like tweaking
  571. 17:00the app and doing stuff to it as well.
  572. 17:02And any that involves having to scale
  573. 17:03and you don't really want to deal with
  574. 17:04the complexities of that. The downside
  575. 17:06of this is that it can get very
  576. 17:08expensive cuz it's like a hotel room,
  577. 17:09right? You get like all these like fancy
  578. 17:11things as a premium option here. You may
  579. 17:13be locked into like a specific vendor
  580. 17:15and their services, and it's not
  581. 17:17private. What happens in the hotel
  582. 17:19doesn't stay in the hotel, potentially.
  583. 17:21Some examples of a cloud options would
  584. 17:23be like AWS, Google Cloud, Azure. I'm
  585. 17:25going to put on screen now some other
  586. 17:26examples of the cloud option as well.
  587. 17:28Now, the second option you can consider
  588. 17:30when you're hosting your hyper-specific
  589. 17:32app is a VPS, a virtual private server.
  590. 17:35This is a fixed virtual machine that
  591. 17:37you're renting out 24/7 for a flat fee.
  592. 17:39It's kind of like renting out an
  593. 17:41apartment. You have a fixed monthly
  594. 17:43amount that you're renting it for and
  595. 17:45during that time it's your
  596. 17:46responsibility to keep it tidy and not
  597. 17:48disturb your neighbors. And you can also
  598. 17:50like decorate it and do what you want to
  599. 17:51it within reason. This is good if you're
  600. 17:53going to be building a bunch of
  601. 17:54different hyper-specific apps that are
  602. 17:57all like relatively low traffic, as in
  603. 17:59you're going to be the only person using
  604. 18:01the app or maybe just like a couple
  605. 18:03other people. It's customizable, you
  606. 18:04manage it yourself, and you have a
  607. 18:06little bit more privacy as well cuz what
  608. 18:07you store inside, like data for example,
  609. 18:10is a little bit more private. Do you
  610. 18:11want to make a caveat here that's not
  611. 18:13completely private though because
  612. 18:14technically your virtual machine is
  613. 18:16still physically sitting on somebody
  614. 18:19else's data server, right? Like the
  615. 18:20hardware still belongs to somebody else.
  616. 18:22So, if they like really, really, really
  617. 18:24wanted to go take a look, like your
  618. 18:25landlord if they really, really, really
  619. 18:26want to come in and look at your
  620. 18:27apartment, they can. They still have the
  621. 18:29keys. Like they probably won't, but they
  622. 18:31could. So, yes, that is one of the cons
  623. 18:33and you do need to manage things
  624. 18:34yourself. You got to clean up after
  625. 18:35yourself. And even if you don't have the
  626. 18:37app running, you still have to pay for
  627. 18:38that rental fee. So, I think this is
  628. 18:40more worth it if you do have multiple
  629. 18:41apps that are going to be running or you
  630. 18:43are actually going to be like constantly
  631. 18:45using the apps that you built. Some
  632. 18:46examples of VPS providers include
  633. 18:48Hostinger, Hetzner, Digital Ocean. Put
  634. 18:50some other examples on screen now. Now,
  635. 18:52finally, the third option is hosting
  636. 18:55your hyper-specific app on your own
  637. 18:57hardware. This is a physical machine
  638. 18:59that you would just buy one time and it
  639. 19:01would just like run on your desk.
  640. 19:03Continuing with our property analogy,
  641. 19:04this is like truly owning a home. The
  642. 19:06pros like you buy it once, it's truly
  643. 19:08yours. You really have like complete
  644. 19:10privacy. Nothing is going to leave your
  645. 19:12local machine, and you exactly know what
  646. 19:15you want to do with it. Fully
  647. 19:16customizable to whatever you want it to
  648. 19:18do. And after you buy it one time, the
  649. 19:20only thing that you really need to pay
  650. 19:21for is like just the electricity for
  651. 19:23running your hardware. But, just like
  652. 19:25owning a home, there are also downsides.
  653. 19:27Because you own everything, you also are
  654. 19:29responsible for everything. If something
  655. 19:31breaks, you got to fix it yourself,
  656. 19:33completely. Something needs updates, got
  657. 19:35to do it yourself. Your home internet
  658. 19:37could also be a bottleneck if it's not
  659. 19:39fast enough. Physical failures,
  660. 19:41downtime, things like that. This
  661. 19:42technically could also be the cheapest
  662. 19:44option for you if your laptop is
  663. 19:46relatively modern, because you can
  664. 19:48actually use your laptop to be running
  665. 19:50your app. But, the issue is that if
  666. 19:52you're like doing other stuff
  667. 19:53simultaneously, it might make your app
  668. 19:55really slow or just like completely
  669. 19:56shuts down. Or if you close your laptop,
  670. 19:59then it would stop working. So, that's
  671. 20:00why people often choose to buy separate
  672. 20:02hardware to run their apps. Like, that
  673. 20:04machine is going to be dedicated to
  674. 20:06running your hyper-specific app, and
  675. 20:08it's going to be running 24/7. The very
  676. 20:10popular option is to buy a Mac mini for
  677. 20:12this, but I'm going to put on screen now
  678. 20:13some other options that are available as
  679. 20:15well. Cool. Now that you've have hosting
  680. 20:18done, you officially have a properly
  681. 20:20functioning hyper-specific app. Isn't
  682. 20:22that cool? I'm genuinely so excited to
  683. 20:24have been able to share this this
  684. 20:25workflow with you guys. I use it so
  685. 20:27much, and it really is my favorite AI
  686. 20:30workflow right now. I hope you're going
  687. 20:31to be building your own hyper-specific
  688. 20:33apps soon, too, if you haven't already.
  689. 20:35Let me know in the comments what you
  690. 20:36want to build now. Thank you so much for
  691. 20:38watching until the end of this video,
  692. 20:39and I will see you guys in the next
  693. 20:41video or live stream.

About this transcript

This page contains the full transcript of My Favourite AI Workflow Nobody Talks About by Tina Huang, generated from the public captions YouTube serves with the video. The transcript has 4,847 words across 693 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.