WEBVTT 00:00:00.000 --> 00:00:03.419 Picture this. You're building a massive Lego 00:00:03.419 --> 00:00:05.820 castle on your desk. Okay. It's got everything. 00:00:06.160 --> 00:00:08.939 Drawbridges, towers, the works. It's your masterpiece. 00:00:09.039 --> 00:00:11.480 I'm with you. But then halfway through, you realize 00:00:11.480 --> 00:00:13.759 you need to build a Lego car. Right. But your 00:00:13.759 --> 00:00:16.000 desk is full. So to build the car, you'd have 00:00:16.000 --> 00:00:18.559 to, what, smash a turret? Or sweep the whole 00:00:18.559 --> 00:00:21.320 castle into a bin just to clear some space. It 00:00:21.320 --> 00:00:23.960 is the worst feeling. You lose all that momentum. 00:00:24.420 --> 00:00:27.379 But what if you just buy second desk? And that's 00:00:27.379 --> 00:00:29.300 it. That's the core of what we're talking about 00:00:29.300 --> 00:00:33.060 today. In the real world, a new desk is expensive. 00:00:33.520 --> 00:00:36.600 In software, it's surprisingly cheap. And yet 00:00:36.600 --> 00:00:38.820 we're all still smashing our castles. We are. 00:00:39.200 --> 00:00:41.689 Welcome to the deep dive. I've been thinking 00:00:41.689 --> 00:00:44.070 a lot about how we interact with these AI tools. 00:00:44.570 --> 00:00:47.070 We tend to treat them like fancy search engines. 00:00:47.130 --> 00:00:50.310 Like a Q &A box? Exactly. When really the source 00:00:50.310 --> 00:00:52.189 material argues we should be treating them like 00:00:52.189 --> 00:00:55.090 junior partners. It's a total mental model shift. 00:00:55.909 --> 00:00:58.409 I see so many developers get frustrated. They're 00:00:58.409 --> 00:01:00.789 staring at code, waiting for the AI, copy -pasting, 00:01:00.969 --> 00:01:03.500 getting errors, and they just say... This is 00:01:03.500 --> 00:01:05.719 slower than me doing it myself. It's interesting 00:01:05.719 --> 00:01:07.879 you say that, because today's guide argues that 00:01:07.879 --> 00:01:10.239 if it feels slower, you're working linearly. 00:01:10.340 --> 00:01:12.959 One thing at a time. And the fix, and this is 00:01:12.959 --> 00:01:15.379 kind of the big theme today, is parallelism. 00:01:15.719 --> 00:01:18.060 Right. Breaking out of that input weight output 00:01:18.060 --> 00:01:20.599 loop. So let's go back to that second desk idea. 00:01:20.760 --> 00:01:22.719 The guide talks about this productivity killer 00:01:22.719 --> 00:01:25.379 called the waiting game. Oh, yeah. This is the 00:01:25.379 --> 00:01:28.560 invisible time sack. You ask the AI for a function, 00:01:28.819 --> 00:01:31.540 takes 30 seconds. You run a test suite, that 00:01:31.540 --> 00:01:34.180 takes a minute. And in that gap. You check Slack. 00:01:34.939 --> 00:01:36.980 You look at Twitter. You totally lose your train 00:01:36.980 --> 00:01:39.239 of thought. It's the context switch that kills 00:01:39.239 --> 00:01:41.939 you, not the weight itself. So the solution is 00:01:41.939 --> 00:01:44.560 parallel work. For, you know, the command line 00:01:44.560 --> 00:01:47.939 power users, the guide brings up Git work trees. 00:01:48.480 --> 00:01:50.140 I've heard of these, but can you break down what 00:01:50.140 --> 00:01:52.939 that actually is? Sure. Think of it like this. 00:01:53.340 --> 00:01:56.079 Your project folder is your one desk. If you 00:01:56.079 --> 00:01:58.379 want to switch from a bug to a feature, you have 00:01:58.379 --> 00:02:00.920 to switch branches. And that changes all the 00:02:00.920 --> 00:02:03.480 files. Their whole desk transforms. But with 00:02:03.480 --> 00:02:05.260 WorkTrees, you can have the same project in two 00:02:05.260 --> 00:02:07.379 different folders at the same time. So I'd have, 00:02:07.379 --> 00:02:10.699 like, folder A for the login feature and folder 00:02:10.699 --> 00:02:13.360 B for a database bug, and they're both open. 00:02:13.780 --> 00:02:16.159 Precisely. You don't close one to open the other. 00:02:16.349 --> 00:02:19.789 You have two desks. You push code to the AI on 00:02:19.789 --> 00:02:22.310 desk one. And while it's thinking, you just swivel 00:02:22.310 --> 00:02:24.530 over to desk two and keep working. OK. But what 00:02:24.530 --> 00:02:26.430 if I'm not living in the terminal? What if I'm 00:02:26.430 --> 00:02:29.009 just using the Claude web interface? Same principle. 00:02:29.469 --> 00:02:31.949 Stop using one long chat for everything. Open 00:02:31.949 --> 00:02:35.229 five tabs. And this part is crucial. Name them. 00:02:35.389 --> 00:02:38.110 Name them what? Log and fix. Database logic. 00:02:38.789 --> 00:02:42.110 Tab three can be stupid questions. Stupid questions? 00:02:42.650 --> 00:02:44.870 Yeah. Everyone needs a safe space. That's where 00:02:44.870 --> 00:02:46.770 you ask, wait, how does a switch statement work 00:02:46.770 --> 00:02:49.229 again? You don't want that basic stuff cluttering 00:02:49.229 --> 00:02:51.990 up your complex architecture discussion. But 00:02:51.990 --> 00:02:53.750 doesn't that, I mean, doesn't splitting your 00:02:53.750 --> 00:02:55.870 focus like that just create chaos? I feel like 00:02:55.870 --> 00:02:58.189 my brain would just melt. It sounds chaotic, 00:02:58.210 --> 00:03:00.949 I know, but it's actually cleaner for the AI. 00:03:01.689 --> 00:03:03.969 You have to remember these models have a context 00:03:03.969 --> 00:03:06.990 window. They can only remember so much. Right. 00:03:07.370 --> 00:03:10.229 If you mix database queries and UI design and 00:03:10.229 --> 00:03:13.650 CSS tweaks in one thread, the AI gets confused. 00:03:14.030 --> 00:03:17.169 It starts making things up, splitting them, keeps 00:03:17.169 --> 00:03:20.530 the AI's memory pure. So it keeps the AI's memory 00:03:20.530 --> 00:03:22.789 distinct and manageable. Exactly. You're acting 00:03:22.789 --> 00:03:25.569 less like a coder and more like a product manager 00:03:25.569 --> 00:03:28.349 running a team of specialists. Being an orchestrator 00:03:28.349 --> 00:03:31.270 means you need a plan, which brings us to the 00:03:31.270 --> 00:03:34.009 next point, the architect mindset. The source 00:03:34.009 --> 00:03:36.389 is pretty harsh here. It calls most of our prompts 00:03:36.389 --> 00:03:39.349 vending machine prompts. Right. Make me a snake 00:03:39.349 --> 00:03:42.030 game. Fix this bug. You put a coin in, you want 00:03:42.030 --> 00:03:44.009 a product out. And then you get mad when the 00:03:44.009 --> 00:03:46.789 code is messy? Of course. But you'd never tell 00:03:46.789 --> 00:03:49.370 a human contractor, build me a house and just 00:03:49.370 --> 00:03:50.969 walk away. You'd end up with a bathroom in the 00:03:50.969 --> 00:03:53.289 kitchen. Exactly. So the fix is what they call 00:03:53.289 --> 00:03:55.889 plan mode. Before you write a single line of 00:03:55.889 --> 00:03:58.590 code, you open a new chat and you say, act as 00:03:58.590 --> 00:04:01.610 a senior software architect. You ask for a blueprint, 00:04:02.030 --> 00:04:05.449 a file structure, libraries, data schemas, security 00:04:05.449 --> 00:04:07.810 risks, all of it. You're front loading the thinking, 00:04:08.229 --> 00:04:10.389 but doesn't that just delay the actual building? 00:04:10.590 --> 00:04:13.710 I want to see code. It delays the start, but 00:04:13.710 --> 00:04:17.310 it massively accelerates the finish because changing 00:04:17.310 --> 00:04:20.589 a text description is free. Changing 500 lines 00:04:20.589 --> 00:04:22.769 of code because you chose the wrong database 00:04:22.769 --> 00:04:25.639 library. That's really expensive. There's a tactic 00:04:25.639 --> 00:04:27.819 in here I really liked getting a second opinion 00:04:27.819 --> 00:04:30.319 on that plan. This is my favorite trick. So you 00:04:30.319 --> 00:04:34.040 get the plan from chat A, then you copy it, open 00:04:34.040 --> 00:04:37.699 a totally new chat B, and you say, hey, I'm thinking 00:04:37.699 --> 00:04:39.959 of building this. critique it, tear it apart. 00:04:40.139 --> 00:04:42.800 It's like an adversarial review. It is. The second 00:04:42.800 --> 00:04:45.420 AI has zero attachment. It's fresh. It'll spot 00:04:45.420 --> 00:04:47.240 things the first one best because it's not, you 00:04:47.240 --> 00:04:49.240 know, high on its own supply. So why is that 00:04:49.240 --> 00:04:51.839 planning phase so non -negotiable? Because entropy 00:04:51.839 --> 00:04:55.399 is the default state of all code. Without a plan, 00:04:55.899 --> 00:04:59.360 complexity just spirals. With a plan, you're 00:04:59.360 --> 00:05:01.379 executing a checklist. Fixing text is cheap. 00:05:01.550 --> 00:05:03.889 Fixing code is expensive. That's the bottom line. 00:05:04.089 --> 00:05:06.029 OK, so we have our parallel desks. We have our 00:05:06.029 --> 00:05:08.509 blueprints. But then you hit the wall of repetition. 00:05:08.910 --> 00:05:10.850 Everyone's felt this. You start a new chat. You 00:05:10.850 --> 00:05:14.089 have to remind the AI. I use Python. I like short 00:05:14.089 --> 00:05:17.209 lines of code. Don't use this library. It's like 00:05:17.209 --> 00:05:20.389 that movie 50 First Dates. Assistant amnesia. 00:05:20.509 --> 00:05:23.370 It's a huge friction point. So the solution here 00:05:23.370 --> 00:05:27.870 is a file called ClaudeEatE .md. How does a simple 00:05:27.870 --> 00:05:30.959 text file fix amnesia? Well, most of these AI 00:05:30.959 --> 00:05:33.459 coding tools can look at your project files now. 00:05:33.920 --> 00:05:36.439 So if you create a file in your main folder called 00:05:36.439 --> 00:05:44.120 Claude EMD... That's the start? Your tech stack? 00:05:44.279 --> 00:05:46.899 Sure. But the real high leverage stuff is lessons 00:05:46.899 --> 00:05:48.540 learned. Give me an example. Okay, let's say 00:05:48.540 --> 00:05:51.199 you spend four hours on a bug where the attainment 00:05:51.199 --> 00:05:53.839 gateway crashes if an address has a special character. 00:05:54.000 --> 00:05:57.540 You fix it. Normally, that knowledge just lives 00:05:57.540 --> 00:06:00.319 in your head. But with this system, you go to 00:06:00.319 --> 00:06:06.300 clodee .md and you write, lesson. Payment API 00:06:06.300 --> 00:06:09.699 fails on special characters. Always sanitize 00:06:09.699 --> 00:06:12.120 input. And the next time I ask the AI to write 00:06:12.120 --> 00:06:14.920 payment code. It reads that file first. It knows 00:06:14.920 --> 00:06:17.060 not to make that mistake. And if it slips up, 00:06:17.379 --> 00:06:19.100 you don't just fix the code. You actually scold 00:06:19.100 --> 00:06:22.279 it. You say, you ignore the instructions. Update 00:06:22.279 --> 00:06:25.019 clodee .md so you never do this again. So how 00:06:25.019 --> 00:06:26.920 does this change your relationship with the tool? 00:06:27.180 --> 00:06:30.160 It stops being a fresh chat every time and starts 00:06:30.160 --> 00:06:32.319 kind of reading your mind. It accumulates wisdom. 00:06:32.500 --> 00:06:34.939 It's moving from a blank slate to institutional 00:06:34.939 --> 00:06:37.500 memory. Exactly. The AI becomes a colleague that's 00:06:37.500 --> 00:06:39.279 been on the project as long as you have. Let's 00:06:39.279 --> 00:06:41.600 move to automation. The guide mentions custom 00:06:41.600 --> 00:06:44.540 skills or magic spells. This sounds pretty complex. 00:06:44.620 --> 00:06:46.480 It's actually not. We all have these boring, 00:06:46.480 --> 00:06:49.180 repetitive tasks. Check for unused variables, 00:06:49.459 --> 00:06:51.620 summarize what the team did on Slack. The idea 00:06:51.620 --> 00:06:53.480 is just to turn those into shortcuts. Like tech 00:06:53.480 --> 00:06:56.160 death. Right. In a really sophisticated setup, 00:06:56.699 --> 00:06:59.019 you can program a command like techdet to run 00:06:59.019 --> 00:07:01.240 a script. But even if you're just in the web 00:07:01.240 --> 00:07:03.839 chat, you can keep a text file on your desktop. 00:07:04.319 --> 00:07:07.519 Call it myprompts .txt. So when I want a really 00:07:07.519 --> 00:07:09.199 tough code review, I don't type out the whole 00:07:09.199 --> 00:07:11.779 thing. I just copy paste my spell. You got it. 00:07:11.779 --> 00:07:14.019 It just lowers the cognitive load. If asking 00:07:14.019 --> 00:07:16.129 for a code review is a hassle, You're not going 00:07:16.129 --> 00:07:18.350 to do it. If it's a quick copy paste away, you'll 00:07:18.350 --> 00:07:20.310 do it every time. So is this just about typing 00:07:20.310 --> 00:07:23.189 faster or is it something more? It's about reducing 00:07:23.189 --> 00:07:25.589 that cognitive load. It turns these repetitive 00:07:25.589 --> 00:07:28.629 chores into single word triggers. It lowers the 00:07:28.629 --> 00:07:30.889 barrier to doing things the right way. Exactly. 00:07:31.089 --> 00:07:34.019 Speaking of barriers, let's talk debugging. The 00:07:34.019 --> 00:07:37.519 red test. You run your code. The screen fills 00:07:37.519 --> 00:07:40.279 with red error text. And the human reaction is 00:07:40.279 --> 00:07:43.040 to squint at it, scroll up, try to find the one 00:07:43.040 --> 00:07:45.560 important line. Stop. Stop playing detective. 00:07:46.220 --> 00:07:48.480 Humans are terrible at parsing thousands of lines 00:07:48.480 --> 00:07:51.420 of logs. We see what we expect to see. AIs are 00:07:51.420 --> 00:07:53.899 just logic engines. They're literal. So what's 00:07:53.899 --> 00:07:56.279 the workflow then? You copy the entire error 00:07:56.279 --> 00:07:58.740 log. I don't care if it's 500 lines of gibberish. 00:07:59.160 --> 00:08:02.350 You paste it in and just say Fix the failing 00:08:02.350 --> 00:08:04.470 tests. You don't even try to give it a hint. 00:08:04.529 --> 00:08:06.850 Like, I think it's the database. No, because 00:08:06.850 --> 00:08:09.350 you might be wrong. And if you give it a hint, 00:08:09.550 --> 00:08:12.589 you're biasing the AI. You're contaminating the 00:08:12.589 --> 00:08:15.649 evidence. Let it look at the raw data. The source 00:08:15.649 --> 00:08:19.670 also mentions a term, MCP, Model Context Protocol. 00:08:19.829 --> 00:08:22.310 What does that actually mean? It sounds like 00:08:22.310 --> 00:08:25.149 jargon, but think of it as giving the AI hands. 00:08:25.769 --> 00:08:28.709 Normally, the AI is a brain in a jar. It can 00:08:28.709 --> 00:08:31.800 only talk. MCP lets the AI reach out and touch 00:08:31.800 --> 00:08:34.399 your tools. It can read your terminal, access 00:08:34.399 --> 00:08:37.059 your Slack, look at GitHub issues directly. So 00:08:37.059 --> 00:08:39.440 instead of me pasting the logs, the AI just goes 00:08:39.440 --> 00:08:41.519 and looks at them itself. Exactly. You can say, 00:08:41.720 --> 00:08:43.779 why is the goal failing? And it goes and checks 00:08:43.779 --> 00:08:45.779 the server logs for you. It connects the dots 00:08:45.779 --> 00:08:48.519 way faster. Why does this work better than a 00:08:48.519 --> 00:08:51.000 human searching through logs? Because it's analyzing 00:08:51.000 --> 00:08:53.980 logic instantly. It finds things we miss, like 00:08:53.980 --> 00:08:56.500 a missing semicolon or an inverted true false. 00:08:56.759 --> 00:08:59.720 It's unbiased pattern recognition versus human 00:08:59.720 --> 00:09:01.860 guessing. That brings up a really interesting 00:09:01.860 --> 00:09:04.960 tactic called grill me, which feels like the 00:09:04.960 --> 00:09:08.399 opposite of asking for help. Yeah. Usually we 00:09:08.399 --> 00:09:11.799 want the AI to be nice to us. Grill me is asking 00:09:11.799 --> 00:09:15.179 the AI to be mean. How so? You say, I'm about 00:09:15.179 --> 00:09:18.000 to commit this code, act as a strict security 00:09:18.000 --> 00:09:21.360 auditor, grill me on the flaws. You're asking 00:09:21.360 --> 00:09:23.779 for a stress test. And it will find them. It'll 00:09:23.779 --> 00:09:26.320 say, you didn't handle what happens if the internet 00:09:26.320 --> 00:09:29.580 connection drops. Or, this loop will freeze the 00:09:29.580 --> 00:09:32.220 browser with 10 ,000 items. It catches all the 00:09:32.220 --> 00:09:34.460 edge cases we're too optimistic to see. Then 00:09:34.460 --> 00:09:37.340 there's the scrap it method. This one, I think 00:09:37.340 --> 00:09:39.539 I'd struggle with this. Knowing everything you 00:09:39.539 --> 00:09:42.059 know now, scrap this and implement the elegant 00:09:42.059 --> 00:09:44.299 solution. It hurts, doesn't it? It's the sunk 00:09:44.299 --> 00:09:46.440 cost fallacy. I spent three hours on this, I 00:09:46.440 --> 00:09:48.860 have to make it work. But the guide argues the 00:09:48.860 --> 00:09:51.940 second draft is always better. Always. The first 00:09:51.940 --> 00:09:54.340 draft is for you and the AI to learn the problem. 00:09:54.559 --> 00:09:56.860 Once you understand it, that code you wrote is 00:09:56.860 --> 00:10:00.440 just baggage. If you tell the AI, OK, lesson 00:10:00.440 --> 00:10:02.820 learned, delete it, write it clean, you get a 00:10:02.820 --> 00:10:04.559 solution that's half the length and twice as 00:10:04.559 --> 00:10:07.419 fast. It just feels so counterintuitive asking 00:10:07.419 --> 00:10:10.919 the AI to delete its own work. But the restart 00:10:10.919 --> 00:10:13.179 carries all the lessons you learned and ditches 00:10:13.179 --> 00:10:15.340 all the baggage. Killing your darlings to lock 00:10:15.340 --> 00:10:18.519 in the lesson. Ruthlessly. Let's touch on the 00:10:18.519 --> 00:10:20.440 environment itself. There's a mention of things 00:10:20.440 --> 00:10:23.120 like ghosty and voice mode. Is this just about 00:10:23.120 --> 00:10:25.919 making things look nice? Well, ghosty is a terminal. 00:10:26.419 --> 00:10:29.419 And yeah, good readability helps. But voice mode, 00:10:30.399 --> 00:10:32.799 that's the real game changer. Really? I type 00:10:32.799 --> 00:10:35.019 pretty fast. You don't type as fast as you think. 00:10:35.700 --> 00:10:38.159 But more importantly, when you type, you summarize. 00:10:38.549 --> 00:10:41.870 You cut corners because typing is work. When 00:10:41.870 --> 00:10:44.169 you speak, you ramble, and I mean that in a good 00:10:44.169 --> 00:10:47.289 way. You provide so much more context. So instead 00:10:47.289 --> 00:10:49.070 of typing, fix this function, you're saying, 00:10:49.190 --> 00:10:50.990 OK, so I'm trying to sort this list, but I'm 00:10:50.990 --> 00:10:52.889 worried about the user data being null, and we 00:10:52.889 --> 00:10:55.029 need to make sure. Exactly. You're dumping your 00:10:55.029 --> 00:10:57.769 entire mental state into the AI. It's just high 00:10:57.769 --> 00:10:59.889 bandwidth communication, and it gives the AI 00:10:59.889 --> 00:11:02.509 a much richer context to work with. The guide 00:11:02.509 --> 00:11:04.970 also uses a house building analogy for something 00:11:04.970 --> 00:11:08.419 it calls subagents. How does that work? So think 00:11:08.419 --> 00:11:10.580 about building a house. You've got your general 00:11:10.580 --> 00:11:13.740 contractor, but you also have a plumber, an electrician, 00:11:13.940 --> 00:11:16.240 a roofer. You wouldn't ask the electrician to 00:11:16.240 --> 00:11:19.059 plumb the toilet. So a subagent is like a specialist. 00:11:19.340 --> 00:11:22.379 Yes. For a massive task, you tell the main AI, 00:11:22.559 --> 00:11:25.440 break this down, use subagents. One agent might 00:11:25.440 --> 00:11:27.860 just go read the documentation for a new library. 00:11:28.419 --> 00:11:31.059 Another might write the SQL queries. They all 00:11:31.059 --> 00:11:33.240 report back to the main boss. And what's the 00:11:33.240 --> 00:11:35.980 real benefit of using these subagents? It keeps 00:11:35.980 --> 00:11:39.299 the main AI from getting tired. It stops its 00:11:39.299 --> 00:11:41.500 context window from filling up with noise. So 00:11:41.500 --> 00:11:44.080 specialization keeps the context clean. Right. 00:11:44.340 --> 00:11:46.799 We've talked a lot about building faster. But 00:11:46.799 --> 00:11:49.539 there's a fear here. If the AI writes the SQL 00:11:49.539 --> 00:11:53.779 and the AI debugs the logs, are we just forgetting 00:11:53.779 --> 00:11:56.460 how to code? Yeah, the de -skilling fear. And 00:11:56.460 --> 00:11:59.600 it's valid, but only if you use the tool lazily. 00:11:59.840 --> 00:12:02.220 If you just say, do it, and paste the result 00:12:02.220 --> 00:12:04.519 without thinking, then yes, you will become a 00:12:04.519 --> 00:12:06.320 worse coder. So how do you flip that script? 00:12:06.659 --> 00:12:09.100 You turn the AI into a tutor. The source calls 00:12:09.100 --> 00:12:12.279 this the educational shift. If the AI writes 00:12:12.279 --> 00:12:15.399 a complex SQL query, don't just use it. Ask it, 00:12:15.700 --> 00:12:18.440 why did you use a left join here instead of an 00:12:18.440 --> 00:12:20.580 inner join? You force it to explain its work. 00:12:21.059 --> 00:12:23.740 And you can use visual aids. You can ask for 00:12:23.740 --> 00:12:26.580 an ASCII diagram. It'll literally draw a flowchart 00:12:26.580 --> 00:12:28.480 with text characters to show you how the data 00:12:28.480 --> 00:12:30.419 is moving. And what about spaced repetition? 00:12:30.879 --> 00:12:33.809 Yeah. Quiz me on React hooks. Use that downtime. 00:12:34.429 --> 00:12:36.289 Instead of scrolling Twitter while your code 00:12:36.289 --> 00:12:39.710 compiles, have the AI grill you on concepts you're 00:12:39.710 --> 00:12:42.009 weak on. You actually end up learning faster 00:12:42.009 --> 00:12:44.490 because you have a senior engineer who never 00:12:44.490 --> 00:12:46.690 gets tired of explaining things. So does relying 00:12:46.690 --> 00:12:49.809 on AI make us forget how to code? Not if you 00:12:49.809 --> 00:12:52.009 use it as a senior teacher or a study buddy. 00:12:52.389 --> 00:12:54.649 So active curiosity prevents that skill from 00:12:54.649 --> 00:12:57.090 fading. You have to use it to create curiosity, 00:12:57.330 --> 00:12:59.139 not replace it. We're going to take a quick break. 00:12:59.200 --> 00:13:01.100 When we come back, we're going to distill all 00:13:01.100 --> 00:13:04.700 this down. If you only do one thing from this 00:13:04.700 --> 00:13:07.120 list, what should it be? We've covered a lot. 00:13:07.340 --> 00:13:11.039 Parallel work trees, CLOUDE .md, adversarial 00:13:11.039 --> 00:13:13.440 grilling, voice mode. It feels like a completely 00:13:13.440 --> 00:13:16.460 new operating system for development. But if 00:13:16.460 --> 00:13:19.600 we zoom out, what's the one big idea here? The 00:13:19.600 --> 00:13:22.399 shift is from thinking of it as a tool to thinking 00:13:22.399 --> 00:13:26.120 of it as a partner. Meaning? Well, you don't 00:13:26.120 --> 00:13:28.230 give a hammer a memory file. You don't ask your 00:13:28.230 --> 00:13:30.870 hammer for a plan. But you treat AI like a hammer. 00:13:30.990 --> 00:13:33.049 You get really limited results. You have to treat 00:13:33.049 --> 00:13:35.629 it like a brilliant but junior colleague. It 00:13:35.629 --> 00:13:38.090 needs context. It needs clear instructions. It 00:13:38.090 --> 00:13:41.250 needs a dedicated workspace. It does. It's about 00:13:41.250 --> 00:13:43.029 leveraging the fact that it can do the heavy 00:13:43.029 --> 00:13:45.370 lifting, which frees you up to do the high -level 00:13:45.370 --> 00:13:47.730 thinking. Your brain is for architecture, business 00:13:47.730 --> 00:13:50.409 logic, creativity. And the AI is for syntax, 00:13:50.669 --> 00:13:53.070 boilerplate, and error checking. When you get 00:13:53.070 --> 00:13:55.629 that balance right, coding stops being a grind. 00:13:55.919 --> 00:13:59.159 it starts feeling like magic again. So for the 00:13:59.159 --> 00:14:00.759 listener who's maybe sitting at their desk right 00:14:00.759 --> 00:14:03.279 now, don't try to do all 10 of these things at 00:14:03.279 --> 00:14:05.519 once. No, you'll burn out. Just pick one. And 00:14:05.519 --> 00:14:07.559 which one should that be? I'd say start with 00:14:07.559 --> 00:14:10.919 cliude .md. Just create that text file on your 00:14:10.919 --> 00:14:13.879 project. Put your tech stack in it. Add just 00:14:13.879 --> 00:14:16.000 one lesson you learn the hard way. And just watch 00:14:16.000 --> 00:14:18.700 how that changes the conversation when the AI 00:14:18.700 --> 00:14:21.379 stops having amnesia. That's the actionable step. 00:14:21.899 --> 00:14:24.059 And I'll leave you with this thought. If the 00:14:24.059 --> 00:14:26.740 AI is handling the syntax, the debugging, the 00:14:26.740 --> 00:14:29.440 boilerplate, maybe we aren't just coders anymore. 00:14:29.620 --> 00:14:32.000 Maybe this tool lets all of us finally become 00:14:32.000 --> 00:14:34.240 architects. That's the goal. Let the AI handle 00:14:34.240 --> 00:14:36.179 the boring stuff so you can get back to the joy 00:14:36.179 --> 00:14:38.740 of building. Thanks for diving in with us. See 00:14:38.740 --> 00:14:39.299 you in the next one.