They did it... — Transcript

Full transcript

1. 0:00So, I've been working on a project
2. 0:01recently that made me rethink how all of
3. 0:04software is going to be built. I was
4. 0:06setting up a new PC and I needed to read
5. 0:09a PDF. And the first thing that I did
6. 0:11was try to open it, of course, but
7. 0:14Windows doesn't have a good built-in PDF
8. 0:16reader. And so, reluctantly, I had to go
9. 0:20to Adobe Acrobat. Now, I know there's a
10. 0:22few other alternatives out there, but
11. 0:23Acrobat is kind of the standard. So, I
12. 0:26went to download it. I installed it and
13. 0:28I hated every second of it. It's
14. 0:30bloated. It comes with a bunch of
15. 0:31features I don't need. It's trying to
16. 0:34push other products. It's even trying to
17. 0:36get me to log in. And all the while I'm
18. 0:39sitting here and all I'm trying to do is
19. 0:41read, maybe edit a PDF. This is really
20. 0:43simple stuff. In that moment, rather
21. 0:46than suffer with this piece of software,
22. 0:48I decided to make my own. And I know
23. 0:50this sounds really basic like okay a PDF
24. 0:53reader very basic software but it
25. 0:55actually has massive implications that I
26. 0:58want to talk about in this video because
27. 1:00then I started building other pieces of
28. 1:01software that I need and not only that
29. 1:03there are other developers out there who
30. 1:05are building entire software suites like
31. 1:08somebody literally cloned the entire
32. 1:10Adobe suite including Premiere and After
33. 1:13Effects. And so I feel like we're
34. 1:14getting into this world where consumers
35. 1:17are going to be building their own
36. 1:19software. And there are a few things
37. 1:20that needed to come together all at the
38. 1:22same time to make this happen. And over
39. 1:25the last month, I am now convinced they
40. 1:28have. Number one, the models had to get
41. 1:31good enough. And I'm going to get back
42. 1:32to that in a second. Number two, they
43. 1:34had to be cheap enough. If you're having
44. 1:36to burn tens of millions of tokens to
45. 1:40build a piece of software that you can
46. 1:41go buy for 10, 20, $30, why would you do
47. 1:44it? And then three, it had to be easy
48. 1:46enough. And the models wrapped in agents
49. 1:49have become so autonomous that it very
50. 1:51much is that easy. So this has been
51. 1:53growing over the last few months. the
52. 1:55models have gotten so good and Opus 5.5
53. 1:59was kind of the peak, the turning point
54. 2:02at which I can just tell it to make a
55. 2:05piece of software or to clone a piece of
56. 2:07software and it would go off and do so.
57. 2:10And it wasn't just the model quality
58. 2:12that allowed it to do that. It was also
59. 2:13the fact that the price of the tokens
60. 2:16and the cost to complete the task had
61. 2:20dropped tremendously. And so there's
62. 2:23this calculus going on at all times. How
63. 2:27much does it cost to build a custom
64. 2:30piece of software for what you need and
65. 2:32to maintain it versus how much does it
66. 2:34cost to buy a piece of software often
67. 2:37times in a subscription model. So you're
68. 2:40paying this ongoing fee for it. And I
69. 2:42think we've hit this critical point in
70. 2:44which the cost of tokens to build and
71. 2:47maintain a piece of software is now less
72. 2:49expensive than simply buying it. And the
73. 2:52cool thing is when you're building it
74. 2:54yourself, you get the exact features you
75. 2:56need. Even if the feature wouldn't have
76. 2:59applied or been attractive to anybody
77. 3:02else, it doesn't matter. You build it
78. 3:04for yourself. So the models have now
79. 3:06gotten to the point where they're good
80. 3:07enough. I mean, let me give you a few
81. 3:10examples. So I've been doing model
82. 3:12testing for a couple years now. It
83. 3:15started way back where I just said,
84. 3:16"Okay, make the game snake in Python."
85. 3:20And that was like dead simple. And the
86. 3:22models just kept getting better where
87. 3:24that snake game was too easy. So I moved
88. 3:27on to harder things. And one of those
89. 3:29kind of more well-known harder things
90. 3:31was build a Rubik's cube simulation. And
91. 3:34the models got so good at that. And so I
92. 3:37had to come up with these new tests to
93. 3:40test the latest models on. And what I
94. 3:42decided to do was just simply rebuild
95. 3:45software from the ground up. and I would
96. 3:47point it at Microsoft Excel. One of the
97. 3:50most complex pieces of software on the
98. 3:53planet, one of the most used pieces of
99. 3:56software on the planet. I would just say
100. 3:58a simple prompt, go clone Excel and it
101. 4:01would do pretty well for a while. When
102. 4:04we were on like GPT5 model and prior to
103. 4:06Fable coming out, they would be able to
104. 4:08get pretty close. there would be a lot
105. 4:10of rough edges and it wouldn't be
106. 4:12feature complete but you know it would
107. 4:15look like Microsoft Excel and then these
108. 4:17latest generation of models dropped and
109. 4:19it would just simply clone Excel like
110. 4:21end to end not over multiple days
111. 4:23either. I mean I'm talking about within
112. 4:25a matter of 90 minutes or maybe a couple
113. 4:28hours it would fully clone Excel every
114. 4:31feature every piece of functionality. I
115. 4:33would make it browserbased and in fact
116. 4:35I'll go ahead and drop my latest version
117. 4:37down below so you can play with it. But
118. 4:38the point is it was able to build it.
119. 4:40And now if you wanted to use Microsoft
120. 4:43Excel, the standalone Microsoft Excel is
121. 4:46$180
122. 4:48for a single license, but you get a
123. 4:51lifetime, which, you know, fine. It's a
124. 4:53very valuable piece of software. $180
125. 4:56over the lifetime of the piece of
126. 4:57software. Or you're going to be paying a
127. 4:59subscription, which is, you know,
128. 5:01between $7 and $10 per month, and you
129. 5:03get Excel and Word and PowerPoint and
130. 5:05everything else. But then I spent maybe
131. 5:07$10 in tokens to rebuild it with
132. 5:11complete feature parody. And now I can
133. 5:14take that and open source it and give it
134. 5:16to the world for free. And it is
135. 5:18becoming more and more apparent that the
136. 5:22value of software is approaching zero.
137. 5:24the cost of software will be zero
138. 5:28because anybody with a $200 a month
139. 5:33subscription to Claude or to Chad GPT
140. 5:36can simply just build any software they
141. 5:38want and release it to the world. And
142. 5:39then somebody else can take that base
143. 5:42and modify it how they want costing a
144. 5:45fraction of what the original buildout
145. 5:47costs and everybody just gets exactly
146. 5:49what they need. And if this sounds
147. 5:51familiar, it's called open source. I
148. 5:54really do believe that all software in
149. 5:56the future is going to be de facto open-
150. 5:58source. And so I'm going to show you
151. 5:59some other projects that I made in a
152. 6:01minute. But the reason I really think
153. 6:03that is because the recent growth and
154. 6:06popularity of decompiling software. And
155. 6:09if that sounds complicated, it's really
156. 6:12not. And I've been building all of this
157. 6:14software lately on my new Nvidia RTX
158. 6:16Spark machine. It is truly phenomenal
159. 6:19and it's become my daily driver laptop.
160. 6:21Let me tell you about it. This is the
161. 6:23new Microsoft Surface Laptop Ultra and
162. 6:26it is unlike other laptops. This is all
163. 6:28powered by the Nvidia RTX Spark. Nvidia
164. 6:31is now making both the CPU and GPU. It's
165. 6:33a Grace CPU, a Blackwell RTX GPU, and it
166. 6:38shares up to 128 gigs of unified memory
167. 6:42and RTX Spark, which is what's powering
168. 6:44it, has CUDA natively, so you have
169. 6:47Nvidia software for accelerating AI
170. 6:49right on the machine. I've been testing
171. 6:51it. I've been running local models. I've
172. 6:53been playing Gears of War. I've been
173. 6:55video editing and kind of all at the
174. 6:58same time. So, the materials that they
175. 6:59used to build this laptop feel
176. 7:01absolutely premium. And even when I'm
177. 7:03running AI and playing games on it, it
178. 7:06doesn't get that hot, and I don't really
179. 7:07understand how they did that. The fan is
180. 7:09relatively quiet. It comes with a 15-in
181. 7:12touchscreen, so I can play with my
182. 7:13little Rubik's Cube simulation all day
183. 7:15long. So, go check out the Nvidia RTX
184. 7:17Spark laptops. I'll drop a link down
185. 7:19below. And so besides for the software
186. 7:21that I was building for myself, I
187. 7:23started seeing other people decompile
188. 7:25software and probably the most
189. 7:27well-known one which I covered in a
190. 7:29video is called Artcraft. It is the
191. 7:31entire Adobe suite of products,
192. 7:34open-source and completely free. Here's
193. 7:36PhotoCraft. I mean, look at this. It
194. 7:38looks exactly like Photoshop. Here's
195. 7:40Vectorcraft for creating vectors. Here's
196. 7:43Filmcraft, which is basically Adobe
197. 7:45Premiere. Here it is. It looks
198. 7:47beautiful. And so what this guy did was
199. 7:49simply point his AI at an existing piece
200. 7:51of software and said, "Iterate until you
201. 7:54clone it." Exactly. And so this was one
202. 7:56of the most exciting projects I had seen
203. 7:58in a while. Okay, so it's definitely
204. 8:00happening now. And I'm going to show you
205. 8:01a few pieces of software that I've been
206. 8:04building myself. But first, how did we
207. 8:06actually get here? So a couple years
208. 8:07ago, I made this video called SAS is
209. 8:09dead. And it proposed this thesis that
210. 8:11all software was going to collapse down
211. 8:13into agents. agents just being able to
212. 8:15accomplish whatever like the actual
213. 8:18application wasn't even needed anymore.
214. 8:20And then the other thing is kind of a
215. 8:22layer of database like the actual ground
216. 8:24truth which needs to be stored in a more
217. 8:26traditional database. It can't be done
218. 8:28by AI. And so it's going to be those two
219. 8:30layers. And that was specifically in the
220. 8:32enterprise, but I never really thought
221. 8:33that consumers would be able to build
222. 8:35their own software. And maybe not they
223. 8:38weren't going to be capable to do it,
224. 8:40but it wouldn't make sense to do. But
225. 8:42then, you know, fast forward to just a
226. 8:44few weeks ago and that is not the case
227. 8:46anymore. Like the math to build your own
228. 8:49software now favors building your own
229. 8:51software versus buying. And so, let me
230. 8:53show you Folio. This is my PDF reader
231. 8:56editor. You can eign with it. You can
232. 8:59use AI to ask questions about the PDF.
233. 9:02And if there's any other functionality I
234. 9:04want, I can just simply add it in here.
235. 9:06And it's lightweight. It's lightning
236. 9:08fast. It's completely free. It's open
237. 9:10source. I'll drop a link down below so
238. 9:11you can get it and use it if you want.
239. 9:13You can fork it, make it your own. So,
240. 9:16look here. I'll click here and, you
241. 9:18know, simple. It's a PDF reader. You can
242. 9:21draw on it. You can add notes to it. You
243. 9:23can ask questions to it. You can do a
244. 9:25whole bunch of cool stuff. And it just
245. 9:27works. And I'm not going to pay for this
246. 9:29software again. Uh there's no reason to
247. 9:31download Adobe Acrobat. And then there
248. 9:33was this piece of software I used for a
249. 9:35while called One of 10. It allowed me to
250. 9:38see how other people's videos were
251. 9:39performing. It gave me stats on those
252. 9:41videos, stats on different channels. And
253. 9:43then they just started adding all this
254. 9:45functionality that I never needed. And
255. 9:47now they're making me authenticate, give
256. 9:49them my information. It started to
257. 9:52clutter up YouTube. Like they would take
258. 9:54up 30% of the real estate of the screen.
259. 9:56They just have a bunch of features that
260. 9:58I will never need. So I built my own.
261. 10:00This is called Outlier Tracker. And now
262. 10:02I can see on each video. It gives me
263. 10:05some stats on the video. I can track the
264. 10:07video. I can do a channel analysis. It's
265. 10:09the exact functionality I needed from
266. 10:11one of 10. And now it's mine. And this
267. 10:13took like 30 minutes to build. And
268. 10:15here's that Excel clone that I
269. 10:16mentioned. It is full Excel in the
270. 10:18browser. I mean, it works perfectly.
271. 10:21It's absolutely beautiful. It has all
272. 10:23the functionality, every single piece of
273. 10:26functionality that you would ever want.
274. 10:28And so, where does this fall short? I
275. 10:30think one of the common arguments
276. 10:32against individuals, consumers building
277. 10:34their own software and also enterprise
278. 10:36building their own software is that it's
279. 10:38not enough just to build it, you also
280. 10:40have to maintain it and you have to
281. 10:42think about security and keeping it up
282. 10:44to date. And there's a reason for that.
283. 10:46there have been increasing amount of
284. 10:48hacks. And so if you're building, you
285. 10:51know, what people would consider slop
286. 10:53software or vibecoded software, maybe
287. 10:55it's going to be full of a bunch of
288. 10:56holes and you have to maintain the
289. 10:59software over time, but I don't know if
290. 11:01that's necessarily true. I think AI is
291. 11:04writing better and better software and
292. 11:06the models getting better means that
293. 11:08they will be more hardened and more
294. 11:10secure. And there's this graph that you
295. 11:12can plot in which the cost of just
296. 11:15running an agent all the time constantly
297. 11:19updating the software constantly looking
298. 11:21for security issues is actually less
299. 11:23than buying the software from someone
300. 11:24else. But let's say you don't even want
301. 11:26to do that. I mean you can go and
302. 11:29download an open source piece of
303. 11:30software because I guarantee there are a
304. 11:32lot of other developers who are just
305. 11:34creating software and giving it out for
306. 11:36free like Artcraft and they are
307. 11:38maintaining it. they are just
308. 11:39continuously throwing tokens at it. And
309. 11:41as the cost of tokens continues to
310. 11:44decrease to basically zero, there's
311. 11:48almost like this just desire to build
312. 11:51and it's essentially free to build and
313. 11:54you can put stuff out there and have
314. 11:55people use it and it's exciting and just
315. 11:59kind of the love of the craft of
316. 12:01building. And so I really do think we're
317. 12:03in this renaissance of software where
318. 12:05every individual is just going to build
319. 12:08the software that they need. Exactly.
320. 12:10And the one last thing I want to mention
321. 12:13is Omari, which is a new operating
322. 12:16system by the guy who created Ruby on
323. 12:18Rails. And they have AI embedded in its
324. 12:21DNA to the point where you can customize
325. 12:24the operating system to be exactly what
326. 12:26you want it to be at all times. And
327. 12:28that's super exciting to me. I mean, you
328. 12:30can more or less do some of that with
329. 12:33Windows right now, but to fully
330. 12:35customize the operating system at all
331. 12:37times, I think it's quite exciting to
332. 12:39think maybe there is room for a new
333. 12:41operating system. So, I encourage you go
334. 12:44check out open- source, try to build a
335. 12:46little app for yourself, see how far you
336. 12:48get, and then see how far you can go.
337. 12:50And one of the best open source models
338. 12:51on the planet right now is Deepseek V4.1
339. 12:54Flash. I did a full video about that.
340. 12:56You should go check that out right