YouTube2Text

Claude Code Can Be Your Second Brain — Transcript

by Every · 14,569 words · 1,950 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Noah Brier might have the coolest Claw
  2. 0:02code setup I've ever seen. He rigged a
  3. 0:04home server in his basement, put his
  4. 0:06Obsidian vault in it, and then runs
  5. 0:08Claude code on top so he can think,
  6. 0:10research, write, and even ship code
  7. 0:12right from his phone. Today, he shows us
  8. 0:14how he uses Claude Code as a true second
  9. 0:16brain, a thinking partner that asks him
  10. 0:18sharp questions, pulls research from his
  11. 0:20whole note archive and the web, and even
  12. 0:22keeps a running log of what he's learned
  13. 0:24and what his best ideas are. And he
  14. 0:26walks us through his whole stack and his
  15. 0:28whole workflow. If you want to learn how
  16. 0:29to use claude code as a true second
  17. 0:31brain, this is the episode to watch.
  18. 0:33Let's dive in.
  19. 0:36[Music]
  20. 0:49This podcast is supported by Google. Hi
  21. 0:51folks, Paige Bailey here from the Google
  22. 0:53DeepMind Devil team. For our developers
  23. 0:55out there, we know there's a constant
  24. 0:57trade-off between model intelligence,
  25. 0:59speed, and cost. Gemini 2.5 Flash aims
  26. 1:02right at that challenge. It's got the
  27. 1:03speed you expect from Flash, but with
  28. 1:05upgraded reasoning power. And crucially,
  29. 1:07we've added controls like setting
  30. 1:09thinking budgets, so you can decide how
  31. 1:11much reasoning to apply, optimizing for
  32. 1:12latency and costs. So try out Gemini 2.5
  33. 1:15Flash at a studio.google.com, and let us
  34. 1:18know what you build.
  35. 1:19>> Noah, welcome to the show.
  36. 1:21>> Thanks for having me.
  37. 1:23I'm excited to have you. Uh it's really
  38. 1:26good to uh get to chat. This is our
  39. 1:28first interview in probably like 5
  40. 1:31years. You were one of the for people
  41. 1:32who don't know uh you were one of the
  42. 1:34first uh super organizers interviewees.
  43. 1:37That was the newsletter that turned into
  44. 1:39every and um I I love the way that your
  45. 1:43brain works. uh you have this like uh
  46. 1:45really interesting taste for tools for
  47. 1:48thought and back in the day you were
  48. 1:50using Evernote in all these really
  49. 1:52interesting ways. Um you were uh the
  50. 1:54co-founder of a really cool startup uh
  51. 1:56called Percolate uh and then another one
  52. 1:58called Variance and now you're running
  53. 2:00Alfic which is an AI strategy
  54. 2:02consultancy
  55. 2:04and um I'm just really excited to see
  56. 2:06how your um mind has uh started to use
  57. 2:12these AI tools now that now that they're
  58. 2:14uh now that they're you know working so
  59. 2:16well. And I know you have some pretty
  60. 2:18cool cla code stuff to show us. So yeah,
  61. 2:20thanks for coming on. Thanks for having
  62. 2:22me. Yeah, I'm super excited. That was a
  63. 2:24a fun interview all those years back.
  64. 2:26>> It was. It was really really fun. Um, so
  65. 2:30I want to just like dive right into the
  66. 2:32the like the thing that I think is so
  67. 2:34cool about what you're doing. So I know
  68. 2:36you have a whole like vibe coding setup
  69. 2:38that you built for yourself. Can you
  70. 2:40talk us through that?
  71. 2:42>> Yeah, I wouldn't uh I'm not sure about
  72. 2:44actually the vibe coding part of it. Um
  73. 2:47uh I have a a sort of fairly heavy duty
  74. 2:50clawed code set up. Um but actually uh
  75. 2:55mostly not for code. Um so uh since
  76. 2:59those days of uh super organizers like
  77. 3:02many people um uh I've abandoned
  78. 3:05Evernote um and uh switched over to
  79. 3:09Obsidian. And one of the big advantages
  80. 3:12with Obsidian as a note-taking platform
  81. 3:15is that it's a bunch of markdown files
  82. 3:16and a bunch of folders. Um, and uh, they
  83. 3:20can then be synced with Git and you can
  84. 3:25do lots of other fun kinds of things.
  85. 3:27And so actually probably my number one
  86. 3:30claude code use is using it as a tool to
  87. 3:35interact with my notes. Um, and so that,
  88. 3:40uh, I've got a fairly serious cloud code
  89. 3:42setup that I use with Obsidian. And, um,
  90. 3:45my most recent obsession,
  91. 3:48um, has been, uh, standing up a server
  92. 3:51in my house so that I could also use
  93. 3:55Claude Code on my phone.
  94. 3:57>> This is incredible. I want to I want to
  95. 3:59go through all of this. So, um, where
  96. 4:02should we start? Should we do should we
  97. 4:03do how you use cloud code as sort of
  98. 4:05this like research assistant notes
  99. 4:08organizer notetaker thing or should we
  100. 4:10start with uh how you use it on your
  101. 4:11phone?
  102. 4:13>> Um we can use it we can start uh with
  103. 4:15just the sort of general part of it that
  104. 4:17that might be the sort of easiest. The
  105. 4:19phone is really just an extension of
  106. 4:21that same thing. I um I would say sort
  107. 4:24of generally and this is something I
  108. 4:26feel like not enough people talk about
  109. 4:27with AI is like one of the things I find
  110. 4:29really extraordinary about it is the
  111. 4:32ability for me to um work really
  112. 4:37productively
  113. 4:38on my phone and that's been like a huge
  114. 4:41huge change because so much of what I do
  115. 4:43is sort of writing or coding and the
  116. 4:46phone is definitely not the best place
  117. 4:47for that and you know even the phone
  118. 4:49wasn't always the best place for doing
  119. 4:50research and thinking Um, I felt like my
  120. 4:53computer was a better place for it,
  121. 4:54which is why I've been such a sort of
  122. 4:55notetaker. And, um, you know, I have
  123. 4:58found whether it's like claude code and
  124. 5:01obsidian or, um, I mean, even claude
  125. 5:04code and code, right? So, like the other
  126. 5:06piece of it is being able to then, you
  127. 5:08know, if you see something go wrong,
  128. 5:09being able to sign in on your phone and
  129. 5:11have Claude code push a small update to
  130. 5:14something because you just realize it
  131. 5:16while you're out is amazing. But then
  132. 5:19even like uh um you know I use quite a
  133. 5:22bit of uh uh Grock voice mode and you
  134. 5:25know I find that that as a sort of like
  135. 5:28alternative way of working through
  136. 5:29problems. I have a Tesla so now it's
  137. 5:31baked into the Tesla. Um and uh you know
  138. 5:34it's just I and you know obviously all
  139. 5:37the sort of other Chad GBT and Claude
  140. 5:38and all these things of just being able
  141. 5:39to sort of like go and do research and
  142. 5:41really think and and explore things
  143. 5:45in this device that's always been useful
  144. 5:47but like not useful for deep work I
  145. 5:49think is probably um something most
  146. 5:52people would agree with is that the
  147. 5:53phone has not been the best place to
  148. 5:56kind of do deep coding and research work
  149. 5:58and I feel like it it it's really
  150. 6:00changed my ability to do that. Wait, I
  151. 6:02got to stop you. So, you're using Grock
  152. 6:05voice mode and and is that specifically
  153. 6:07because it's built into your Tesla or
  154. 6:09using it um in situations where you
  155. 6:11could also use, for example, CHBT voice
  156. 6:13mode?
  157. 6:14>> No, I'm using it because it's way better
  158. 6:15than any of the other voice modes. And I
  159. 6:17will fight anybody who says anything
  160. 6:18different.
  161. 6:19>> Okay. No, tell me like what is what do
  162. 6:21you like about it? Why is it better? I
  163. 6:23to be fair um uh OpenAI launched their
  164. 6:27real-time API which may or may not be
  165. 6:29baked into chat GBT voice. Now I it's
  166. 6:33not totally clear but the old voice mode
  167. 6:35was based on 40 and I just found it to
  168. 6:37be completely unusable. Um and uh uh
  169. 6:41Gemini's voice mode I just didn't find
  170. 6:43to be smart enough. Um and I just found
  171. 6:46Grock's voice mode to be significantly
  172. 6:48smarter than anybody else's. Um, you
  173. 6:50know, I'm using Grock
  174. 6:532, three, four, I don't even remember
  175. 6:55what whatever the latest
  176. 6:56>> the latest one. Yeah.
  177. 6:57>> Yeah. Not the super I don't have the
  178. 6:59most expensive account.
  179. 7:00>> Super heavy or whatever.
  180. 7:01>> I don't have super heavy. Um, but I just
  181. 7:03find it to be much better. It does tool
  182. 7:05calling way better than any of the other
  183. 7:06ones. That's that's I found to be a
  184. 7:08major weakness of the voice models is
  185. 7:09that they don't do great tool calling
  186. 7:11and research. And um, uh, Gro seems to
  187. 7:14have solved that. So, no, even before it
  188. 7:16was loaded in my Tesla, um I dropped my
  189. 7:18daughter off at summer camp this summer
  190. 7:20up in New Hampshire. So, I had a 5-h
  191. 7:22hour drive on my own and I spent like
  192. 7:24two hours researching and essentially
  193. 7:26like working through a piece. Um, and I
  194. 7:31did it by just like connecting it to
  195. 7:34Bluetooth and just sort of sitting there
  196. 7:35in the car. And, um, I found it to be by
  197. 7:38far the uh the best of the voice modes.
  198. 7:41I I hope these other models catch up
  199. 7:45there cuz I would I would love more
  200. 7:47really good voice modes. I mean, I had a
  201. 7:49mind-blowing session this weekend and uh
  202. 7:51I'm giving a talk and I'm sort of have
  203. 7:53some ideas. I think it's generally going
  204. 7:55to be about transformers eating the
  205. 7:57world and um so I was sort of catching
  206. 7:59myself up on self attention and exactly
  207. 8:01how it works. and um I did like an hour
  208. 8:04session and it really I I it like was by
  209. 8:07far the sort of best
  210. 8:10explanation I've ever read for it and or
  211. 8:13ever heard I guess and so yeah I' I've
  212. 8:15just found it to be a a kind of pretty
  213. 8:17extraordinary product.
  214. 8:18>> I do love voice mode for that. Um it's
  215. 8:21sort of like it's the it's a podcast
  216. 8:24made specifically for you about whatever
  217. 8:25you're curious about and that's really
  218. 8:27cool. I went up to I I drove up state
  219. 8:30this weekend and I've been reading um
  220. 8:33I've been reading the Iliad and so I had
  221. 8:36it on audiobook and then I had some
  222. 8:37questions as I was driving and so I
  223. 8:39unfortunately use CHBT voice mode
  224. 8:41because I didn't know about Grock. So I
  225. 8:43wish that uh Grock's voice mode so I
  226. 8:45wish that uh we had had this
  227. 8:47conversation before then.
  228. 8:48The thing about TGBT voice mode is yeah,
  229. 8:51I think when it first came out it was
  230. 8:52cool, but it just hasn't gotten as smart
  231. 8:55as the models are are. And they they
  232. 8:58gave it this new personality that I had
  233. 9:00to get used to where every time you ask
  234. 9:01it a question it goes like, "Oh, yeah."
  235. 9:03Uh-huh. Well, you know, and it's like
  236. 9:05it's just this like weird Gen Z thing
  237. 9:06that it feels like it's has a little bit
  238. 9:10too much onwe or something like doesn't
  239. 9:12actually care about you. I don't know
  240. 9:13what that is. So, I had to get used to
  241. 9:15that. Rock has a stoner mode for when
  242. 9:17it's worth. Yeah.
  243. 9:20I will say uh the car version is very
  244. 9:23interesting to me. Like this was in the
  245. 9:25most recent Tesla release like a couple
  246. 9:27weeks ago. And um you know I had been
  247. 9:30doing that same thing you did where you
  248. 9:33just plug your phone in and you put on
  249. 9:34Bluetooth and you you know do your best
  250. 9:36to make it work. And it's very
  251. 9:38interesting to just like have a voice AI
  252. 9:39button and it syncs back to your regular
  253. 9:43Grock, but it you can't get you can't
  254. 9:46rejoin old chats. So, it's just like,
  255. 9:49hey, I it's just like but you know, I
  256. 9:51mean, these things are significantly
  257. 9:52better than Siri and all of these other
  258. 9:55things and particularly I mean, you
  259. 9:56know, there's no comparison if you
  260. 9:58actually have something more than just a
  261. 9:59a single question you want an answer to,
  262. 10:02right? like if you actually want to have
  263. 10:03a conversation about the Iliad or you
  264. 10:05know about Transformers and self
  265. 10:07attention like um I don't know it's
  266. 10:10pretty amazing to just be able to sort
  267. 10:11of like hit this button and and yeah use
  268. 10:14that car time. I I mean I I was on my
  269. 10:16way somewhere last week and I was um you
  270. 10:20know I was like having it research uh I
  271. 10:22was going back to the um Walter
  272. 10:24Benjamin. I was I was I I have this sort
  273. 10:26of idea to write a piece about um how uh
  274. 10:30um the reactions to every new technology
  275. 10:33are essentially elitist um critiques of
  276. 10:36it and that you know it's always like oh
  277. 10:38no everybody's going to be able to like
  278. 10:39do this thing that only we used to be
  279. 10:41able to do and so um I was in the car
  280. 10:43and I was thinking about this and so I
  281. 10:45had it go and I was like okay you know I
  282. 10:47know I it's been years since I read the
  283. 10:49Walter Benjamin uh
  284. 10:50>> mass production of images one yeah
  285. 10:52>> uh yeah that that one And um so yeah,
  286. 10:55then I'm having conversations about that
  287. 10:56and I'm like who are Walter Benjamin's
  288. 10:59contemporaries? And then I'm like into
  289. 11:00all these, you know, it's just like I
  290. 11:02don't know. That's a it that's amazing.
  291. 11:05>> It's the best.
  292. 11:06>> It's the best. Yeah. I don't know.
  293. 11:09>> Um Okay. And so you're filling your
  294. 11:12brain with all these things from voice
  295. 11:13mode, which I love. Uh, but tell us
  296. 11:15about your your second brain setup or I
  297. 11:17don't know how you I don't know how you
  298. 11:18refer to it, whether whe you think that
  299. 11:19second brain is appropriate for this,
  300. 11:21but I want to know how you're using
  301. 11:22cloud code to uh take notes and do
  302. 11:25research and all that kind of stuff.
  303. 11:26>> Yeah. So, um, uh, you know, I could just
  304. 11:29open it up. Maybe that's the easiest
  305. 11:30thing to walk you through it. I'll I'll
  306. 11:32start on my computer and then we can do
  307. 11:34the phone. The computer's just a way
  308. 11:35easier a way easier share here. So, um,
  309. 11:38all right. So, um,
  310. 11:42uh, this is what I was working on
  311. 11:44before. Um, uh, but essentially, you
  312. 11:47know, this is just claude code and it's
  313. 11:49just sitting on top of my Obsidian. So,
  314. 11:52you know, if I jump out here and I just
  315. 11:54do like, you know, you can see I'm I'm
  316. 11:57following the the Parah method. Um, and
  317. 12:01uh uh you know, I've just got everything
  318. 12:03sort of organized in here and and put in
  319. 12:05the places that they need to be. Well,
  320. 12:07let me just uh let me just stop you for
  321. 12:08people who are listening. So, okay. So,
  322. 12:10we're looking at we're looking at cloud
  323. 12:12code. It sounds seems like you have
  324. 12:14cloud code running in your Obsidian
  325. 12:16vault and there's some kind of uh it's
  326. 12:20adding something to an existing it looks
  327. 12:22like it's adding something to an
  328. 12:23existing note. Is that that's what
  329. 12:25that's what's going on? That's what
  330. 12:26we're looking at.
  331. 12:26>> Yeah. So, in this particular one I'm I'm
  332. 12:28working on this talk. So, I I'm putting
  333. 12:30on my conference in two weeks. I'm
  334. 12:32giving this talk about marketing and AI
  335. 12:34and sort of what's going on. And um I'm
  336. 12:36uh if we sort of jump back a second um
  337. 12:40I've been doing these conferences uh
  338. 12:42called Brand BRXnd.AI and they're about
  339. 12:45marketing and AI and um I did one in
  340. 12:47February in um LA and my talk in LA was
  341. 12:52about this uh I'm sure you've seen it.
  342. 12:54It was the Office of Strategic Services
  343. 12:56which was the precursor to the CIA wrote
  344. 12:58this manual called the Simple Sabotage
  345. 13:00Field Manual. Um, and it was essentially
  346. 13:02a manual to help citizen saboturs in
  347. 13:04Nazi occupied territories um, sort of
  348. 13:07like quietly sabotage um, the the Nazi
  349. 13:11occupation. And um so it was like you
  350. 13:14know there's a whole bunch of stuff for
  351. 13:15bluecollar workers that's like if you're
  352. 13:17a janitor you should leave a bucket of
  353. 13:19oily waste around and accidentally drop
  354. 13:21a cigarette in there so that you know it
  355. 13:23it will um but then there's this amazing
  356. 13:27um set of recommendations for white
  357. 13:29collar workers and they're like um
  358. 13:32always refer things to committee um uh
  359. 13:35always revisit previously made decisions
  360. 13:38um make sure that like if somebody is
  361. 13:40trying to make a decision you should
  362. 13:42suggest that they don't act with too
  363. 13:44much haste um uh less like we be
  364. 13:47embarrassed and and so it's like um you
  365. 13:50know uh my talk was about kind of how
  366. 13:54one hope I have is that AI um can kind
  367. 13:56of like sidestep a lot of the
  368. 13:59bureaucracy that exists inside large
  369. 14:00organizations because it sort of has
  370. 14:03this um kind of uh goo like effect where
  371. 14:07it can kind of fit into any crevice or
  372. 14:09crack because it can act as this fuzzy
  373. 14:11interface and it doesn't really care
  374. 14:12about the sort of input output. Um and
  375. 14:15so
  376. 14:16the sort of next part of the story is I
  377. 14:18after the conference I realized that um
  378. 14:21uh that manual was in the public domain.
  379. 14:23So I hired a designer and I printed 300
  380. 14:26copies and I wrote a new forward for it.
  381. 14:29>> And so we're giving this away at the
  382. 14:30conference and so my talk is sort of
  383. 14:32trying to tie all these ideas together.
  384. 14:33Right? So, I'm I'm trying to pull from
  385. 14:35the sabotage manual and then I was doing
  386. 14:37a bunch of research into wild Bill
  387. 14:39Donovan who started the OSS. And the OSS
  388. 14:42was sort of the precursor to the both
  389. 14:44the CIA and the um uh uh special forces.
  390. 14:48And so anyway, I'm I'm writing this talk
  391. 14:50and so I've got a project inside my
  392. 14:52Obsidian, which is the beginning of the
  393. 14:55research for this project. And I'm
  394. 14:57pulling in sort of like chats and
  395. 14:58articles and all these things. And then
  396. 15:00I'm constantly kind of talking to the
  397. 15:03the AI in here and giving it new ideas.
  398. 15:06So I'm like, "Oh, I need some
  399. 15:07conclusions. Here's sort of my first
  400. 15:09thought on conclusions." And I'm having
  401. 15:11it like note down the conclusions. And
  402. 15:13then at the end of each day, I have the
  403. 15:15AI write up the changes that I sort of
  404. 15:17like the things I learned that day that
  405. 15:19are going to help me push this talk
  406. 15:20along. And um so that's what you're
  407. 15:22looking at right here is sort of this is
  408. 15:24all part of this um work that I've been
  409. 15:27doing where I've been feeding it. Um I
  410. 15:31was working on sort of what are some of
  411. 15:32the conclusions I want it to be. And so
  412. 15:34this is all sitting in this uh this is
  413. 15:37all sitting in in my Obsidian a project
  414. 15:40specifically for that talk.
  415. 15:42>> Okay. So let me get a a clearer sense of
  416. 15:44this. This is really interesting. So you
  417. 15:47have a project uh when you when you have
  418. 15:49a new thing you're you're giving a talk
  419. 15:51you make a new folder and then uh as
  420. 15:54you're thinking about stuff you're
  421. 15:56working with cloud code inside of the
  422. 15:59folder
  423. 16:00and uh you're researching stuff and then
  424. 16:03saying like I want you to take notes on
  425. 16:04it. Um, in this particular case, you're,
  426. 16:07you know, that a component of your talk
  427. 16:09is the conclusions section. And so
  428. 16:11there's one particular markdown file
  429. 16:13that like you're just going back and
  430. 16:15forth with it and having it add
  431. 16:16conclusions. But like what else is in
  432. 16:18that folder? So is it like there's a
  433. 16:20body there's a body note and then
  434. 16:22there's an intro note or is it like
  435. 16:24>> So one of the big things here is that
  436. 16:26I'm in thinking mode, not writing mode
  437. 16:28yet. Um, and so, uh, there's some stuff
  438. 16:31in here where I've specifically told, I
  439. 16:34think it's in the front matter actually,
  440. 16:35where I've told claude code like, don't
  441. 16:38help me write anything right now. And I
  442. 16:39I generally find this to be a big thing
  443. 16:41with all these models is like they
  444. 16:43immediately jump to wanting to help you
  445. 16:45with the artifact. Um, and you know,
  446. 16:49when you're just in thinking mode, you
  447. 16:50have to be very explicit in like, hey, I
  448. 16:52just want you to help me think and ask
  449. 16:55me questions. And so, um, yeah, what you
  450. 16:57can see here is like there's there's a
  451. 16:58bunch of files in here. I've got chats.
  452. 17:00That's where I'm literally like taking
  453. 17:02chats I'm having in other things. And,
  454. 17:04um, I'm just like using the Obsidian web
  455. 17:07clipper to pull the whole chat in. I've
  456. 17:09got daily progress. That's where I'm
  457. 17:10having the AI actually like look through
  458. 17:12all the notes that came out that day and
  459. 17:14like help me think through the progress.
  460. 17:17Um, and then I've got research. That's
  461. 17:18where I've got a bunch of like articles
  462. 17:20and PDFs and stuff that I've pulled in
  463. 17:22so far and been reading about. And then
  464. 17:24there's a bunch of other kind of random
  465. 17:25notes along here where I've been, you
  466. 17:27know, just using it to kind of help me
  467. 17:30um think and and so yeah, I was in the
  468. 17:32midst of um one I've got this conclusion
  469. 17:35note. So, you know, I I sort of felt
  470. 17:37like I had blocked out the big kind of
  471. 17:39um the big themes of the talk, but I was
  472. 17:42like, okay, I need to figure out what am
  473. 17:43I going to say at the end? And you know,
  474. 17:44essentially what I'm going to say at the
  475. 17:45end is about a lot of the stuff I've
  476. 17:47learned over the last few years of
  477. 17:48working with these large brands on AI
  478. 17:51projects. And um so I was starting to
  479. 17:53get it to the conclusions. And so yeah,
  480. 17:55I'm just kind of like trying I'm really
  481. 17:57piecing all this stuff together right
  482. 17:59now. That's kind of what's happening.
  483. 18:00>> And give me a sense of like when this
  484. 18:02folder was empty, what did you start
  485. 18:04with?
  486. 18:06>> Um so I think I started with um the the
  487. 18:11I started with uh telling it like I'm in
  488. 18:14thinking mode. I'm not in writing mode.
  489. 18:16Um, here are my past few talks that I've
  490. 18:19given at brand to give you a sense of
  491. 18:20the sort of style that I have and I
  492. 18:23here's the kind of general idea and the
  493. 18:25big points I want to make, right? Like
  494. 18:26I'm giving away this book. So I want to
  495. 18:27talk about simple sabotage field manual
  496. 18:29and I have this notion like I have this
  497. 18:31it's kind of just a title. It's like
  498. 18:32transformers are eating the world. It's
  499. 18:33this idea that like one of the very
  500. 18:36interesting things happening with these
  501. 18:37models is they're sort of displacing a
  502. 18:40whole bunch of specialized code in
  503. 18:42places. Um, and so I sort of want to
  504. 18:44talk about that. that and then I've got
  505. 18:45these conclusions. And so the first
  506. 18:47thing I said was like, hey, just go look
  507. 18:48through all of the rest of my, you know,
  508. 18:50probably 1,500 things in my Obsidian and
  509. 18:54go see anything else you can find that
  510. 18:56um might be of value to this talk um of
  511. 19:00the existing things I have. And so just
  512. 19:03go kind of pull those into the research
  513. 19:05folder at the beginning to kind of like
  514. 19:07jumpstart this process.
  515. 19:09>> Got it. And you're starting are you
  516. 19:11starting Claude in this folder or are
  517. 19:13you starting it in your full Obsidian
  518. 19:15vault so that it can access all that
  519. 19:16stuff?
  520. 19:18>> Um, no. I'm So I'm starting it in the
  521. 19:21full Obsidian vault. So like if we um
  522. 19:24like this is coming if I step out of
  523. 19:26here, right? Um we're in the root
  524. 19:29directory. All this stuff is in the root
  525. 19:30directory. My Obsidian
  526. 19:31>> I get Um, and my Obsidian setup is also
  527. 19:34like a little more intense for what it's
  528. 19:35worth because like um I've also realized
  529. 19:38like you can add a package.json to add a
  530. 19:42bunch of like custom code commands to
  531. 19:45your Obsidian that you can then run and
  532. 19:46then you could use those code commands
  533. 19:48in slash commands and all of these other
  534. 19:50things. So, you know, there are a bunch
  535. 19:52of other kind of moving pieces in here,
  536. 19:54but generally it's a fairly
  537. 19:55straightforward um I mean it's it's a
  538. 19:58I'm trying to use parah and some other
  539. 20:00kind of bits and pieces.
  540. 20:02>> So, for people who are who are listening
  541. 20:04or watching and are like we just went
  542. 20:05through a bunch of stuff really fast.
  543. 20:07So, the basic gist is Obsidian is just
  544. 20:09like a a notetaker note-taking app uh
  545. 20:11that runs it's all local and so
  546. 20:13everything that all the notes you take
  547. 20:15like they they exist in essentially text
  548. 20:17files on your computer organized by
  549. 20:18folder. Um, and when you're starting
  550. 20:21Cloud Code, uh, one way to do it would
  551. 20:24be to start Cloud Code in the folder for
  552. 20:25the particular project that you have.
  553. 20:27But it sounds like what you're doing is
  554. 20:29instead you're starting in the root
  555. 20:30directory where all of your Obsidian,
  556. 20:32uh, notes live. And the advantage of
  557. 20:34that is
  558. 20:36um, Cloud Code has some like sandboxing
  559. 20:39things where it's like it's not really
  560. 20:40supposed to like run commands outside of
  561. 20:43the folder it was started in. It can run
  562. 20:44commands inside of any subfolder, but it
  563. 20:47sounds like what you're doing. So, it
  564. 20:48has access to your entire Obsidian. It
  565. 20:50can do a bunch of stuff. And you've also
  566. 20:51added a package.json which lets it run,
  567. 20:54you know, custom software custom
  568. 20:56software commands basically. Um, that's
  569. 20:59really really interesting. Okay. And do
  570. 21:02you do you find cuz I've sort of like
  571. 21:05had this as a twinkle in my eye to like
  572. 21:06have it go find relevant stuff for me.
  573. 21:09Do you find that it's actually relevant
  574. 21:10and interesting? Cuz I think sometimes
  575. 21:12when I've done this kind of thing before
  576. 21:13with language models, they're like, "Oh
  577. 21:16yeah, like this random thing is relevant
  578. 21:18because XYZ." Like it doesn't feel like
  579. 21:21I can understand why it picked it as
  580. 21:23being relevant, but if it really knew
  581. 21:26who I am and like what I think is
  582. 21:28interesting, it definitely would not
  583. 21:29have. Um, do you do you find that that's
  584. 21:32the case or have you figured out a way
  585. 21:33to make it relevant? Imagine going from
  586. 21:35idea to live website in minutes without
  587. 21:38a single line of code. Framework lets
  588. 21:39you do just that. Their no code platform
  589. 21:41looks, feels, and works like a design
  590. 21:44tool, but it outputs real functional
  591. 21:46websites that look like they were done
  592. 21:47by a developer. Framer is the design
  593. 21:49first no code website builder that lets
  594. 21:51anyone shift a productionready website
  595. 21:53in minutes. At every we have only 15
  596. 21:55full-time employees, but we have five or
  597. 21:57six businesses that we run internally.
  598. 21:58So, we need to build a lot of websites
  599. 22:00all the time. That's what we use framer
  600. 22:02for. Framer AI builds entire websites
  601. 22:04from a single prompt. Type modern SAS
  602. 22:06landing page for a productivity app and
  603. 22:08watch it generate a structured site with
  604. 22:11a responsive layout, clean sections, and
  605. 22:13pre-filled copy. After that, refine,
  606. 22:15style, and publish. It's also really
  607. 22:17powerful. You can create buttery smooth
  608. 22:19effects and micro interactions with just
  609. 22:21simple sliders. Do you need to reach a
  610. 22:22global audience? One click lets you
  611. 22:25translate your website, every page, and
  612. 22:27every slug into any language. And when
  613. 22:29you're ready to launch, you just hit
  614. 22:30publish once. Behind the scenes, Framer
  615. 22:32handles hosting, blazing fast load
  616. 22:34times, and SEO optimization. This lets
  617. 22:36you own your entire journey from concept
  618. 22:38to live site. No developer required.
  619. 22:40Ready to make a site that looks
  620. 22:42handcoded without hiring a developer?
  621. 22:44Launch your site for free at framer.com
  622. 22:46and use the code dan to get your first
  623. 22:48month of pro on the house. Rules and
  624. 22:50restrictions may apply. And now back to
  625. 22:52the episode.
  626. 22:53>> I think by and large, yes, I agree with
  627. 22:55you. I think in this case relevance is a
  628. 22:57little simpler since like ultimately
  629. 23:00this talk is sort of the the things I
  630. 23:03were as was asking it to look for I've
  631. 23:05done a bunch of thinking and research
  632. 23:07around. So it's like I'm not asking it
  633. 23:08to make large conceptual leaps to
  634. 23:12relevance. It's like go find all this
  635. 23:14like I'm want to talk about the simple
  636. 23:15sabotage field manual. It can literally
  637. 23:17just do a like find for all the times
  638. 23:20all the articles and things I've got in
  639. 23:22my Obsidian about that. And so it
  640. 23:26relevance is uh yeah it's kind of a
  641. 23:28loaded term right um and I I agree with
  642. 23:32I I I agree with what you're saying. I
  643. 23:34think this is what I'm asking it to do
  644. 23:35is much more simple which is like
  645. 23:37amongst this set of things go find all
  646. 23:40the notes that I've already researched
  647. 23:42that kind of brought me to be thinking
  648. 23:43about these things to begin with.
  649. 23:45>> Got it. And then once you had it do all
  650. 23:47that research, did you have it do any
  651. 23:49sort of summary to like let like sort of
  652. 23:51stimulate you to be like, "Okay, here
  653. 23:52are the here are some jumping off points
  654. 23:55based on what you've done before." What
  655. 23:56was your next step once you once
  656. 23:57>> No. So my next step is I actually have
  657. 24:00an agent in here. So if we go to I'll do
  658. 24:03continue for now. Um
  659. 24:06>> and so for people who are listening, so
  660. 24:07you're just starting up Claude. You're
  661. 24:09using the continue flag. So uh you're
  662. 24:12you're starting claude by continuing the
  663. 24:14last session that you were in. And now
  664. 24:16you've got and you cloud code gives us
  665. 24:18ability to do sub agents. So those are
  666. 24:20like little mini clouds that you can
  667. 24:21spawn. And you have a
  668. 24:22>> you have a thinking partner one
  669. 24:24>> thinking partner sub agent. Okay. How
  670. 24:25does that work?
  671. 24:26>> Yeah. And so this is the whole thing
  672. 24:28where I'm like hey you're a
  673. 24:30collaborative thinking partner
  674. 24:32specializing in helping people explore
  675. 24:33complex problems. Um your role is to
  676. 24:36facilitate thinking. Um, and basically
  677. 24:39don't try to write the thing. Um, and so
  678. 24:42after I had that initial set of things,
  679. 24:44I flipped to this and it's like, okay,
  680. 24:47let's get into a flow. Ask me the kinds
  681. 24:49of questions, help me think through it.
  682. 24:51Um, you know, this is also where I've
  683. 24:53got a chats folder in here. So, it's
  684. 24:54like it's not just happening here. I was
  685. 24:56also having like a I've got a whole um
  686. 25:00uh sorry, I'm just backing out. So like
  687. 25:03if we go into um
  688. 25:11if we go into chats like these are a
  689. 25:14whole bunch of the chats that I had
  690. 25:16>> um
  691. 25:17>> like with the interviewer. No, these are
  692. 25:19chats I was having with chatb and Claude
  693. 25:21and Grock and all of these different
  694. 25:22things that I went and just grabbed the
  695. 25:24full transcript of. Um, so you know, I
  696. 25:27was also having all these other
  697. 25:28conversations and then, you know, I'm
  698. 25:30specifically telling the interviewer
  699. 25:31like review all these other things. So
  700. 25:33actually you know I think the first one
  701. 25:35there's one of these conversations is I
  702. 25:38originally had this idea about
  703. 25:39transformers are eating the world and
  704. 25:41you know the sort of notion there is
  705. 25:43like um you know there was some research
  706. 25:45that came out I think a few months ago
  707. 25:46that um uh they had found a um they were
  708. 25:49able to sort of outperform some
  709. 25:51specialized uh time series modeling um
  710. 25:55uh models with transformers and like you
  711. 25:57know I think there's really interesting
  712. 25:59stuff you know there's a story about
  713. 26:00Tesla removing 300,000 lines code with a
  714. 26:04neural network and you know I've just
  715. 26:05got kind of got these bits and pieces
  716. 26:07and one of the ways I work generally is
  717. 26:09like when I have an idea of something to
  718. 26:11write or think about I'll start a a
  719. 26:14thread in chat GBT or claude and I'll
  720. 26:18then save that somewhere and then I'll
  721. 26:20just kind of keep coming back to it when
  722. 26:21I have more ideas. It's like, oh, here's
  723. 26:23another example of transformers doing
  724. 26:26something. And so, one of these
  725. 26:27conversations is actually that thing
  726. 26:29from, you know, probably four or five
  727. 26:31months ago when that kind of idea
  728. 26:33initially came into my head. Maybe when
  729. 26:34I saw the research about time series
  730. 26:36modeling or something
  731. 26:38>> really interesting. Okay. Um, okay. So,
  732. 26:41let's let's keep going. So, uh, you've
  733. 26:45got you got the sub agent and I actually
  734. 26:46I want to like just actually I want to
  735. 26:48pause on that really quick, which is I
  736. 26:50think this is a very common complaint
  737. 26:52that they just dive in and it's a common
  738. 26:54pattern to make a thinking agent and I
  739. 26:57think claude code or claude in general
  740. 26:59is probably the best one for this. So
  741. 27:02this is a thing that we faced with uh
  742. 27:04one of the apps that we've incubated
  743. 27:05called Spiral which is a a aentic ghost
  744. 27:07writer and
  745. 27:10I think we found I found the same kind
  746. 27:12of thing when when I was thinking about
  747. 27:13okay how does a good ghostriter work
  748. 27:16they don't just like you don't say hey I
  749. 27:17want you to write a blog post and
  750. 27:18they're just like cool I made it here it
  751. 27:20is like a good ghostwriter is going to
  752. 27:21get to know you and really you're gonna
  753. 27:23you're going to work together to figure
  754. 27:27out what's in your head about it but
  755. 27:28also shape what's in your head like it's
  756. 27:30not just, oh, I I can see it and like
  757. 27:33I'm they need to get it out of you. Like
  758. 27:35you're you're actually making it
  759. 27:36together. And in order to do that, you
  760. 27:38have to have a really good um inter
  761. 27:41basically interview process to uncover
  762. 27:44things. And that sounds like you've
  763. 27:46found that too. And I think that's
  764. 27:47really really interesting and really
  765. 27:48important for people who are thinking
  766. 27:50about how do I get the best out of AI?
  767. 27:52Actually stop for a second and like let
  768. 27:55it ask it to understand you first.
  769. 27:57>> Yeah. One of the things I say to a lot
  770. 27:59of people is just like I think partially
  771. 28:01because we call it generative, there's
  772. 28:03entirely too much focus on its ability
  773. 28:05to write and not enough focus on its
  774. 28:07ability to read. Um, you know, and it's
  775. 28:09like its ability to read is incredible,
  776. 28:12right? And and I think, you know,
  777. 28:14arguably sort of like much more useful
  778. 28:16on a day-to-day basis. Like we produce
  779. 28:18artifacts far less frequently than we
  780. 28:21just like think about things. And um,
  781. 28:24>> uh, so yeah, I I I do this a lot. This
  782. 28:26is definitely a complaint I have about
  783. 28:27all the models is like, you know, even
  784. 28:29when you very specifically tell it not
  785. 28:31to try to do your work, it still often
  786. 28:34still tries to do your work. And so you
  787. 28:36have to like really really be like, "No,
  788. 28:38I said no." Like I think actually if we
  789. 28:41look at um
  790. 28:43uh so
  791. 28:45here critical when Noah says he's just
  792. 28:48collecting source materials or I do not
  793. 28:50under any circumstances want you to try
  794. 28:52to write it, take this literally. Do not
  795. 28:54create outlines, drafts, or any versions
  796. 28:56of talkswriting. Only gather and
  797. 28:58organize the requested materials.
  798. 29:01>> It's so good. I love it.
  799. 29:03>> Um, yeah, this is like uh but yeah, I
  800. 29:05think, you know, I think we all we all
  801. 29:07experience that and you know, I mean, I
  802. 29:09I do hope over time that that sort of
  803. 29:12gets baked into the models. I think it's
  804. 29:13a very interesting tension that exists
  805. 29:15with the model companies because like
  806. 29:16obviously you know like the sort of a
  807. 29:20lot of the economic input output is sort
  808. 29:22of measured in the artifacts that it
  809. 29:24produces. Um and so I think it's very
  810. 29:26oriented and you know I I suspect that
  811. 29:28part of it is just like that sort of the
  812. 29:30helpful assistant thing has like come to
  813. 29:33be a sort of meme that is probably
  814. 29:35self-ingested.
  815. 29:37Um, but yeah, it's it's um I think for
  816. 29:41those of us who are trying to do more
  817. 29:44interesting things with these models, it
  818. 29:46it becomes a real barrier to work.
  819. 29:48>> Totally. Um, okay. So, I want to think
  820. 29:52about when you're using the thinking
  821. 29:53agent, did you say it like it's is it
  822. 29:56outputting some sort of summary of what
  823. 29:57you've come to into a particular place
  824. 29:59or?
  825. 29:59>> Yeah. So that thinking agent is sort of
  826. 30:01told to as it ask me questions kind of
  827. 30:03make notes about the questions that it's
  828. 30:06asking me and keep a kind of running log
  829. 30:08of what I'm uncovering and how I'm
  830. 30:10thinking about it and all those sorts of
  831. 30:12things.
  832. 30:12>> Got it. And then you know you come back
  833. 30:14the next day and you're like oh I just
  834. 30:16want to go down this rabbit hole on XYZ
  835. 30:18thing about the you know this wild bill
  836. 30:21guy. Um and that you start in a new chat
  837. 30:24maybe with it maybe with the subision
  838. 30:26maybe maybe not and that becomes its own
  839. 30:28new file on that topic.
  840. 30:30>> Yeah. Exactly. So like I I this the um
  841. 30:34the Wild Bill stuff started as like deep
  842. 30:36research in Chad GBT um and I you know
  843. 30:39had it go out and I'm reading the Wild
  844. 30:41Bill book right now. there's a like one
  845. 30:44sort of particularly famous biography of
  846. 30:46him and uh um you know I'm kind of
  847. 30:49thinking about the bits and pieces and
  848. 30:51trying to make and I I think I made a
  849. 30:52kind of interesting connection in there
  850. 30:55um uh where you know a big part of what
  851. 30:59sort of he seems to have been after with
  852. 31:01the OSS and and you know the sort of
  853. 31:03inspiration for the special forces was
  854. 31:05like empowering individuals um you know
  855. 31:08that's sort of like the theme of that
  856. 31:10manual was like obviously empower
  857. 31:12empowering citizen saboturs, but also,
  858. 31:14you know, I think a big part of the
  859. 31:15special forces is like, you know, having
  860. 31:18kind of like incredible operators at the
  861. 31:20edge who, you know, obviously operate
  862. 31:22within a sort of command and control
  863. 31:24hierarchy, but like have a ton of
  864. 31:25autonomy to move and execute
  865. 31:29independently because they're kind of
  866. 31:30they have all the things that they need.
  867. 31:32And um so you know in all of that wild
  868. 31:35bill research I kind of went back to
  869. 31:36this and I was like is is this like an
  870. 31:38interesting way to connect all these
  871. 31:40ideas that like maybe kind of there's a
  872. 31:42you know and again I'm this is still
  873. 31:44early I have not like solidified these
  874. 31:46conclusions this is like you know the
  875. 31:48regular kind of writing process right um
  876. 31:50but it's like well I think there's this
  877. 31:52idea that like you know fundamentally
  878. 31:54well I know you know fundamentally one
  879. 31:56of the big things with transformers is
  880. 31:57it moved us from sequentialbased models
  881. 32:00to models that can act in, you know,
  882. 32:03sort of paralyze their work better. And
  883. 32:06that obviously allowed us to have much
  884. 32:07more powerful and, you know, interesting
  885. 32:09models and has, you know, arguably
  886. 32:11kicked off this entire sort of
  887. 32:13revolution of what's going on. Um, and
  888. 32:15you know, what we both do for a living.
  889. 32:17Uh, and so, you know, I think there's
  890. 32:19this kind of interesting connection. And
  891. 32:20so that was what I was playing with. I
  892. 32:22think here was like oh maybe there's
  893. 32:24this connection between kind of
  894. 32:26sequential processing to this kind of
  895. 32:29parallel and then there's this
  896. 32:30connection to bureaucracy and then
  897. 32:32there's this connection to wild bill who
  898. 32:34seems to be have been very much about
  899. 32:35sort of like working within a system but
  900. 32:37like having autonomy at the edges and so
  901. 32:41that's kind of what I was playing with
  902. 32:42and just kind of taking notes and then
  903. 32:44yeah I would jump out and be like oh
  904. 32:46well actually I haven't figured out a
  905. 32:47conclusion yet let me start the
  906. 32:48conclusion section and I'll just sort of
  907. 32:50get that going on the side but Um, and
  908. 32:52then, you know, I I have a job, so I
  909. 32:55can't be doing this all the time. So,
  910. 32:56it's also like you interrupt yourself,
  911. 32:58and it's really nice to be able to come
  912. 32:59back and be like, you know, um, can you
  913. 33:02catch me up on the last three days of
  914. 33:06research?
  915. 33:07>> Oo, I love that question. That's so
  916. 33:09cool.
  917. 33:12>> And so, yeah, you can just kind of go in
  918. 33:13and be like, can you catch me up on the
  919. 33:14last few days of research? And, um, it's
  920. 33:17just going to go read all the stuff,
  921. 33:19right? Again, it's like I think the
  922. 33:21point you made earlier about the, you
  923. 33:23know, go find relevant sources. It's
  924. 33:24like I find a lot that the difference
  925. 33:27between the people are getting a lot of
  926. 33:30this right now is part of it is just
  927. 33:31like you have a good feel for where the
  928. 33:33edges of the capabilities of these
  929. 33:35models are and you sort of like
  930. 33:37encourage them to work within those
  931. 33:39capabilities. Like this is an incredibly
  932. 33:41easy, you know, it's like we know what
  933. 33:43it's going to do here, right? Like I
  934. 33:44could write all these Unix commands.
  935. 33:45It's just going to go find a bunch of
  936. 33:47files in this directory and it's going
  937. 33:49to look at them by date and it's going
  938. 33:51to look at all the files created in that
  939. 33:52project over the last you know and we
  940. 33:54know it's going to be able to do that
  941. 33:56and so it's saying um you know the major
  942. 33:59breakthrough day was this idea of
  943. 34:00bureaucracy as positional encoding which
  944. 34:03is very much work in progress idea but I
  945. 34:05kind of like it um but you know so it's
  946. 34:07just like it's pretty amazing also to
  947. 34:09just be able to kind of revisit deep
  948. 34:11work like this right where you know you
  949. 34:13know you're going to break your flow um
  950. 34:16you're not and it's like it's often you
  951. 34:18know I find whether it's code or writing
  952. 34:20the hardest part is like just picking it
  953. 34:22up again um because you're you're out of
  954. 34:24it and so just to kind of like kickstart
  955. 34:26that process is sort of amazing um I
  956. 34:29think what I was playing with here is
  957. 34:30this idea that um bureaucracy was
  958. 34:34actually like an innovation right um
  959. 34:36that like we look at bureaucracy as a
  960. 34:37negative and generally we talk about it
  961. 34:39as a negative and I think often it is a
  962. 34:41negative um but you know um bureaucray
  963. 34:45cy was a sort of like huge innovation
  964. 34:47for how companies operate, right? And
  965. 34:49ultimately it sort of represents kind of
  966. 34:51hierarchy and structure and a whole
  967. 34:54bunch of things that are like actually
  968. 34:56like pretty positive for operating at a
  969. 34:59large scale. Um and so um you know again
  970. 35:03my my kind of whole thesis on AI around
  971. 35:05all this bureaucracy stuff is that
  972. 35:07what's interesting about it is that um
  973. 35:09as opposed to kind of past technologies
  974. 35:11which kind of forced you to make a
  975. 35:13decision about whether you wanted to
  976. 35:14kind of like use your existing
  977. 35:18sort of structure and build that
  978. 35:21technology into your existing structure
  979. 35:22or adopt the new structure. most of the
  980. 35:24time the new software required you
  981. 35:26adopted the new structure and that's why
  982. 35:28so many sort of software projects failed
  983. 35:30for so long at least that's sort of part
  984. 35:31of what I think and um you know I think
  985. 35:34part of what's interesting about AI what
  986. 35:35I find so interesting is like that
  987. 35:39you can kind of keep letting everybody
  988. 35:41work in whatever way they want you know
  989. 35:42it's like a classic problem inside large
  990. 35:44companies is like one team wants to use
  991. 35:46a sauna and one wants to use Jur and one
  992. 35:47wants to use linear right and so then at
  993. 35:50some point there's like a there's a
  994. 35:51there's a huge project and they bring in
  995. 35:54some big consulting firm and they decide
  996. 35:55they're going to all centralize on this
  997. 35:57one thing and now twothirds of the
  998. 35:59company is unhappy um and like they've
  999. 36:02all made sacrifices and you know you're
  1000. 36:04sort of in this like um very non ideal
  1001. 36:07state and I think what's really
  1002. 36:10interesting about AI and I I this is a
  1003. 36:12little more sort of theoretical because
  1004. 36:13I think you know we're so early in this
  1005. 36:15is that like I think it's very possible
  1006. 36:17you could just say well everybody just
  1007. 36:19keep doing what we're you're doing we're
  1008. 36:21going to stick sort of some models in
  1009. 36:22the middle they don't care what you use
  1010. 36:24because like it's all just data
  1011. 36:26structures to them and so we can then
  1012. 36:28have this sort of central thing and if
  1013. 36:30you know I when you we talked about
  1014. 36:32percolate at the beginning percolate was
  1015. 36:34a content marketing platform worked with
  1016. 36:36very large companies so it's a
  1017. 36:37enterprise software product and it's
  1018. 36:39like you know at the end of the day this
  1019. 36:41is sort of the fundamental challenge of
  1020. 36:42enterprise software is about like you
  1021. 36:44know adoption and change management and
  1022. 36:46I just think I think and I hope and
  1023. 36:49again this is sort of the the optimist
  1024. 36:51in me that like AI kind of lets us just
  1025. 36:54not worry so much about these things and
  1026. 36:56and rather than trying to make everybody
  1027. 36:58change the ways that they work, kind of
  1028. 37:00let them work in these ways and let AI
  1029. 37:02sort of it's my I I call it my Thomas's
  1030. 37:05English muffin theory of AI, which is
  1031. 37:07that it like gets into the nooks and
  1032. 37:09crannies. Um, and so uh uh yeah, that's
  1033. 37:13so anyway, but I have no idea what
  1034. 37:14bureaucracy is positional encoding means
  1035. 37:16yet. I'm hoping I figure it out in the
  1036. 37:17next two weeks before I have to give
  1037. 37:19this talk. I think no, but I I think the
  1038. 37:22point you just made is is totally right
  1039. 37:23and it's actually not it doesn't have to
  1040. 37:25be theoretical like I I've been seeing
  1041. 37:27this too inside of every and I've been
  1042. 37:29meaning to write about it and the the
  1043. 37:30place that it's been coming up is we
  1044. 37:33have so inside of every we run like six
  1045. 37:37different products um and we have 15
  1046. 37:39people so it's it's like a crazy product
  1047. 37:42to headcount ratio um and what's
  1048. 37:45interesting is I really like doing
  1049. 37:47things in a in a bottomup way so
  1050. 37:49everyone everyone each of the products
  1051. 37:50has its own stack. We're not like
  1052. 37:51centralized into a particular stack.
  1053. 37:54Each you know GM that runs a product
  1054. 37:56like just has made a decision about do I
  1055. 37:58run rails or TypeScript or whatever. And
  1056. 38:01what I'm seeing happen which is very
  1057. 38:03cool is um a lot of the different
  1058. 38:07products are running into similar things
  1059. 38:09they want to solve for. So an easy
  1060. 38:11example is um we have one product called
  1061. 38:14Sparkle which is a little bit like a um
  1062. 38:17uh a finder replacement or spotlight
  1063. 38:19replacement. So it it it organizes your
  1064. 38:21files and then it implements really fast
  1065. 38:24spotlight search.
  1066. 38:25>> I'm a user. I like it.
  1067. 38:27>> Okay. So then you you know um and uh so
  1068. 38:30that's really cool. And uh Agentic
  1069. 38:33Search coming soon. Uh check it out. And
  1070. 38:36uh and we're just building a new
  1071. 38:38product, new GM, new stack uh called
  1072. 38:42parah, which is essentially an in-house
  1073. 38:46council. So it's short for parallegal,
  1074. 38:47not par like you know, Thiago Forte
  1075. 38:49Parah. Um and uh the whole job for PAR
  1076. 38:53is just, you know, take all of your
  1077. 38:55legal files and whenever I have a
  1078. 38:57question and be like, okay, did do we
  1079. 38:59ever sign this contract or what's the
  1080. 39:00employee agreement template or whatever?
  1081. 39:02it just gives you the answer and it's
  1082. 39:03just clouded code sitting on top of a
  1083. 39:04directory. Um,
  1084. 39:07and and a thing that we needed to
  1085. 39:09implement for that is uh this sort of
  1086. 39:12like fast file file search. And what's
  1087. 39:14really interesting is historically if we
  1088. 39:17wanted to reuse the stuff that we
  1089. 39:20learned from implementing sparkles file
  1090. 39:23search that would have to be abstracted
  1091. 39:25out into this modular library that
  1092. 39:28anyone can use and then we have to be on
  1093. 39:29the same platform and like that all
  1094. 39:30those things right and what we did
  1095. 39:33instead is we just added uh Piti who's
  1096. 39:36the developer for for for par right now
  1097. 39:38we just added her to the sparkle repo
  1098. 39:40and I was just like just ask cloud code
  1099. 39:42to figure out how it works and just do
  1100. 39:44your own version. Um, and so you get
  1101. 39:46this like sort of tacit code sharing
  1102. 39:49where we all get better but without
  1103. 39:51having to do the work of uh abstracting
  1104. 39:54and modularizing everything because the
  1105. 39:57the percentage of things that you can do
  1106. 39:58that for are pretty low because it's a
  1107. 40:00it's a heavy lift. And I'm seeing that
  1108. 40:02happen all the time where just having a
  1109. 40:04bunch of repos that are all solving
  1110. 40:05similar problems but in different
  1111. 40:07environments in different ways you
  1112. 40:09everyone gets more productive because AI
  1113. 40:11can kind of translate.
  1114. 40:13I've uh one thing we've done there uh we
  1115. 40:16also we so at le
  1116. 40:21uh very large brands um and so we sort
  1117. 40:24of build all kinds of AI things and um
  1118. 40:27you know so we we've got lots of sort of
  1119. 40:28internal and external repos and we
  1120. 40:30frequently have the same thing and
  1121. 40:32actually I've used the GitHub MCP um a
  1122. 40:35few times for that same purpose which is
  1123. 40:37just like um you know you're just in
  1124. 40:40cursor or cloud code or whatever and
  1125. 40:42you're like hey can you go like look up
  1126. 40:44uh we run we've got an internal tool um
  1127. 40:47called intelligence that just sort of is
  1128. 40:49a wrapper around a whole bunch of like
  1129. 40:51stuff that we use right so it's like got
  1130. 40:54some CRM stuff and it's just like been a
  1131. 40:57fun place to build the things that we
  1132. 40:58need to run our company um but it's also
  1133. 41:01a good place to kind of experiment and
  1134. 41:03explore and figure out solutions to
  1135. 41:04interesting problems and so I'll
  1136. 41:06frequently be like oh can you go like
  1137. 41:08just go look at the intelligence repo
  1138. 41:11and um you look at how I implemented
  1139. 41:13that thing there and take those sort of
  1140. 41:16best practices and just pull them over.
  1141. 41:19Um, and yeah, I think that stuff again
  1142. 41:21that's where I really do believe in this
  1143. 41:24idea. I like one of my whenever we have
  1144. 41:30like a client meeting or something my
  1145. 41:31first the question the icebreaker I
  1146. 41:32always use is what was your aha moment
  1147. 41:34with AI? And um uh mine was
  1148. 41:39uh uh I mean it was probably not the
  1149. 41:42very first, but it's the one that sort
  1150. 41:43of I think was most impactful was I was
  1151. 41:46uh I got access to build a chat GPD
  1152. 41:49plugin. Remember when plugins came out
  1153. 41:50like uh two and a half years ago or
  1154. 41:52something now?
  1155. 41:53>> You mean 50 years ago?
  1156. 41:55>> 50 years ago. The the um uh and you know
  1157. 42:00I I like you. I've I've written a lot of
  1158. 42:03software in my life and I you know you
  1159. 42:05know what you do when you like get
  1160. 42:07access to something new. It's like I got
  1161. 42:08to go read the API docs and figure it
  1162. 42:11out and you know like there's a going to
  1163. 42:14be a contract and you know as long as
  1164. 42:15you follow that contract it works. And I
  1165. 42:18go read the plug-in spec and it
  1166. 42:20basically is like oh you just stick a
  1167. 42:22manifest.json JSON file in the root
  1168. 42:24directory of your application and in
  1169. 42:25that you describe how you want us to
  1170. 42:27send you data and you describe how
  1171. 42:28you're going to send it back to us and
  1172. 42:30then we'll deal with the rest. And I was
  1173. 42:32just like that's amazing. It's also like
  1174. 42:38it it's how the world should work. Like
  1175. 42:40I wish everything worked that way. I
  1176. 42:41wish I didn't always have to adhere to
  1177. 42:43the big company's contract for how to
  1178. 42:46send and receive data. Um, but also like
  1179. 42:49I the thing that really struck me in
  1180. 42:51that moment and it's like been my kind
  1181. 42:52of like rallying cry around all this
  1182. 42:54stuff is that um it's also just like
  1183. 42:55fundamentally counterintuitive in that
  1184. 42:57like I literally have a career's worth
  1185. 42:59of intuition for how to integrate
  1186. 43:02software systems and it flipped it on
  1187. 43:04its head like like like quite literally
  1188. 43:08180 degrees away from my intuition of
  1189. 43:10how software systems should be
  1190. 43:12integrated was this thing. And that I
  1191. 43:16think since then has been my kind of
  1192. 43:18thing for everybody has been like this
  1193. 43:21is just not intuitive for now. And and
  1194. 43:24that's not a bad thing. It just means
  1195. 43:25like you need to build intuition and
  1196. 43:28like that's what we're all just out
  1197. 43:30there trying to do with it, right? And
  1198. 43:32so, you know, when I don't know, I mean,
  1199. 43:33part of what I like about what you're
  1200. 43:35doing and, you know, even just hearing
  1201. 43:37the things you're saying, but like
  1202. 43:38generally what you do with the podcast
  1203. 43:40and what you do with every is like so
  1204. 43:42much of it is like we're all kind of
  1205. 43:44just figuring stuff out for the first
  1206. 43:45time, right? And like um you know, we're
  1207. 43:47like, "Oh, will this work?" And then
  1208. 43:49like all of a sudden you have this new
  1209. 43:51bit of intuition for what these things
  1210. 43:53can do and what a computer
  1211. 43:55that is not deterministic looks like.
  1212. 43:58And um that's I think that's just what
  1213. 44:01we're all doing all the time. And that's
  1214. 44:03why it's so fun. I think
  1215. 44:04>> that's why I love this moment cuz like
  1216. 44:06you just have a weird idea and you're
  1217. 44:07like
  1218. 44:09has anyone done this before? And and
  1219. 44:11it's like no. And it's not a it's not a
  1220. 44:13complicated idea. It's just it's just a
  1221. 44:15new whole new territory, you know.
  1222. 44:18>> I Yes. I I think about that all the
  1223. 44:21time. And I think actually like I think
  1224. 44:23one of the really damaging sort of
  1225. 44:24things out there is that um I think
  1226. 44:28there are a lot of people who think
  1227. 44:29we're way further along in this than we
  1228. 44:32are. And so you know I think
  1229. 44:34particularly the people who are sort of
  1230. 44:36scared you know we run we work with like
  1231. 44:38Fortune 50 companies. And so you know
  1232. 44:42when we're sort of like out there and
  1233. 44:43we're talking to people inside the
  1234. 44:44organization a lot of people feel like
  1235. 44:46they've already been left behind. And
  1236. 44:48it's like, no, you can like literally go
  1237. 44:50sign into Chad GBT and like
  1238. 44:54do something like nobody's thought about
  1239. 44:55doing with this thing yet because
  1240. 44:57there's just so much white space to
  1241. 45:00explore and and you might discover some
  1242. 45:03totally new way of using it and or like
  1243. 45:06totally new trick and um I don't know
  1244. 45:08that's just that's and you know I think
  1245. 45:10to be fair to some people that's sort of
  1246. 45:12very intimidating and I don't think by
  1247. 45:14and large the models do any favors to
  1248. 45:16themselves in helping those people get
  1249. 45:18their feet wet in that like you know I
  1250. 45:20think people go on there and they you
  1251. 45:22know it's like you ask it to write you a
  1252. 45:23poem and then it writes you a poem and
  1253. 45:25you're like okay it wrote me a poem um
  1254. 45:27but uh I don't know that feeling of like
  1255. 45:32that feeling of yeah being it's like
  1256. 45:33being on the frontier right
  1257. 45:35>> totally and yeah I I think that um your
  1258. 45:38point about intuitions and getting
  1259. 45:40intuitions is the big thing and I think
  1260. 45:42people what we don't realize is when
  1261. 45:44you're dealing with something
  1262. 45:44fundamentally new you you can't trust
  1263. 45:49like how you reason about it without
  1264. 45:51experiencing it because you have to
  1265. 45:52build the intuition in order to be able
  1266. 45:53to reason about what it means and and
  1267. 45:55how it fits in and whether it works or
  1268. 45:56not. And we're just not used to that cuz
  1269. 45:59we're used to reasoning about things we
  1270. 46:01already have an intuition for. And um
  1271. 46:04and I think that's why like when you
  1272. 46:06first see maybe when you first saw JBT
  1273. 46:08be like oh my god it can do everything.
  1274. 46:09Like we're not going to have jobs in a
  1275. 46:11year. And now we're like three years in
  1276. 46:12and we're like, "Yeah, it's awesome."
  1277. 46:14And
  1278. 46:16jobs are complicated. There's a lot of
  1279. 46:18complex stuff that we do, you know? Um,
  1280. 46:21and I I love I love kind of that, you
  1281. 46:23know, in order to build the intuition,
  1282. 46:26all you have to do is is use it. And
  1283. 46:28that just by using it, you're already
  1284. 46:30kind of on the edge uh for now. And
  1285. 46:33yeah, I think that's the best.
  1286. 46:34>> There's apparently there's a a German
  1287. 46:36word called fingerpitsengu.
  1288. 46:38of course there is
  1289. 46:39>> building fingertip feeling and that's
  1290. 46:41been my um uh just because I can't
  1291. 46:44resist also um I I I'm in that whole in
  1292. 46:50in sort of the realm that you're
  1293. 46:51discussing like uh I've been trying to
  1294. 46:54do a lot of analogy analogizing right
  1295. 46:56it's sort of like and I think you know
  1296. 46:58that's really hard and um uh you know
  1297. 47:01but my two that have sort of stuck the
  1298. 47:03most one is just um I watch a lot of
  1299. 47:05YouTube with my kids and um Uh, we watch
  1300. 47:08this channel called Veritassium. It's a
  1301. 47:10science channel and
  1302. 47:11>> I love Veritassium.
  1303. 47:12>> Yeah, it's great. Um, and he did one
  1304. 47:14where he built a bike that locks out
  1305. 47:17left if you try to turn right and locks
  1306. 47:19out right if you try to turn left. And
  1307. 47:20what he's proving is that you can't
  1308. 47:22actually turn a bike left unless you can
  1309. 47:25turn it right. Um, which none of us
  1310. 47:27would think about when we ride a bike
  1311. 47:29because it's all just second nature and
  1312. 47:31intuition. But it's also why you can't
  1313. 47:34explain to a child how to ride a bike.
  1314. 47:37um they just have to get on it and feel
  1315. 47:38it and um and so you know I really that
  1316. 47:41video is amazing and um uh it's uh that
  1317. 47:44channel is amazing and and but I I've
  1318. 47:46thought a lot about that and then the
  1319. 47:47other one which is a sort of deeper cut
  1320. 47:50is um uh there's an amazing book about
  1321. 47:53quantum physics called beyond weird um
  1322. 47:56by Philip Ball and um the thesis of the
  1323. 47:59book is basically that like there's
  1324. 48:01nothing particularly strange about
  1325. 48:03quantum physics that we like we have a
  1326. 48:05very good understanding of it like we
  1327. 48:07wouldn't be talking right now. We
  1328. 48:09wouldn't be on computers. We wouldn't
  1329. 48:10have phones if we didn't have like a
  1330. 48:12very good grasp of the mechanics that
  1331. 48:14exist underneath it. And his thesis in
  1332. 48:16the book essentially is that like what's
  1333. 48:18really lacking is the vocabulary because
  1334. 48:20like we all exist in a Newtonian world,
  1335. 48:22not in a quantum one. And so we all have
  1336. 48:25words that reflect the sort of
  1337. 48:28deterministic processes of that macro
  1338. 48:32universe. And I think a lot about that.
  1339. 48:34I have not like fully been able to sort
  1340. 48:36of pull that string all the way to AI.
  1341. 48:38But I feel like there's a real
  1342. 48:40connection there because I think that
  1343. 48:41there's just something really weird
  1344. 48:43about using probabilistic computers.
  1345. 48:44Like we're not used to using things that
  1346. 48:47like you ask them the same question
  1347. 48:48twice and they have different answers.
  1348. 48:50Like that's very strange. We're not used
  1349. 48:52to I'm not used to writing code where
  1350. 48:54you can tell the larger company how you
  1351. 48:57want them to send you data and they can
  1352. 48:59just do it. Like these are not
  1353. 49:02normal things that any of us have lived
  1354. 49:05with in our lifetimes and so of course
  1355. 49:07it takes some time for us to adjust. I
  1356. 49:09think so too and I I actually have a
  1357. 49:12hope that language models by becoming a
  1358. 49:16standard way that we use computers will
  1359. 49:19create that vocabulary. Um because we
  1360. 49:23actually are quite good at dealing with
  1361. 49:26probabilistic nondeterministic things
  1362. 49:28like other humans. Um we've just grown
  1363. 49:31up in a world where
  1364. 49:34because of you know the enlightenment
  1365. 49:35and the scientific revolution and and
  1366. 49:37the tools that came out of that are very
  1367. 49:40much like deterministic we've associated
  1368. 49:43that with uh how we see like that's how
  1369. 49:46we see the world because of those tools
  1370. 49:47and that language and uh there's a whole
  1371. 49:50other part of the way that we see the
  1372. 49:51world which is much more squishy and
  1373. 49:52much more like vibes based that has been
  1374. 49:56um I think de deprioritized uh
  1375. 49:59especially in western culture that now
  1376. 50:01that we have a tool that works that way,
  1377. 50:03I think we'll be able to start seeing
  1378. 50:05that again. And that's one of the
  1379. 50:06beautiful things to me about language
  1380. 50:07models is it opens up that whole world
  1381. 50:09again.
  1382. 50:10>> Yes, I love it. Um
  1383. 50:13>> I do want to go back to cloud code uh
  1384. 50:15and
  1385. 50:17we should do the phone unless there's
  1386. 50:18unless there are other things that you
  1387. 50:20uh that that you want to share on the
  1388. 50:21computer. Um, but the the the thing I
  1389. 50:25want to do before we get there that's
  1390. 50:26just on my mind right now is like you
  1391. 50:28said you said, you know, you watch this
  1392. 50:30with your kids and I'm sort of curious
  1393. 50:32how like what do your kids think about
  1394. 50:34this and how are you dealing with it
  1395. 50:36with your kids? Um,
  1396. 50:38>> yeah. Um, I love that question. Uh, so
  1397. 50:41uh I've got a seven and a 10-year-old.
  1398. 50:43Um, and obviously like I'm pretty kind
  1399. 50:45of deeply embedded in this stuff and so
  1400. 50:47I've sort of exposed them quite a bit to
  1401. 50:49it. um uh you know they don't uh so they
  1402. 50:54will like occasionally use the sort of
  1403. 50:56voice models and they have a pretty good
  1404. 50:58understanding and we'll be in the car
  1405. 50:59and just play games and ask questions
  1406. 51:00with Grock and do those kinds of things.
  1407. 51:02Um this weekend actually for the first
  1408. 51:04time my 10-year-old um uh she was really
  1409. 51:08eager to be every year um uh my wife and
  1410. 51:15her sister and brother and mom and and
  1411. 51:18all the cousins and we all get together
  1412. 51:19and we do Christmas together and so it's
  1413. 51:21too many presents to give to everybody.
  1414. 51:23So we do a kind of like not secret
  1415. 51:25secret Santa where everybody chooses one
  1416. 51:26person and my 10-year-old really wanted
  1417. 51:28to be the one who got to be the chooser.
  1418. 51:31Um, and uh, I encouraged her to uh, vibe
  1419. 51:35code an app to do it. Um, and so I just
  1420. 51:38gave her my phone and Vzero. And um,
  1421. 51:42honestly that was like so amazing to
  1422. 51:45watch. Like not just cuz it was so cool
  1423. 51:47to see her do that and build it. And she
  1424. 51:50went through, she was having so much
  1425. 51:51fun. She did 75 revs on Vzero. Um, uh,
  1426. 51:56so she like really got it going.
  1427. 51:58>> Polished Santa app. Uh, She also like
  1428. 52:03started to get into like really
  1429. 52:05interesting kind of like computer
  1430. 52:08science ideas without knowing it. So in
  1431. 52:10one of the things like the adults give
  1432. 52:12presents to adults and the kids give
  1433. 52:14presents to kids, but she wanted this to
  1434. 52:16be a more generalized app. So she
  1435. 52:18realized that like rather than having
  1436. 52:20adults and kids, you need to call them
  1437. 52:21groups, right? and like you know so
  1438. 52:24she's like getting into data modeling
  1439. 52:26and like all of this I'm like watching
  1440. 52:28this conversation happen and um I just
  1441. 52:32thought that was so awesome. Um, and you
  1442. 52:35know, also just like a real pet peeve of
  1443. 52:37mine right now and I've I've sort of
  1444. 52:39gotten this argument with a bunch of
  1445. 52:40people is like there seems to be a big
  1446. 52:42conversation that like there's a a
  1447. 52:44bubble in vibe coding and like because
  1448. 52:46one company or another might have too
  1449. 52:48high of a valuation. And my take on that
  1450. 52:51is like I just I I could not care less
  1451. 52:53what the valuations of these companies
  1452. 52:55are. I think like fundamentally if
  1453. 52:57there's a tool that could allow a
  1454. 52:5810-year-old to like build an app there
  1455. 53:01can't that can't be a bubble. like I I
  1456. 53:04just like can't see a possibility where
  1457. 53:06that is that. So, um anyway, that's sort
  1458. 53:09of one side of it. The other big one for
  1459. 53:10me that I've been thinking a lot about
  1460. 53:12is um
  1461. 53:14uh is sort of media literacy and
  1462. 53:17education um stuff. So, um you know,
  1463. 53:20both at the sort of schools they go to
  1464. 53:23um and then also I went to NYU and I've
  1465. 53:26sort of um been having more and more
  1466. 53:28conversations with the dean of the
  1467. 53:30school I went to there. Um and uh um
  1468. 53:34there's a lot of fear inside schools
  1469. 53:36right now about AI and about cheating
  1470. 53:38and um there's a big thing you know so
  1471. 53:41some parents in my town they you know
  1472. 53:43they wanted to have more of a
  1473. 53:45conversation about it and um you know as
  1474. 53:48someone who I I've thought about this a
  1475. 53:50lot but I've also just like been I've
  1476. 53:52spent my entire life thinking about sort
  1477. 53:54of technology and its effects on culture
  1478. 53:56and I I think I'm like relatively
  1479. 53:58grounded in these things. of like I've
  1480. 54:00put in good hours of thinking. I I know
  1481. 54:02that for sure. Um and so, you know, my
  1482. 54:04my take on it is like one that sort of
  1483. 54:06you can't hide technology that won't be
  1484. 54:09hidden, right? So, it's like, you know,
  1485. 54:11putting our head in the sand is not the
  1486. 54:13best solution. And then, you know, my
  1487. 54:14bigger one though is like I I was out um
  1488. 54:17a friend of mine asked me to come um
  1489. 54:19talk to a school two years ago about AI
  1490. 54:22um out in LA and um afterwards I was
  1491. 54:25talking to a English teacher there and
  1492. 54:27she was like, "What do I do? Like what
  1493. 54:29do I do about all these kids, you know,
  1494. 54:31using AI and I was like, look, I don't
  1495. 54:33really know what your job is because
  1496. 54:34like I mean being an English teacher for
  1497. 54:3711th graders sounds really much harder
  1498. 54:38than my job." Um, but on a really
  1499. 54:42fundamental level, like I don't actually
  1500. 54:44think your job is to teach these kids to
  1501. 54:46write because that's like a lifelong
  1502. 54:48pursuit. I think your job is to convince
  1503. 54:50them that it's worth learning to write.
  1504. 54:52And so in that way, like I I'm not sure
  1505. 54:53that anything fundamentally changes
  1506. 54:55because of AI. Like I I think that you
  1507. 54:58know and again this is my very
  1508. 54:59optimistic take but like I am I think
  1509. 55:03that there are so many parts of the
  1510. 55:06education system that AI really just
  1511. 55:08exposes the sort of flaws in the way
  1512. 55:10that we teach. like why are there so
  1513. 55:12many tests on these kinds of things
  1514. 55:14instead of encouraging thinking and
  1515. 55:16learning and coming to love to write and
  1516. 55:19research and whatever you know it's like
  1517. 55:20we're so focused on teaching kids the
  1518. 55:22you know five paragraph essay you know
  1519. 55:25while every adult who is a writer has
  1520. 55:28long abandoned that and because like
  1521. 55:31it's all about sort of discovering that
  1522. 55:33you like to write and you you know what
  1523. 55:35your own style is and how to do it and
  1524. 55:37you know it's like I'm sure it's like
  1525. 55:39you know a big part of working with AI
  1526. 55:41to write is like telling it is ignoring
  1527. 55:44it because you're like no that's not me.
  1528. 55:46I'm not into that. Like I like I I I'm
  1529. 55:49totally comfortable saying really here.
  1530. 55:51I know you don't think it's a good idea
  1531. 55:52but like I'm I'm cool with it. Um and so
  1532. 55:55you know I don't know. I I've been sort
  1533. 55:56of having a lot of these different
  1534. 55:58thoughts. I'm actually pitching a class
  1535. 56:00for the fall of 26 at NYU to kind of um
  1536. 56:05the idea for it is uh code is essay and
  1537. 56:08my my sort of point of view is like this
  1538. 56:10sort of opens up this new way to express
  1539. 56:12yourself. um and that you know like we
  1540. 56:17have all these other ways that we
  1541. 56:19celebrate to express ourselves but code
  1542. 56:21has been like long shut off from people
  1543. 56:23because it's but actually it's kind of
  1544. 56:25amazing and it lets you express yourself
  1545. 56:26in all these different kinds of ways and
  1546. 56:28like you know this is what my
  1547. 56:2910-year-old was doing this weekend. So
  1548. 56:30anyway, those are all the bits and
  1549. 56:32pieces as a parent I've been thinking
  1550. 56:33about. The one other thing I will plug
  1551. 56:35is um media literacy I think is a big
  1552. 56:38piece of it. A lot of people are afraid
  1553. 56:39of these models and hallucinations and
  1554. 56:42um there's a book um by a guy named Tim
  1555. 56:45Harford who writes for uh the FT and
  1556. 56:48he's an economist um and he has a book
  1557. 56:50called The Truth Detective which is an
  1558. 56:52adaptation of the data detective but
  1559. 56:54it's written for kids and it's the best
  1560. 56:56media literacy book I've ever read for
  1561. 56:59adults or for children. Um, and I think,
  1562. 57:02you know, a lot of what this AI
  1563. 57:04conversation exposes is like how bad a
  1564. 57:07job we do with helping and arming our
  1565. 57:10kids and our adults to be sort of like
  1566. 57:15truly media and technology literate.
  1567. 57:17And, you know, like being really good at
  1568. 57:20knowing what's real on social media
  1569. 57:22turns out to be also really useful for
  1570. 57:24like differentiating between
  1571. 57:26hallucinations and non-h hallucinations
  1572. 57:28in chat GBD, right? like this is a sort
  1573. 57:30of to me a kind of very central skill
  1574. 57:33that like we need to arm everybody with
  1575. 57:35and um I am sort of way more interested
  1576. 57:38in that with my kids than I am in
  1577. 57:41worrying about them cheap. That was a
  1578. 57:42very long answer to your question. I
  1579. 57:43don't know if it got at what you were
  1580. 57:45looking for.
  1581. 57:45>> No, it's amazing. I that's exactly what
  1582. 57:48I'm looking for and it sort of what it
  1583. 57:50strikes me another way to frame what
  1584. 57:53you're saying is for example rather than
  1585. 57:58learn having them memorize and be
  1586. 58:00quizzed on the 50 what the 50 states are
  1587. 58:03uh
  1588. 58:05asking them to go find the 50 states
  1589. 58:07with CHBT and be able to tell when the
  1590. 58:10AI is giving them the wrong answer
  1591. 58:11because that's a lot more of the the
  1592. 58:14meta skill that they're going to need
  1593. 58:15anyway down the down the line. And
  1594. 58:17again, I'm not a teacher. I'm sure
  1595. 58:19there's lots of teachers who were like,
  1596. 58:21"This is that's crazy for a lot of
  1597. 58:23probably good reasons." But there's
  1598. 58:24something interesting there where it
  1599. 58:26becomes the meta skills become more
  1600. 58:29important than they used to be. And in
  1601. 58:31order to do well at the meta school, you
  1602. 58:33have to also be able to like to some
  1603. 58:35degree do the underlying skill, too. But
  1604. 58:37um we probably could be spending a much
  1605. 58:39more time in the medical school than we
  1606. 58:40are now. And the education system like
  1607. 58:41isn't really set up to do that anyway.
  1608. 58:43>> Yeah. And I I you know even I think even
  1609. 58:45simpler examples is like I I went to a
  1610. 58:47school called Gallatton at NYU and and
  1611. 58:50um when you graduate you have to sort of
  1612. 58:53it's not quite a thesis defense but you
  1613. 58:55have to sort of spend three hours with
  1614. 58:56four professors and and or three
  1615. 58:59professors and sort of explain kind of
  1616. 59:02your line of reasoning around why you
  1617. 59:04studied what you studied. and you need
  1618. 59:06to be prepared to kind of like weave
  1619. 59:08into that defense um uh any of 25 books
  1620. 59:12that you put on your book list. Um and
  1621. 59:15you know I was talking to them and it's
  1622. 59:16like amazing that's like entirely AI
  1623. 59:18proof, right? Like there's no cheating
  1624. 59:20on that. Like you show up in that room
  1625. 59:22and you're either prepared to speak to
  1626. 59:24it or you're not. And whether you're
  1627. 59:25prepared with AI or not is uh it doesn't
  1628. 59:28mean anything, right? like it's it's
  1629. 59:29like like do you can you can you make a
  1630. 59:33an argument in this room and like not
  1631. 59:36everything is going to be that easy to
  1632. 59:38be sort of like cut off but like I don't
  1633. 59:40know there's something really beautiful
  1634. 59:42about that idea right because it's like
  1635. 59:44it's naturally cheating proof because
  1636. 59:47like you're sitting there and and it's a
  1637. 59:49question of like did you actually
  1638. 59:51internalize these things um and I don't
  1639. 59:54know that's way more interesting to me
  1640. 59:56than even like was your essay or any of
  1641. 59:59these other it's like did you did you
  1642. 1:00:01get it like um and uh so yeah I don't
  1643. 1:00:06know um I'm I'm I I'm trying to do my
  1644. 1:00:10best to kind of like um take a balanced
  1645. 1:00:12approach and um try to at least sort of
  1646. 1:00:15like tamp down some of I I live in a
  1647. 1:00:18small town in Connecticut and you know I
  1648. 1:00:20think there's a lot of fear amongst
  1649. 1:00:22parents you know it's like it was mobile
  1650. 1:00:24phones and then it was social media and
  1651. 1:00:26now it's AI and it's like another thing
  1652. 1:00:28that's going going to ruin our kids. And
  1653. 1:00:29um I don't think that that is true. Um
  1654. 1:00:33uh but I think there are things we can
  1655. 1:00:35and should do to encourage it to not be
  1656. 1:00:38true. Like really get them really good
  1657. 1:00:40at the things you're saying. Like how do
  1658. 1:00:42you tell? And again, it's like a
  1659. 1:00:43hallucination is just a form of the same
  1660. 1:00:46kind of misinformation that exists in on
  1661. 1:00:49television and on the internet and in
  1662. 1:00:51social media and everywhere else. And um
  1663. 1:00:54you know just sort of like encouraging
  1664. 1:00:55people to kind of get in touch. There's
  1665. 1:00:56a great part of the truth detective um
  1666. 1:00:59for the kids book he he calls it the
  1667. 1:01:01brain guard and like one of the bits of
  1668. 1:01:03advice he has is when you encounter some
  1669. 1:01:05piece of information if it makes you
  1670. 1:01:07feel really good cuz you agree with it
  1671. 1:01:09then you should be even more skeptical
  1672. 1:01:11of it and he calls it the brain guard
  1673. 1:01:12you know and he's explaining this like
  1674. 1:01:14this is for a 9-year-old and I just
  1675. 1:01:15thought that was like such a beautiful
  1676. 1:01:16way to put it right like that's like
  1677. 1:01:18what that's what you learn to do when
  1678. 1:01:20you get good at being on the internet um
  1679. 1:01:24is that you're like wait I should like
  1680. 1:01:25be more skeptical of this because this
  1681. 1:01:27is like in line with everything I think.
  1682. 1:01:29Let me just doublech check the way like
  1683. 1:01:32I I you know, you get to learn that
  1684. 1:01:34feeling in your gut and get to learn
  1685. 1:01:36when to react to it. And so yeah, that's
  1686. 1:01:39a lot of how I think about it.
  1687. 1:01:41>> That's great. I'm I'm interested. I'm
  1688. 1:01:42going to get I'm going to get that book.
  1689. 1:01:44I'm I'm interested in reading it.
  1690. 1:01:46>> You should. My plan is to read it to my
  1691. 1:01:47kids every year um from now on.
  1692. 1:01:50>> Nice. I love it.
  1693. 1:01:50>> Just refresh it.
  1694. 1:01:52>> Um all right. Now for the moment we've
  1695. 1:01:54all been waiting for. show us how you
  1696. 1:01:56use cloud code on your phone as a second
  1697. 1:01:58brain notetaker.
  1698. 1:02:00>> Okay, so here we go. So I am going into
  1699. 1:02:04an app called Termius.
  1700. 1:02:08>> Termas is just a terminal.
  1701. 1:02:11Um, and uh, what is allowing all this to
  1702. 1:02:14happen behind the scenes is in my
  1703. 1:02:16basement I have a mini PC and on that
  1704. 1:02:20mini PC I have a thing called tail scale
  1705. 1:02:24running and tail scale lets you set up
  1706. 1:02:26these very simple VPNs. So I'm currently
  1707. 1:02:30inside like if I scroll down here you
  1708. 1:02:33see I'm on a VPN. That's my personal
  1709. 1:02:35VPN. I'm not like on an outside VPN. I
  1710. 1:02:38see. Um, so the only way to access this
  1711. 1:02:40machine is through my VPN.
  1712. 1:02:42>> Okay.
  1713. 1:02:43>> Um, and so then when I go in there, um,
  1714. 1:02:46because I sync my Obsidian,
  1715. 1:02:50um,
  1716. 1:02:52with, uh, Git, so I put it, it's on
  1717. 1:02:55GitHub, uh, in a private GitHub. Um, I
  1718. 1:02:58can then sync it back down to here. And
  1719. 1:03:00so then I can just call up Claude.
  1720. 1:03:03And now I'm just in claude code talking
  1721. 1:03:07and thinking and I can just be like um
  1722. 1:03:11what's new in the last two days.
  1723. 1:03:16Um I can access any of my agents. Um I
  1724. 1:03:20can do anything. And again this is in my
  1725. 1:03:22Obsidian but I can use this anywhere,
  1726. 1:03:25right? So I'll be like on the fly. I've
  1727. 1:03:28got other repos in here. Um, you know, I
  1728. 1:03:30I realized like a link was broken on my
  1729. 1:03:34conference site and so I uh um I just uh
  1730. 1:03:39opened the repo. I pulled it down. I
  1731. 1:03:43asked QuadCode to make the changes and I
  1732. 1:03:45was able to do it right here. Um, so
  1733. 1:03:47this has been like completely wild to me
  1734. 1:03:50because again, this is very much in that
  1735. 1:03:52like like on Tuesday of this week, um,
  1736. 1:03:57we had Monday off, Tuesday, I dropped
  1737. 1:03:58the I dropped the kids off at the bus
  1738. 1:04:00and then I, uh, I went and I sat and had
  1739. 1:04:02breakfast and I literally sat on my
  1740. 1:04:04phone and worked on this talk for like
  1741. 1:04:07two hours. Um, and I did it through
  1742. 1:04:10here, right? like on my phone where I
  1743. 1:04:12was like doing real thinking and
  1744. 1:04:14research and pulling things in and
  1745. 1:04:15pasting things in and doing all this
  1746. 1:04:17kind of stuff and you know I'm able to
  1747. 1:04:20do it all and it it just it doesn't seem
  1748. 1:04:24like I could do that kind of thing
  1749. 1:04:26without that. So yeah, this has been
  1750. 1:04:27like a completely revolutionary change
  1751. 1:04:29in my life. Um, and uh, actually one of
  1752. 1:04:33the things I've been doing lately is
  1753. 1:04:34like setting up I've got all these
  1754. 1:04:36friends now who have set up um, like
  1755. 1:04:38little partitions of this mini PC in my
  1756. 1:04:40basement um, so that they can also run
  1757. 1:04:42cloud code on their phone because um, I
  1758. 1:04:45like it so much. Do
  1759. 1:04:47>> does this make you be like, "Oh my god,
  1760. 1:04:49I got to drop everything and just build
  1761. 1:04:51a actual purpose-built notes app that
  1762. 1:04:53has cloud code as a backend for this?"
  1763. 1:04:57>> No. Well, I mean, actually, one of the
  1764. 1:04:58things I've been thinking a lot is like
  1765. 1:05:00maybe I just like everything should just
  1766. 1:05:02run in Linux all the time for me. Maybe
  1767. 1:05:04this is like at least for the short
  1768. 1:05:05term, this is the answer to all of what
  1769. 1:05:07I just need to like not have anything
  1770. 1:05:08anywhere else. Um, no, I mean, I'm I'm I
  1771. 1:05:12I will say I'm I'm sort of pretty out of
  1772. 1:05:14the SAS game these days, so I don't
  1773. 1:05:16often kind of think that I should drop
  1774. 1:05:18everything and and do anything. I find
  1775. 1:05:20this to be a really amazing solution. Um
  1776. 1:05:22and uh um but no, I mean I this but this
  1777. 1:05:27has really this has really like changed
  1778. 1:05:30the way I work and I feel like I can
  1779. 1:05:32just be anywhere and just be on my phone
  1780. 1:05:33and um you know I mean I was out like uh
  1781. 1:05:37I needed a break. Uh it was you know
  1782. 1:05:394:30 I went and sat outside for a while
  1783. 1:05:42and then um we had a project that needed
  1784. 1:05:44to get delivered to a client and a small
  1785. 1:05:46change needed to be made that like I was
  1786. 1:05:48the best suited to make that change. Um,
  1787. 1:05:51and so I just like hopped on my phone. I
  1788. 1:05:53pulled the repo down and I um went into
  1789. 1:05:55Cloud Code. It's like a tiny little
  1790. 1:05:57change, you know, like the way I find
  1791. 1:05:59myself using Cloud Code the most for
  1792. 1:06:01code is that like mostly I'm having it
  1793. 1:06:03do the work I already know how to do.
  1794. 1:06:06You know, I'm like, oh, I like I know
  1795. 1:06:07exa I knew exactly what was going on in
  1796. 1:06:09that situation. Um, like I knew why we
  1797. 1:06:13were having the issue we were having.
  1798. 1:06:14And so it was like I could have gone
  1799. 1:06:16back to my computer and opened up uh
  1800. 1:06:18cursor and done it in cursor either by
  1801. 1:06:21hand or with cursor, but it was like I
  1802. 1:06:23just I was like I told Cloud Code
  1803. 1:06:24exactly where to look. It was like and I
  1804. 1:06:26first confirmed that the problem was
  1805. 1:06:28what I thought the problem was and then
  1806. 1:06:29I just had it push a solution and it
  1807. 1:06:31pushed a PR and then I was done.
  1808. 1:06:33>> Amazing.
  1809. 1:06:33>> And I was still sitting outside by the
  1810. 1:06:35pond.
  1811. 1:06:39>> I love that. Um yeah, I've I've
  1812. 1:06:41definitely had that experience. I've
  1813. 1:06:43never I've not done it on my phone. I'm
  1814. 1:06:44like I have my laptop out with me, you
  1815. 1:06:46know, by the pond or by a lake or
  1816. 1:06:48whatever. But you're inspiring me. I
  1817. 1:06:50have a Mac Mini in the office that I've
  1818. 1:06:52been meaning to set up. So, I think this
  1819. 1:06:53is going to give me
  1820. 1:06:53>> one of the other, by the way, um one of
  1821. 1:06:56my other I'll just take you out of here
  1822. 1:06:58for one sec actually just to show you
  1823. 1:06:59this. One of my other big ahas um
  1824. 1:07:03recently has been um
  1825. 1:07:06uh building um clawed code helpers for
  1826. 1:07:12um doing uh
  1827. 1:07:15uh basically like uh setup work. So it's
  1828. 1:07:19like I'm not I'm like been playing with
  1829. 1:07:21Linux lately. I'm playing with this
  1830. 1:07:22Omari which is DHH's Linux distribution
  1831. 1:07:26and I'm not like super comfortable in
  1832. 1:07:28here. And so I got this whole this is a
  1833. 1:07:31claude code project specifically to help
  1834. 1:07:33me configure this box and it's like so
  1835. 1:07:36nice because I'm like oh how do you do
  1836. 1:07:38this in Linux again or like what's the
  1837. 1:07:40neoim command or like can you change
  1838. 1:07:42this can you help me install this plugin
  1839. 1:07:44for neo or whatever it is or and so now
  1840. 1:07:46I have one on my Mac too where it's like
  1841. 1:07:48um can you clean up all the homebrew
  1842. 1:07:50things or like I switched from like
  1843. 1:07:52which Python package manager I was using
  1844. 1:07:54and it was like that would have been
  1845. 1:07:55super overwhelming for me and I was like
  1846. 1:07:57I want to switch from
  1847. 1:07:58pip to using UV. Can you like just make
  1848. 1:08:01that happen? And it like just did all
  1849. 1:08:02this stuff for me. And it knows all my
  1850. 1:08:04preferred settings. And so, um, I
  1851. 1:08:06actually have a version of this where
  1852. 1:08:07like now I've got it so tuned that, um,
  1853. 1:08:10if I want to launch a new box for doing
  1854. 1:08:13something, it'll just have all my
  1855. 1:08:14settings ready to go and then I can log
  1856. 1:08:16into Claude Code and the Claude code can
  1857. 1:08:18then set up anything else that didn't
  1858. 1:08:20get set up in the initial process.
  1859. 1:08:22>> That is amazing. That's wild.
  1860. 1:08:25Um,
  1861. 1:08:26>> this is my happy place.
  1862. 1:08:28>> I I can see that. Are do you have any
  1863. 1:08:30like um uh are there are there any big
  1864. 1:08:33projects like this that you've been
  1865. 1:08:34itching to do or itching to try out?
  1866. 1:08:36>> Uh, no, not really. I mean, I've been
  1867. 1:08:39having a ton of fun that my my server
  1868. 1:08:40stuff has been a ton of fun. I've been
  1869. 1:08:42doing a lot of that. Um I'm uh I'm I
  1870. 1:08:46like I said I mean this is kind of a
  1871. 1:08:47joke and kind of not is like I'm super
  1872. 1:08:49interested in um like doing more I
  1873. 1:08:53because Claude code has become such a
  1874. 1:08:55sort of integral part of my life like
  1875. 1:08:57I'm very interested in the command line
  1876. 1:08:58and I found myself installing more and
  1877. 1:09:00more things into the command line um and
  1878. 1:09:02like doing more and more work. So, I've
  1879. 1:09:04been using like uh um Simon Wilson has a
  1880. 1:09:07LLM command line tool and like been
  1881. 1:09:10doing more and more stuff sort of in
  1882. 1:09:12that and then like also layering that
  1883. 1:09:14back into cloud code. So, it's like I
  1884. 1:09:16did my newest Claude code little
  1885. 1:09:19Obsidian tool is it I have an
  1886. 1:09:22attachments folder in Obsidian where all
  1887. 1:09:24the PDFs and images and stuff in any
  1888. 1:09:26note go, but inevitably they have
  1889. 1:09:29terrible names, right? I mean, um, and
  1890. 1:09:31so this goes through very similar to,
  1891. 1:09:34um, uh, Sparkle and but just in that
  1892. 1:09:37Obsidian folder and it it renames them
  1893. 1:09:39all and then it also puts them in like a
  1894. 1:09:42meta data. It puts them in a table in
  1895. 1:09:44the attachments folder and then it
  1896. 1:09:46renames all the attachment links back.
  1897. 1:09:48So it just like cleans everything up. It
  1898. 1:09:50just does it through Gemini Flash. Um,
  1899. 1:09:53and so it's like I don't know, stuff
  1900. 1:09:54like that's kind of amazing. Um, so I
  1901. 1:09:58don't I'm just like I'm having the time
  1902. 1:09:59of my life just building and tinkering
  1903. 1:10:02and you know I mean this is just on the
  1904. 1:10:04side and I get to do the same thing. I
  1905. 1:10:06mean we work with like Amazon and Meta
  1906. 1:10:08and PayPal and all these big companies
  1907. 1:10:10and you know we're just like building
  1908. 1:10:12amazing stuff all the time.
  1909. 1:10:13>> I love that. I love the energy. If
  1910. 1:10:15people are interested in uh following
  1911. 1:10:19you or working with you uh where should
  1912. 1:10:21they find you?
  1913. 1:10:23>> Yeah. Um, so Alfic is alle.com. A l
  1914. 1:10:27pic.com.
  1915. 1:10:30Um, and then I also run this thing
  1916. 1:10:31called Brand B RX N D.AI to make it
  1917. 1:10:34particularly confusing. Um, that's a
  1918. 1:10:36conference. We've got the conference
  1919. 1:10:37coming up on September 18th in New York
  1920. 1:10:39City. You should come if you're around.
  1921. 1:10:41It'll be really fun. We're going to talk
  1922. 1:10:42about marketing and AI. Um, and I also
  1923. 1:10:44write a newsletter there at
  1924. 1:10:46newsletter.brnd.ai
  1925. 1:10:48about sort of specifically at this
  1926. 1:10:50intersection of AI and marketing. Um,
  1927. 1:10:52those are kind of the best places to
  1928. 1:10:54find me these days.
  1929. 1:10:55>> Awesome, Noah. Always a pleasure.
  1930. 1:10:58>> Pleasure is all mine. Thank you, Dan.
  1931. 1:11:03[Music]
  1932. 1:11:08Oh my gosh, folks, you absolutely,
  1933. 1:11:10positively have to smash that like
  1934. 1:11:12button and subscribe to AI and I. Why?
  1935. 1:11:15Because this show is the epitome of
  1936. 1:11:17awesomeness. It's like finding a
  1937. 1:11:18treasure chest in your backyard, but
  1938. 1:11:20instead of gold, it's filled with pure
  1939. 1:11:22unadulterated knowledge bombs about chat
  1940. 1:11:25GPT. Every episode is a roller coaster
  1941. 1:11:28of emotions, insights, and laughter that
  1942. 1:11:30will leave you on the edge of your seat,
  1943. 1:11:32craving for more. It's not just a show,
  1944. 1:11:35it's a journey into the future with Dan
  1945. 1:11:37Shipper as the captain of the spaceship.
  1946. 1:11:40So, do yourself a favor, hit like, smash
  1947. 1:11:43subscribe, and strap in for the ride of
  1948. 1:11:45your life. And now without any further
  1949. 1:11:47ado, let me just say Dan, I'm absolutely
  1950. 1:11:50hopelessly in love with you.

About this transcript

This page contains the full transcript of Claude Code Can Be Your Second Brain by Every, generated from the public captions YouTube serves with the video. The transcript has 14,569 words across 1,950 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.