YouTube2Text

How to Actually Code with AI (ft. Matt Pocock) — Transcript

by NetworkChuck · 24,631 words · 1,326 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Would you say that anyone needs to learn a  specific programming language right now? Nah.
  2. 0:04I think that software engineering has always  been easier than people think it [music] is,
  3. 0:08right? Coding has always been easier than  people think. I would have relatives coming
  4. 0:12to me and say, "Yo, you must be so smart.  You're a coder. You know, you know how
  5. 0:14all that stuff fits [music] together." You  know, but you just learn it. I do not have
  6. 0:18a software development background. Yet, I can  now do it all. If I ask AI to code something,
  7. 0:22it can code it all day. But I can feel the  limitations and I think everyone else does.
  8. 0:26Engineers tend to produce better products than  non-engineers right now because they have the
  9. 0:32domain expertise. They [music] know what to  ask for. They have the right language. Let's
  10. 0:37take a high school kid. They want to become a  software developer. And in this day and age,
  11. 0:41it's like what does that mean? Should  they? What I [music] would say is
  12. 0:49[music]
  13. 0:54first of all, thank you for making the time to  do this. No worries. I know you've been busy.
  14. 0:58Like you just launched your new uh AI crash  course and your new skills like was it 1.3
  15. 1:03you released? That's right. Yep. That's awesome.  Y congrats. I've got a I've got a baby coming in
  16. 1:10uh late November and I'm trying to get a course  out before then. It's unusual. I have this hard
  17. 1:16of a deadline, you know, but the baby is coming.  Let's go ahead and get started. So, first for
  18. 1:22people who don't know who you are, which I'll  kind of give you an intro. We can just kind of
  19. 1:25cold start here. Um you are Matt PCO and the where  I've heard about you was initially your skills,
  20. 1:30which I think is what most people um that's their  gateway drug to Matt PCO and um everyone uses
  21. 1:36them. It's like the de facto thing to use for  AI coding now. But everyone just kind of adopts
  22. 1:42your skills to their own workflow without exactly  looking at everything you do. I think that's kind
  23. 1:47of my my that that was kind of my approach. But  then I saw that you had a course and I'm like I'm
  24. 1:52going to buy the course. So I bought your course  and it's awesome. Um but it's also challenging and
  25. 1:58which is why I wanted to have this conversation  with you. So um real quick, if there's anything
  26. 2:01I missed about you, can you go ahead and tell the  audience who you are and what you're about? No, I
  27. 2:05mean I right now I'm the AI guy. I suppose I'm the  skills guy. Um I've not always been this person.
  28. 2:10Uh before I was the TypeScript person actually.  So I was a hu influencer in the TypeScript space.
  29. 2:17I was teaching Typescript courses and that came  out from uh working in open source on very very
  30. 2:22very challenging TypeScript projects like XATE.  I don't know if you've heard of Xate. It's a tiny
  31. 2:26little library. Well, not actually that tiny  but it's about state machines and state charts
  32. 2:31in Typescript. And I was trying to make that type  safe and that was a blooming nightmare. Uh taught
  33. 2:36me a lot about TypeScript and I released a course  about that and went full-time teaching based on
  34. 2:40that course which was called total typescript  and that meant that when a AI came around I was
  35. 2:46looking for my next project. Basically I didn't  know what I was going to do and AI came and I
  36. 2:51realized ah this seems interesting. took me a  while to get there because I was pretty skeptical,
  37. 2:56but I have risen through my skepticism and now  yeah, I'm definitely one of the skills people
  38. 3:04pushing their skills out there and just trying to  write down interesting processes I find and things
  39. 3:09that work and shipping them as markdown files.  So, the reason I'm so excited to talk with you
  40. 3:16is I think I'm in a place of a lot of other people  where I I have now been given a magic wand. I can
  41. 3:23do whatever I want, but I need to go to Hogwarts  to learn how to use this thing. [laughter] And I
  42. 3:31uh I I'm coding like crazy. I'm probably coding  as we speak. I I I'm sure I have an agent doing
  43. 3:36something, but I know I'm not doing it well.  And I know there's a way to do it well. And and
  44. 3:42and where I'm really curious, and I think we're  going to end up talking about this at some point,
  45. 3:46is where is coding sitting right now? how much can  we give away to the agent and trust it versus how
  46. 3:53and I think you the term you use is putting agents  in a straight jacket which I love that because
  47. 3:56it's like they do go off the rails and they do get  kind of crazy. So that's why I'm excited. I think
  48. 4:01a lot of people in my audience are in the same  boat. Like I do not have a software development
  49. 4:05background yet I can now do it all. I don't have  to know how to write any code but I can feel the
  50. 4:11limitations and I think everyone else does. Like  I made this app but it's on my computer. How do
  51. 4:16I get other people to use it? I don't know what's  the best way to do it. let me ask my AI agent and
  52. 4:20then they spin up Versell or they do something  random and it's it's their opinion and it's not
  53. 4:24good. So what I would love to know and we I guess  we can just go ahead and jump into this. Can you
  54. 4:30tell me what actually you tell me what what is  the best way to approach software development
  55. 4:36right now with AI coding? Cuz like you you teach  the course, you're the master. How can how can
  56. 4:41we step into this? I don't like this framing. I  don't like the framing of me being the guru and
  57. 4:46you guys being the students. You know what I mean?  Like that's that's it's that's not what I'm going
  58. 4:49for at all. Like I you know I've got a bald head,  right? I live near Oxford. I've got glasses. That
  59. 4:53doesn't make me smart. Doesn't make me smarter  than you. Right. And I've got the British accent
  60. 4:57too which helps. But it it helps so much. You  have no idea. So much right. [laughter] So okay,
  61. 5:04Johnow in his book a philosophy of software design  talks about the difference between tactical and
  62. 5:10strategic programming. Right? Tactical versus  strategic. Tactical is like the on the ground
  63. 5:15sergeant trying to win the battle and the  strategic stuff is the general trying to win
  64. 5:21the war, right? The long view. And whenever,  you know, preai, whenever you were working,
  65. 5:26you would try to be working with a bit of tactical  and a bit of strategic all the time, right? You'd
  66. 5:32be thinking about, okay, how does this change  make a difference in the long term? And also,
  67. 5:37how does this change like how does it accomplish  my short-term goals? How does it help me win
  68. 5:41the battle of today? Now AI is really good at  tactical programming and terrible at strategic
  69. 5:48programming. It's just garbage at it. It because  of the way it's trained and because of the way
  70. 5:53like you think about an AI in a benchmark, right?  It has a task to fulfill and it's being judged on
  71. 6:00that specific task. In other words, its focus  is 100% tactical. Whereas it's not you're not
  72. 6:06going to get better scores on the benchmark by  thinking long term and thinking strategically.
  73. 6:10And even how would you grade that, you know,  it's very difficult to grade that because your
  74. 6:15strategic mistakes don't usually come back  at you until like 9 months down the line,
  75. 6:19you know. So to in order to get good at working  with AI, you need to get good at the strategic
  76. 6:26stuff because AI can take the tactical stuff,  can eat it for you, but unless you're careful,
  77. 6:32it'll just focus on the stuff for now and won't  care about the stuff long term. Okay. And I I do
  78. 6:39feel that because you're right, like if I if I ask  AI to code something, it can code it all day and
  79. 6:45it it will even go in a very bad direction because  it just wants to be that agreeable assistant for
  80. 6:50me. But strategy is where it's weak. Um would you  say that's the the biggest weakness you see right
  81. 6:56now with people jumping into vibe coding? Totally.  I mean with vibe coding is a people have like
  82. 7:04there's this term called semantic diffusion where  like words stop meaning the thing they originally
  83. 7:08meant right and so vibe coding is a stupid um  term that now means all sorts of different things
  84. 7:13I take the original definition which is coding  without looking at the code without thinking about
  85. 7:18the code with considering that the code is just  like a an artifact we don't care about now vibe
  86. 7:23coding I don't think has ever worked if it's the  only thing you do right if you're only vibe coding
  87. 7:30It means that your code will slowly get worse and  worse and worse because again this tactical agent
  88. 7:35is just thinking about how to solve the current  problem. It's not thinking about the long term and
  89. 7:39it gets worse and worse and worse. And because the  code is the environment your agent operates in,
  90. 7:46the environment your agent operates in slowly  gets worse. It's like the dog meme of the house
  91. 7:50burning around them, right? And the dog is just  sort of sat there going, "Oh, okay." Right? the
  92. 7:54agent just just watching the house fall down as  they're trying to get their jobs done. So, you
  93. 8:00need someone to keep the house in order and that  person is probably at the moment you. I do think
  94. 8:05that agents might be getting better at this, but  they still need to be prompted and told to think
  95. 8:12about strategy, to care about strategy long term,  and right now they just don't. So, you you don't
  96. 8:20call your course a a vibe coding course. And I and  I see why you would steer away from that because
  97. 8:24that that vibe coding definition still has that  negative connotation even though everyone is kind
  98. 8:29of vibe coding if you if you um if you call it AI  assisted vibe coding. Vibe coding has a negative
  99. 8:36connotation. Uh do you think that developers right  now all of them should be using AI? Like are they
  100. 8:45foolish to not use AI? Yeah, it's ridiculous not  to use AI because so much of our job before was
  101. 8:52tactical programming, right? It was it was just  mostly tactical and you would get these nice uh
  102. 8:57relaxing breaks where you could just focus on the  tactical stuff, just solve the problem today. Oh,
  103. 9:01I get to kind of just go into a nice flow state  and just solve it. But now you can pay less than
  104. 9:06minimum wage for someone else to do it, right?  It's unreasonable for you to be spending your
  105. 9:10time doing that. And to be honest, the stuff that  always interested me most as you know a developer,
  106. 9:16as someone who's interested in increasing velocity  on our team, increasing the um just the quality
  107. 9:23of code that we're putting out, the in quality  of features for people. The strategic stuff was
  108. 9:27always the most interesting stuff for me. Anyway,  so I I mean I love this new age. I don't mind that
  109. 9:32tactical programming is dead. I'm fine with it.  You know, just get me thinking about the long
  110. 9:36view. let me just sort of cook and create this  beautiful um environment for my agents to operate
  111. 9:41in. I find that stuff really exciting. M okay. So  that's that's where I want to live and I'm hoping
  112. 9:47that throughout this conversation you will give  me and the audience a direction on where we can
  113. 9:52start actually becoming I guess becoming software  engineers in a way cuz like right now I I don't
  114. 9:58see a place where I'm going to actually go out  and hire a software engineer. Like I know I can
  115. 10:02do it myself. I just know there's a few things I'm  missing, a few gaps I need to fill in. Um would
  116. 10:07you say that's where where people can be right  now? Like can we just become software engineers
  117. 10:11with a few like a course that you produce? I  think that um software engineering has always been
  118. 10:17easier than people think it is, right? Coding has  always been easier than people think, you know?
  119. 10:21Uh you know, I would have relatives coming to me  and say, you must be so smart. You're a coder. You
  120. 10:26know, you know how all that stuff fits together,  you know? But like it's just you just learn it,
  121. 10:30right? you know, you just spend like a couple of  weeks learning it and you get better over time and
  122. 10:34you make mistakes and you know, and you've got  all these tools to help you out and now agents
  123. 10:38can help you too. And it's always been easier than  it looks. And strategic programming is really just
  124. 10:45starting to think and care about different things  than you're used to caring about. And honestly,
  125. 10:51so much of strategic programming is just  caring a bit more than the person next to you,
  126. 10:57right? is caring about the quality of the thing  you're producing, of the code, the environment or
  127. 11:02the agent operates in. And so so much of that  like I could always tell which of the my team
  128. 11:09members were strategic programmers that had that  long-term view because they just cared more. And
  129. 11:14I think honestly if you're just interested in it,  if you want to produce better code and you want
  130. 11:19your agents to do it, then that's the first place  to start. It's just, you know, can I can I swear
  131. 11:24on this? I'm not sure. Can I curse? Yeah. Yeah.  Go ahead. Yeah. Okay, you have to give a you know
  132. 11:27what I mean? You just got to give a [ __ ] And so  that's the first place to start. Hey, time for a
  133. 11:32coffee break. And this coffee break is sponsored  by Threat Lucker and their learning event Zero
  134. 11:36Trust World. Did you know about this? It's a thing  and it's kind of sick. It's happening February
  135. 11:4017th through 19th, 2027. Gosh, is it really  already 2027 coming up? At Zero Trust World,
  136. 11:45you're going to get access to hands-on hacking  labs where you get firstirhand experience hacking
  137. 11:49stuff. And it's okay if you're new to hacking.  They have every skill level. They're also going
  138. 11:53to have mainstage talks, breakout sessions,  and technical deep dives. These are experts
  139. 11:57in the field sharing their knowledge with you. And  speaking of experts, you're going to be around a
  140. 12:01bunch of them at this event. Networking is one  of the main reasons you want to go to any type
  141. 12:04of learning event. You'll be spending time with  people who are dealing with the same threats that
  142. 12:07you are. And it's people who are practitioners  and also executives. You're going to be making
  143. 12:11some connections and those sessions you take can  be continuing education credits. This event has so
  144. 12:16much and it grows every year. And the afterparty,  I heard, is something you don't want to miss. Why?
  145. 12:20You got to go and find out. So now I know you  want to go. I got a link below. You'll get $200
  146. 12:24off or use the code ZTW networkchuck 27. So go to  ztw.com and use the code ZTW networkchuck 27. And
  147. 12:32thank you to Threat Locker for sponsoring this  podcast and thank you for taking a coffee break
  148. 12:35with me. Okay. So I I know I didn't prep you for  this um and you can say no, but would you be able
  149. 12:41to walk me through an example of how you approach  a coding problem and with with your skills? Yeah,
  150. 12:47totally. So what you want me to screenshot? Yeah.  Yeah, that'd be great. Yeah. I think that's that's
  151. 12:52the one thing I think most people are missing.  Anyone can jump into Codex now and go make me this
  152. 12:57thing. And Codex will start doing lots of things  using lots of words people have never seen before.
  153. 13:01They assume it knows what it's doing. And then  they spit out some code and you're like that's
  154. 13:05cool, but it's not strategic. And I want people  to kind of feel what that feels. Oh yeah. See
  155. 13:11what that feels like as you do this. Totally. So  essentially my skills have evolved into a process,
  156. 13:18right? And that process is different depending on  the kind of task that I'm trying to do. And uh I
  157. 13:27think to answer your question is how am I thinking  strategically is that the strategic thinking is
  158. 13:32sort of blended into every stage of the process.  It's not just what am I solving today. It's how
  159. 13:37my how is the thing I'm solving going to be um uh  you know going to benefit this project long term.
  160. 13:45So one I'm I'm inside a repo right now which is  I have a sort of course video manager repo which
  161. 13:52is essentially this huge great big project which  is like a video editor. It's an AI writer. It is
  162. 13:59um essentially where I produce all of my videos,  all of my content. And the first place to start is
  163. 14:05by thinking very very carefully about the language  you use when you're working with agents. Now,
  164. 14:12this comes from a software practice called domain  driven design. Right? I've actually literally I've
  165. 14:16been filming this part of my course today. So,  I've got the book next to me. This is Eric Evans
  166. 14:20domain driven design. Absolutely. Um it's a bit of  a tome, right? But is the first couple of chapters
  167. 14:25that talk about the importance of language in  development is so so cool. And so with a couple of
  168. 14:31my skills, I've got a skill called uh grill with  docs, let's say. and grill with docs. What it does
  169. 14:37is it essentially quizzes you, grills you about  the thing that you're about to build. You might
  170. 14:42have heard of my grill me skill, which is like  my sort of most famous skill. Um, which is a sort
  171. 14:47of simpler version of this. So, let's say I go in  with Grill with Docs. I'm going to say, I need to
  172. 14:52improve the teleprompter part of my video editor.  I think it's just gotten a little bit cluttered
  173. 14:57and I think there's some stuff I no longer use in  there. Can you help me out? So, I'm just going to
  174. 15:02give it a vague prompt. Basically, I'm going to  send this off and it's going to do a couple of
  175. 15:05things. It's going to first of all explore the  codebase and figure out what we're even dealing
  176. 15:11with and what we're looking at. And it's going  to look in this glossery here for stuff that
  177. 15:15I've defined in terms of what is actually in my  application and what the entities are. So inside
  178. 15:21here, you can see there's language, there's course  structure, right? So I have a course and then
  179. 15:26we have sections inside courses. We have course  versions, right? because courses change over time
  180. 15:31and I keep track of them. And then lessons are  single learning units within a section. We have
  181. 15:37uh draft versions, whether it has changes, pending  versions, and all of this stuff is language that
  182. 15:42I've developed with the agent over time. And  that pays off because it means, you know, like
  183. 15:47how Opus 5 was just garbage to talk to, right? I  didn't have that much of a problem with it because
  184. 15:53it was using the terms in these this glossery,  right? It was using terms that I'd already agreed
  185. 15:58on. It wasn't that verbose because it was talking  the language of my application already. And so,
  186. 16:05and by the way, are you using Whisper Flow right  now? I'm using Whisper Flow. Yeah. Exactly. Okay.
  187. 16:10Love Whisper Flow. I use it all the time.  Yeah. I'm a huge dictation fan. Like, um,
  188. 16:13if you're not dictating, you're just going way  too slow, basically. So, it's got a sub agent
  189. 16:19sort of like looking in the background and sort of  peeking at stuff. And then what it's doing is it's
  190. 16:24quizzing me about the thing we're about to build,  right? So, this is essentially a grilling session.
  191. 16:28and it's going what does the teleprompter read  from? When you film today, what's on screen? Now,
  192. 16:32this stuff is actually it's probably better if I  just wait for the explore to finish because it's
  193. 16:36kind of asking me stuff that it can figure out in  the code. When do I use it? What's on screen at
  194. 16:41once while you're reading? What clutter have you  noticed? So, it's asking me sort of like vague
  195. 16:45questions that I can provide answers to. And  we're sort of slowly getting aligned here. So,
  196. 16:50this is actually more tactical, right? This is  about how do you get the agent to do what you want
  197. 16:55it to do? How do you align with the agent before  you begin? Because I think what most people do is
  198. 17:00they just prompt the agent. And they don't they  underestimate that there's a communication gap,
  199. 17:07right, between you and the agent. The agent  is just like a developer that you're hiring
  200. 17:12in order to get the work done. And they've been  dropped into this world like the guy from Momento,
  201. 17:17right, with no memory whatsoever. Just totally  blank slate. How the hell am I going to like
  202. 17:22uh figure out what I need? So it immediately dives  into the codebase and asks in in this case asks
  203. 17:27you good questions because usually what they'll  do is they'll just kind of their way through and
  204. 17:33they'll just sort of say oh yeah sure that sounds  good and I'll build something like what you want.
  205. 17:37Whereas if you force it to ask you questions if  you try to get over that communication barrier
  206. 17:42it's much more effective and you're going to  get the result that you want instead of the
  207. 17:46result that you sort of hope for. And I I feel  that often. um I I won't be thorough enough in
  208. 17:53the initial ask and I'll get the result of it.  Actually, just before this meeting, I I have I
  209. 17:59built a a tool to help me do interviews and keep  track of my documents and I noticed the feature. I
  210. 18:05didn't ask it to build. It's just there. And I'm  like, I kind of like that, but it wasn't in the
  211. 18:08original design. It just made the choice for me,  which I'm like, uh I'm not sure if I like that it
  212. 18:14did that. [laughter] Exactly. Yeah. And you have  this sort of when you're communicating something,
  213. 18:19you have this like you kind of know some parts of  what you want, but maybe you don't know everything
  214. 18:24you want, right? And maybe it feels clear to you,  but it probably isn't that clear to an agent,
  215. 18:30right? I've been a teacher for uh since I was 21  years old. So, well, that that's 14 years I've
  216. 18:37been a teacher. I'm quite good at knowing when  someone understands what I'm saying or not. Right?
  217. 18:44But even me, I will try to prompt something. I  will get complacent and it will just build me
  218. 18:50something out of scope or like won't build the  thing that I wanted. You know what I mean? So
  219. 18:54there's always that communication gap and that's  not going to go away if the model gets smarter,
  220. 18:58right? That's just always there because it's  the human that's the in the wrong essentially.
  221. 19:02So what we've got now is we've got our grillings  come back and it's quizzing me on all of these
  222. 19:07sources and it's saying, "Okay, it's using my  language." So we've got beats and scripts and
  223. 19:13animatics. Does the script still autoscroll? Is  the status area at the top end space etc. So this
  224. 19:19is kind of like to answer your original question.  This is the tactical stuff coming through. We are
  225. 19:24fixing the problem and it's benefiting from this  nice glossery that I've been building and working
  226. 19:29on as I've been working on this app over the last  6 months. So when you say grill with docs because
  227. 19:35I'm very familiar with grill me. I use the grill  me skill. I'm not familiar with grill with docs.
  228. 19:39So you're saying the docs part is this document  that you've created or is there more docs that
  229. 19:44we're not seeing? I'll show you. So this is my  skills repo and if we take a look at the grill
  230. 19:51me skill. So if we take a look at the grill me  skill, it's essentially very short now. It's just
  231. 19:58call the skill tool with grilling. So it's going  to find you this grilling skill and call it. And
  232. 20:04this is the grilling part of the grill with docs  and grill me. Now what grill with docs looks like
  233. 20:13is it looks like this. It says use grilling  but also use this second skill called domain
  234. 20:18modeling. And domain modeling is this extra  docs. It's this glossery and also this extra
  235. 20:24little line about architectural decision records  as well which is something I've added in here.
  236. 20:29ADRs what they do is they essentially document the  stuff that you can't document in code. You know,
  237. 20:36code is like the current view of all of the  decisions that went into building this product,
  238. 20:42right? But it's only the present. It doesn't  tell you about the past. And if you don't have
  239. 20:47like a small record about the past, then  the agent will just continually make the
  240. 20:51same mistakes over and over. And so these  ADRs, this little list of markdown files,
  241. 20:56essentially they're quite small. I don't know  if I've got any in in this repo. Docs ADR. Yeah,
  242. 21:02here we go. So these ones here. Uh yeah, I've  got what 31 ADRs just knocking about in the repo.
  243. 21:09And these essentially just track the important  decisions I've made that are hard to reverse,
  244. 21:14expensive to take back. And it means that the  agent just has a lot richer context when it goes
  245. 21:19to build features and grills me. Okay. So when  you're grilling or when it's grilling you, it will
  246. 21:24go through those it's instructed to go through  those ADRs. It'll go through your glossery. Um,
  247. 21:29is that something that you had to like within that  skill you you have it written in there go to the
  248. 21:34ADRs, go to the glossery. Is that built into the  skill or is that something you added custom for
  249. 21:38you and this one repo? Um, this is built into the  skill and it's also it's basically in every skill
  250. 21:44I have it says reference the glossery. Use the  glossery when you're building specifications. Use
  251. 21:49it when you're building modules too. So the code  will actually be using the same language as when
  252. 21:54I speak about the code, right? That makes the repo  incredibly easy to navigate because everything it
  253. 22:00just pulls up everything about a course version  and the database is aligned. The uh uh codebase,
  254. 22:05the front-end codebase is aligned, everything just  reads properly. And it's amazing how many repos
  255. 22:12you'll see and that I've worked with in my career  where the front end will call, let's say we have a
  256. 22:18refund or something, the front end will call it  a reversal. the email that sends will call it a
  257. 22:24pending chargeback and the code will say something  completely different, right? Code will call it a
  258. 22:30refund in the database. This sort of language, it  just makes the app so much harder to understand,
  259. 22:35so much harder to work on for you and for the  agent. And I think we underestimate again these
  260. 22:41communication barriers between you and the agent.  And if you use a shared language, it just feels
  261. 22:46like the agent is in your brain. You know what I  mean? It's just a totally different experience.
  262. 22:50So when you let's say you initially started  this project and you're starting from scratch,
  263. 22:56are you by hand typing out or are you dictating  the glossery and telling it what to make or are
  264. 23:02you kind of just giving the agent a general idea  and letting it create the glossery for you. So the
  265. 23:09cool thing about this is you really like if you're  thinking, okay, I want to use this on my project,
  266. 23:14but I don't I I don't want to have to go through  the whole rigma roll of having to set all this up,
  267. 23:18right? That sounds like a pain in the ass. What  you can do is you can just start grilling and it
  268. 23:23will mint the terms and quiz you about the right  terms as you go. And that way you can just sort of
  269. 23:29slowly build it outwards just focused on the hot  spots that you're actually working on. And then
  270. 23:34it will say, "Okay, would you like to update the  code so that it matches the glossery?" And slowly
  271. 23:38slowly it's like you're peeling like a little  window of sanity in your vibe coded slot base,
  272. 23:43right? It's just beautiful. And like as as you  do it, it just feels like magic because you just
  273. 23:49realize, God, I understand what's happening in  my project now. Everything reads as if it's like
  274. 23:55as if I wrote it. So, it's hard to describe  without actually walking through and doing it
  275. 24:00and seeing how this benefits you long term,  but it's the first strategic thing that I'd
  276. 24:04recommend folks do. So, for anyone just like  watching this now, and just like by the way,
  277. 24:09what you said, I'm sure someone's like, "Oh, that  sounds awesome. and that's what I need right now
  278. 24:13for this mess I'm coding. What's the quickest way  to apply that? Is it just to go out to your skill
  279. 24:20repo and just say install or use it? Point your  agent in that direction. What's the best way to
  280. 24:25do it? Yep. You just go to aihero uh.dev/skills  and I've got the um essentially a full set of uh
  281. 24:33documentation on all of these skills. And for your  next feature, just use grill with docs instead of
  282. 24:38uh let me just get the screenshot so it's open.  What was that? What [laughter] was that? I think
  283. 24:45that's tracking how much time I spend on Twitter.  That's what that was. And it's Guess what, guys?
  284. 24:50The uh the verdict is not good. Um so this is  Grill with Docs. So essentially, you just install
  285. 24:56the skill and you get started with the skill.  Okay. Okay. That's that's pretty simple. Okay.
  286. 25:02That's awesome. Um and by the way, I had the same  problem with Twitter. I had to um uninstall it
  287. 25:08from my phone and I'm now coding a thing that will  keep me off Twitter but keep me up to date. So,
  288. 25:13it's a thing. Anyways, it's a bad bad uh bad  addiction. So, once we've once we've grilled and
  289. 25:21we get to that point where the the AI understands  what we want and it understands the feature,
  290. 25:26what's next? So, you've really got two choices. It  depends if the thing you're building is massive.
  291. 25:33If the thing you're building is uh relatively  small and you think it can fit inside the same
  292. 25:39context window as in the grilling, then you go to  the implement skill. And the implement skill is
  293. 25:45essentially you turn the idea that you've just  grilled about into code. Now when I talk about
  294. 25:51context window, I'm sort of talking about I'm not  really talking about the 1 million token context
  295. 25:56window that Opus has. I consider there to be a  smart zone and a dumb zone inside these context
  296. 26:03windows which is as you add tokens to the context  window. The model makes uh dumber and dumber
  297. 26:10decisions sort of happens slowly as essentially  you add tokens and the relationships between the
  298. 26:17tokens increase quadratically. you get this sort  of crazy attention degradation where it's having
  299. 26:22to hold all this information in its head but  also listen to the most important stuff in a
  300. 26:27room that's getting full of people right it's  very very difficult task and I don't models are
  301. 26:34getting better at it but they're still not great  at it so I get start getting nervous around 150k
  302. 26:39tokens right really that's early very early right  because I I want the good stuff right I want my
  303. 26:45models making smart decisions and I don't want  them like if I like ask a follow-up question,
  304. 26:51I don't want to be paying again for the um you  know 300,000 tokens that I've got in my context
  305. 26:57window, right? I really like optimizing for small  context windows. It keeps things very nimble,
  306. 27:02keeps things very fast as well because actually  latency increases the more tokens you add,
  307. 27:07you know. So for me, it just feels like a general  principle of health keeping my agent in the smart
  308. 27:12zone and keeping it very very early. Now the  actual number I think doesn't matter. It's more
  309. 27:17about the instinct of context paranoia of keeping  things small, keeping the agent uh smart. And so
  310. 27:24if you feel like your thing you're building is  going to be bigger than 150,000 tokens, then you
  311. 27:32create a spec. So you take the conversation that  you've just had in Grillwood docs and you turn
  312. 27:37it into a specification. And this specification  lives either in a markdown file or it lives in
  313. 27:42your GitHub repo as an issue. And then that spec  sort of defines where you're going, right? And
  314. 27:48from there once you've figured out where you're  going and you can kind of go, okay, I can check
  315. 27:52the final product against this specification,  then you turn that spec into tickets. And those
  316. 27:59tickets are the thing that the agent is going to  walk through and actually implement one by one.
  317. 28:04So each ticket is going to be a session. So each  ticket gets implements on it and then you can use
  318. 28:11my code review to do a sort of adversarial review  on the thing that was built to check that it
  319. 28:15matches up against that spec. So that is the whole  flow, right? So you've got two choices. Either you
  320. 28:21go down this spec and tickets route or you just  implement the thing that you grilled. Okay. Um I
  321. 28:28have a few questions on context. How do you what  what are your tactics to manage like sometimes
  322. 28:36when you when you're doing a grill me I can  imagine is if you have a big code base it could
  323. 28:41fill up that context window very quickly. So how  do you have sub agents doing I mean we saw a bit
  324. 28:48the sub agent was doing explorations are using sub  agents to do a lot of that to manage the context.
  325. 28:52So yes, I think I think there there so there's  like four things there, right? The first thing
  326. 28:59is I think people equate big code bases as  big code bases always use more context, right,
  327. 29:05to explore. I don't think that's true. I think  that hard to navigate, poorly organized code bases
  328. 29:12are harder to explore. And you can have a small  codebase that's actually hard to navigate, right?
  329. 29:17And you can have large code bases that are easy  to navigate. And so we can talk later about how
  330. 29:22to improve navigability of your codebase because  that's again something that we've got here. Well,
  331. 29:26you said there were four things that you do to  manage the or some tactics to manage context. Yes.
  332. 29:32And I can't remember what they were. [laughter]  Oh no, you compacted your context. Oh no, I've
  333. 29:39compacted at the wrong [laughter] moment. Damn it.  Damn it. What is going on with me? Um I've been
  334. 29:46filming all day and my brain is mushed. That's  That's why I completely understand that. That is,
  335. 29:50by the way, thank you for taking the time to do  this interview after filming all day. I know that
  336. 29:54you spend all your tokens in that those moments.  Um, now I did have another question on that. So,
  337. 30:00when you're doing Grill Me, uh, which agent are  you using? Are you using Fable? Are you using
  338. 30:05Opus 5.5? What do you trust with that? Yeah, I'm  using Opus 5.5. And I think I remember one of the
  339. 30:13points I wanted to make which was I think there's  an epidemic of people using too high an effort
  340. 30:19right I think most folks are using way too high  effort either like high x high or max whatever
  341. 30:26and what that means is when they do an exploration  the agent like if you think about effort an effort
  342. 30:33like producing um let's say a painting right or a  animation you know you get opus 5.5 five and you
  343. 30:40maybe want it on like a higher effort because  it's going to put more effort in and produce a
  344. 30:45better painting or produce a better animation. But  if you think about the task of exploration like
  345. 30:51like is is more effort going to get you a better  exploration, you know what I mean? Sure, it might
  346. 30:56sort of look in more files, but does more files  it really just needs to read the right files,
  347. 31:01you know, like and find the right information  and give you that information back. So I think
  348. 31:06people are using way too high efforts to explore  their code bases when mostly what I just use for
  349. 31:11pretty much everything is Opus 5.5 median, right?  That's just really Yeah. So for me, I'm like and
  350. 31:17I I think there's still massive confusion on this  because I think effort well I want my AI model to
  351. 31:22do his best every time. I that's that's what what  I would expect from a human like do your best
  352. 31:28always explore that with 110%. Um so I keep mine  on extra high. Yeah. And I think I think that is
  353. 31:35a big mistake because what you're doing there is  you're just telling it to spend tokens on tasks
  354. 31:39where it doesn't need to spend tokens and so you  end up getting into the dumb zone faster because
  355. 31:44it's just and and it's going to perform worse  because it spent all of its effort on too early
  356. 31:49a stage. So yeah, Opus 5.5 medium is what I'm on  and I'm even thinking of dropping that right to
  357. 31:55low genuinely. Wow. So, are there any moments when  you're specking something out that you feel the
  358. 32:02need to go to a higher reasoning or higher model?  Not particularly. I think maybe research tasks are
  359. 32:07interesting one, especially when it's the web,  right? Because the web is very hard to navigate.
  360. 32:12It's hard to find information in. And so, I  think probably high XI is a good example of that.
  361. 32:18when I'm doing performance testing, I think, or  maybe debugging where it needs to try lots and
  362. 32:24lots of different things, I don't want it to stop  early, that's when I might go to a higher effort
  363. 32:29level. But in practice, usually I just default  to medium bis because I I'm thinking about other
  364. 32:34things and it's not something I'm constantly  twiddling or tweaking. Gosh, that is crazy. And
  365. 32:40I'm I'm probably going to switch to medium after  this just to see how it feels. Um, and I imagine
  366. 32:45it's a lot faster and more token efficient. That  would be that would be fun. So Opus 5.5 is rapid
  367. 32:51as well. Like it's fast anyway. And so when  you reduce the it's it's a very good model,
  368. 32:56guys. It's a really good model. Do you switch back  and forth between uh Anthropic and OpenAI? I've
  369. 33:02got an anthropic subscription and I I just haven't  really picked up a codeex model or like I haven't
  370. 33:09had a fiddle around with it. I tend not to minmax  between models that much. I tend to think like I I
  371. 33:16feel I think this is a little bit sort of I I like  holding contrary opinions, right? It's it's fun to
  372. 33:23hold an opinion that most people don't hold. I  tend to think that if you keep the model static
  373. 33:27and you stop the room around you from burning  down, you're going to get much better results. In
  374. 33:33other words, if you fix the environment your agent  operates in, there's often so much lowhanging
  375. 33:37fruit there that people just underestimate  basically and they think, "Ah, the new model
  376. 33:42will fix all my problems, right? Or if I crank  it up another effort level, then I'll get better
  377. 33:46results." And sometimes you will, but you'll  probably just burn a ton of tokens there. I think
  378. 33:50instead of waiting for the next model to fix all  your problems, why not just fix all your problems?
  379. 33:55That's my take. Interesting. Okay. saying you're  saying that going through a method like you have a
  380. 34:01methodology with your skills and getting your code  base organized that will probably fix the problems
  381. 34:06that you're hoping a new agent that's smart enough  to just do that for you would do. Yeah, exactly.
  382. 34:11And we can touch on a skill that does specifically  that if you like. Yeah. And if you want to keep
  383. 34:17keep on context. Yeah, let's let's stay on context  before we um start a new session. We're just using
  384. 34:22a bunch of AI stuff. It's I'm so AI pilled. That's  how I think now. I'm like, hold on, we start a new
  385. 34:28session because this context is already full. And  it's so funny how humans are kind of like that.
  386. 34:32Um, so we we've done we we can take a large idea.  If it's a large idea, it'll put it into tickets,
  387. 34:38which I imagine at that point you're having one  orchestrator agent divvy that up between sub
  388. 34:44aents and they're working that in like work trees.  Is that is that what's happening? Yeah, exactly.
  389. 34:48Um there is a skill here called implement spec  which basically you pass in the spec that you've
  390. 34:53just created in a new session and it goes and  implements it all with parallel sub aents doing
  391. 34:58it in work trees merging all back together until  you get one mega PR ready for you to implement or
  392. 35:03ready for you to review. Wow. Okay. That's I like  that. And I just can't get over the fact that you
  393. 35:09um you don't AI max [laughter] because like I do  you only have one anthropic subscription because
  394. 35:16that would surprise me. Just one. That is  crazy. 200x. Yep. My goodness. Um, well,
  395. 35:23I don't I'm not token efficient, so I have three  of those bad boys. And I also have probably a
  396. 35:28couple uh codeex ones, too. But I'm just always  testing stuff. But the way you approach this,
  397. 35:34how simple. Well, it's it's not exactly  simple, but it's just very straightforward.
  398. 35:38It's refreshing. Yeah. I'm I think that again it's  this sort of slightly wanting to have a different
  399. 35:46view than everyone else. I was just sort of  surprised by how well this works and by how
  400. 35:52token efficient it is actually. Um I think a lot  of people would say it's not token efficient you
  401. 35:57know like uh sure implementing sub aent like  people have different appetites for how many
  402. 36:01tokens they want to spend and that's fine. Um, but  yeah, I think just keeping it relatively simple,
  403. 36:08having a clear coherent philosophy about the way  you want to use AI is important. And I find the
  404. 36:14smart zone dumb zone thing has been very helpful  in clarifying my decision-m and understanding. And
  405. 36:19this is not my idea, by the way. This is Dex Hy um  came up with this who you should have on actually.
  406. 36:24He's very very clever with this stuff. And um  yeah, just having that as like a guiding principle
  407. 36:30has been really nice just to clarify my thinking  about how to use it. Now, with um with creating
  408. 36:38those tickets, I'm sure the agent is smart enough  to divvy up those tickets to where they're small
  409. 36:43enough that the sub agent itself won't get into  the dumb zone. Correct. Yes. They're not amazing
  410. 36:49at it, but yeah, they do a good enough job. Um,  and I think of it as having like if you imagine
  411. 36:54each ticket should be around 150,000 tokens, then  that's about right. Okay. How do you keep track of
  412. 37:02the tickets that's created? I guess it creates it  inside the repo, right? So, it creates worked on
  413. 37:08just as GitHub issues basically. So, you can just  go on the [clears throat] GitHub issues. They're
  414. 37:11tracked as sub issues underneath each spec.  And so, then you can just see the spec being
  415. 37:16um completed as it's being worked through. Okay.  So then after after it all completes and the
  416. 37:22work trees are finished and it's it's merging all  the code, what's next? Yeah. So I do two things.
  417. 37:28So what we do is we go on a code review. So the  implement spec actually calls code review on your
  418. 37:35um the code that it's done. And here two things  happen. So it does a standards review and a spec
  419. 37:42review. So spec review is it compares the code  against the original spec and it checks whether
  420. 37:48it did the right thing right did it build the  thing that I wanted or has it gone over scope
  421. 37:53or under scope and then with standards it reads  the standards files that you have in your repo so
  422. 37:59let's say you have a a codingstandards.mmd file  and I can pull one up for you if you like these
  423. 38:05coding standards.mmd this is full of information  on how I want my applications to be built right
  424. 38:11so I use effect which is a very cool TypeScript  library and it has a ton of stuff in here for
  425. 38:18all the information I want to review. So every  any in Typescript is a leak. Uh specific stuff
  426. 38:24about course versions, you know, this is really  dense stuff. I think a lot of people would have
  427. 38:30this in their agents.mmd file, right? I imagine  maybe is this something you do as well? You know
  428. 38:38what's funny is I don't even pay attention to  that. I trust the agent with it. Yep. Right.
  429. 38:45You just think, okay, I'm not worrying about  these steering files or not worrying about what
  430. 38:49the agent is reading or thinking about the agent's  environment. Yeah. I'm more like I I I definitely
  431. 38:53tell my agent, I want to have a standard, but I  want you to create the standard for me and I don't
  432. 38:58want to have to look at it. Got you. Yeah. I think  that's um I would probably describe that as vibe
  433. 39:05coding. I think I think that's vibe coding, right?  Because it's not really caring about, you know,
  434. 39:10I want you to be good, but I don't care how you're  good, you know? [laughter] Yeah. It's it's like
  435. 39:15that that meme I or some kind of Tik Tok I saw  a while back. It's just like think faster. Deep
  436. 39:20research. Do it. Do just be smart. Make it happen.  Like it's just making wishes and and um yeah,
  437. 39:26ultra ultra think you know um ultra think ultra  think do it right until you're um you know or
  438. 39:31I'll fire you or something. You know what I mean?  Just like so that is what we're that's the second
  439. 39:36axis that we're comparing on. Is it built correct  according to my coding standards? because I sort
  440. 39:41of assume that when my agents implement something,  they're going to get stuff wrong. And for me,
  441. 39:46that's fine because I've got a code review step  that goes again and does this adversarial review
  442. 39:51and checks it against the standards. And so  once that's done, then I'll go and human review,
  443. 39:58right? And I don't do this for everything. I do it  for some things, for very important work. And if I
  444. 40:03was like being professionally paid to like build  an application, then I would probably do it for
  445. 40:07most things really. So describe to me what the  human review looks like. Is it you actually in
  446. 40:13the code looking at I mean is it line by line?  Is it just like an overview or using an agent
  447. 40:18to help you look at it? Yeah. So what I do is  I have a PR skill which is I think in reference
  448. 40:25skills here which basically builds me a PR body  that does a couple of things. It's this is a nice
  449. 40:31skill by the way because you can just install  this and the agent will call it. It's unlike
  450. 40:35my other skills. has just sort of improved your  PRs and it does a few things. It gives me like a
  451. 40:40sort of visual summary of the thing that's being  built. It gives me evidence that it's working. So,
  452. 40:46it has to take like a screenshot or have like  execution based evidence. And then it tells me
  453. 40:52the merge danger. What is the danger of me merging  this? Whether it's a one-way door or a two-way
  454. 40:57door. So, if it's a one-way door, then it's a door  we can't walk back through, right? we merge it and
  455. 41:02then there's data loss or like you know an email  gets sent out to 100,000 people where if it's a
  456. 41:08two-way door then we can merge it and come back  and then also it's blast radius too so whether
  457. 41:14um whether if it goes wrong it's going to be  a big bad thing or a small thing and so I use
  458. 41:20this merge danger as basically like do I need to  review this because if it's a small blast radius
  459. 41:25and it's a two-way door then I probably don't need  to like review it that hard you know what I mean
  460. 41:29I can probably just like test it in production  So, wherever if it's a one-way door with a big
  461. 41:34blast radius, you better believe I'm reviewing  not necessarily line by line, but I'm looking at
  462. 41:39um I'm trying to break it. So, that's kind of  what I do for human review. And I like human
  463. 41:45review because it's not only a chance to make sure  the stuff I'm shipping is high quality, but I then
  464. 41:52take the stuff that I'm saying in human review and  I pull that back into my coding standards into my
  465. 41:59uh environments that my agents in. Right? So I  start thinking, how can I improve my environment
  466. 42:05so I never have to write this comment again? And  that's what I'm thinking when I'm reviewing. Now
  467. 42:10is that like an automatic thing? So when you're so  let's say you're reviewing it human review you're
  468. 42:16looking at the code what is your method for giving  so I imagine you're not going to fix it yourself
  469. 42:21you'd probably just tell the agent to fix it so  it has that context um is it automatic to take
  470. 42:26what you're saying and add that to your your your  coding standards or do you have to tell it that
  471. 42:31to actually initiate the skill there is a skill  for that so we have a retro skill so this looks
  472. 42:37back on a session looks back on a PR and suggests  improvement ments based on what the agent did,
  473. 42:43right? So, it's another look back on the agent  session, on your review comments, on everything
  474. 42:48that just happened. Let's improve it next time.  M okay. Okay. And I'm sure we'll get to this
  475. 42:54because it's more of an architectural and design  question, but like your coding standards, is that
  476. 43:00just something that you as a software engineer  with that background just know what you want to
  477. 43:05have? And how would someone who doesn't have that  background even go about doing that? Yeah. Um,
  478. 43:11this is this is where the strategic stuff comes  in, right? Because this is I don't want this to
  479. 43:16happen next time. I'm thinking about the long  view. And you need to be able to recognize what
  480. 43:20good code bases and bad code bases look like. And  that's really hard. I like I'm I'm we want to dive
  481. 43:28into that now because we totally can. Um, let's  wait a moment because I and I'm like I'm like yes,
  482. 43:35but also no because I know we we probably want  to keep going through this the different skills
  483. 43:39we have as we're adding features because I I do  have a question like once we once we review it
  484. 43:45and it looks good. Let's say it's a two-way door.  Blast radius is small. It's it's an easy thing
  485. 43:50to deploy. Um, what's next? Do do you deploy it?  Does your agent deploy it? What's your approach? I
  486. 43:58mean what's next is I press merge or I say okay if  there's a bunch of them I just say okay guys merge
  487. 44:04all these five PRs or something and you figure  out the right order you merge them in together
  488. 44:08and then we're kind of done right you know we've  done the loop and now we're back to wanting to um
  489. 44:12ship some more features either do another grilling  session or just like plain ship something. Um,
  490. 44:18so that is the loop. We've covered the entirety  of the main flow where you grill with something,
  491. 44:22you either then to spec and tickets, implement it,  code review, and then retro so that it's better
  492. 44:27next time. So I think that that's probably the  the part where most people stop cuz anyone can
  493. 44:34go Matt PCO's a smart dude. He's got skills that  will make my agent smarter. I'm just going to copy
  494. 44:40paste and go. But the next part we're going to  talk about is the actual thing the thinking that
  495. 44:45needs us to understand architecture design and  a little bit of coding. So walk me through that.
  496. 44:52Okay. So before we think about we like we've got  the tactical stuff kind of done right. We now know
  497. 45:00how day-to-day we're going to be um shipping code.  But how do you then take the long view and think
  498. 45:06how do I um improve this over time and get my  agents to a point where I can trust them? And the
  499. 45:12point of this is that uh Lauren Tan um who goes  by Potato, who works at Grockbot, who's uh kind of
  500. 45:19famous now for shipping like 2,500 PRs a month or  something, she talks about how do you essentially
  501. 45:25scale up your trust in your agents? Because the  more you can trust your agents, you sort of walk
  502. 45:32up this trust ladder. And if you don't really  trust your agents, then you're never going to
  503. 45:37be able to scale beyond like four agents at once,  you know, because you're [clears throat] going to
  504. 45:41be watching them everything they do, checking  all of their work, human reviewing every part
  505. 45:45of it. And even if you walk through this whole  flow, you're still going to be doing that, right?
  506. 45:49So her question is, if you can go up the trust  ladder and you can trust maybe some PRs to be
  507. 45:56merged without human review, you know, the two-way  doors, how are you going to get your agents to
  508. 46:01show you evidence that they've done their work?  How are you going to constrain them so that they
  509. 46:06do better work every time? And if you've got  essentially a really great environment for those
  510. 46:12agents to operate in, then you can start trusting  them to levels where you can have hundreds of
  511. 46:16agents operating at the same time comfortably.  and trust that they're doing good work. And so
  512. 46:22all of this is about trust. And what you need is  a codebase that's a good codebase. And if you read
  513. 46:30like the classics that I've got on my shelf, they  tend to converge around one definition of a good
  514. 46:35codebase, which is a codebase that's easy to make  changes in. A codebase where if you make a change,
  515. 46:41it's probably not going to screw anything up.  And that's really hard to do actually because
  516. 46:47or actually I don't know if it's necessarily  hard to do but you have to care about that in
  517. 46:51order to make your codebase easy to change. And so  what that means for me is that you need a ton of
  518. 46:59feedback loops so that the agent can grade its  own work and can verify that what it has done
  519. 47:05is good. And you asked me before about effort.  I would say that the point where I really want
  520. 47:11a high effort level is when it's verifying its  own work. When it's checking that what it's done
  521. 47:16actually functions and it's clicking around a real  application where it's running like performance
  522. 47:23traces on its code to make sure it hasn't degraded  anything, that's where I want it because if I can
  523. 47:28trust that the agent has done that work and has  verified its own stuff and that the codebase is
  524. 47:32easy to make changes in, then that is stuff I can  just like merge without thinking about it. I can
  525. 47:37just go, "Oh yeah, that looks good. merge it.  You know, I don't have to worry about the code.
  526. 47:42Are you there? Like, do you trust your agents  right now? I'm in an environment where I can,
  527. 47:48right? I can absolutely because I'm not  working on a medical application. I'm not
  528. 47:53working on a finance application. My domain is I'm  building personal software for myself, right? So,
  529. 47:58I have been in those situations before. I've  worked on, you know, banking finance applications.
  530. 48:02And in those situations, I would not or or I would  be looking very very carefully at at certainly the
  531. 48:11one-way doors, right? I would probably be asking  my agents to verify everything from a security
  532. 48:16point of view really heavily as well. I would be  burning a lot of tokens on verification so that
  533. 48:21I could get to the point where I trusted them.  [clears throat] So, what are some examples of
  534. 48:27feedback loops you have right now that build your  trust in your agents? Great question. So there are
  535. 48:33a ton of them. And by the way, um I have this  really cool resource. It's called the AI coding
  536. 48:38dictionary. I've not really talked about it very  much. And it's essentially every term that I use
  537. 48:42is somewhere on this graph, right? So we have uh  model providers, context windows, and all of them
  538. 48:49have a definition associated with them. We have  uh somewhere in here will be automated checks.
  539. 48:57And so automated checks are essentially what I'm  about to talk about which is strict typing um you
  540. 49:04know a fast test suite and a bunch of other stuff.  So the main automated and this is this is free by
  541. 49:10the way you just you know access it there's a  GitHub repo if you want to point your agents
  542. 49:12at it. Very useful for just like this is kind of  like the glossery that I showed you earlier but
  543. 49:18applied to AI coding right this is where I got  obsessed with it. So all of my resources use the
  544. 49:23same language. Now, automated checks essentially  you want strong types. So, you want to be using
  545. 49:29things like TypeScript and not JavaScript because  types tell your agent um very very quickly that
  546. 49:36it's made a mistake. Then you want tests. So,  you want actual runtime tests that verify that
  547. 49:43it's doing what it thinks it should be doing  at runtime and that you don't break anything.
  548. 49:48You want things like llinters as well that uh for  those who don't know a llinter is essentially a
  549. 49:55program that checks your code to make sure that  it looks right that you can do all sorts of things
  550. 50:00with it like check that you're using the right  functions check that you're importing from the
  551. 50:03right modules that you're not using bad patterns  that the authors of the framework know are bad and
  552. 50:09are trying to protect you against. So llinters  are very important too and what you want is to
  553. 50:15get the most value out of those you can. So you  want to use like types really effectively. This
  554. 50:21is why I have let's say a this little rule  inside my coding standards here says don't
  555. 50:28use the any type in Typescript because I know  as a developer that if you use the any type
  556. 50:34it's actually a cheat. So it's trying to cheat  around the type system and it's going to reduce
  557. 50:38the quality of my type checking further down. So  I am trying to improve those automated checks so
  558. 50:46that they work better and catch more errors next  time. Interesting. Okay. Now these all these
  559. 50:53automated checks the llinter the um the CI the  testing is this something that anyone can easily
  560. 51:00implement? What's the process for that? You just  you just have to care, right? You just tell your
  561. 51:05agent about them. You say, "I want to constrain  my application more. I want you to find harsh lint
  562. 51:11rules." And this is something that retro does  as well. So what retro will do is it will say,
  563. 51:17"Okay, we we found like a comment in um the PR  um and it tries to find ways that it can prevent
  564. 51:24it deterministically with an automated check  next time." So every time it sort of ratchets
  565. 51:29up the power of these automated checks every  time you run retro or find something. And so
  566. 51:35eventually you get to the point where the agent  is so boxed in. They're in their straight jacket,
  567. 51:40right? They literally cannot make a mistake. And  what you worry about then is of course improving
  568. 51:46the speed of those checks to make sure that you  can still ship work. But agents are really good
  569. 51:49at that as well. So all of these individual parts  agents are good at. They're good at finding lint
  570. 51:54rules to constrain themselves. They're good at  um making those tests run really quickly. It's
  571. 52:00just you have to care. You have to tell them to  do that. And if they if they don't know to do it,
  572. 52:05then they won't care about it either. Is that  a skill you have? I'm thinking about building
  573. 52:11a skill in this direction. Actually, I do have  a skill um which we can talk about in a minute
  574. 52:17that sort of prompts the agent to improve its  own environment a bit. But essentially, this is
  575. 52:23retro. That's what retro does is the more often  you run retro, the more sessions you run it on,
  576. 52:28um, the better your codebase is going to be.  Okay. Okay. So, let's say let's say we've got the
  577. 52:37average Joe. They have an idea for an application.  Let's say it's a website that, I don't know,
  578. 52:41makes ordering coffee and tea better and easier.  You can find the cheapest tea and coffee in the
  579. 52:45world. Doesn't matter what the idea is. How  would they approach that idea? So, obviously,
  580. 52:51Cloud Code is your favorite. I think it's it's  very good too. It's one of the best harnesses.
  581. 52:55So cloud code. Um would you say they need  something like VS Code or can they just stay in
  582. 52:59the terminal? What's your what's your preference  on that? I would recommend VS Code. It's nice to
  583. 53:05be able to easily access the code. I don't always  use it to like read code or think about code.
  584. 53:10I would say that what what my stack is and it's  very approachable like it's free is GitHub is VS
  585. 53:18Code is Git obviously and Claude code or whatever  agent you like it doesn't really matter what the
  586. 53:24agent is to be honest as long as it can use skills  and it can use um uh my skills u then you're good
  587. 53:30to go and pretty much every harness out there is  you know can use skills because they're a common
  588. 53:37standard. Okay. So, doesn't matter what the what  the harness is really. Um, and I think most of
  589. 53:42them use sub agents. Like everything's kind of  the same now. They've all copied each other. Uh,
  590. 53:46so once they have that set up, what are the  things they have? So, when we get to the
  591. 53:52strategic thinking, when they're building their  app, obviously they'll they'll start talking with
  592. 53:56their AI. Do you have skills that ensure that they  do that? Is that what the grill me is for? Like,
  593. 54:02is that does that ensure they have that strategic  thinking they need, or is there something more
  594. 54:06they need to know? So it is baked into different  parts of the process. So let's say when you call
  595. 54:12to spec, it's going to ask you sort of what shape  of modules you want, which might be a little too
  596. 54:18technical for some people, but just sort of keeps  you in the code a little bit. I don't necessarily
  597. 54:23think you need like a like full awareness  all the time of everything that's happening,
  598. 54:29right? You can people talk about like software  factories and they talk about a dark factory,
  599. 54:35right? which is kind of like, you know, you have  your factory, your agents are working in there,
  600. 54:38and you never turn the lights on, right? You  have no idea what's going on. I don't like the
  601. 54:41dark factory. I do think you need to look in there  and care, but you can have little dimmer switches,
  602. 54:47right, for different rooms and stuff. There are  rooms in your codebase that you never really need
  603. 54:51to walk into, right? You can just sort of have  a little peek in there. And I think one thing we
  604. 54:57should talk about before we like go here is that  there is another skill that I think really helps.
  605. 55:03So, I sort of talked earlier about um wanting  to get the most out of your automated checks.
  606. 55:09Mhm. And [clears throat] when you one thing that I  think people really underestimate is that when you
  607. 55:15write tests for your application, so specifically  the things that test your application at runtime
  608. 55:21that actually run parts of your code and like  verify that they work. Tests are really hard.
  609. 55:26They're really hard to get right. And this  is this has been true, you know, before AI.
  610. 55:31The reason tests are hard is it's really easy  to write a test that really doesn't verify very
  611. 55:36much. It can like, you know, you look at like a  line of code and it just says um the the amount
  612. 55:45of characters you can have in a tweet is 240 or  something. I can't remember what the actual number
  613. 55:49is, but let's say 240. And then you'll have a test  that just checks that number and checks that that
  614. 55:56number is 240. Right? The test is just repeating  the implementation. It's not actually like what
  615. 56:02you want from your tests is you want a big chunk  of code that I'm going to call a module and you
  616. 56:07want that module to basically be hidden from the  tests. The stuff inside the module is hidden from
  617. 56:12the tests because that means it can just test it  via a very narrow little um sort of entry point
  618. 56:19to that module. And it means that you can change  everything inside that module without needing to
  619. 56:24alter the tests. [snorts] Because if I like have a  single line of code that's tied to a single test,
  620. 56:29then every time I change that line of code  or I restructure it or do something with it,
  621. 56:33then that test will have to change as well.  And it's just sort of endless crappy churn.
  622. 56:37Whereas if I have these big what I call deep,  what what John Asterout calls deep modules,
  623. 56:42then you can actually sort of turn the lights  off inside that module. You don't have to care
  624. 56:47about what's in there. As long as the tests happen  from outside and everything sort of seems to work,
  625. 56:52then it's probably going to be fine. And I have  a skill that can basically turn your codebase
  626. 56:59into this shape. It's called improve codebase  architecture. So you can fire this off at any
  627. 57:05repo and it will give you a visual report on  the bad stuff that it finds and it will suggest
  628. 57:13improvements for making it better and then you can  take the result of that and turn that into a spare
  629. 57:18tickets and run it through the normal loop.  Ah, so using this skill, it will it obviously
  630. 57:23will show you the the holes and the errors and  all the bad things inside your your codebase,
  631. 57:27but it'll also help you build those those tests.  And are we talking about unit tests? Is that the
  632. 57:32same thing? Yeah, unit tests. Exactly. Okay.  Which I I as I was learning coding before AI,
  633. 57:37I never got to that point, but I heard  it was always the worst part of coding,
  634. 57:41just creating really good tests. Totally.  It's really really hard. But agents agents
  635. 57:46make it easy to type out tests, but they don't  necessarily think about structuring the codebase
  636. 57:51so that it's easy to change, right? So that those  tests actually catch things that are important.
  637. 57:57It's very easy to write tiny tiny little tests  when what you need are tests that actually like
  638. 58:03take out huge chunks of the application and put it  all under test at once. So that's what this skill
  639. 58:08does. Okay. Okay. And that I think that would  be a massive help for a lot of people. I mean,
  640. 58:14I'm going to run this skill on a few of my code  bases right after this interview just to see what
  641. 58:18happens. Um, okay. So, we we've we can produce  good tests and that's a good place for us to go to
  642. 58:26um to make sure our code base is is healthy and  and it's um it's and you said it was we could be
  643. 58:33able to dim the the lights or turn lights off on  a module as long as the tests are really good,
  644. 58:38right? Yep, that's right. you have like  this feeling of um I don't know when when
  645. 58:45you work on a codebase that's been vibe coded I  always feel this terror right this I don't know
  646. 58:51what's in there all right and I think it's bad  and what you start to feel is right how will I
  647. 59:00know if something goes wrong right will I have to  eventually like look at this code and review this
  648. 59:03code like how am I going to get to that whereas  if you have this if you have improved codebase
  649. 59:09architecture and your codebase starting to be  put into these big beautiful modules. You can go,
  650. 59:14okay, I understand the shape of this and my app  understands how to use these modules as well,
  651. 59:19how to like um leverage them and how to call their  functions and things like this. And you have these
  652. 59:25lovely big chunks of stuff that you don't need  to review that's all using your domain language,
  653. 59:30all using your glossery terms and stuff. So, you  know exactly what's going on in there. and you
  654. 59:34realize, okay, I can start thinking of my app's  codebase in these big chunks, right? In like,
  655. 59:41okay, that's the bit where we do authentication,  that's the bit where we do um transcripts,
  656. 59:46that's the bit where like the whole video editor  interaction lives in one big chunk and it's all
  657. 59:51testable. So, when once you have that, you start  realizing, okay, cool. I don't need to worry that
  658. 59:56much about this codebase because it's structured  in a way where it's more robust. Hey, real quick,
  659. 1:00:01let's take a coffee break, you and me. [snorts]  This coffee break is sponsored by Coveron. Look
  660. 1:00:06at Zach. We talk a lot about AI on this channel.  I mean, everyone's talking a lot about AI. And
  661. 1:00:11while AI has made you pretty good at your job,  you know who else has made pretty good at their
  662. 1:00:14jobs? Scammers. They can clone a voice, write  a perfect fishing email, spin up a fake fake
  663. 1:00:19website in minutes, actually maybe seconds.  And they can do that at scale. And they're not
  664. 1:00:23just coming after you. They're coming after your  family, your parents, your kids, everybody. And
  665. 1:00:26that's where Coveron comes in. Their family plan  watches the dark web for you because that's where
  666. 1:00:30all that stuff ends up. So, they're watching  the dark web and they will alert you when you
  667. 1:00:34or anyone in your family has like a new account  set up. And if something does slip through, you
  668. 1:00:38got real people you can contact for help. They'll  help you clean it up. Plus, up to $2 million for
  669. 1:00:43identity theft and up to 100 grand for online  fraud, cyber extortion, and home title fraud. One
  670. 1:00:48scam can cost you everything and they're moving  fast. Protect yourself right now. Get 71% off this
  671. 1:00:54Black Friday deal. Coveron.com/chuckoveron.  That's coveron.com/chuckcoveron.
  672. 1:01:01Okay. So, do you think we're at a place now  where we can get to the how do you structure
  673. 1:01:05your codebase? How do you structure your project?  Yes. I think like there's no one like genius way
  674. 1:01:15to do it. Essentially, you use these deep modules.  You if you have I what I I really want to actually
  675. 1:01:23just run this and um because it's very very easy  to run, right? because we can just literally go
  676. 1:01:28claude and then I'm just going to run improve  codebase architecture. And what this does is it's
  677. 1:01:34going to um pull in the codebase design skill  which sort of um if you want to just read that
  678. 1:01:39that's got all of my uh like how I like to design  code bases in there and it's finding the hotspots.
  679. 1:01:45So it's going to find my recent commits and look  inside the repo to uh explore that and explore
  680. 1:01:51what might need to be improved. And then we're  gonna get a HTML report back of the things that
  681. 1:01:57it wants to change. So once this is done, I'll be  able to like exactly tell you how to do it. There
  682. 1:02:04is I think what people need to get is there's no  one perfect codebase, right? All you can do is
  683. 1:02:10just improve it over time and get it better over  time. And what I do is I have a cron job that runs
  684. 1:02:16on each of my repos where once a day it will run  improve codebase architecture and it will give
  685. 1:02:21me a top suggestion back and then it will go and  make a pull request based on improving that thing.
  686. 1:02:27And because those are two-way doors then I don't  really need to worry about it. I just know that
  687. 1:02:31it's making decent improvements and slowly pulling  the code into where it needs to be. And so you're
  688. 1:02:37essentially just trying to fight back against  the wave of slop, right? because agents will
  689. 1:02:42make tactical mistakes and strategic mistakes and  you just need to constantly be pushing it to this
  690. 1:02:47position where they use these deep modules.  Uh when you have these cron jobs running,
  691. 1:02:55do you trust it enough just to fix things without  your approval? Yeah, I what I do is I do sampling,
  692. 1:03:03right? I will just sort of check in every five  PRs or so. And this this works for my application
  693. 1:03:08because it's again not finance, not medical. So  I will do sampling. I will just sort of check
  694. 1:03:14that the thing is running. I will occasionally um  retro it. So just say retro look at the last five
  695. 1:03:20of these and see if they're actually reasonable.  See if they make sense. See how we can improve
  696. 1:03:23them next time. And so yeah, basically I do.  Hm. Interesting. Now do you do you use like the
  697. 1:03:31Cloud Code app on your phone? Do you do you do  things remotely? Do you text it? Use Telegram?
  698. 1:03:36What what's your stack for that? So, I have been  like in all sorts of like I've tried all sorts
  699. 1:03:43of setups basically. I've like my recent one is I  liter I literally tore it down today which I built
  700. 1:03:48a uh a custom Discord bot for myself. Kind of like  my own Hermes basically. I put it up in the cloud.
  701. 1:03:53I was like texting from my phone and stuff. I  think phones are cigarettes basically. Phones are
  702. 1:03:57poison. Um so I tried to make my phone as boring  as possible. So I'd sort of go away from that. I
  703. 1:04:01didn't want to do it anymore. So right now, as  of now, I'm just using claude code and using um
  704. 1:04:06GitHub actions for some scheduled jobs as well. So  this is an open source repo. GitHub actions allow
  705. 1:04:12you to run cron jobs and I just run claude inside  the GitHub action and it updates and creates pull
  706. 1:04:18requests for me. And agents know GitHub actions  really well. It's actually free for open source
  707. 1:04:24repositories and so it's a really good option for  me. Okay. I I love what you said that phones are
  708. 1:04:30poison. And what's funny is I I um I have it  to where I can use iMessage for my agent right
  709. 1:04:36now and it's I was just thinking this morning  as I was on my walk. I'm like this is so fun
  710. 1:04:41just being able to text my agent to do whatever I  wanted to do and it it does feel freeing. So I I
  711. 1:04:46find it interesting that you're like the opposite.  You're like no phones are poison. I want to make
  712. 1:04:50my phone as boring as possible. Why? Well, because  I had that too, right? I would have Whisper Flow
  713. 1:04:57running on my phone. I would be chatting to it  on Discord as I sort of went, you know, for my
  714. 1:05:01walk and stuff. Um, but this is more, this is just  a personal thing really. Like I think we'll look
  715. 1:05:07back on this generation and think of it in the way  that we thought about cigarettes in the 50s, you
  716. 1:05:11know, like you know, we're giving, you know, like  phones to 13-year-old kids and stuff, you know,
  717. 1:05:15like it's it's miserable. Like these kids like I  it's just, you know, life is there to be lived,
  718. 1:05:24you I I I I just I don't having this constant  access to information and uh especially the social
  719. 1:05:32side of it as well and this constant hankering of  people wanting to pull you in and scroll basically
  720. 1:05:39just scroll scroll scroll scroll a bit more.  I just hate it. I'm with you there. And I I I
  721. 1:05:45I like the idea of keeping my agents close to  me, being able to text them. But like I said,
  722. 1:05:50like just this past weekend, I deleted Twitter.  I deleted or X I deleted Tik Tok and YouTube and
  723. 1:05:55everything off my phone because it was sucking  that that time away and it's and it's so funny
  724. 1:06:00when you take that stuff away you notice you  kind of have that itch like I'm bored for these
  725. 1:06:053 seconds I have to wait let me go check no I  don't I can't I don't have it and you you feel
  726. 1:06:09that urge start to like it's and it's nasty. I've  even gone further which is I've uh removed the web
  727. 1:06:14browser and I've removed the email from my phone  as well which is apparently something you can do
  728. 1:06:19on Android and so it is as boring as possible.  And what I noticed was I would hope that or I
  729. 1:06:25hoped that when I did that I would get all this  extra inspiration. I'd feel so much more energy.
  730. 1:06:30I would like become this whole new person  and then nothing happened. Right? Essentially
  731. 1:06:36nothing h nothing rushed in to fill that gap.  You know what I mean? It turns out that yeah,
  732. 1:06:40these things are really addictive and you you  know I've been trying to read more books, I guess,
  733. 1:06:46but it nothing is as good or feels as good as  checking Tik Tok, right? Or checking Twitter. So,
  734. 1:06:54you just have to kind of live with that.  [laughter] It's like a you're you're talking like
  735. 1:06:59a former addict or something like I'm always a  drug addict and there's nothing better than that,
  736. 1:07:02but it's ruining my life. [laughter] Absolutely.  Uh so, what is this? Well, this is the report
  737. 1:07:08that we just got back from uh improved codebase  architecture. So, what we're essentially seeing
  738. 1:07:14here is a before and after of each of the um  modules. So, what we have is a problem. Submit and
  739. 1:07:22duplicate course contain 150 near identical lines.  So, there's some duplicated code inside our repo.
  740. 1:07:29And what it's basically proposing is instead of  this horrible sort of setup where we have a bunch
  741. 1:07:34of shallow thin modules on the left hand side,  we say, okay, all of these different parts of
  742. 1:07:39our application are all going to rely on a single  deep module and we can then test that module. And
  743. 1:07:45so this one feels like a like a total win for me.  So this is cool, but honestly I think and I speak
  744. 1:07:53for myself as well as probably my audience and  many people watching. I'm like, um, I see that I
  745. 1:07:59found stuff and it's very interesting, but also  I probably have to ask AI about a lot of these
  746. 1:08:03things. Like, what does that mean? What does that  mean? Why did you show me that? What is this? No,
  747. 1:08:07absolutely. Absolutely. So, this kind of maybe  goes into another section of our our conversation,
  748. 1:08:13but um I'll ask it now and maybe you can tell  me where we should ask or answer it. But for
  749. 1:08:20people who are wanting to get into this world  and what it means to be a software developer,
  750. 1:08:25what is that knowledge that you still need to  learn? What's the thing? Because obviously it's
  751. 1:08:29not it's not tactical coding anymore, but I look  at this and I'm like, I don't know this, so I
  752. 1:08:34probably should learn this. What What is it? So,  there's a skill for that. Um, [laughter] so, uh,
  753. 1:08:42do you have a t-shirt with that? Like, that should  be a t-shirt. [laughter] I have a teach skill,
  754. 1:08:46right? I'm a teacher. I was bored on the tube  one day and I like I thought, why don't I just
  755. 1:08:51write down everything that I know about teaching  um, and turn that into a skill. And what I would
  756. 1:08:56say is that let's say you have improved codebase  architecture. You get a report and you look at
  757. 1:09:00the report and you think, I can't read that. I  don't understand what's going on there. I don't
  758. 1:09:04understand the terms it's using. Um, I've thought  very carefully about the terms that I've I've used
  759. 1:09:09in the skill so that they point out to um books  and they point out to well-known methodologies.
  760. 1:09:15For instance, let me just I've got a big pile of  books next to me. This is one of the books that
  761. 1:09:20I recommend called uh working effectively with  legacy code by Michael Feathers. Um, wonderful
  762. 1:09:26wonderful book. Teaches a lot of this stuff. And  what teach can do is it you basically just say you
  763. 1:09:32pass it the report you've just received. you say,  "I don't understand this. Teach me about this."
  764. 1:09:36And it goes and finds a like a workspace on its  own. It keeps track of what you know and then it
  765. 1:09:43quizzes you and builds these again HTML lessons so  that you learn things slowly. So it's essentially
  766. 1:09:49a build your own course for uh whatever you  want. So it taught me how to um do a Rubik's cube
  767. 1:09:56actually. So this was how I tested it and uh I  can now do a Rubik's cube which is wonderful. Um,
  768. 1:10:02but you can also I've tried it out by teaching  strategic programming as well and it's really
  769. 1:10:06really good at it because my um these reviews here  they're basically like they use a common language.
  770. 1:10:13So they talk about locality. So you probably  have no idea what locality is. A lot of software
  771. 1:10:18developers don't know what locality is. It means  how if I make a change to a piece of code, how
  772. 1:10:25likely am I that all of the changes are going to  be in the same location? In other words, are all
  773. 1:10:29of the related pieces of code together in one file  or in a couple of files? So, locality is good. You
  774. 1:10:35want locality? And yeah, like you just basically  give it this and you say, I want to understand
  775. 1:10:43this. Teach me about it. And then it's going to do  a great job. So, you're saying, which by the way,
  776. 1:10:48that teach me skill is such a fun idea. I love  that. So, you're saying that it's important for
  777. 1:10:54us to know all this. um at what point is it not  important for us to know this? Like are the models
  778. 1:11:01going to get so good to where it doesn't matter?  So it comes down to caring again, right? And all
  779. 1:11:14you need to know is you need to know enough  strategic programming to be dangerous and to
  780. 1:11:18know that it's effective and that improves things  over time, right? because the models are still not
  781. 1:11:24going to care about it, right? I maybe there's a  world in which they are um prompted to do this.
  782. 1:11:31And I I've been thinking about this myself as  well, but for now it seems to me that people who
  783. 1:11:37are really good at something are going to be able  to get models to be really good at that thing,
  784. 1:11:43right? M so engineers tend to produce better um  products than non-engineers right now because
  785. 1:11:51they have the domain expertise they know what to  ask for they have the right language right um you
  786. 1:11:58know you might say uh what's a good example let's  say that you have a problem in your application
  787. 1:12:05where the you have some sort of flashing stuff  coming in when you load the app right it sort
  788. 1:12:10of shifts around in terms of layouts right and I  know what would cause that. I would say um fix the
  789. 1:12:18hydration error on the homepage, right? Or fix the  uh layout shift on the homepage. Whereas someone
  790. 1:12:25else might need to say, okay, the when I load the  layout page or the homepage, um it loads and then
  791. 1:12:31it shifts everything around. It just allows me to  communicate more effectively and prompt the agent
  792. 1:12:36in a better way because I know the domain language  because I'm an expert in it and I've seen that
  793. 1:12:41kind of thing before and I can communicate what  I want much much better. H that does make sense.
  794. 1:12:49Um and I'm sure I'm sure you've seen this term  uh the bitter pill engineering idea. Uh bitter
  795. 1:12:56lesson or bitter lesson. Yeah, I I forget it's  it's then been used interchangeably, but the idea
  796. 1:13:01that all this all the um architecture, all the  the the straight jacketing we're doing um will
  797. 1:13:08be replaced by better, smarter models and we just  need to kind of get out of their way because they
  798. 1:13:13might know better than us how to architecture a  given thing. But you're saying that it's it's the
  799. 1:13:17caring that matters the most. It's the caring,  right? And I always say like I'm not a pundit,
  800. 1:13:23right? I have no idea where things are going. you  know, I'm just in ox like rural Oxfordshire. I'm
  801. 1:13:28not in Silicon Valley. I don't know anything.  Um I'm my job is not to predict the future. My
  802. 1:13:33job is to tell you how to get these things working  now. And they're very effective now. Um but yeah,
  803. 1:13:40I and I think there probably is some sort of  harness, some way of doing this that will make the
  804. 1:13:47agents care, right? Because you can get the agents  to care. It's just you need to care yourself. So,
  805. 1:13:53that's something I've been thinking about  and working on too is how do you get agents
  806. 1:13:56to think strategically? And I think that probably  is something you can do in the harness, just not
  807. 1:14:02um it doesn't have to be in the model. But for  now, this is what I consider the state-of-the-art
  808. 1:14:08way to do this. Okay. [clears throat] And then  once you've I think what a lot of people don't
  809. 1:14:16understand how to do right now is they don't  understand the architecture or rather the
  810. 1:14:22infrastructure like where to actually put this  app they're building um do they use a database
  811. 1:14:28what type of database um do they put it on AWS or  Google cloud like all these things are big things
  812. 1:14:35that normally are done by DevOps engineers or  or software engineers who have that background.
  813. 1:14:40Now, how should someone approach that? Especially  if it's someone who like only has done vibe
  814. 1:14:46coding. Let's say they they're watching your  course and they're coming into it. Do they
  815. 1:14:50need to learn all that stuff or can they just  ask the agent to do it? And do you have skills
  816. 1:14:54that help with that? I don't have skills that  help with that. And I would say we don't really
  817. 1:14:59need skills to help with that. What's a little bit  of a history lesson which is over the last 10 or
  818. 1:15:0420 years DevOps has moved from need like being  something that you needed a degree in to like
  819. 1:15:12figure out how to do you know like uh hosting uh  self-hosting your own Linux box or something and
  820. 1:15:17doing really complicated stuff or stuff that could  really go wrong and really bite you into a set of
  821. 1:15:22products that have been built by Google cloud by  AWS um especially you know serverless platforms
  822. 1:15:27if you want to do some googling that really take a  lot of that stress away from you and has resulted
  823. 1:15:33in products like Versel for instance who I used  to work for and um Netlefi and all this stuff that
  824. 1:15:40basically take all the pain away from you right  and what you really need to do is just figure
  825. 1:15:46out where your data is going to live whether it's  going to be in a database you should probably just
  826. 1:15:51use like a Postgress database if you don't know  what you're doing and once you've got that and
  827. 1:15:55there are many services that will host and manage  your Postgress database for you planet scale
  828. 1:16:00Superbase. Once you've got your data, you can  basically deploy your app anywhere. You know,
  829. 1:16:06just Netleify, Versel, whatever you want,  and it'll be fine. You know, like this has
  830. 1:16:12gotten so much easier that an agent really can  just do it for you and it will be good. So,
  831. 1:16:18as long as you're running some security checks and  you're doing relatively regular security reviews,
  832. 1:16:23which agents are also pretty good at, you'll be  fine. It's a lot easier, just like coding. It's
  833. 1:16:29a lot easier than most people think it is. Yeah.  And I can speak to that. Like I I'm very much a
  834. 1:16:36home labber. I've got Proxmox and all kinds  of things in my environment. But as soon as
  835. 1:16:41I felt comfortable enough to give the keys over  to my agent, I can simply text him and go, "Hey,
  836. 1:16:45I want to deploy this app, make a new VM, make  it happen, just do all the things. Give it an IP
  837. 1:16:50address, give it a domain name, and boom, it just  does it." And it's magic. It feels like magic. But
  838. 1:16:56I did need to have a bit of that domain expertise  to even know how to tell it to do that and to give
  839. 1:17:02it those keys. Um, but you're saying that a lot  of the DevOps things um, so things like databases
  840. 1:17:10and and and Versell hosting now they've abstracted  that away to where you don't really have to know
  841. 1:17:14it. Your agent can do it. Do you is that where  you host your stuff? Is it is it Versel? Yeah,
  842. 1:17:19I'm on Versell. Yeah. M like I would say just  use the thing that's easiest to use basically
  843. 1:17:27especially when you're getting started. I wouldn't  worry about you know like going into AWS manually
  844. 1:17:32or figuring out all that stuff. There's stuff you  can like there's foot guns there right which is
  845. 1:17:36why these companies like Versel exist. So like  they just sort of make that easier and you don't
  846. 1:17:42have to worry about you know manually deploying or  anything. You just push to your GitHub repo and it
  847. 1:17:46deploys for you. Um it really is very very simple.  Um, you can even use something like GitHub pages
  848. 1:17:51or something. And all of this stuff is even when  it comes to like domains and um, uh, DNS records
  849. 1:17:58and things like that. All of that is really easy  for agents to figure out information on. And often
  850. 1:18:03um, these sites like, you know, Cloudflare as  well. Um, you know, again, it's just something you
  851. 1:18:09have to care about. You just basically say to your  agent, make this deploy as good as possible. Um,
  852. 1:18:14give me everything I might need in terms of  security. um uh distributed denial of service
  853. 1:18:19attacks, DOS attacks. Um try to make it secure  and just give me the basic level of what I need.
  854. 1:18:24And agents are really good at that. Time for a  quick coffee break. I want to talk to you about
  855. 1:18:27something. You've probably seen one of these  before. A video of a celebrity or a CEO giving
  856. 1:18:31away some kind of scammy thing. Actually, I just  saw that with the Roadster announcement. You know,
  857. 1:18:36Elon's flying car. I was trying to find  the live stream and I actually just found
  858. 1:18:40a scammer live stream that had a fake Elon  sitting there talking. To me, it looked fake,
  859. 1:18:45but for many people, it would not. And there  were like 30,000 people watching that stream,
  860. 1:18:48and he was pumping some crypto scam, but it looks  stinking real. Voice, face, everything. Like,
  861. 1:18:53I was showing it to my family. They're like, "Oh,  are you sure? Are you sure it's not him?" It's
  862. 1:18:57a deep fake. Scammers are literally using them  all the time. And in a survey of 7,000 people,
  863. 1:19:01that's their biggest concern. Deep fake scams are  so bad now, and one in seven have claimed they've
  864. 1:19:06fallen for one. So, here are some red flags you  should look for. when the lips are a little bit
  865. 1:19:09off from the words. They have some weird blinking  and an offer that's like too urgent or too true.
  866. 1:19:14I mean, come on. Most of these announcements  we see from founders are like pre-order in 3
  867. 1:19:18months. They're not like buy this right now.  So, when you see that, pause, don't click,
  868. 1:19:21don't send any money, don't share it, and then  check it cuz Bit Defender has a free app called
  869. 1:19:27Real Check. You can drop in a link from Tik Tok,  YouTube, Instagram, Facebook X, upload the clip,
  870. 1:19:32and it tells you if the video looks altered and  how likely it is to be a scam. And honestly,
  871. 1:19:36not every AI video is a scam. Real Check will  help you discern which ones are. Now, you may
  872. 1:19:41be thinking, "Well, Chuck, I don't get fooled by  these scams." Well, what about your mom and your
  873. 1:19:45grandma and your dad and your brother and your  sister? Do they know enough to spot these things?
  874. 1:19:48Give them a tool that will help them do it. Run  suspicious clips through Bit Defender Realcheck to
  875. 1:19:52see if it's a deep fake meant to scam you before  you click or send money or share it. Links in the
  876. 1:19:57description. Bitdefend.me/realchecknc.  That's bitdefend.me/realchecknc.
  877. 1:20:04And it's free. So like what do you have to lose?  Just try it. What do you do for security? Um
  878. 1:20:10because I know that that's one thing everyone  actually most people don't think about but
  879. 1:20:14they should is all these AI agents are just making  things very quickly. They want to make it work and
  880. 1:20:19that's the primary objective. But then security  is a whole realm. Do you have skills for security?
  881. 1:20:23No. I'm the wrong person to ask because um no  the project the projects I work on are personal
  882. 1:20:28software essentially and I have a platform team  that handles my site where we might have some
  883. 1:20:32security concerns. Uh so no I don't don't think  about it that much but I know it's a concern.
  884. 1:20:37What I would recommend is if you are interested  in security is to add something to code review so
  885. 1:20:43that it does instead of just um to access instead  of just thinking about standards and spec you also
  886. 1:20:50do a security review and many harnesses have a  built-in security review um skill. So that's what
  887. 1:20:56I recommend. H have you had any security issues  with anything you've coded specifically with? No,
  888. 1:21:06but I haven't had those cons. I I I've definitely  screwed up in many many ways, but I don't think
  889. 1:21:13I've ever uh especially preai, but I don't think  I've ever caused a security issue. I don't think.
  890. 1:21:22That's good. That's good. We can trust you now.  No, I'm just kidding. Um, now with your with your
  891. 1:21:30skills, how do you make sure what what's your  method for making sure the skills are good?
  892. 1:21:36Because, you know, anyone can make a skill, but  how do you test them? How do you make sure they're
  893. 1:21:41they're effective? Yeah. A lot of people ask me  this and especially how do you like eval your
  894. 1:21:46skills, right? So for those folks who never heard  of evals, they're basically benchmarks that you
  895. 1:21:50run against some system by pushing a bunch of data  through the system and then seeing what happens.
  896. 1:21:56So you might eval coding agent by saying okay, is  it better with this model or this model by running
  897. 1:22:02like um a benchmark of something? You you get it  to do the same task every time and you see if it
  898. 1:22:08um succeeds or fails. So that could be a 0%  score or 100% score on the eval. Now skills
  899. 1:22:15are incredibly hard to eval incredibly hard and  incredibly expensive to eval. How would I eval
  900. 1:22:22my like grill with docs or something? You know,  like how would you eval u domain modeling? How
  901. 1:22:28do I know if I've chosen the right term? A lot  of these skills are just sort of augmenting what
  902. 1:22:32the model does anyway and what the harness does  anyway. Trying to kind of get out of the harness's
  903. 1:22:37way and just give it basically trying to get it to  care about different things. How do you eval that?
  904. 1:22:43And if essentially I have the fortune of having  a ton of data going through my skills anyway,
  905. 1:22:51right? Because I have, you know, you know,  millions of users all around the world who are
  906. 1:22:55using the skills and reporting issues back. I've  got very active Discord where people talk about
  907. 1:22:59skills all the time. So I'm sort of running evals  except my evals are people, you know, they're just
  908. 1:23:04So what I do is I listen to them and I react  and I try to improve my skills all the time.
  909. 1:23:10Um, but I don't burn tokens trying to build evals  that might or might not give me good data. So
  910. 1:23:18that's the decision I've chosen. If I was in  a model lab and I was being paid to do this
  911. 1:23:22and I had unlimited tokens, I would build evals.  Okay. And which I think even those eval wouldn't
  912. 1:23:30be as helpful as what you're experiencing now  with people. Exactly. if I didn't like I'm very
  913. 1:23:35grateful to my users because they do report stuff  really really effectively and that's the actual
  914. 1:23:40stuff I need right if I try to synthetically  create that with evals it would be worse so
  915. 1:23:46when they do give you feedback how do you go about  implementing that how do you verify it how do you
  916. 1:23:51test it uh I have an agent that um looks at it and  looks at the issue it tries to recreate that issue
  917. 1:23:58so it runs it in like a dummy repo and then it  says yes this seems good and then and it creates
  918. 1:24:03a PR. So I have that on a schedule running um  every day I think because I get quite high volume
  919. 1:24:09of issues. Um but I also get a ton of crap issues  as well. So that agent also filter stuff out. Uh
  920. 1:24:15did we ever talk about like what happens after the  PR like do you deploy the changes yourself to your
  921. 1:24:21infrastructure or do you let the agent do it? Um I  use a continuous integration platform. So on let's
  922. 1:24:27say Versel for instance, it will just whenever  GitHub detects a change to the default branch, it
  923. 1:24:34will just ping off to Versel and then Versel will  redeploy my application. We'll build it. We'll put
  924. 1:24:40it in the cloud. So that's something that's been  solved for yonsks right before agents. And so it's
  925. 1:24:45just not something I think about. And it's and  it's always deterministic. So it always happens
  926. 1:24:50the same way every time. And I don't burn any  tokens to do it, right? It just does it by itself.
  927. 1:24:56That's yeah and I didn't think about that  like yeah CI/CD has already largely solved
  928. 1:25:00that problem. I think so many people don't  even think about CI/CD. They just think maybe
  929. 1:25:05uh they think about just having their agents log  in and deploy the code itself. Um so you would say
  930. 1:25:10that people should think about CI/CD and implement  that. Oh yeah. I mean if you have a choice between
  931. 1:25:14determinism and non-determinism, why would you  ever pick non-determinism, right? Why would you
  932. 1:25:20do like why bother? Because determinism works  the same way every time, right? your agents
  933. 1:25:25can always go in and fix if things go wrong. It  costs you no tokens and it just costs you CPU
  934. 1:25:30cycles or something, you know, just costs you this  resource which is incredibly cheap. So, always
  935. 1:25:34pick determinism. And this is this is true too  when you're doing like a retro or something when
  936. 1:25:39you think, should I change this into like a rule  where my agent is going to code review against it
  937. 1:25:44or should I change it into a lint rule that is  deterministic and runs the same way every time.
  938. 1:25:50always always choose determinism when you have  a choice and most people don't realize they have
  939. 1:25:55that choice. Yeah, I I don't think they do. And  and even just me recently, I just started doing
  940. 1:26:01some CI/CD. I I um host I self-host Git Git T or  Gideia, I'm not even sure how to say it. Um and I
  941. 1:26:08I use some of their CI and action functions, which  is it's great. Um, now for you right now, I take
  942. 1:26:16it you mainly probably do mostly teaching, but  you do also do some of your own development. Do
  943. 1:26:22you have any other projects that you're constantly  working on that you're maybe going to sell or make
  944. 1:26:26public, make it a SAS? Nah, I'm not a big SAS  guy. Basically, I'm like I'm in this teaching
  945. 1:26:31thing as like a lifestyle choice basically. Um,  I have a young son and I want to spend as much
  946. 1:26:36time with him as possible. Um, being at home  here, like getting to choose my own projects,
  947. 1:26:42choose what I care about and choose my own time is  incredibly important to me. As as you can imagine,
  948. 1:26:48I've been approached by a fair few organizations  saying, "We would love to hire you and come work
  949. 1:26:52for us." But I'm basically unhirable at the  moment. like until my like uh my son and I've
  950. 1:26:58got another one coming uh go to school then like  and I have a maybe a bit more time you know until
  951. 1:27:05that happens I'm going to stick exactly where I am  so I'm I'm not doing the SAS thing but I am a bit
  952. 1:27:10of an open source boy so I have a project called  S castle which is a kind of effort to um do agent
  953. 1:27:18orchestration in Typescript using essentially  allows you to drop in any agent and run them
  954. 1:27:25in sandboxes. So like in Docker sandboxes or uh  Podman or Versell or actually whatever you want.
  955. 1:27:31And then I've got another one which is sort of  falling a bit by the wayside at the moment which
  956. 1:27:35is called Everlight. And Everite is actually  an eval framework amazingly. So I I worked on
  957. 1:27:40this like a year and a bit ago I think. Yeah, it's  really um fallen by the wayside to be honest. But
  958. 1:27:46essentially it's like a a visual eval framework  that you can run locally and so yeah it's um it's
  959. 1:27:52open source for me and kind of that's that's where  the bulk of my work is going and also into this
  960. 1:27:59video editor that I've been building and that I  use for everything. I'm fascinated by that because
  961. 1:28:04I also build lots of video editing workflows for  myself, you know, because we make videos. Um,
  962. 1:28:10we should chat about that in a separate discussion  because I'm sure we can go get very nerdy very
  963. 1:28:14quickly and not many people are going to care  about it. Oh yeah. Um, now with your open source,
  964. 1:28:18I love that you called yourself an open source  boy. And I I think for many people, they would
  965. 1:28:23love to not only start their own open source  projects, but contribute to open source, but they
  966. 1:28:27don't know how. And they never could until really  now. But how do you do that? Well, like could they
  967. 1:28:34take your skills and start to contribute or even  start their own open source project and what would
  968. 1:28:39they need to like what would they need to be aware  of that isn't already like what we haven't already
  969. 1:28:44discussed? Yeah, I mean what I'd say is that this  was a different conversation about 18 months ago
  970. 1:28:51than it is now because now we are living in a  world where it's incredibly easy for anyone to
  971. 1:28:58just ping up a slot PR to a repo, right? to just  say I want this change made here's a slot PR and
  972. 1:29:05very popular projects are now closing their PRs to  new contributors so for instance uh if we look at
  973. 1:29:14uh HonoJS I think they just did this the other  day um Hono is one of the most popular kind of
  974. 1:29:20like server frameworks in JavaScript and you can  no longer create PRs against Hono unless you're
  975. 1:29:26a collaborator right because of just the level  and I've closed uh PRs on my skills for instance
  976. 1:29:31as well because I want my skills to be very  carefully curated and it's a rather popular
  977. 1:29:36repo. It's nearly up to 300,000 stars now and  that is, you know, people just pinging crap
  978. 1:29:40into it all the time and I just didn't want  to be notified about it. So, open source now
  979. 1:29:46is all about providing issues and providing good  evidence, right? Because it's so cheap now to turn
  980. 1:29:52those issues into pull requests and to just get an  agent to work through them and to figure it out.
  981. 1:29:57So the higher quality uh repros that you can do is  great and like that will teach you a lot about how
  982. 1:30:05to communicate to people working open source and  how to communicate to other developers and how to
  983. 1:30:09communicate to AI. But of course you can build  your own thing, right? And the cool thing about
  984. 1:30:16having all of this code available to you is that  your agent can read it. You know your agent can
  985. 1:30:21just like take a look inside this codebase. You  can say for slashte me about the Hono codebase.
  986. 1:30:26What can I learn based on what I've told you  about my experience already? What is cool
  987. 1:30:31about this codebase? What are they getting  right strategically? How are they deploying
  988. 1:30:34their infrastructure? What infrastructure do they  have? What do they need? You know, and all of this
  989. 1:30:39stuff, you should just look at these cool projects  and just point your agent at them and say, "Teach
  990. 1:30:43me about this." And then like again, open source  is a lot um it's a lot simpler than people think
  991. 1:30:49it is, you know? It's really just like deploying  something to uh npm or something, you know. So,
  992. 1:30:56npm you can just use uh like a um a change set  manager to just sort of do versioning and things
  993. 1:31:04like this. Very very easy to learn, very simple  to get um to get right, you can just do it within
  994. 1:31:09GitHub actions, make it nice and secure and you  go and agents know how to do it. So, if you want
  995. 1:31:14to build an open source um piece of software,  you can and agents will enable you to do it.
  996. 1:31:22It's interesting that you you mentioned that  people are closing their PRs to only verify
  997. 1:31:27contributors. Um, do you think that they'll go  even further and just cut off contributors because
  998. 1:31:33you can just have AI do everything for you now?  Why even bother with the messiness of humans? I
  999. 1:31:39think um some projects do need teams behind them.  You know, you still need teams of people like
  1000. 1:31:45making decisions and um some you know, some more  managing triage and contributing and that kind of
  1001. 1:31:50stuff. But I think yeah essentially what you're  saying when you cut off PRs is only these five
  1002. 1:31:5712 two people can contribute to this project using  their agents and we are sort of and I think that's
  1003. 1:32:05sort of always been true like anyone who's done  um any long-term open source will have encountered
  1004. 1:32:13the everyone who's contributed to open source has  met a BDFL a benevolent dictator for which is the
  1005. 1:32:21person in charge of the open source project who  is wants it to succeed but they also only want
  1006. 1:32:27their point of view in the repo and they are the  one who is the the gatekeeper for stuff entering
  1007. 1:32:33it. So contributing to open source has always been  weird, always been political, always been a hard
  1008. 1:32:39thing to do well. Um, it's why I have a huge  amount of respect for people that I know who
  1009. 1:32:44are amazing at it and who contribute to all sorts  of open source projects, but I think it's harder
  1010. 1:32:49to get into than it's ever been. H so what what do  you think is the and actually I think we're going
  1011. 1:32:56to get into some of the last things I wanted to  ask you here but uh I think before the benefit of
  1012. 1:33:04contributing to open source was that it would show  it would be a free place to uh get experience and
  1013. 1:33:09show off what you can do in order to help you  land a job and actually make money from that
  1014. 1:33:13because obviously open source there's no money  in that. Is that still valuable to to be able
  1015. 1:33:18to show off what you can do now that it's mostly  AI doing that? AI has eaten tactical programming,
  1016. 1:33:24but it can't do strategic programming. So,  you need to show evidence that you can program
  1017. 1:33:28strategically. You need to show evidence that you  can um manipulate infrastructure, that you can
  1018. 1:33:34build a beautiful repository, right? That you  can improve your agents environment, that you
  1019. 1:33:39can steer it properly. And I think that strategic  programming skill is actually more valuable than
  1020. 1:33:45it's ever been. You know, I I genuinely think that  I think it's like because now you have a fleet of
  1021. 1:33:51agents, tactical agents that can just go off and  do whatever you want them to do. And so if you're
  1022. 1:33:57able to raise their level and care about improving  them, then it's honestly a rid like it's so much
  1023. 1:34:04better than when you were working with humans.  I've always cared about this, right? Even when
  1024. 1:34:08I was working with um with people, I've always  wanted to make the project better and make us
  1025. 1:34:12all work faster. I've always cared about that,  I think, more than maybe my colleagues had who
  1026. 1:34:16just wanted to clock in and go home. So, not all  of them, of course, but like just being able to
  1027. 1:34:24have those skills has just felt like a ridiculous  uplift in the amount I'm able to trust my agents.
  1028. 1:34:30And I think that is an incredibly valuable skill  because that can take you from a you know 5x agent
  1029. 1:34:36user into a 100x 200 200x agent user you know okay  so for someone who let's let's take the let's take
  1030. 1:34:46a high school kid they want to become a software  developer and in this day and age it's like what
  1031. 1:34:52does that mean? Uh what what advice would you give  them? How could they become a software developer?
  1032. 1:34:58Should they? and what do they have to learn? So,  I'm not I don't again I'm not I can't predict
  1033. 1:35:05the future. I've not been on the job market for  yonsks, right, since 20 or something maybe. Um,
  1034. 1:35:12so I don't know what it's like out there. So, your  mileage may vary in this advice, but what I would
  1035. 1:35:19say is demonstrate that you care and that you care  about the strategic stuff. you might not even need
  1036. 1:35:26like great expertise in it. But if you're using  language like deep modules, if you're thinking
  1037. 1:35:32about okay, how am I going to improve things next  time? If you're constantly thinking about not only
  1038. 1:35:37improving the code, but improving the next code  that comes through, then I think that is enough
  1039. 1:35:43to start de and you demonstrate that through  projects, right? That you're actually shipping
  1040. 1:35:48stuff that may be assisting you in your own job  hunts. You know, that's the classic one. um you
  1041. 1:35:52know, you build a little uh CRM for yourself or  something and you make it as beautiful as you can.
  1042. 1:35:56You make it as performant as you can. You really  care about it and you deploy it somewhere that
  1043. 1:36:00people can click through it and actually view  it and and fiddle around with it and go, "Wow,
  1044. 1:36:04okay, this this kid knows what he's doing and he's  doing it all with agents." That is what I'd say.
  1045. 1:36:10And it is very possible to demonstrate that and  demonstrate huge amounts of work. You just have
  1046. 1:36:15to care about the strategic layout. Okay. So you  would say would you say that anyone needs to learn
  1047. 1:36:24a specific programming language right now? Nah.  Uh programming languages like you know that's
  1048. 1:36:30tactical stuff right you know actually getting  good at a programming language you know that's
  1049. 1:36:36you know it's it's important to be able to read it  to understand what module boundaries are but you
  1050. 1:36:42can just get agents to just teach you about it and  to teach you about it in pseudo code. What I would
  1051. 1:36:47recommend is start um start working through some  of these classic books, you know, like a Michael
  1052. 1:36:52Feathers book or something. Actually buy the book  and start obviously it's going to be pretty dense,
  1053. 1:37:00right? There's going to be words in there that you  don't understand, but then use the teach skill or
  1054. 1:37:04something. So, it pulls that down information  for you so you can start actually going, "Oh,
  1055. 1:37:08I see how I would apply this in real life." You  know, this book is from 2003 and I I think it
  1056. 1:37:13was wrapped in plastic in on my like bookshelf  or something and I just thought, oh, like I've
  1057. 1:37:19been prompting agents to design good code bases,  but this guy was doing that before there were even
  1058. 1:37:24agents, right? He's been writing how to what a  good codebase is. And so I pulled it down and I
  1059. 1:37:30just I couldn't believe what I was reading because  it just felt so relevant to what I was looking at.
  1060. 1:37:36and I was he was able to put into words the stuff  that I sort of felt and I just started plugging
  1061. 1:37:44that stuff into agents and that's how improved  codebase architecture came about. Um so care about
  1062. 1:37:52it read books about it use the teach skill to get  your way there and that's how to do it. What other
  1063. 1:38:00books should people read? Philosophy of Software  Design by John Asterout. Pragmatic Programmer by
  1064. 1:38:06David Thomas and Andrew Hunt. Um Michael Feathers  is good. Refactoring by um what's his name? Martin
  1065. 1:38:14Fowler. Uh Extreme Programming by Kemp Beck. Those  are probably the big ones. Wow. And so would you
  1066. 1:38:20say that people should read these to become  effective at AI programming? Yeah. Uh if you
  1067. 1:38:27actually do the reading, it puts you way ahead  of everybody else in the field. genuinely like
  1068. 1:38:33um so yeah definitely I I think I do think that  experience is the only way to you know get there
  1069. 1:38:39but it really helps when you have the experience  and you have the words to explain that experience
  1070. 1:38:45because I think like just understanding the  language and the categories of things and like
  1071. 1:38:51okay this is more important than this is that's  what strategic programming is all about. It's all
  1072. 1:38:57about finding balance between lots of different  things that are all competing. You know, like,
  1073. 1:39:02okay, I want my agents to ship features really  well, and so I'm going to write 4,000 tests
  1074. 1:39:07for them. Oh, but those 4,000 tests are actually  slowing things down because they actually make the
  1075. 1:39:12agent run really slowly. How do we winnow those  down? How do we put those in a CI so that it can
  1076. 1:39:16rely on them later? How do we shard them so that  they've got, you know, running in four different
  1077. 1:39:21um nodes instead of just one? You know, all of  that that might sound confusing and complicated,
  1078. 1:39:26but all of those are just words and you just learn  the words and then you prompt the agent and then
  1079. 1:39:31the agent can care about it. So, what you're  doing is you're just giving the agent permission
  1080. 1:39:35to care about the strategic layer and you're  learning the words in order to do that. Okay,
  1081. 1:39:41that makes sense. And it makes me wonder and  I know you don't you're not in the job market,
  1082. 1:39:46you can't really look into the future, but how  effective right now are computer science degrees?
  1083. 1:39:51I would ask how effective have they ever been  actually [laughter] because I I didn't get a CS
  1084. 1:39:57degree, right? I got a drama degree, you know, I  was a voice coach for six years and I just taught
  1085. 1:40:00myself this stuff in like 2015 and started moving  over. Wow. Um so yeah, I don't know. I don't know
  1086. 1:40:06if that they've ever been that effective. Um the  the issue is people who care and people who don't.
  1087. 1:40:13That's interesting. And I I'm sure many people  will be watching this and they're in a computer
  1088. 1:40:17science degree. Um, what would you tell them?  Care, you know, just just care about the the
  1089. 1:40:27stuff you're shipping. Care about your users,  care about the other devs on your team, even
  1090. 1:40:31though they're agents, and care to learn more,  care to be curious, and yeah, uh, enjoy because
  1091. 1:40:40it is fun. [laughter] And this is so fascinating  because like it's fun to talk with someone who I
  1092. 1:40:46didn't know you didn't have a competi degree  because that's that's even more interesting.
  1093. 1:40:49You're selftaught and I I think that's been the  best way to learn anything for a while now. AI
  1094. 1:40:54has made that even more obvious. Um for you if you  were to hire a developer to join your team, should
  1095. 1:41:03you need one? Actually, that's probably a two-part  question. What would make you need one? Why would
  1096. 1:41:08you need one? And what would you look for? what  would they what skills would they need to have
  1097. 1:41:12that would impress you? Um I mean I I hate to  harp on this again but it is it is that strategic
  1098. 1:41:19thinking right it is the environment it is the all  of that and using the right language and caring
  1099. 1:41:23about the right things. Um, for me, I do think  there are bottlenecks all over the process still,
  1100. 1:41:30right? This is not just um, you can't just  infinitely accelerate and just like, you know,
  1101. 1:41:36one person can ship 2,500 PRs a month, but if you  have two people doing that, that's still 5,000
  1102. 1:41:42PRs a month. You know what I mean? That's still  doubling the amount of work as long as both of
  1103. 1:41:46them are enabled and both of them can work on cool  stuff. And honestly, the the thing that really
  1104. 1:41:53differentiates good products these days is not  the amount of tokens that have been spent on them,
  1105. 1:41:58but the amount of attention that has been given  to the product, the amount of use that it's had,
  1106. 1:42:03and the the smart things people observe about the  product in order to make it better. So, it's the
  1107. 1:42:08testing and the QA and the planning and figuring  out strategically what needs to be done next to
  1108. 1:42:13improve the product. How and noticing problems  and and essentially product management and product
  1109. 1:42:19design and that is something you need a human  for, you know, or you need a human at least to
  1110. 1:42:26tell an agent to care about it, you know. So yeah,  that's what I'd say is that people now are all
  1111. 1:42:32about directing agents and using their attention  and their care to make the agents care about the
  1112. 1:42:39right things. Okay. And it sounds the word care  about a thousand times and I wasn't planning to.
  1113. 1:42:45I think I've just found a new word that I like. So  yeah. Yeah. I think uh [laughter] and I think that
  1114. 1:42:50is important because as we've kind of noticed  uh through this, you can't just go in and just
  1115. 1:42:56write code. You can't just go in and go, I'm going  to make this thing work. You have to really care
  1116. 1:43:00about what it's going to be. You have to think  about it's it's like thinking about a child,
  1117. 1:43:06like what do I want you to be? What how what can I  expose you to? What are the standards I'm going to
  1118. 1:43:09set for you? What's the rules for life I'm going  to give you? Um that's interesting. Uh now I'm
  1119. 1:43:15curious how much of this strategic programming um  how to use AI agents and I'm going to allow you to
  1120. 1:43:23shill your course a bit here because I I I love  your course. How much of that is in the courses
  1121. 1:43:27you create right now? Yeah, that's like I'm saying  all this stuff matters and as I'm saying it,
  1122. 1:43:33I'm trying to find ways to teach people about  it. You know, I'm saying software fundamentals
  1123. 1:43:38still matter and so people's follow-up is okay,  what software fundamentals, how do I learn them,
  1124. 1:43:43how do I apply them? And so I've got a course  coming out in November. Um I haven't released
  1125. 1:43:48this information, but it's coming out November  9th. is like a two-week cohort course that's just
  1126. 1:43:53timed perfectly so that um I go on paternity leave  like right afterwards and uh so stressing trying
  1127. 1:44:00to get it done and all of that is about using  my skills sure but skills are essentially just
  1128. 1:44:06pointers to a process and the process I'm trying  to teach is that you need to understand how to
  1129. 1:44:12improve your agents environment. It's essentially  the the end point is where you create a software
  1130. 1:44:17factory that does a lot of this stuff for you. So  a lot of the caring is sort of automated in these
  1131. 1:44:23cron jobs in these routines that automatically try  to make your agent and its environment better. As
  1132. 1:44:29I was saying it's sort of like you know ideally  the harness would improve the environment as you
  1133. 1:44:35go and that's what we're trying to get to by the  end of the course. Interesting. And I I've I just
  1134. 1:44:41heard the term software factory for the first time  last week and I know it's been thrown around for a
  1135. 1:44:45minute. What exactly is that? Well, on the AI  coding dictionary, we have a definition for a
  1136. 1:44:51software [laughter] factory. There is a definition  for that. Um, and it's essentially I I won't pull
  1137. 1:44:57it up, but what it does is if you imagine that  producing work involves several points at which
  1138. 1:45:07the human needs to interact with the work as it's  being done. So, it might be that you need to tell
  1139. 1:45:14the agent to do something. That's the first  point. Then you might do a grilling session
  1140. 1:45:19and you might um approve the plan. That's the next  point. Then it might go all the way through until
  1141. 1:45:24the work needs to be merged. Now what a software  factory does is it tries to take the human out of
  1142. 1:45:30more points in that process. So it might be that  instead of you um needing to uh initiate the work,
  1143. 1:45:39it might be that there's some trigger that  happens. Let's say a new issue is created
  1144. 1:45:43on your repo that pings down to an agent and the  agent decides whether to initiate some work based
  1145. 1:45:48on that. Or it might be there's a slow database  warning in your um in your service that checks
  1146. 1:45:54your database that again goes down to an agent.  The agent decides whether to um proceed with that.
  1147. 1:46:00And then you might have ones where okay that might  not even need a human in there all the way up to
  1148. 1:46:05human review. But maybe you don't do human review  for things. Maybe you use that oneway two-way
  1149. 1:46:10door thing and actually only one-way doors need a  human review and the agent just goes and merges it
  1150. 1:46:17itself. So the software factory is basically just  taking humans out of more points so that you can
  1151. 1:46:23accelerate the development of software more.  And the issue with that is that the more you
  1152. 1:46:29accelerate it, you need breaks as well. You need  to be able to pull the software factory back so
  1153. 1:46:33it doesn't just become a slob cannon. which again  is where the strategic layer, the thinking about
  1154. 1:46:38the environment really makes a big difference.  Okay. And that that sounds great. Um but it also
  1155. 1:46:46sounds really hard to do. Yeah. Yeah, I think  so. I think um there are you're basically having
  1156. 1:46:53to plug together a bunch of services in order to  do it. I think as um the tools and the harnesses
  1157. 1:47:00improve and make those things easier then you know  you know like people are obsessed with like loop
  1158. 1:47:06engineering like um few months ago or whatever  this is basically loop engineering what you're
  1159. 1:47:11doing is you know a loop is one way to do it where  you just schedule the agent to just like pull the
  1160. 1:47:16GitHub issues and just sort of initiate work on  its own. Um but you can have multiple different
  1161. 1:47:20ways of doing it essentially. So this is just loop  engineering by another name essentially and it's
  1162. 1:47:25it's very exciting and it means that you can churn  out a ton of work and really improve the quality
  1163. 1:47:30of what you're working on. Uh and that's what  software factory is. Interesting. Okay. And then
  1164. 1:47:36you I'm curious about how you approach so like  for like right right now if I went to your course
  1165. 1:47:43website is that a platform or is that something  you guys designed? Yeah, that's something we
  1166. 1:47:47built from scratch. Um I work with a team called  badass.dev. dev and they work with a few course
  1167. 1:47:54creators and they uh build this uh course platform  from scratch basically. Oh dang, that's pretty
  1168. 1:47:59cool. So you're not you're not running the AI  coding for that right now? Uh no. What are some
  1169. 1:48:05do you have any projects you're working on right  now? Uh the open source stuff? No, just S castle.
  1170. 1:48:12just um trying to get this course built and trying  to uh yeah ship this video editor that's uh always
  1171. 1:48:19always constantly improving. Okay. Okay. I can't  resist. Tell me about this video editor. I I've
  1172. 1:48:26seen you post about it a few times and I'm like  that looks interesting. Okay. He's he's having
  1173. 1:48:30it create demos for him before he records it.  That's interesting. So what is this thing? Oh,
  1174. 1:48:36it's it's outrageous. So it's [laughter] uh it is  it is outrageous. It's uh uh what it is. It came
  1175. 1:48:44down to this idea that I hate editing video. And  so what I wanted was I wanted to as I recorded,
  1176. 1:48:50I would essentially leave a little gap and then  it would cut that and that would go be transcribed
  1177. 1:48:57automatically and just appear as a clip in  the finished editor essentially. So there's
  1178. 1:49:02no editing. So I edit as I record. And from there  that developed into, okay, I need like a I use a
  1179. 1:49:09lot of diagrams in my setup. I'm going to build  like a custom diagram playground as well that
  1180. 1:49:13links to the clip so I can see from the clip  which diagram I was viewing that so I can go
  1181. 1:49:17back in time and just edit it and stuff. And then  maybe I Oh, yeah, Elgato teleprompter looks quite
  1182. 1:49:22nice. Maybe I need a teleprompter, too. Um, oh,  maybe I can customize the teleprompter based on
  1183. 1:49:26what I'm seeing. Oh, maybe I can get an agent to  write scripts. Maybe I can get an agent to write
  1184. 1:49:30animatics as well. Opus 5.5 looks really good with  actually building out um custom videos for you. So
  1185. 1:49:36why don't I watch the video I'm about to record  before I record the video. That seems good. And
  1186. 1:49:40then from there I you know it just grows and grows  and grows and like and now like every edit I make
  1187. 1:49:45get automatically synced to my course platform. Um  I have you know version history. I have linting. I
  1188. 1:49:51have uh AI writing. Like it's just a a ridiculous  wonderful project. And it's it's open source too.
  1189. 1:49:57So, it's in course video manager. We're 2,500  commits deep now. Um, and it's just all in here,
  1190. 1:50:03you know. So, if you want to uh use this as like I  I don't take um pull requests and um like I don't
  1191. 1:50:10even take issues on this actually. Uh but if you  want to just have a look at it, it's there. Yeah,
  1192. 1:50:14I'll take a look. And I I I do something  not quite as streamlined as you, but I do
  1193. 1:50:19um have it to where when I finish recording, my AI  has been trained enough back and forth with me to
  1194. 1:50:26where it knows how to edit me. It knows what kind  of cuts I like. It knows that when I cuz I I I've
  1195. 1:50:31been doing this for a long time. I used to edit  my own stuff. Now I have editors. So like when
  1196. 1:50:34I say something, I'll freeze so I can remember  what to say again and we don't have to do a weird
  1197. 1:50:39cut and then I'll keep talking. Um so it knows  to watch for that. So when I finish recording,
  1198. 1:50:44I'll just tell, hey, go go go to the place we  where we have our files recorded, offload it,
  1199. 1:50:50ingest it, and what that means is it will um  transcribe it. It'll use the Gemini video API to
  1200. 1:50:56watch it so we know exactly what's happening. And  then um it will start to actually produce an edit,
  1201. 1:51:02and it will edit it to like a very clean draft. We  coded our own frame to IO. I'll watch it to make
  1202. 1:51:08sure the story is right and type and then it will  export an XML for Adobe Premiere where my editors
  1203. 1:51:14just open up and there's the draft ready to go and  they can make it look pretty. Um, it's it's very
  1204. 1:51:19cool, but I want to get some more where you are.  That sounds very interesting. Uh, it's just so
  1205. 1:51:24great. It's so great. Um, yeah. I mean, it's, you  know, it's there if you want to have a have a play
  1206. 1:51:29around with it. Um I I just wanted a project that  I knew that I would constantly use and constantly
  1207. 1:51:36update and like so it's just always in the back of  my head that okay I'm going to work on this going
  1208. 1:51:39to work on this and I've you know I've got agents  right now like playing around with it improving
  1209. 1:51:43the environments that the code operates in. This  is my main playground really for trying out my
  1210. 1:51:47ideas. Now you you mentioned that you have AI  scripting. How do you feel about that? Do do
  1211. 1:51:54you do you uh speak word for word from what the AI  scripts for you? No, I mean I did a drama degree,
  1212. 1:52:01right? So like I'm good at improvising  words and I just use the script as like a
  1213. 1:52:07uh it has like commands that I can copy and paste  in order to set up the environment and stuff and
  1214. 1:52:11like go to the right commit in the the history for  the project I'm working on. Um so mostly it's just
  1215. 1:52:17a guide essentially. So I'm always improvising  what I'm saying. Interesting. So I I mean I create
  1216. 1:52:23courses as well. So I'm I'm I'm very curious about  your process. Uh is AI pretty much part of every
  1217. 1:52:29process of your course creation from like ideation  to to episode or um video management and all that?
  1218. 1:52:36So, well, one thing I do and one thing uh that I I  actually recommend to anyone who has an audience,
  1219. 1:52:42who has people pinging them all the time is I  have a a personal wiki which is for my audience.
  1220. 1:52:49And every 3 hours or so, I have a chron job that  um looks at all my tweets people have sent me,
  1221. 1:52:56all my mentions, uh all my Discord messages, all  my Slack messages, my Gmail, basically every sort
  1222. 1:53:01of input I can find and then writes that or sort  of ingests it into uh Carpathy style LLM wiki.
  1223. 1:53:09So that means that I can go into this, it's  just a repository, it's private obviously,
  1224. 1:53:13and I can just ask it, hey, what is my audience  interested in? Right? what um questions do they
  1225. 1:53:19have about these skills? And from there, I can  create learning goals from that. So, things that
  1226. 1:53:25people are interested in learning. From there, I  have to actually get off AI. I can't use it for
  1227. 1:53:30this next bit because it's too hard. I print off  those learning goals onto these little car bits
  1228. 1:53:35of card, right? Oh, I love And I I like rearrange  them physically in space until I realize, okay,
  1229. 1:53:41this feels like a lesson. This feels like a  lesson. This feels like a lesson. figure out
  1230. 1:53:45that can take a couple of hours actually just  off computer off AI um just for one section
  1231. 1:53:51and my courses often have eight sections. So and  from there I go back and plug that back in so it
  1232. 1:53:57turns into lessons and turns into videos and then  I start writing the scripts and getting agents to
  1233. 1:54:04dry run it as well. I create these animatics and  then I'm ready to record. So you know it it can
  1234. 1:54:10be months planning a course. It really can. It's  very very slow. That's fascinating. So, I mean,
  1235. 1:54:15two things that just jump out at me. First,  the the wiki idea is such a good idea. Um,
  1236. 1:54:19I'll probably steal that. Uh, the second one  is uh that you have to go off offline to do
  1237. 1:54:25this kind of thinking, which like I've got pen and  paper and this is I find the best thinking happens
  1238. 1:54:31with this. And I do use AI quite a bit to bounce  ideas off of and like try to form an idea more,
  1239. 1:54:37but it's it's those very human moments where I'm  disconnected that things the magic happens. So
  1240. 1:54:44I'm curious how often do you do you go back to  that kind of thing where you're disconnected?
  1241. 1:54:49It's mostly on tasks where there is a lot of  information and I need to keep it all in my head
  1242. 1:54:54at the same time because if I'm like planning if  I am like talking to an agent and we're having a
  1243. 1:55:02back and forth about a plan and maybe shaping the  plan in certain ways then it is going to write a
  1244. 1:55:08ton of stuff right in response to me and I lose so  much time by trying to get over that communication
  1245. 1:55:14gap right constantly trying to go okay what do  you mean And and in doing that, I lose context,
  1246. 1:55:22right? I lose information that I could if I just  slowed down and I just kept everything in my head
  1247. 1:55:29at a speed that works for me. L, you know, writing  stuff on pen and paper. It sort of slows you
  1248. 1:55:34down so that you can consider things at the right  speed. It's it's me paced, right? It's paced at a
  1249. 1:55:39human speed. Whereas agents just work too fast for  that. So for specifically planning, that's what I
  1250. 1:55:47have to do. And I've not found a better way to do  that with agents. That it's it's interesting you
  1251. 1:55:51mentioned it that way. When you're talking to an  agent, you're like trying to bounce an idea off of
  1252. 1:55:55it or just like try to do anything. It does feel  too fast and you're it'll say a bunch of stuff
  1253. 1:56:01and it's so dense you're like or too slow as well,  right? Or you're just waiting for a response for
  1254. 1:56:06like 2 minutes and you think, "Oh, I just check  Twitter." And then half an hour later you go, "Oh,
  1255. 1:56:09what did the agent say?" You know? Right. Oh,  yeah. Yeah. Like like 7 7 o'clock at night and
  1256. 1:56:14you're like, "Oh, I would did ask that question  earlier. at night and I don't even care about that
  1257. 1:56:19anymore. Exactly. You just got to get out of the  conversation. Just keep it in your own head for
  1258. 1:56:22a while and then I usually just brain dump it back  to the agent. Um just, you know, use whisper flow,
  1259. 1:56:27dictate out a ton of stuff and then um then it  just pulls it into its own structure and its own
  1260. 1:56:33data and then I just usually chuck the cards away  because I don't don't need to go back to paper
  1261. 1:56:37there because for small edits, for little stuff,  once you've got the main structure, AI is really
  1262. 1:56:42good. Do you use AI in a way to where because I  know we've been looking at coding agents and and
  1263. 1:56:49getting the coding agents to know your repos and  your code, but do you use anything for your life?
  1264. 1:56:54Do you have like a Hermes agent or any kind of  agent that like has the context of you? Um, yes.
  1265. 1:56:59This audience wiki has is sort of a bit of a blend  of like me plus my audience, right? Because those
  1266. 1:57:04things are often closer than you think. And so,  yeah, I I have that repo sort of serves as like my
  1267. 1:57:10home base for all of my agents. And so I have like  a daily standup with my agents where I say, "Okay,
  1268. 1:57:15what tasks need to be done today?" Um, I have a  like a pixel watch that I dictate stuff into and
  1269. 1:57:22that goes on my to-do like an agent triage to-do  list and the agent then pulls it into um it was
  1270. 1:57:29like uh deciding every minute of the day for me  like scheduling out all that stuff. And I have
  1271. 1:57:34like a little um little thing in front of me. This  is called a terminal that basically Yeah, they're
  1272. 1:57:39kind of they're all right. I don't know if they're  any good, but I've got two of them actually,
  1273. 1:57:43one downstairs and one upstairs. And uh I was just  sort of ignoring it. So, I'm I'm still trying to
  1274. 1:57:48find the right uh interface for the agent to help  me run my life, you know, but it's a it's a me
  1275. 1:57:54problem, not not an agent problem. The agents are  really good at it. I find the same thing. I'm I'm
  1276. 1:57:59trying to figure out the best way to like like  help me schedule my day or help me do this and
  1277. 1:58:03help me work with my team and help it's it's I  don't think anyone's quite solved it yet. Well,
  1278. 1:58:08it's really hard to figure out. No. And I think  it's actually much harder than coding. Actually
  1279. 1:58:13using it for knowledge work is much harder than  coding because coding is so verifiable, right?
  1280. 1:58:18You can use these automated checks to actually  make sure that the thing you're doing is correct,
  1281. 1:58:23that you have high quality on the output. But  with knowledge work, with planning a course, how
  1282. 1:58:29do you how do you know that you've built a better  course than you did two months ago? You know, it's
  1283. 1:58:33really hard. There's no way to verify that. How  do you know that this video script is better than
  1284. 1:58:39um your previous one? So, yeah, I think knowledge  work is really difficult, way harder than people
  1285. 1:58:45think it is. And I think we're going to be hitting  that over the next few years as we try to force
  1286. 1:58:50AI into all of these holes. And I love that this  entire call you everything coding related you're
  1287. 1:58:57like it's easier than you think with knowledge  work you're like it's actually harder than people
  1288. 1:59:02realize. [laughter] Yeah, maybe I'm wrong there  today. Maybe it's just because I know coding more
  1289. 1:59:07than knowledge work. Maybe. No, I think I think  you're right though. It's that very very distinct
  1290. 1:59:10point you made is that coding is verifiable and  that you can learn it and it's it's deterministic
  1291. 1:59:16and you can test it and like you you know it's  good, it works or I mean that's that's why agents
  1292. 1:59:21are amazing at math problems and can solve math  problems way better than humans because every
  1293. 1:59:27part of it is verifiable. All the inputs are text  and all the outputs are text. But it's also why
  1294. 1:59:32they still can't draw hands, right? Because like  it's just different inputs than they're used to,
  1295. 1:59:39right? They can't write a convincing tweet that we  don't know is is AI, right? Because that is just
  1296. 1:59:44a much harder job because it's so so much less  verifiable. And it's something I don't even really
  1297. 1:59:49care the AI can do. I think there's there's human  things that I want to keep human that just make us
  1298. 1:59:53who we are. And communication is like one of those  things that make us who we are. Absolutely. Yeah.
  1299. 1:59:57I play a ton of board games. Um keep agents out of  my board games. I don't want to see any AI art. I
  1300. 2:00:03don't want to hear any AI music. I don't want to  uh I get a thousand AI tweets a day and I hate
  1301. 2:00:08them, you know. [laughter] Yeah. Yeah. I think we  feel the same way. Yes. Absolutely. Well, we are
  1302. 2:00:14near the Actually, no, we are at the end of time.  Um tell me where can people find you and with your
  1303. 2:00:20courses where should they start? Go to aihero.dev.  It's the place. Um find my skills. Uh install my
  1304. 2:00:27skills, enjoy them. Um, if you have any questions,  go to the Discord, find me on Twitter. Um,
  1305. 2:00:31you know, I'm pretty active in all those places  and I'm on YouTube as well. Um, recently crossed
  1306. 2:00:35400,000 subscribers, which I'm very pleased about.  Congrats. And, uh, yeah, look me up. I'm there. Do
  1307. 2:00:41you uh do you have like a wait list for your new  course? Uh, there's a newsletter. Um, but yeah,
  1308. 2:00:47sign up and I'll see you there. Cool. Awesome.  Well, Matt, this has been so fun and informative
  1309. 2:00:53and you've given me so many ideas. Thank you  for taking the time to not only teach me but my
  1310. 2:00:58audience. This has been very very cool. Appreciate  it, pal. Hey, made it to the end of the podcast.
  1311. 2:01:03At the end of my videos, I like to pray for you,  my audience. Kind of weird, I know, but just go
  1312. 2:01:08with it. You're not here for um I mean, you are  here for a reason. It's not an accident. And yeah,
  1313. 2:01:15let's just go with it. Uh I put it at the end, so  you have the option to just, you know, duck out.
  1314. 2:01:19The information has already been given, but if you  want to chill out and stay here with me right now,
  1315. 2:01:23I'd like to pray for you and your life and your  family. Ready? 1 2 3 pray. Uh God, I thank you for
  1316. 2:01:28the person on the other side of this screen, this  camera. I thank you for them, who they are. And
  1317. 2:01:34um I thank you that they stuck to the end of this  and that they are learning about coding and and um
  1318. 2:01:40technology and just all this AI stuff, which can  be stressful, but also is very life-giving if we
  1319. 2:01:45approach it the right way. So God, I ask in your  name that you shift any anxiety, any stress uh
  1320. 2:01:52about AI or the future, about jobs. I I ask that  you would just flip that switch in their brain to
  1321. 2:01:59be joyful and excited and let them be at peace  about it and just give them energy and and give
  1322. 2:02:06them ideas and a path forward. If they're looking  for a job right now, I pray you provide that job.
  1323. 2:02:11If they're looking to start a company, give them  that idea and the and the steps forward. Bless
  1324. 2:02:15them right now. Bless them and their family. Watch  over their families. Um give them peace and joy in
  1325. 2:02:23their lives right now. And help them to take the  knowledge from this video and apply it well. I
  1326. 2:02:29ask this in your name, Jesus. Amen. Thank you for  letting me do that. I'll catch you guys next time.

About this transcript

This page contains the full transcript of How to Actually Code with AI (ft. Matt Pocock) by NetworkChuck, generated from the public captions YouTube serves with the video. The transcript has 24,631 words across 1,326 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.