YouTube2Text

Yup, Claude Code Plan Mode is here: The Senior Engineer's Workflow — Transcript

by IndyDevDan · 3,945 words · 569 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Let's open claw code.
  2. 0:03Shift tab. Shift tab. Bam. We're in plan
  3. 0:07mode. The debate is over. Claw Code is
  4. 0:10taking over the engineering world. The
  5. 0:12Claw Code team continues their red hot
  6. 0:16streak as the most cracked engineering
  7. 0:19team. How are they able to do this? The
  8. 0:21creator of Claw Code said it himself.
  9. 0:24Generally at Anthropic, we have this
  10. 0:25product principle of do the simple thing
  11. 0:27first. The team has stealth launched
  12. 0:30plan mode. It's simple and powerful. How
  13. 0:33does plan mode work and why does it
  14. 0:35exist? There are two powerful ways to
  15. 0:38use plan mode. We're going to break down
  16. 0:41right now. One of these techniques takes
  17. 0:42a little more time but pays you
  18. 0:44dividends. You want to make sure you're
  19. 0:46using the right method so you don't
  20. 0:48waste your valuable engineering time.
  21. 0:50Let's master claw code plan mode in 20
  22. 0:53minutes.
  23. 0:59No one knows clawed code better than
  24. 1:02claw code. Let's ask it to tell us about
  25. 1:05plan mode. Reflect on your system
  26. 1:08prompt. Hi, can you explain plan mode?
  27. 1:12So, with certain features, you can be
  28. 1:13sure the system prompt has information
  29. 1:16about it. Here, we're quite literally
  30. 1:17asking Claude code to tell us about its
  31. 1:20new plan mode. You can see incoming
  32. 1:21tokens streaming in. And we look at
  33. 1:23that. Here is the details on plan mode
  34. 1:26right out of the system prompt. Plan
  35. 1:28mode is a special operating mode where I
  36. 1:30research and analyze, create a plan,
  37. 1:32present the plan for approval, wait for
  38. 1:34confirmation. So, this is really
  39. 1:35important and this is interesting. I
  40. 1:37didn't know it had an exit plan mode
  41. 1:39tool. This might be a fresh feature that
  42. 1:42just rolled out. The cloud code team's
  43. 1:44engineering velocity is mind-blowing.
  44. 1:46I'm restricted from making any file
  45. 1:47edits, running commands, creating or
  46. 1:50deleting, or making commits. And here we
  47. 1:52get a nice breakdown of when this is
  48. 1:54useful. We're working with AI here, so
  49. 1:56we can just ask stupid zero judgment
  50. 1:58questions that might have already been
  51. 2:00explained in the previous prompt, right?
  52. 2:01Ask stupid questions, validate your
  53. 2:03assumptions, learn fast, and move even
  54. 2:06faster. Can plan mode create or modify
  55. 2:09any files? Let's say like plan MD, can
  56. 2:13you modify that? So, I'm just being
  57. 2:15super super clear with cloud code. I
  58. 2:17want it to repeat things back to me.
  59. 2:19After all, we are operating in plan
  60. 2:22mode, so this is what it's for. No, plan
  61. 2:24mode cannot create or modify any files,
  62. 2:25including plan.MD. So, this is really
  63. 2:28about building up the context window
  64. 2:30inside of cloud code. There's no
  65. 2:32actions. There's no mutations happening
  66. 2:35here, right? So, let's be totally clear
  67. 2:37here. So I'll say so the purpose of plan
  68. 2:40mode read files understand the codebase
  69. 2:43by loading
  70. 2:45your context window with important
  71. 2:47information
  72. 2:49before we start working then switch out
  73. 2:52and edit correct me if I'm wrong. I like
  74. 2:56to use this correct me if I'm wrong to
  75. 2:59decrease sickle fancy basically decrease
  76. 3:01the bias in the model and present a more
  77. 3:04neutral question format. These models
  78. 3:06love to tell you exactly what you want
  79. 3:08to hear because it's good for business
  80. 3:09for these AI labs. So, you have to be
  81. 3:11really clear. So, I like to use this.
  82. 3:12Correct me if I'm wrong. It's just a
  83. 3:14little bit more neutral. You can see
  84. 3:15cloud code validating exactly what we
  85. 3:17were saying. Yes, you're exactly right.
  86. 3:18That's a perfect understanding of plan
  87. 3:20mode's purpose. It's research and
  88. 3:22planning phase where I load, analyze,
  89. 3:24design, present, then after you approve,
  90. 3:27exit, execute, and then we run all the
  91. 3:29commands so on and so forth. Right? This
  92. 3:31is really interesting. Notice how claw
  93. 3:33code is converging on the senior
  94. 3:36engineers developer workflow. First, we
  95. 3:38understand the feature by gathering the
  96. 3:40right information. We build up the state
  97. 3:42space of the problem and the solution.
  98. 3:45And only then do we start building.
  99. 3:47Before plan mode, we'd have to do
  100. 3:49something like this, right? You have
  101. 3:50your long prompt and then we say
  102. 3:52something like this, right? Don't write,
  103. 3:53update, or delete anything. Just read
  104. 3:56and plan. So with two shift tabs, we can
  105. 3:59immediately just enable that prompt
  106. 4:02modification. Planning mode means you
  107. 4:04can focus on priming your agent with the
  108. 4:07right context before getting work done.
  109. 4:09Great. So we have this new plan mode.
  110. 4:11Let's put it to work.
  111. 4:15I'm going to open up cursor here and
  112. 4:17let's enhance our previous proof of
  113. 4:20concept codebase, the infinite agentic
  114. 4:22loop. In this codebase, you can have
  115. 4:24claw code generate infinite versions of
  116. 4:26a solution to your problem until it
  117. 4:29breaks. Let's go ahead open up a new
  118. 4:31terminal window here. Fire up clog.
  119. 4:33We'll make sure we're running the
  120. 4:35state-of-the-art model opus 4. I'm going
  121. 4:37to run a prime command to get our
  122. 4:39context window loaded with key
  123. 4:41information about this codebase. You can
  124. 4:43check out the prime command inside of
  125. 4:45the custom slash command. This is a
  126. 4:47really simple one. We're just running
  127. 4:48get ls files and we're having cloud code
  128. 4:51read a couple of files. So you can also
  129. 4:52see we have claw.md added to this
  130. 4:55directory. So claw code is now primed.
  131. 4:56I'll hit shift tab tab. And now we're in
  132. 5:00plan mode. In the beginning I mentioned
  133. 5:02there are two ways to use plan mode. You
  134. 5:04can plan then write code or you can plan
  135. 5:07write the plan to a spec file and then
  136. 5:10have cloud code execute on the spec that
  137. 5:12you created. Four out of five times. I
  138. 5:15recommend the second technique
  139. 5:17specifically for mid to largesized
  140. 5:19features which you should be aiming for.
  141. 5:22We have more than enough compute and
  142. 5:23intelligent models in great tools. So
  143. 5:26aim big, hand off more work to your
  144. 5:28agent. Do more in single swings to save
  145. 5:31time and understand how far you can
  146. 5:34really push these models and these
  147. 5:36tools. So right now our infinite agentic
  148. 5:38loop set up in this codebase. If we open
  149. 5:41up source or source infinite, you can
  150. 5:43see they're generating single files. And
  151. 5:45in our previous video, we looked at
  152. 5:46these UIs, we can quickly just take a
  153. 5:48look at one of these. If I open up UI
  154. 5:50hybrid one, copy the path to this, and I
  155. 5:54just type Chrome this inside of bash
  156. 5:57mode, which you can trigger inside of
  157. 5:59cloud code with the bang or the
  158. 6:01exclamation. You know, you can take a
  159. 6:02look at this. Basically, we're just
  160. 6:03generating infinite UIs that could do
  161. 6:05many different things. If I go to two,
  162. 6:08you can see here's another one. Here's
  163. 6:10another one. Right? So, we just
  164. 6:13generated all of these interesting user
  165. 6:15interfaces.
  166. 6:17Right? Here's another kind of wild one.
  167. 6:19Here's another. Right? We can go, you
  168. 6:22know, we had tons and tons of these
  169. 6:23generated. If if you open up these
  170. 6:24directories, you can see all of them.
  171. 6:26You can check these out yourself. Okay.
  172. 6:27So, we ran a prompt that generated a ton
  173. 6:29of UIs inside of a single HTML file.
  174. 6:32Okay. So, very cool. What we're going to
  175. 6:33do here is update our lower level prompt
  176. 6:36that we passed into our infinite agentic
  177. 6:39loop prompt and we're going to have
  178. 6:40cloud code plan out exactly what we want
  179. 6:43done here. So what we're doing here is
  180. 6:44effectively metaprompting. I'm writing a
  181. 6:46prompt that's going to write a prompt
  182. 6:48that is a plan aka a spec. What I want
  183. 6:51to do here is just something simple. I
  184. 6:53want to showcase the power of plan mode.
  185. 6:55And instead of writing single
  186. 6:57self-contained HTML files, what we'll do
  187. 7:00is we'll break them up into their own
  188. 7:01isolated directory with a CSS.js and
  189. 7:05HTML file. Okay? And then we're going to
  190. 7:07embed all that in a new plan. So let's
  191. 7:10go ahead and get to work there. And all
  192. 7:11we need to do is be concise and write a
  193. 7:14great prompt. So I'll say let's plan
  194. 7:17specs V3. And I'll say V4. So I'm going
  195. 7:19to update that. That's what we want to
  196. 7:22generate. We want to build on specs v3.
  197. 7:26I'm using clause autocomplete here with
  198. 7:27the at@ symbol, but instead of writing a
  199. 7:30singlehtml
  200. 7:32file split into CSS.js.html,
  201. 7:37we also want to store these three files
  202. 7:40unique directory. All right. And that's
  203. 7:43it. So in plan mode, Claude Code is now
  204. 7:46going to get to work filling up its own
  205. 7:48context window with the information it
  206. 7:50needs to plan this workout. Okay, so
  207. 7:53we're going to kick this off. I'm going
  208. 7:54to leave open the specs directory.
  209. 7:56Nothing is going to get created here,
  210. 7:58right? This is all going into Cloud
  211. 7:59Codes context window. It's not going to
  212. 8:01create or modify anything, right? Just
  213. 8:03like it told us here, just like we
  214. 8:05learned by prompting and asking
  215. 8:07questions against Cloud Code. So let's
  216. 8:09go ahead and take a look at this. It's
  217. 8:10reading context. It's figuring things
  218. 8:12out. It's understanding what it needs to
  219. 8:14do to build out this V4 spec, right? So,
  220. 8:18it has a clear understanding of
  221. 8:19requirements and now it's going to
  222. 8:20present a plan. And so, this is quite
  223. 8:22interesting, right? It's planning inside
  224. 8:24of its own context window. It's just
  225. 8:27looking at your codebase, gathering
  226. 8:29information, reading files, right? It's
  227. 8:31all read. There's no mutations, there's
  228. 8:34no create, there's no update, there's no
  229. 8:36deletes. It's a very safe mode. And very
  230. 8:39importantly, this mirrors the beginning
  231. 8:41of a senior engineer's process, right?
  232. 8:45We plan, work first, we architect, we
  233. 8:47think, we plan, and only then do we
  234. 8:50build. So let's see what we have here.
  235. 8:52It's walking through based on my
  236. 8:53analysis. Here's a comprehensive plan.
  237. 8:55So this looks decent. You can see it has
  238. 8:57this key changes from V3 to V4. Each
  239. 9:00iteration creates its own all CSS,
  240. 9:02alljs. Perfect. So this looks good. Now
  241. 9:05I want to ask a question here about the
  242. 9:07implementation. There's one key detail.
  243. 9:08I want to make sure every version that
  244. 9:11we're going to generate with the
  245. 9:12infinite agentic loop gets their own
  246. 9:14directory with those three files. So I'm
  247. 9:16just going to be super clear here and
  248. 9:18showcase you know how you can use plan
  249. 9:20mode. So we're basically talking with
  250. 9:22the cloud 4 opus model running on claw
  251. 9:25code to be clear. showcase what the
  252. 9:28output durest
  253. 9:32mark directory looks like after running
  254. 9:36three agents. And so output directory is
  255. 9:39going to be part of our arguments, part
  256. 9:42of our variables inside of our infinite
  257. 9:44agentic loop prompt. And so I'm
  258. 9:46referencing this variable here, which is
  259. 9:48where all the iterations will be saved.
  260. 9:50So I'm going to kick this off. I want to
  261. 9:51be super clear about what's going on.
  262. 9:53And importantly, this conversation we're
  263. 9:55having is getting added to the context
  264. 9:58window. Right? So, check this out. This
  265. 9:59is exactly what I wanted to see. Right?
  266. 10:01Scroll up. You can see output dur right
  267. 10:03there. However, that gets defined in the
  268. 10:06infinite prompt, right? As a variable.
  269. 10:08We then are getting our own separate
  270. 10:11versions. And this is exactly what I
  271. 10:13want to see. I'm just proving things out
  272. 10:15uh with cloud code, right? I'm planning
  273. 10:17with cloud code with powerful
  274. 10:19intelligence. all my abilities, all your
  275. 10:21abilities, they're now augmented by
  276. 10:24cloud code, by Agenta coding tools, and
  277. 10:26by these models. So, make sure you can
  278. 10:28tap into it, right? Planning mode really
  279. 10:31helps us tap into this. So, let's go
  280. 10:32ahead and kick this off. I'm going to
  281. 10:34get out of plan mode. So, I'm going to
  282. 10:35run shift tab, shift tab. I want to get
  283. 10:38into autoaccept yolo mode. And now we'll
  284. 10:41let cloud code rip build the v4 spec
  285. 10:44mirror v3. and be clear about the
  286. 10:47improvements
  287. 10:49we are making. And I'll throw in a think
  288. 10:51hard to trigger the reasoning model
  289. 10:54here. This is awesome. So, we're going
  290. 10:55to get our new V4 spec out of this.
  291. 10:58Okay. So, the important thing here is
  292. 11:00that we're having Cloud Code write a new
  293. 11:03plan. There's a big problem with if we
  294. 11:05just said, you know, build out write the
  295. 11:07code for this, right? We can't see the
  296. 11:09full plan without consistently prompting
  297. 11:11Claude code. And remember, agents are
  298. 11:13ephemeral. What does that mean? They're
  299. 11:15temporary. They're stateless. Their
  300. 11:17context window will be blown away. So,
  301. 11:19we can improve on plan mode by having it
  302. 11:22write out the full plan into the specs
  303. 11:24directory, one of the three essential
  304. 11:26directories for agentic coding. Okay.
  305. 11:29And then we can once this plan is
  306. 11:31written here, we can review it, we can
  307. 11:34iterate on it, we can tweak it, and we
  308. 11:36can commit it to our repository, which
  309. 11:38has many downstream positive side
  310. 11:40effects. We'll dig into future videos.
  311. 11:42Make sure you're subscribed so you don't
  312. 11:43miss out on actionable agentic coding
  313. 11:46ideas like this. As you can see, week
  314. 11:48after week, there's a lot of value we
  315. 11:51can create with powerful agentic coding
  316. 11:53tools like cloud code. So, we're going
  317. 11:54to get the spec here written. It's
  318. 11:56activated the reasoning model. We use
  319. 11:57the think hard information dense keyword
  320. 12:00encoded by the cloud code team. Where
  321. 12:02was that? Let me search for that. Up
  322. 12:03here. Yeah, right here. Right. I said
  323. 12:05think hard and this triggers the
  324. 12:08reasoning model. Right now, Claude 4
  325. 12:10Opus is running its thinking process,
  326. 12:13making sure that everything's super
  327. 12:15clear. You can see very importantly,
  328. 12:16we're mirroring again another IDK right
  329. 12:19there. We're mirroring the V4 structure,
  330. 12:21but adapt it for the multifile approach.
  331. 12:24This is key. And while it's loading
  332. 12:25here, we can open up, you know, UI V3.
  333. 12:28Again, definitely check out the previous
  334. 12:29video to really understand the infinite
  335. 12:31agentic prompt and what's really going
  336. 12:33on here. But, you know, you can see just
  337. 12:35at a high level what exactly is going
  338. 12:36on, right? We're generating uniquely
  339. 12:38themed UI components that combine
  340. 12:40multiple UIs into one elegant solution.
  341. 12:43So that's what we're doing there. And
  342. 12:44you can see we just got V4 created for
  343. 12:46us here. 360 lines. Let's go ahead and
  344. 12:48open it up and make sure it makes sense.
  345. 12:50Okay. So themed hybrid UI specification
  346. 12:53V4 looks awesome. Builds on V3 with a
  347. 12:56critical architectural improvement,
  348. 12:58separation of concerns, new modular file
  349. 13:00structure. Great. And so I just want to
  350. 13:02quickly look through this plan and I
  351. 13:04want to see this key detail right here.
  352. 13:06output requirements directory structure
  353. 13:08UI hybrid iteration number very
  354. 13:11important right this differentiates it
  355. 13:13from v3 where we just use the iteration
  356. 13:15number from the infinite agentic loop
  357. 13:17and then we just create thehtml here
  358. 13:19we're creating the directory fantastic
  359. 13:22looks great I'm catching cloud code here
  360. 13:24before it makes a mistake I'm going to
  361. 13:25hit escape uh this is one of the you
  362. 13:27know great reasons to see the thinking
  363. 13:29of your model it was about to create an
  364. 13:31updated infinite command which we don't
  365. 13:33need at all this is good I'm going to
  366. 13:34stop it right here then I'm going to
  367. 13:36boot up another agent here. Open up a
  368. 13:38new cloud code agent and we're just
  369. 13:40going to kick this off. Right? So, this
  370. 13:41plan looks good. It has all the details
  371. 13:44from V3, right? We just built on the V3
  372. 13:47spec. And so, now we're going to kick
  373. 13:48off this infinite agentic workflow with
  374. 13:50our new plan. Just to kind of hop back
  375. 13:52into that idea from last week, this is
  376. 13:55the power of having higher order
  377. 13:57prompts. What does that mean? It's just
  378. 13:59like higher order functions where in
  379. 14:01certain programming languages, you can
  380. 14:03pass a function into a function. We're
  381. 14:05doing that same thing here with auntic
  382. 14:07coding and with prompt engineer. We're
  383. 14:09passing a prompt into a prompt, right?
  384. 14:13The infinite prompt takes a prompt as a
  385. 14:15parameter that it can operate in. Okay,
  386. 14:18there's a ton of value here. We're just
  387. 14:20scratching the surface of what can be
  388. 14:21done with higher order prompts. More on
  389. 14:23that later. Again, make sure to
  390. 14:25subscribe so you don't miss that. Let's
  391. 14:26go ahead and kick this off. Right, so
  392. 14:27there's V3. Let's close that and let's
  393. 14:29kick off V4 in our new Cloud Code
  394. 14:32instance. Check the model. We always
  395. 14:34like that big beefy opus model. And
  396. 14:37let's go ahead and kick this off, right?
  397. 14:39So slash um infinite. This is a custom
  398. 14:42slash command inside of the directory.
  399. 14:43We'll hit tab. And now we need a couple
  400. 14:45of arguments. So we can go ahead and
  401. 14:47sync these up. We need the spec file v4.
  402. 14:52We need our output directory source
  403. 14:55group. And then we need count. How many
  404. 14:58iterations are we going to run? Uh let's
  405. 15:00just fire off 10. We're going to have 10
  406. 15:01new versions here created. Let's go
  407. 15:02ahead and fire this off. Now our
  408. 15:03infinite agentic loop is going to read
  409. 15:06the spec output to the source group
  410. 15:09based on whatever is inside of our spec
  411. 15:11and it's going to create 10 versions.
  412. 15:12Let's go ahead and let this rip. And
  413. 15:14we're doing all this thanks to planning,
  414. 15:17right? This is all thanks to plans. We
  415. 15:19have plans on plans, right?
  416. 15:20Custom/comands. Then we have our specs,
  417. 15:23right? Another prompt. We had, you know,
  418. 15:24cloud code in plan mode, right?
  419. 15:27Thinking, building up context. So much
  420. 15:30of what we do as engineers in the
  421. 15:32generative AI age in this kind of second
  422. 15:34phase is all about building up
  423. 15:37information sets and organizing
  424. 15:38information for our agents and really
  425. 15:41understanding what our agents can do.
  426. 15:43There we go. You can see we have two
  427. 15:45waves. We're going to run in two
  428. 15:46batches. Wave one is going to create
  429. 15:48five agents. Wave two is going to create
  430. 15:49five agents as well. You can see their
  431. 15:51themes. And now cloud code and opus mode
  432. 15:54is going to kick off five agents in
  433. 15:57parallel. Again, once you finish this
  434. 15:59video, check out the previous one to
  435. 16:01really understand the infinite agentic
  436. 16:04loop prompt. Here's the magical moment
  437. 16:06we've all been waiting for. We're going
  438. 16:08to get five parallel agents kicked off
  439. 16:11and they're going to start working for
  440. 16:13us. There we go.
  441. 16:18So, you can see we have that new
  442. 16:19directory and we also have those five
  443. 16:22new agents running in parallel here.
  444. 16:24Let's go and just open all these
  445. 16:26directories up preemptively and we're
  446. 16:28going to be able to watch all of these
  447. 16:30parallel cloud code agents work together
  448. 16:33to accomplish every version of this
  449. 16:36task. Okay? And so if somehow you've
  450. 16:38been watching this channel and you
  451. 16:40haven't been convinced of the importance
  452. 16:42of the fundamental principle of AI
  453. 16:45coding, great planning is great
  454. 16:47prompting. You should absolutely be
  455. 16:49convinced now. Okay. a leading AI lab
  456. 16:52ghost launched and used the tab key for
  457. 16:56the feature for quick access. It might
  458. 16:58seem like I'm micro analyzing, but this
  459. 17:00is what great engineering is, right?
  460. 17:02Every detail matters. Every product
  461. 17:04decision matters and every product
  462. 17:06decision says something. All right, so
  463. 17:08you can see five agents running in
  464. 17:10parallel doing their own isolated pieces
  465. 17:13of work based on our plans. um you know
  466. 17:16principal ad coding members and
  467. 17:17subscribers of the channel you've
  468. 17:19already known about planning and
  469. 17:20architect editor prompt chains and you
  470. 17:23know that key principle that the plan is
  471. 17:26the prompt right we've known this for
  472. 17:27some time now you already have this
  473. 17:29deployed into production you're already
  474. 17:31using this in cloud code every single
  475. 17:33day this is just a new you know mode of
  476. 17:36operating for us as engineers now right
  477. 17:38but if you're just hearing about this
  478. 17:39principle of a coding the plan is the
  479. 17:41prompt check out principled AI coding to
  480. 17:44learn the other Seven principles of AI
  481. 17:46coding link in the description. This is
  482. 17:48fantastic, right? This is working for us
  483. 17:50on our behalf. We have five agents here.
  484. 17:52Everyone's up to two tool uses. It's
  485. 17:54probably going to be three each and then
  486. 17:56we'll be able to dive in and actually
  487. 17:57look at what's going on here. Okay, so
  488. 17:59it looks like hybrid v2 is coming around
  489. 18:02the corner here. This is exactly what
  490. 18:04we've detailed in our plan. And thanks
  491. 18:06to Cloud Code in plan mode here, right,
  492. 18:09shift tab, shift tab, uh, we were able
  493. 18:11to very quickly load up its context
  494. 18:13window with a couple key details, right?
  495. 18:16Very concise prompts. You don't have to
  496. 18:18type a lot. You don't have to do a lot
  497. 18:20with these great tools, but you do need
  498. 18:22to know exactly what to say and what to
  499. 18:24ask. We focus on these key ideas of
  500. 18:26writing concise prompts on the channel
  501. 18:28all the time. All right, so it looks
  502. 18:30like all of our versions are getting
  503. 18:31filled out here. You can see task two
  504. 18:33complete, task three complete. Both
  505. 18:36agents had four tool uses each. They
  506. 18:38both consumed about, you know, 40k of
  507. 18:40Opus tokens. That means there's a lot of
  508. 18:43value in these tokens, right? Uh 3
  509. 18:45minutes each. Not too bad. All right.
  510. 18:47And all of our other tasks are now
  511. 18:48complete. These should all be, you know,
  512. 18:50ready for us to operate in. Now, our
  513. 18:52agent, you know, running about 400
  514. 18:54seconds here. 5.2K tokens on the top
  515. 18:56level agent. Um and now we're going to
  516. 18:58kick off wave two. So, I'm going to stop
  517. 19:00it here. I I want these tokens for some
  518. 19:03other engineering work. But you can see
  519. 19:04it's going to kick off wave two here.
  520. 19:06I'm going to go ahead and stop it here.
  521. 19:07And let's just go ahead and look at our
  522. 19:08results here. Right. So I can type um
  523. 19:11Chrome. Let's open up one of these
  524. 19:13files. Copy the reference with command
  525. 19:16shift R. And I'll just paste that here.
  526. 19:18This is going to open up our brand new
  527. 19:20index script style. Still with that key
  528. 19:23prompt of generating interesting new UI.
  529. 19:25So let's go ahead and kick this off and
  530. 19:27check this out. Right. Uh nature's
  531. 19:28answers. Uh plant your search seed. Very
  532. 19:31interesting. All right. So, we have this
  533. 19:32interesting search bar here. We have a
  534. 19:34filter. Interesting UI there. Let's go
  535. 19:35on to number two. And remember, these
  536. 19:37were all generated in parallel. So, this
  537. 19:39is another interesting one that's going
  538. 19:41to validate as we type. And phone
  539. 19:43number, credit card. You see we have
  540. 19:45complete progress at the bottom. This is
  541. 19:47a really clean minimalist UI. Looks
  542. 19:49great. Let's look at V3. Okay. And V4.
  543. 19:53Okay. Very cool. We have like a music
  544. 19:55player in this kind of retro neon style.
  545. 19:59Okay. So there's more versions here. You
  546. 20:01get the point. Um the prompt or should I
  547. 20:03say the prompts and this prompt system
  548. 20:06uh that we have with the infinite
  549. 20:07agentic loop. It's doing all this work
  550. 20:09for us. This is very powerful. This is
  551. 20:11all thanks to planning as a you know key
  552. 20:15engineering concept that's been embedded
  553. 20:17inside of Cloud Code, the best agentic
  554. 20:20coding tool in the game right now. This
  555. 20:22codebase is going to be available to you
  556. 20:24in the description just as last week.
  557. 20:26I'm going to add this new spec here. And
  558. 20:28I'll leave source group in here as well.
  559. 20:30Planning is a key element of software
  560. 20:32engineering. First, we think, then we
  561. 20:34plan, and then we build. All with our
  562. 20:38agentic coding tools. There's a tectonic
  563. 20:41shift happening in the engineering world
  564. 20:42right now. The winning engineers and
  565. 20:44teams are setting up their code bases
  566. 20:46and tools, not for themselves, but for
  567. 20:49their agents. You know where to find me
  568. 20:51every single Monday. Stay focused and
  569. 20:54keep building.

About this transcript

This page contains the full transcript of Yup, Claude Code Plan Mode is here: The Senior Engineer's Workflow by IndyDevDan, generated from the public captions YouTube serves with the video. The transcript has 3,945 words across 569 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 translate the transcript into 100+ languages.

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, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.