WEBVTT 00:00:00.000 --> 00:00:03.439 Have you ever given an AI a complex coding task? 00:00:03.980 --> 00:00:07.780 You press enter and you just sit there. You watch 00:00:07.780 --> 00:00:10.460 your terminal freeze up while you visibly age. 00:00:10.660 --> 00:00:13.259 Yeah, it's a very specific type of modern agony. 00:00:13.480 --> 00:00:17.230 It really is. Well, welcome to the Deep Dive. 00:00:17.309 --> 00:00:20.190 Today, we're exploring a massive upgrade to Cloud 00:00:20.190 --> 00:00:23.129 Code. It's called Ultraplan. Right. It's a completely 00:00:23.129 --> 00:00:26.050 quiet update, but, you know, a huge deal. Exactly. 00:00:26.550 --> 00:00:29.289 We're looking at how Anthropic rewired the coding 00:00:29.289 --> 00:00:31.750 process. They're taking us out of that agonizing 00:00:31.750 --> 00:00:35.649 local queue. They moved us to a parallel cloud 00:00:35.649 --> 00:00:37.789 architecture. We're going to examine the hidden 00:00:37.789 --> 00:00:39.950 compute costs behind this, too. And we'll look 00:00:39.950 --> 00:00:41.810 at a real -world dashboard test. Plus, we have 00:00:41.810 --> 00:00:43.909 to cover the crucial limitations you must know. 00:00:44.990 --> 00:00:47.090 I got to admit something first, though. I still 00:00:47.090 --> 00:00:49.070 wrestle with the anxiety of a frozen terminal 00:00:49.070 --> 00:00:51.670 myself. I just sit there wondering if my whole 00:00:51.670 --> 00:00:53.770 system just crashed. Oh, absolutely. You just 00:00:53.770 --> 00:00:56.049 stare at that blinking cursor. You're entirely 00:00:56.049 --> 00:00:57.929 locked out of your own machine. You just hope 00:00:57.929 --> 00:00:59.549 the AI is actually doing something productive. 00:00:59.829 --> 00:01:02.530 Exactly. You feel totally helpless. So let's 00:01:02.530 --> 00:01:05.750 unpack why Ultraplan even exists. It all comes 00:01:05.750 --> 00:01:08.409 down to the bottleneck of local planning. Right. 00:01:08.469 --> 00:01:11.189 So look at standard local planning first. It 00:01:11.189 --> 00:01:14.959 uses one linear agent. That single program. works 00:01:14.959 --> 00:01:18.840 step by step inside your terminal. It just types 00:01:18.840 --> 00:01:20.420 out its internal thought process. It's doing 00:01:20.420 --> 00:01:22.120 this right in front of you. Yeah, and that's 00:01:22.120 --> 00:01:24.540 a major problem for complex builds. It produces 00:01:24.540 --> 00:01:27.560 a massive wall of scrolling text. You can't easily 00:01:27.560 --> 00:01:29.840 review that text at all. You certainly can't 00:01:29.840 --> 00:01:31.659 change its mind before it starts writing code. 00:01:31.939 --> 00:01:34.560 Because by the time it finishes, you're already 00:01:34.560 --> 00:01:37.640 committed. The train has left the station. Exactly. 00:01:37.760 --> 00:01:40.299 I mean, if you make a bad structural choice early 00:01:40.299 --> 00:01:43.620 on, you pause. You step back. A local AI agent 00:01:43.620 --> 00:01:45.280 doesn't do that. It just keeps going forward. 00:01:45.319 --> 00:01:47.200 It tries to course -correct mid -build, right? 00:01:47.239 --> 00:01:49.260 Yeah, and that's exactly when your project gets 00:01:49.260 --> 00:01:52.280 incredibly messy. It writes spaghetti code to 00:01:52.280 --> 00:01:54.879 fix its earlier bad assumptions. It's incredibly 00:01:54.879 --> 00:01:56.719 frustrating to watch. And I guess that brings 00:01:56.719 --> 00:01:59.799 us to Ultraplan. Anthropic decided to ship the 00:01:59.799 --> 00:02:03.719 entire planning job to their cloud. Yes, they 00:02:03.719 --> 00:02:05.700 used their cloud container runtime for the heavy 00:02:05.700 --> 00:02:08.479 lifting. It introduces a multi -agent architecture, 00:02:09.000 --> 00:02:11.680 meaning several AI programs working together 00:02:11.680 --> 00:02:15.300 simultaneously to solve one problem. Start on. 00:02:15.439 --> 00:02:18.479 Instead of one agent struggling alone, the cloud 00:02:18.479 --> 00:02:21.460 spins up a team. It creates three parallel exploration 00:02:21.460 --> 00:02:25.139 agents, and it adds one dedicated critique agent. 00:02:25.379 --> 00:02:29.000 And they all use Opus 4 .6. That's Anthropic's 00:02:29.000 --> 00:02:31.900 most powerful AI model available right now. Right. 00:02:31.960 --> 00:02:34.879 Beat. Think about the philosophical difference 00:02:34.879 --> 00:02:37.270 here. yeah local planning is like having one 00:02:37.270 --> 00:02:40.129 really smart person but they're at a cramped 00:02:40.129 --> 00:02:43.210 messy desk sweating over a notebook by themselves 00:02:43.210 --> 00:02:46.069 getting total tunnel vision but ultraplan is 00:02:46.069 --> 00:02:48.229 like a collaborative working session they get 00:02:48.229 --> 00:02:50.889 a spacious room with a massive whiteboard they 00:02:50.889 --> 00:02:53.409 explore multiple architectural paths at the exact 00:02:53.409 --> 00:02:57.479 same time whoa Beat. Imagine three Opus 4 .6 00:02:57.479 --> 00:02:59.840 agents exploring your entire code base at the 00:02:59.840 --> 00:03:02.159 exact same time. It's staggering. The critique 00:03:02.159 --> 00:03:04.180 agent acts like a senior developer then. It steps 00:03:04.180 --> 00:03:06.280 in and tightens the whole plan up. And the final 00:03:06.280 --> 00:03:08.560 output completely changes your developer experience. 00:03:08.860 --> 00:03:10.979 Instead of a chaotic terminal output, you get 00:03:10.979 --> 00:03:13.810 an organized web document. Yeah, hosted on Anthropix 00:03:13.810 --> 00:03:15.729 interface. Yeah. You see the project context 00:03:15.729 --> 00:03:18.069 right at the top. You see a list of files it 00:03:18.069 --> 00:03:20.569 intends to create. It's a proper technical brief. 00:03:20.789 --> 00:03:23.030 It even shows the final verification steps it'll 00:03:23.030 --> 00:03:25.389 run. Sometimes it includes structural diagrams 00:03:25.389 --> 00:03:28.169 of the proposed architecture too. And here's 00:03:28.169 --> 00:03:30.509 the absolute best part. You're no longer just 00:03:30.509 --> 00:03:33.030 a passive observer. Right. You can actually leave 00:03:33.030 --> 00:03:37.250 inline comments on this web document. You can 00:03:37.250 --> 00:03:40.349 even use emoji reactions on specific parts. You 00:03:40.349 --> 00:03:42.509 do all of this before you approve anything. It 00:03:42.509 --> 00:03:45.090 separates the thinking phase from the doing phase. 00:03:45.389 --> 00:03:47.530 So does this mean the cloud actually writes the 00:03:47.530 --> 00:03:49.610 final code for you? No, it just builds the blueprint. 00:03:49.849 --> 00:03:52.509 The code generation still happens locally. Got 00:03:52.509 --> 00:03:55.229 it. So the cloud plans, but your local terminal 00:03:55.229 --> 00:03:57.590 builds it. Precisely. It's a much safer way to 00:03:57.590 --> 00:03:59.569 build software. It feels like a much more mature 00:03:59.569 --> 00:04:02.449 workflow. But how do we actually invite this 00:04:02.449 --> 00:04:05.270 multi -agent team to our project? It requires 00:04:05.270 --> 00:04:07.669 a very specific approach. You must explicitly 00:04:07.669 --> 00:04:11.310 use the slash command forward slash Ultraplan. 00:04:11.469 --> 00:04:13.629 Or I guess you can just include the word Ultraplan 00:04:13.629 --> 00:04:16.069 in your prompt. Right, but here's the one strict 00:04:16.069 --> 00:04:19.230 rule you cannot ignore. It only works from the 00:04:19.230 --> 00:04:22.839 CLI. Meaning the text -based command line interface 00:04:22.839 --> 00:04:26.160 on your computer. Yes. If you try to run this 00:04:26.160 --> 00:04:29.720 command inside the Claw desktop app. Or the VS 00:04:29.720 --> 00:04:31.959 Code extension. Right. And nothing special will 00:04:31.959 --> 00:04:35.220 happen. It just quietly defaults to normal sequential 00:04:35.220 --> 00:04:38.600 local planning. That feels like a huge trap for 00:04:38.600 --> 00:04:40.800 casual users. People probably think the whole 00:04:40.800 --> 00:04:43.180 feature is broken. Oh, it's a very common mistake. 00:04:43.379 --> 00:04:46.019 You've got to open Claude code directly through 00:04:46.019 --> 00:04:48.160 your system's terminal. So if you do trigger 00:04:48.160 --> 00:04:50.079 it right from the terminal, what's the workflow? 00:04:50.379 --> 00:04:52.360 You type your prompt. You trigger the command. 00:04:52.899 --> 00:04:55.079 Claude immediately ships the job out to the web. 00:04:55.370 --> 00:04:57.069 And your local terminal doesn't freeze up at 00:04:57.069 --> 00:04:59.069 all. Not at all. Your terminal stays completely 00:04:59.069 --> 00:05:01.269 free. You can keep running your local server. 00:05:01.430 --> 00:05:03.750 You can check your Git logs. Meanwhile, you review 00:05:03.750 --> 00:05:06.350 the AI's plan on the anthropic web interface. 00:05:06.990 --> 00:05:09.250 You read through the context and leave your critical 00:05:09.250 --> 00:05:12.209 comments. You essentially act as the final gatekeeper. 00:05:12.629 --> 00:05:15.189 Then, when it looks perfect, you click approve. 00:05:15.850 --> 00:05:19.350 And boom! The approved plan teleports right back 00:05:19.350 --> 00:05:21.670 to your terminal. It's ready to execute. Exactly. 00:05:21.670 --> 00:05:24.730 It completely removes that agonizing waiting 00:05:24.730 --> 00:05:26.850 game. What happens to my local environment while 00:05:26.850 --> 00:05:28.629 the cloud is doing all this heavy lifting? It 00:05:28.629 --> 00:05:31.389 stays completely unlocked and usable so you can 00:05:31.389 --> 00:05:33.730 keep working. Terminal stays free. You review 00:05:33.730 --> 00:05:36.829 everything separately on the web. Spot on. It 00:05:36.829 --> 00:05:40.810 changes everything about the daily grind. We're 00:05:40.810 --> 00:05:43.870 going to take a quick break. Sponsor. All right. 00:05:43.910 --> 00:05:46.259 And we're back. So we have our terminal back, 00:05:46.439 --> 00:05:48.720 which feels great. Definitely. But does this 00:05:48.720 --> 00:05:51.720 actually save time in the real world? Let's look 00:05:51.720 --> 00:05:54.300 at a side -by -side test from our sources. Yeah, 00:05:54.360 --> 00:05:56.620 they ran a test to build out a complex dashboard. 00:05:57.399 --> 00:05:59.720 And it wasn't just a simple script. It needed 00:05:59.720 --> 00:06:02.879 MRR tracking. Meaning a metric tracking your 00:06:02.879 --> 00:06:05.519 monthly recurring revenue. Right. It also needed 00:06:05.519 --> 00:06:08.019 ARR tracking. A metric tracking your predictable 00:06:08.019 --> 00:06:11.279 annual recurring revenue. Exactly. Plus, it required 00:06:11.279 --> 00:06:13.920 a dynamic churn rate calculation. That's the 00:06:13.920 --> 00:06:15.959 percentage of paying customers who cancel their 00:06:15.959 --> 00:06:18.560 service. Yes. And it needed functional light 00:06:18.560 --> 00:06:21.860 and dark mode UI toggles. So a very standard 00:06:21.860 --> 00:06:25.680 but heavy full stack build. Both local and Ultraplan 00:06:25.680 --> 00:06:28.639 modes started at the exact same time. The results 00:06:28.639 --> 00:06:31.579 were genuinely surprising to read. The Ultraplan 00:06:31.579 --> 00:06:34.660 side was incredibly fast. It generated the complex 00:06:34.660 --> 00:06:37.379 plan in the cloud. It got human approval. It 00:06:37.379 --> 00:06:39.240 teleported the blueprint back to the terminal. 00:06:39.459 --> 00:06:41.920 And it actually finished building the working 00:06:41.920 --> 00:06:44.680 dashboard locally. Wow. It finished the entire 00:06:44.680 --> 00:06:47.100 project before the standard local planner even 00:06:47.100 --> 00:06:49.740 finished its initial planning phase. Two sec 00:06:49.740 --> 00:06:53.160 silence. Better planning literally equals faster 00:06:53.160 --> 00:06:55.639 execution. It creates a much clearer roadmap 00:06:55.639 --> 00:06:57.879 from the start. Claude wanders significantly 00:06:57.879 --> 00:07:00.879 less. It doesn't panic and rebuild sections mid 00:07:00.879 --> 00:07:03.779 -task. It compounds the time advantage beautifully. 00:07:04.139 --> 00:07:06.339 But we really have to talk about the token tradeoff 00:07:06.339 --> 00:07:07.920 here. Right. There's something in the source 00:07:07.920 --> 00:07:10.620 material about token usage. At first glance, 00:07:10.699 --> 00:07:13.079 this new workflow looks incredibly efficient. 00:07:13.399 --> 00:07:16.040 It does. During that dashboard test, standard 00:07:16.040 --> 00:07:18.899 local planning chewed through about 131 ,000 00:07:18.899 --> 00:07:21.459 local tokens. That's a lot of highly visible 00:07:21.459 --> 00:07:24.490 local compute power. But the Ultraplan local 00:07:24.490 --> 00:07:28.389 execution used only 82 ,000 tokens. Yeah, and 00:07:28.389 --> 00:07:30.529 that's a very big illusion. It looks cheaper 00:07:30.529 --> 00:07:33.230 on your terminal screen, but it's not. Ultraplan 00:07:33.230 --> 00:07:35.449 simply shifts the heavy compute burden to the 00:07:35.449 --> 00:07:38.050 cloud. You just don't see the massive token burn 00:07:38.050 --> 00:07:40.569 on your local CLI anymore. Right. Think about 00:07:40.569 --> 00:07:42.550 what's happening under the hood. You have three 00:07:42.550 --> 00:07:45.990 instances of Opus 4 .6 exploring your code base. 00:07:46.459 --> 00:07:49.220 Plus, a critique agent evaluating their work. 00:07:49.360 --> 00:07:51.800 That's an incredibly heavy, expensive compute 00:07:51.800 --> 00:07:54.339 load. The planning step consumes significant 00:07:54.339 --> 00:07:56.819 resources. They're just hidden from your view. 00:07:56.920 --> 00:07:59.959 So the 82 ,000 tokens is just the typing cost. 00:08:00.300 --> 00:08:03.500 But compute is definitely not free. If Anthropic 00:08:03.500 --> 00:08:06.160 is eating the cost of three Opus agents, how 00:08:06.160 --> 00:08:08.319 does that work financially? It works because 00:08:08.319 --> 00:08:10.759 they wall it off behind a hard paywall. You can't 00:08:10.759 --> 00:08:13.560 use standard API billing for Ultraplan. Oh, so 00:08:13.560 --> 00:08:15.620 if you try to trigger it through a pay -as -you 00:08:15.620 --> 00:08:18.480 -go API setup, it just rejects it. Correct. It 00:08:18.480 --> 00:08:21.300 strictly requires a Cloud Pro, Max, or Team subscription. 00:08:21.579 --> 00:08:24.779 So are we actually saving tokens or just hiding 00:08:24.779 --> 00:08:27.199 the massive cost off screen? We're absolutely 00:08:27.199 --> 00:08:28.980 just hiding the cost inside your monthly cloud 00:08:28.980 --> 00:08:31.319 subscription. We hide the heavy token cost in 00:08:31.319 --> 00:08:33.759 the cloud subscription. Exactly. You're paying 00:08:33.759 --> 00:08:35.700 for all that thinking in a different way. Okay. 00:08:35.779 --> 00:08:37.860 So before you rely on this tool daily, there 00:08:37.860 --> 00:08:40.179 are strict requirements. There are also weird 00:08:40.179 --> 00:08:42.519 quirks that'll trip you up. We should definitely 00:08:42.519 --> 00:08:45.169 start with the get rule. Your local project must 00:08:45.169 --> 00:08:48.009 be synced to Git or GitHub. Right. The cloud 00:08:48.009 --> 00:08:51.590 planner absolutely cannot inspect a purely local, 00:08:51.789 --> 00:08:54.750 untracked project. If you have a random folder 00:08:54.750 --> 00:08:57.330 of messy code on your desktop, it won't work. 00:08:57.490 --> 00:08:59.769 If you don't push your changes, the cloud planner 00:08:59.769 --> 00:09:02.549 works completely blind. It makes sense from a 00:09:02.549 --> 00:09:04.929 security standpoint, though. The cloud agents 00:09:04.929 --> 00:09:07.370 can't magically access your physical hard drive 00:09:07.370 --> 00:09:09.830 directly. That would be a massive privacy nightmare. 00:09:10.330 --> 00:09:13.629 Git acts as the synchronized, secure middleman. 00:09:14.149 --> 00:09:16.590 Then there's what the source calls the custom 00:09:16.590 --> 00:09:19.929 skills problem. Yeah, the AI makes rapid assumptions. 00:09:20.250 --> 00:09:22.690 The source gave a highly specific example of 00:09:22.690 --> 00:09:25.549 this flaw. A user wanted architectural diagrams 00:09:25.549 --> 00:09:29.250 rendered in an Excalibur style. Meaning a popular 00:09:29.250 --> 00:09:32.049 digital whiteboard tool with a hand -drawn style. 00:09:32.370 --> 00:09:35.309 Right, but Ultraplan just guessed the user's 00:09:35.309 --> 00:09:38.710 intent. It ignored Excalibur entirely and used 00:09:38.710 --> 00:09:41.809 Mermaid instead. Mermaid being a purely text 00:09:41.809 --> 00:09:44.850 -based tool used for generating technical charts. 00:09:45.110 --> 00:09:47.230 Exactly. So the technical output was functionally 00:09:47.230 --> 00:09:49.870 correct. Yes. But it was absolutely not what 00:09:49.870 --> 00:09:52.350 the user wanted visually. The cloud AI doesn't 00:09:52.350 --> 00:09:55.490 know your internal design standards. It defaults 00:09:55.490 --> 00:09:57.990 to what's computationally easiest for it. But 00:09:57.990 --> 00:10:00.210 the fix is actually very simple. You use that 00:10:00.210 --> 00:10:02.409 interactive web review step. You just leave a 00:10:02.409 --> 00:10:04.889 direct inline comment on the plan. You specify 00:10:04.889 --> 00:10:07.470 your exact internal tools. Right. You type a 00:10:07.470 --> 00:10:10.129 comment like, please use... Scala draw for all 00:10:10.129 --> 00:10:12.549 database diagrams. Let's outline the other weird 00:10:12.549 --> 00:10:15.830 quirks. First, there's a hard compute cap. Yes. 00:10:16.269 --> 00:10:19.409 You face a strict 30 -minute cloud compute cap 00:10:19.409 --> 00:10:22.149 per planning session. Which, honestly, is plenty 00:10:22.149 --> 00:10:24.590 of time for most standard application features. 00:10:24.809 --> 00:10:27.649 True. But you might encounter random authentication 00:10:27.649 --> 00:10:30.929 errors, too. They pop up occasionally, but they 00:10:30.929 --> 00:10:32.870 usually disappear if you just retry the command. 00:10:33.169 --> 00:10:36.240 Web token transparency is also very low. You 00:10:36.240 --> 00:10:37.899 don't really know how much cloud compute you're 00:10:37.899 --> 00:10:40.059 burning. Honestly, the whole thing still feels 00:10:40.059 --> 00:10:43.000 slightly preview -ish. It's powerful, but rough 00:10:43.000 --> 00:10:45.159 around the edges. There's a great pro tip from 00:10:45.159 --> 00:10:47.899 the source here. Keep your planning back and 00:10:47.899 --> 00:10:50.259 forth in the cloud environment. But you should 00:10:50.259 --> 00:10:53.399 open a clean local session strictly for the execution 00:10:53.399 --> 00:10:55.620 phase. Right, so the context window doesn't get 00:10:55.620 --> 00:10:58.100 cluttered with discarded ideas. If it remembers 00:10:58.100 --> 00:11:00.720 all the bad ideas, it might weave them into the 00:11:00.720 --> 00:11:03.139 final code. Definitely. Why does a cloud planner 00:11:03.139 --> 00:11:06.000 care if my code is on GitHub? Because the cloud 00:11:06.000 --> 00:11:09.419 agents need a secure bridge to actually see your 00:11:09.419 --> 00:11:12.000 files. The cloud agents need a bridge to see 00:11:12.000 --> 00:11:14.759 your local files. Precisely. No bridge means 00:11:14.759 --> 00:11:17.600 no multi -agent plan. With all those quirks in 00:11:17.600 --> 00:11:20.279 mind, should you use it for every prompt? Absolutely 00:11:20.279 --> 00:11:22.720 not. You really need to know when to skip this 00:11:22.720 --> 00:11:25.299 workflow entirely. Let's break down those specific 00:11:25.299 --> 00:11:29.000 use cases. When do you completely ignore Ultraplan? 00:11:29.240 --> 00:11:32.320 You must skip Ultraplan for small, isolated bug 00:11:32.320 --> 00:11:35.860 fixes. Don't use it for simple spelling typos. 00:11:35.879 --> 00:11:39.100 Don't use it for quick localized CSS style changes 00:11:39.100 --> 00:11:41.940 either. It's way too heavy of a process for that. 00:11:42.080 --> 00:11:44.080 You don't need a multi -agent whiteboard session 00:11:44.080 --> 00:11:46.620 to change a button color. It's completely unnecessary 00:11:46.620 --> 00:11:50.090 overhead. But you should absolutely use Ultraplan 00:11:50.090 --> 00:11:52.970 for deep structural changes. Use it for complex, 00:11:52.990 --> 00:11:55.470 multi -file features. Use it when you're doing 00:11:55.470 --> 00:11:57.789 full application builds from scratch. Basically, 00:11:57.950 --> 00:12:00.370 when the planning quality directly impacts the 00:12:00.370 --> 00:12:02.789 execution quality. If the architecture matters, 00:12:03.190 --> 00:12:06.059 use the cloud. I really love the Abraham Lincoln 00:12:06.059 --> 00:12:08.159 quote they included in the source material. Right. 00:12:08.220 --> 00:12:10.919 Give me six hours to chop down a tree and I will 00:12:10.919 --> 00:12:13.899 spend the first four sharpening the axe. It perfectly 00:12:13.899 --> 00:12:17.440 describes this new AI coding workflow. Jumping 00:12:17.440 --> 00:12:20.100 straight into rapid execution is no longer a 00:12:20.100 --> 00:12:23.080 viable strategy for complex builds. The AI types 00:12:23.080 --> 00:12:25.879 fast, but if it types the wrong thing, you just 00:12:25.879 --> 00:12:28.419 fail faster. The architectural plan needs to 00:12:28.419 --> 00:12:30.940 be flawless before you strike the tree. Is there 00:12:30.940 --> 00:12:33.580 any real danger in using Ultraplan for a tiny 00:12:33.580 --> 00:12:36.960 bug fix? Not a danger, but it's overkill and 00:12:36.960 --> 00:12:39.379 wastes your limited cloud compute time. It's 00:12:39.379 --> 00:12:41.659 overkill and wastes your 30 -minute cloud compute 00:12:41.659 --> 00:12:44.759 time. Exactly. Save the heavy multi -agent lifting 00:12:44.759 --> 00:12:47.419 for the truly heavy engineering projects. Two 00:12:47.419 --> 00:12:50.059 sec silence. Let's step back and look at the 00:12:50.059 --> 00:12:53.559 big idea here. The future of coding agents isn't 00:12:53.559 --> 00:12:56.559 just about typing syntax faster. No, it's moving 00:12:56.559 --> 00:12:59.600 rapidly toward deliberate, highly structured 00:12:59.600 --> 00:13:02.820 planning workflows. It's deeply about mandatory 00:13:02.820 --> 00:13:06.299 human review occurring before any execution happens. 00:13:06.519 --> 00:13:09.080 The process is shifting. It's moving away from 00:13:09.080 --> 00:13:12.139 blind experimentation. It's moving toward cleanly 00:13:12.139 --> 00:13:14.850 separated thinking and doing. The AI swarm does 00:13:14.850 --> 00:13:17.149 the heavy architectural thinking. You step in 00:13:17.149 --> 00:13:19.169 and do the high -level directing. Then the local 00:13:19.169 --> 00:13:21.909 machine does the tedious doing. It's a profound 00:13:21.909 --> 00:13:24.840 shift in software engineering. Deep. So here's 00:13:24.840 --> 00:13:26.419 a provocative thought for you to take away and 00:13:26.419 --> 00:13:30.320 mull over. If AI is evolving into a team of planners 00:13:30.320 --> 00:13:33.759 rather than just a fast typist, how does your 00:13:33.759 --> 00:13:36.200 role change? It's a deeply important question 00:13:36.200 --> 00:13:38.259 about our fundamental identity as developers. 00:13:38.600 --> 00:13:40.740 Are you even a coder anymore or are you becoming 00:13:40.740 --> 00:13:42.720 an art director for software? An art director 00:13:42.720 --> 00:13:45.379 doesn't paint every single brushstroke. They 00:13:45.379 --> 00:13:48.220 set the grand vision. They review the messy drafts, 00:13:48.220 --> 00:13:50.460 make sure your repo is synced, spin up Ultraplan 00:13:50.460 --> 00:13:52.929 and see how it feels to manage the team. It's 00:13:52.929 --> 00:13:54.970 a whole new professional mindset to embrace. 00:13:55.210 --> 00:13:58.029 It really is. Remember, you don't have to just 00:13:58.029 --> 00:13:59.929 sit there and watch your terminal freeze anymore. 00:14:00.210 --> 00:14:02.850 Send the heavy work up to the cloud. Grab a coffee. 00:14:02.990 --> 00:14:05.289 And review the architectural blueprint like a 00:14:05.289 --> 00:14:07.090 true boss. Take care out there.