WEBVTT 00:00:00.000 --> 00:00:02.680 Have you ever fallen down a massive rabbit hole 00:00:02.680 --> 00:00:06.360 of AI tutorials? Oh. You know, only to realize 00:00:06.360 --> 00:00:08.359 you learned the entirely wrong thing. Welcome 00:00:08.359 --> 00:00:12.080 to the deep dive. Yeah, picture this absolute 00:00:12.080 --> 00:00:15.039 nightmare scenario. You're staring at 20 hours 00:00:15.039 --> 00:00:18.600 of free video courses on Anthropic Academy. You 00:00:18.600 --> 00:00:21.739 casually click the very first one. And you end 00:00:21.739 --> 00:00:24.359 up wasting an entire weekend grinding through 00:00:24.359 --> 00:00:26.719 an eight -hour developer API course. Oh, wow. 00:00:27.079 --> 00:00:29.059 Right? Then Monday morning hits, you suddenly 00:00:29.059 --> 00:00:30.800 remember you're actually just a marketer. It 00:00:30.800 --> 00:00:34.020 happens way too often. Beat and Thropic Academy 00:00:34.020 --> 00:00:37.460 just dropped over 15 free courses. As of April 00:00:37.460 --> 00:00:40.939 2026, these are fully live. Which is huge. It 00:00:40.939 --> 00:00:42.939 is. More importantly, they're built by the actual 00:00:42.939 --> 00:00:45.539 engineers, shipping clod. Our mission today is 00:00:45.539 --> 00:00:47.399 mapping out the seven core courses. We'll tell 00:00:47.399 --> 00:00:49.420 you exactly what to skip. Right. And we'll give 00:00:49.420 --> 00:00:51.439 you the precise learning path for your specific 00:00:51.439 --> 00:00:53.820 role, recovering everyone from non -technical 00:00:53.820 --> 00:00:56.359 beginners to advanced builders to sex silence. 00:00:57.600 --> 00:00:59.740 But before we dissect the actual frameworks, 00:00:59.920 --> 00:01:02.039 we need to examine the platform itself. Exactly. 00:01:02.259 --> 00:01:04.620 Why spend time here? Why not just watch a 10 00:01:04.620 --> 00:01:06.980 -minute YouTube tutorial? Because the landscape 00:01:06.980 --> 00:01:10.109 just moves way too fast. I mean... YouTube is 00:01:10.109 --> 00:01:12.750 full of outdated hacks. Yeah, that's true. The 00:01:12.750 --> 00:01:15.349 Anthropic catalog is cleanly split into three 00:01:15.349 --> 00:01:18.670 distinct tracks. You have AI fluency for non 00:01:18.670 --> 00:01:20.969 -technical folks. Right. You have a developer 00:01:20.969 --> 00:01:24.670 track for the API and MCP. And finally, a role 00:01:24.670 --> 00:01:27.189 -based track. That's for educators and nonprofits. 00:01:27.629 --> 00:01:30.450 But the real hook is the authorship. Built by 00:01:30.450 --> 00:01:33.150 the engineers who actually make the model. Yes. 00:01:33.329 --> 00:01:36.189 These lessons reflect what's strictly true today. 00:01:36.379 --> 00:01:38.239 They don't reflect what worked six months ago. 00:01:38.519 --> 00:01:40.659 The interface changes too rapidly for outdated 00:01:40.659 --> 00:01:43.859 content. They're also completely free. You just 00:01:43.859 --> 00:01:45.980 need an email address. Right. And after the final 00:01:45.980 --> 00:01:48.299 quiz, you get a non -expiring LinkedIn certificate. 00:01:49.299 --> 00:01:53.439 But I have to push back on that. Do hiring managers 00:01:53.439 --> 00:01:55.680 actually care about another digital badge? It 00:01:55.680 --> 00:01:57.900 feels a bit like vanity metrics to me. Right. 00:01:57.980 --> 00:02:00.620 Usually they totally are. But a certificate direct 00:02:00.620 --> 00:02:03.760 from the source proves something highly specific. 00:02:03.900 --> 00:02:06.379 What's that? It proves you didn't just copy -paste 00:02:06.379 --> 00:02:08.740 an influencer's prompt. You actually took the 00:02:08.740 --> 00:02:11.400 time to learn the official underlying mechanics. 00:02:11.879 --> 00:02:14.560 I suppose that's fair. It proves you understand 00:02:14.560 --> 00:02:16.979 the physics of the tool. You aren't just guessing 00:02:16.979 --> 00:02:20.560 at the interface. Which leads to an obvious baseline 00:02:20.560 --> 00:02:23.599 rule for us. Why strictly avoid third -party 00:02:23.599 --> 00:02:27.060 tutorials right now? Official docs reflect today's 00:02:27.060 --> 00:02:30.460 reality, not last year's outdated code. Two secs 00:02:30.460 --> 00:02:33.979 silence. So we know the source is reliable. But 00:02:33.979 --> 00:02:36.680 diving straight into the deep end usually leads 00:02:36.680 --> 00:02:39.479 to serious burnout. Oh, absolutely. We need to 00:02:39.479 --> 00:02:42.180 start with the absolute baseline. There's a foundation 00:02:42.180 --> 00:02:44.300 everyone has to take, regardless of technical 00:02:44.300 --> 00:02:47.620 skill. It's a course called Claude 101. It takes 00:02:47.620 --> 00:02:50.159 about an hour to complete, and it fundamentally 00:02:50.159 --> 00:02:52.819 changes how you view the interface. It's not 00:02:52.819 --> 00:02:55.280 just a chat window anymore. Not at all. It covers 00:02:55.280 --> 00:02:57.580 the actual mechanics of connectors, for instance. 00:02:57.719 --> 00:02:59.800 Connectors let Claude pull directly from your 00:02:59.800 --> 00:03:02.639 Gmail, Drive, or Notion. Let's actually unpack 00:03:02.639 --> 00:03:04.379 how those work into the hood. Yeah, it's not 00:03:04.379 --> 00:03:06.719 just handing Claude your password. Right. It 00:03:06.719 --> 00:03:09.460 securely indexes your workspace text. It reads 00:03:09.460 --> 00:03:12.520 the metadata. That way it can reference specific 00:03:12.520 --> 00:03:14.699 documents without hallucinating. But there are 00:03:14.699 --> 00:03:16.819 definitely weaknesses here. The reading, right? 00:03:16.960 --> 00:03:19.280 Yeah. The course is very reading -heavy. It's 00:03:19.280 --> 00:03:22.159 kind of light on actual hands -on doing. And 00:03:22.159 --> 00:03:24.780 some features discussed are strictly for paid 00:03:24.780 --> 00:03:27.680 tier users. Free tier folks will hit some annoying 00:03:27.680 --> 00:03:30.280 paywalls. But you still desperately need this 00:03:30.280 --> 00:03:32.340 baseline. You need to understand the ecosystem 00:03:32.340 --> 00:03:36.560 because the next skill is the absolute MVP of 00:03:36.560 --> 00:03:40.060 the entire platform. AI fluency. Exactly. This 00:03:40.060 --> 00:03:42.719 core set is entirely on a specific mental model. 00:03:43.099 --> 00:03:45.900 They call it the 4D framework. It's designed 00:03:45.900 --> 00:03:48.259 to stop hallucinations before they even happen. 00:03:48.479 --> 00:03:51.580 beat. The first D is delegation. You have to 00:03:51.580 --> 00:03:53.800 figure out exactly what tasks to hand off. The 00:03:53.800 --> 00:03:56.439 second is description, supplying the exact context 00:03:56.439 --> 00:03:59.360 the model needs to succeed. The third D is discernment. 00:03:59.719 --> 00:04:02.520 This means actively checking the tone and factual 00:04:02.520 --> 00:04:05.699 accuracy. And the final D is diligence. You must 00:04:05.699 --> 00:04:08.460 fully own the final shipped result. The course 00:04:08.460 --> 00:04:10.780 is fantastic because it's heavily exercise -based. 00:04:11.080 --> 00:04:13.479 Let's look at a concrete example from their materials. 00:04:13.520 --> 00:04:15.340 Let's see how this actually works. Say you're 00:04:15.340 --> 00:04:17.360 writing a product launch announcement. Most people 00:04:17.360 --> 00:04:20.139 just type write a launch post. Right, and they 00:04:20.139 --> 00:04:23.319 get generic garbage. Instead, you use a single 00:04:23.319 --> 00:04:26.319 prompt, forcing Claude to use all four Ds. You 00:04:26.319 --> 00:04:28.860 tell Claude what you're writing. First, you ask 00:04:28.860 --> 00:04:30.800 it to list which parts you should write yourself. 00:04:31.000 --> 00:04:34.220 That handles delegation. Second, you ask it for 00:04:34.220 --> 00:04:37.220 three questions about your brand voice. That 00:04:37.220 --> 00:04:40.339 forces description. Third, you make it score 00:04:40.339 --> 00:04:44.100 its own drafted output, checking tone and accuracy. 00:04:44.600 --> 00:04:47.540 That's discernment in action. Finally, you ask 00:04:47.540 --> 00:04:50.459 it to explicitly flag any claims that require 00:04:50.459 --> 00:04:54.379 your manual proof. That sets up diligence. It's 00:04:54.379 --> 00:04:57.060 a brilliant constraint. It forces you to actually 00:04:57.060 --> 00:05:00.060 verify the facts instead of just skimming. I'll 00:05:00.060 --> 00:05:02.220 be honest here. I still wrestle with prompt drift 00:05:02.220 --> 00:05:05.399 myself. Really? Yeah. You get comfortable. You 00:05:05.399 --> 00:05:07.819 get kind of lazy. You start assuming the AI just 00:05:07.819 --> 00:05:10.670 knows what you mean. We blindly trust AI outputs 00:05:10.670 --> 00:05:12.910 far too often. Oh, it happens to absolutely everyone. 00:05:13.009 --> 00:05:14.750 You get complacent when it works perfectly three 00:05:14.750 --> 00:05:18.069 times in a row. It's a dangerous habit. But it's 00:05:18.069 --> 00:05:20.250 human nature to take the path of least resistance. 00:05:20.949 --> 00:05:23.790 So why is diligence the step everyone seems to 00:05:23.790 --> 00:05:26.410 skip? It is much easier to blame the AI than 00:05:26.410 --> 00:05:29.720 yourself. Two -Sec Silence. So you've mastered 00:05:29.720 --> 00:05:32.000 talking to Claude. You're actively using the 00:05:32.000 --> 00:05:35.339 4Ds. But keeping everything inside a chat window, 00:05:35.519 --> 00:05:37.819 it's ultimately limiting. It is. You eventually 00:05:37.819 --> 00:05:40.000 need Claude to talk directly to your data. And 00:05:40.000 --> 00:05:42.839 that brings us to the MCP ecosystem. MCP stands 00:05:42.839 --> 00:05:46.079 for Model Context Protocol. It sounds intimidating. 00:05:46.220 --> 00:05:48.519 But it's really not. Let me define it simply. 00:05:49.060 --> 00:05:52.379 An open standard letting AI connect safely to 00:05:52.379 --> 00:05:54.699 outside data sources. That's a perfect summary. 00:05:55.279 --> 00:05:57.459 But let's clarify the mechanism. How does it 00:05:57.459 --> 00:06:00.139 actually connect without exposing your whole 00:06:00.139 --> 00:06:03.639 hard drive? Think of MCP like stacking Lego blocks 00:06:03.639 --> 00:06:06.800 of data. Instead of writing custom brittle code 00:06:06.800 --> 00:06:09.399 to teach Claude how to read your specific database, 00:06:10.100 --> 00:06:12.040 MCP provides a standardized socket. You just 00:06:12.040 --> 00:06:14.060 plug it in. Claude instantly knows how to read 00:06:14.060 --> 00:06:16.959 the data securely. The Intro to MCP course breaks 00:06:16.959 --> 00:06:18.819 this down beautifully. Yeah, it covers three 00:06:18.819 --> 00:06:20.720 essential primitives. You learn about tools, 00:06:21.040 --> 00:06:23.199 resources, and prompts. But it doesn't just leave 00:06:23.199 --> 00:06:25.680 you with dry theory. You actually build a small, 00:06:25.680 --> 00:06:28.079 local server. You build a specific tool called 00:06:28.079 --> 00:06:31.120 List Today Tasks. It reads a local JSON file 00:06:31.120 --> 00:06:33.740 on your machine. It securely parses that file. 00:06:34.199 --> 00:06:37.180 Then it returns only today's tasks directly back 00:06:37.180 --> 00:06:40.000 to Claude. You see the actual data flow working 00:06:40.000 --> 00:06:42.439 in real time. It demystifies the whole process. 00:06:42.519 --> 00:06:44.620 Once you finish that, there's a second course. 00:06:44.920 --> 00:06:48.480 MCP advanced topics. It dives into incredibly 00:06:48.480 --> 00:06:51.379 dense concepts. Custom transport layers, precise 00:06:51.379 --> 00:06:54.579 sampling, complex file system. But I have to 00:06:54.579 --> 00:06:57.339 play devil's advocate here. If the advanced course 00:06:57.339 --> 00:07:01.040 holds the powerful stuff, why not just jump straight 00:07:01.040 --> 00:07:03.759 there? Why force developers to build a basic 00:07:03.759 --> 00:07:06.819 local server first? Because you absolutely must 00:07:06.819 --> 00:07:09.019 feel the real friction. If you just watch the 00:07:09.019 --> 00:07:11.439 advanced videos, the theory is incredibly dense. 00:07:11.620 --> 00:07:13.459 It's abstract. It won't stick in your brain. 00:07:13.600 --> 00:07:16.519 Right. you'll experience massive conceptual drift. 00:07:17.160 --> 00:07:19.019 You need to build and use a real server for at 00:07:19.019 --> 00:07:20.839 least a week first. You have to see the errors. 00:07:20.959 --> 00:07:23.379 You have to debug the connection locally. Friction 00:07:23.379 --> 00:07:25.279 is exactly where the actual learning happens. 00:07:25.480 --> 00:07:27.740 Let's make sure this is crystal clear. Why force 00:07:27.740 --> 00:07:29.740 developers to build before taking the advanced 00:07:29.740 --> 00:07:32.100 course? Complex theory evaporates completely 00:07:32.100 --> 00:07:34.759 without real -world friction. To sack silence. 00:07:35.579 --> 00:07:37.620 So connecting external data is only half the 00:07:37.620 --> 00:07:40.430 battle. Right. If you're a developer, Pulling 00:07:40.430 --> 00:07:42.889 a list of tasks isn't enough. You need Claude 00:07:42.889 --> 00:07:45.269 to actually write and push the code. That's where 00:07:45.269 --> 00:07:49.269 Claude Code in Action comes in. Sponsor. And 00:07:49.269 --> 00:07:51.649 speaking of taking action, the Deep Dive is brought 00:07:51.649 --> 00:07:54.529 to you by our amazing sponsors who help us keep 00:07:54.529 --> 00:07:57.069 analyzing these technical frameworks week after 00:07:57.069 --> 00:08:02.589 week. Welcome back to the Deep Dive. Before we 00:08:02.589 --> 00:08:05.310 get into the massive enterprise APIs, let's look 00:08:05.310 --> 00:08:08.269 at Cloud Code in action. This course is incredibly 00:08:08.269 --> 00:08:10.290 practical for actual software development. It 00:08:10.290 --> 00:08:12.709 totally moves you past those basic chat interfaces. 00:08:12.829 --> 00:08:15.069 You learn how to set up custom slash commands, 00:08:15.430 --> 00:08:17.449 you get direct MCP integration right in your 00:08:17.449 --> 00:08:20.149 terminal, and you learn proper GitHub workflow 00:08:20.149 --> 00:08:22.509 setups. But the most valuable part is the focus 00:08:22.509 --> 00:08:25.230 on structured planning modes. When you're managing 00:08:25.230 --> 00:08:28.110 a large code base, you can't just wing it. Absolutely 00:08:28.110 --> 00:08:30.480 not. These modes make Cloud Code far more reliable 00:08:30.480 --> 00:08:33.399 on complex tasks. It teaches a crucial two -step 00:08:33.399 --> 00:08:36.379 pattern for coding, beat. Let's walk through 00:08:36.379 --> 00:08:38.720 the exact mechanism of this pattern. Say you 00:08:38.720 --> 00:08:41.000 need to refactor a user authentication module. 00:08:41.159 --> 00:08:43.440 You want to add Google Outh alongside email. 00:08:43.620 --> 00:08:45.860 The instinct is to just say, rewrite the auth 00:08:45.860 --> 00:08:48.559 module. But you don't do that. You ask Claude 00:08:48.559 --> 00:08:51.179 to use the plan command first. Exactly. You command 00:08:51.179 --> 00:08:53.500 it to list every single file that requires changes. 00:08:53.639 --> 00:08:55.620 You ask for the specific order of operations. 00:08:55.899 --> 00:08:59.100 You identify the security risk areas. And you 00:08:59.100 --> 00:09:01.779 explicitly forbid it from writing any actual 00:09:01.779 --> 00:09:04.279 code yet. Then you act as the senior engineer. 00:09:04.919 --> 00:09:07.639 You review that plan thoroughly. You push back 00:09:07.639 --> 00:09:10.500 on any weak architectural decisions. You refine 00:09:10.500 --> 00:09:12.980 the logic. Once the plan has walked solid, you 00:09:12.980 --> 00:09:15.690 approve step one. only step one. You instruct 00:09:15.690 --> 00:09:18.049 quad to make the change and immediately run tests. 00:09:18.509 --> 00:09:21.129 You force it to stop and explicitly ask permission 00:09:21.129 --> 00:09:24.029 before moving to step two. This methodical step 00:09:24.029 --> 00:09:26.230 -by -step approach is totally non -negotiable 00:09:26.230 --> 00:09:28.590 for serious engineering work. There is a serious 00:09:28.590 --> 00:09:31.990 catch though. The prerequisites. Yeah. You absolutely 00:09:31.990 --> 00:09:34.990 must know basic terminal navigation and Git. 00:09:35.309 --> 00:09:37.529 If you don't know what a pull request is, you're 00:09:37.529 --> 00:09:39.710 gonna drown. If you're a coding novice, skip 00:09:39.710 --> 00:09:43.149 this course entirely. It assumes a working baseline 00:09:43.149 --> 00:09:45.850 of programming knowledge. You'll get hopelessly 00:09:45.850 --> 00:09:48.570 lost trying to debug terminal errors. But for 00:09:48.570 --> 00:09:51.830 engineers, this planning methodology is gold. 00:09:52.070 --> 00:09:54.830 Let's summarize why it works so well. Why use 00:09:54.830 --> 00:09:57.950 this strict two -step planning pattern? It stops 00:09:57.950 --> 00:10:01.429 the AI from hallucinating a massive broken rewrite. 00:10:01.710 --> 00:10:04.450 To sex silence. We've covered the foundation. 00:10:04.870 --> 00:10:08.009 We've covered local coding. Now we move to the 00:10:08.009 --> 00:10:10.529 absolute heavyweights. The big stuff. If you're 00:10:10.529 --> 00:10:13.049 building enterprise apps or scalable systems, 00:10:13.629 --> 00:10:16.370 basic scripts aren't enough. You need the core 00:10:16.370 --> 00:10:19.370 API and reusable agent frameworks. This is where 00:10:19.370 --> 00:10:22.009 we hit the beast. Building with the Claude API, 00:10:22.330 --> 00:10:25.549 it's roughly eight hours long. 84 distinct lectures. 00:10:25.649 --> 00:10:28.269 10 quizzes. It's practically a university module. 00:10:28.330 --> 00:10:30.730 And it covers incredibly powerful technical features. 00:10:30.950 --> 00:10:33.029 For example, it teaches prompt caching in great 00:10:33.029 --> 00:10:35.169 detail. Let's explain the mechanism here. Sure. 00:10:35.389 --> 00:10:37.830 When you send a massive prompt, the model normally 00:10:37.830 --> 00:10:40.409 processes every single word. That processing 00:10:40.409 --> 00:10:43.070 burns expensive compute power. Prompt caching 00:10:43.070 --> 00:10:45.830 changes this dynamic entirely. Right. It stores 00:10:45.830 --> 00:10:48.370 the heavy static parts of your prompt directly 00:10:48.370 --> 00:10:51.110 in the context window. It's like keeping a massive 00:10:51.110 --> 00:10:53.570 instruction manual open on a desk. You don't 00:10:53.570 --> 00:10:55.450 have to re -read the whole book every time you 00:10:55.450 --> 00:10:58.970 ask a question. You only pay to process the new 00:10:58.970 --> 00:11:01.789 dynamic questions you ask against it. It saves 00:11:01.789 --> 00:11:04.690 a fortune at scale. It also covers extended thinking 00:11:04.690 --> 00:11:07.590 and adaptive thinking. This is where you allow 00:11:07.590 --> 00:11:10.490 Claude to reason much longer internally before 00:11:10.490 --> 00:11:14.220 it outputs a single word. Whoa. Imagine scaling 00:11:14.220 --> 00:11:17.480 to a billion queries where the AI dynamically 00:11:17.480 --> 00:11:20.399 decides exactly how much computing power it needs 00:11:20.399 --> 00:11:22.340 to think through a problem before answering. 00:11:22.759 --> 00:11:25.820 That is mind -blowing. The implications for complex 00:11:25.820 --> 00:11:29.840 reasoning are massive. Let's look at the specific 00:11:29.840 --> 00:11:32.139 sauce pricing prompt from the course. It shows 00:11:32.139 --> 00:11:34.419 why this matters. Okay. Imagine you're evaluating 00:11:34.419 --> 00:11:36.980 two distinct sauce pricing models. Plan A is 00:11:36.980 --> 00:11:39.779 $29 a month. It has a 5 % monthly churn. And 00:11:39.779 --> 00:11:42.139 you're getting 400 new signups per month. Exactly. 00:11:42.500 --> 00:11:46.080 Now Plan B is $49 a month. It has a 3 % monthly 00:11:46.080 --> 00:11:49.100 churn. But it offers a 20 % annual discount. 00:11:49.340 --> 00:11:53.379 Plus a 14 -day trial with a 25 % conversion rate. 00:11:53.559 --> 00:11:56.960 Right. You need to ask the AI which plan reaches 00:11:56.960 --> 00:11:59.700 $1 million in annual recurring revenue first. 00:12:00.120 --> 00:12:02.779 And you need the exact month. If you rush the 00:12:02.779 --> 00:12:05.960 AI on this math, it totally fails. It hallucinates 00:12:05.960 --> 00:12:08.059 the compounding logic. It gives you estimates 00:12:08.059 --> 00:12:10.779 that miss by thousands of dollars. But when you 00:12:10.779 --> 00:12:13.600 enable expended thinking... The mechanism changes. 00:12:13.919 --> 00:12:16.480 It actually maps out the month -over -month compounding 00:12:16.480 --> 00:12:19.340 math internally before replying. The output is 00:12:19.340 --> 00:12:22.200 pristine. It gets the exact month you hit 1 million 00:12:22.200 --> 00:12:25.399 ARR perfectly right. It completely holds up when 00:12:25.399 --> 00:12:27.559 you verify it in a complex spreadsheet. That's 00:12:27.559 --> 00:12:30.039 the difference between a useless guess and a 00:12:30.039 --> 00:12:32.200 boardroom -ready metric. After mastering the 00:12:32.200 --> 00:12:35.840 API, there's one final developer course, Introduction 00:12:35.840 --> 00:12:38.610 to Agent Skills. Interestingly, it's the shortest 00:12:38.610 --> 00:12:40.389 course they offer. It teaches you how to write 00:12:40.389 --> 00:12:43.129 a simple skill .md file. It's super lightweight. 00:12:43.309 --> 00:12:45.789 Just standard YAML front matter and a plain markdown 00:12:45.789 --> 00:12:48.330 body. The mechanism is elegant. You write your 00:12:48.330 --> 00:12:50.590 complex enterprise -level rules exactly once 00:12:50.590 --> 00:12:52.870 in this file. You never have to keep pasting 00:12:52.870 --> 00:12:54.690 those massive instructions into every single 00:12:54.690 --> 00:12:57.250 chat. Claude automatically loads the skill when 00:12:57.250 --> 00:12:59.789 the context matches. It cuts your administrative 00:12:59.789 --> 00:13:01.909 overhead in half. But let's circle back to that 00:13:01.909 --> 00:13:05.149 crucial API reasoning concept. Why explicitly 00:13:05.149 --> 00:13:08.529 give the AI more time to think? Rushing complex 00:13:08.529 --> 00:13:11.809 math guarantees incredibly costly real -world 00:13:11.809 --> 00:13:15.879 mistakes. To sex silence. We've unpacked a massive 00:13:15.879 --> 00:13:18.539 catalog today, but knowledge without application 00:13:18.539 --> 00:13:21.659 is completely useless. Absolutely. Let's synthesize 00:13:21.659 --> 00:13:25.120 this into actionable, highly specific roadmaps. 00:13:25.470 --> 00:13:28.250 You need to know exactly what to click when this 00:13:28.250 --> 00:13:30.850 deep dive ends. Your path depends entirely on 00:13:30.850 --> 00:13:33.389 your daily role. Let's start with the non -technical 00:13:33.389 --> 00:13:35.950 folks, the marketers, the writers, the operators. 00:13:36.190 --> 00:13:38.049 If that's you, you take exactly two courses. 00:13:38.370 --> 00:13:41.289 You take Claude 101 for basic orientation, then 00:13:41.289 --> 00:13:44.230 you take AI Fluency to master the 4D framework, 00:13:44.289 --> 00:13:47.129 and then you absolutely must stop. Seriously, 00:13:47.370 --> 00:13:49.750 go use the tool in real life for three straight 00:13:49.750 --> 00:13:51.809 weeks. Don't touch the developer track. Wait 00:13:51.809 --> 00:13:54.049 for the upcoming Claude Cowork course in April 00:13:54.049 --> 00:13:57.090 2026. It's going to focus purely on practical 00:13:57.090 --> 00:13:59.009 content workflows. If you're a developer, your 00:13:59.009 --> 00:14:02.029 path is much longer. You lightly skim Claude 00:14:02.029 --> 00:14:05.570 101. You complete AI fluency to solidify your 00:14:05.570 --> 00:14:08.809 prompting basics. Then you tackle that massive 00:14:08.809 --> 00:14:12.090 eight -hour API course. After the API, you take 00:14:12.090 --> 00:14:14.750 intro to MCP. You build that test server, then 00:14:14.750 --> 00:14:17.629 you hit MCP Advanced Topics. Finally, you polish 00:14:17.629 --> 00:14:20.169 it off with Claude code and agent skills. But 00:14:20.169 --> 00:14:22.350 if you're a security professional, your sequence 00:14:22.350 --> 00:14:24.490 is slightly different. Yeah, you start with AI 00:14:24.490 --> 00:14:26.750 fluency to understand shared responsibility, 00:14:27.429 --> 00:14:30.250 then Claude 101. Then intro to MCP to understand 00:14:30.250 --> 00:14:32.720 how external connections are actually made. Security 00:14:32.720 --> 00:14:35.519 pros also have to take the core API course. You 00:14:35.519 --> 00:14:37.860 have to understand the specific mechanics of 00:14:37.860 --> 00:14:40.500 pumped injection risks, and you must understand 00:14:40.500 --> 00:14:43.259 the security implications of AGG. Right. Let's 00:14:43.259 --> 00:14:45.799 define our AGG quickly for anyone unfamiliar, 00:14:46.399 --> 00:14:48.679 letting AI safely search your private text documents 00:14:48.679 --> 00:14:53.100 for specific answers. Beat. Finally, security 00:14:53.100 --> 00:14:55.539 folks close out with MCP Advanced. You have to 00:14:55.539 --> 00:14:57.460 understand those deeper attack surfaces before 00:14:57.460 --> 00:14:59.700 you let your team deploy anything. The overarching 00:14:59.700 --> 00:15:03.080 theme here is extremely strict pacing. Don't 00:15:03.080 --> 00:15:04.860 binge it all at once. Tasting is everything. 00:15:05.120 --> 00:15:07.940 Let's reinforce that final rule right now. Why 00:15:07.940 --> 00:15:10.500 demand that beginners stop after just two courses? 00:15:10.919 --> 00:15:14.120 You desperately need actual practice before deeper 00:15:14.120 --> 00:15:17.779 theory sticks. Two secs silence. We've covered 00:15:17.779 --> 00:15:20.419 a tremendous amount of ground today. Let's zoom 00:15:20.419 --> 00:15:22.919 out and look at the big picture here. Yeah, Anthropic 00:15:22.919 --> 00:15:25.600 Academy isn't just about mindlessly collecting 00:15:25.600 --> 00:15:28.580 digital certificates to post online. It's about 00:15:28.580 --> 00:15:31.639 finding the exact mental frameworks that apply 00:15:31.639 --> 00:15:34.179 directly to your daily work. Frameworks like 00:15:34.179 --> 00:15:36.759 those 40s are universally applicable. And it's 00:15:36.759 --> 00:15:39.480 about learning those core technical primitives. 00:15:39.870 --> 00:15:43.169 Things like MCP sockets or Markdown skill files. 00:15:43.389 --> 00:15:44.850 You're learning them straight from the engineers 00:15:44.850 --> 00:15:47.009 who built them. You apply them directly to your 00:15:47.009 --> 00:15:49.889 specific daily tasks. You should absolutely sign 00:15:49.889 --> 00:15:52.370 up for a free account today. Try running that 00:15:52.370 --> 00:15:55.789 4D launch announcement prompt we discussed. See 00:15:55.789 --> 00:15:57.690 the difference that structured diligence makes 00:15:57.690 --> 00:16:00.629 in real time. Yeah, exactly. As you explore these 00:16:00.629 --> 00:16:02.389 courses, I want to leave you with a final thought 00:16:02.389 --> 00:16:05.389 to mull over. We saw how Claude can evaluate 00:16:05.389 --> 00:16:07.889 its own output using the diligence framework. 00:16:08.090 --> 00:16:10.250 We saw how we can dynamically decide how long 00:16:10.250 --> 00:16:14.470 to think using adaptive thinking. If Claude can 00:16:14.470 --> 00:16:16.389 eventually evaluate its own output for accuracy 00:16:16.389 --> 00:16:18.710 using the Diligence Framework to decide how long 00:16:18.710 --> 00:16:21.190 to think using adaptive thinking, at what point 00:16:21.190 --> 00:16:23.549 does the user transition from being a prompt 00:16:23.549 --> 00:16:25.769 engineer to simply being a manager of digital 00:16:25.769 --> 00:16:26.450 employees?