WEBVTT 00:00:00.000 --> 00:00:03.600 We have to fundamentally change how we talk to 00:00:03.600 --> 00:00:06.360 artificial intelligence. That habit we've all 00:00:06.360 --> 00:00:10.779 built up for two decades of just keyword stuffing, 00:00:11.660 --> 00:00:13.939 treating that little box like a search bar. It's 00:00:13.939 --> 00:00:16.019 actually hurting our productivity now. We really 00:00:16.019 --> 00:00:18.440 need to transition from searching to delegating. 00:00:18.579 --> 00:00:21.199 It's a total mindset shift. The moment you start 00:00:21.199 --> 00:00:25.100 treating the AI like a highly capable, high -stakes 00:00:25.100 --> 00:00:27.280 assistant, the entire game just changes. And 00:00:27.280 --> 00:00:29.899 the benefit here is just simple time arbitrage. 00:00:30.399 --> 00:00:33.140 I mean, if you invest maybe two focused minutes 00:00:33.140 --> 00:00:37.159 writing a clear, structured prompt, you can bypass 00:00:37.159 --> 00:00:39.340 30 minutes of tedious cleanup on the backend. 00:00:39.500 --> 00:00:41.799 And that, right there, is how we eliminate the 00:00:41.799 --> 00:00:44.320 edit tax. Welcome to the Deep Dive. Look, if 00:00:44.320 --> 00:00:46.740 the overwhelming constant change in the AI space 00:00:46.740 --> 00:00:48.899 has you feeling some serious information overload, 00:00:49.460 --> 00:00:51.939 you are definitely not alone. Every single week, 00:00:52.000 --> 00:00:53.880 there's a new model, a new feature set. But the 00:00:53.880 --> 00:00:55.820 real distinction, I think, is that AI mastery 00:00:55.820 --> 00:00:58.799 in 2026 isn't about which tool you subscribe 00:00:58.799 --> 00:01:01.020 to. It's about your core communication skill. 00:01:01.140 --> 00:01:03.710 It's really about clarity. And that's our mission 00:01:03.710 --> 00:01:05.829 for this deep dive. We're going to cut through 00:01:05.829 --> 00:01:08.530 all that noise to focus on the one skill that 00:01:08.530 --> 00:01:11.170 actually separates regular users from power users. 00:01:11.730 --> 00:01:13.549 We'll show you that foundational shift. We'll 00:01:13.549 --> 00:01:15.569 introduce you to the golden square for a perfect 00:01:15.569 --> 00:01:18.510 prompt, help you match the job to the right AI 00:01:18.510 --> 00:01:20.930 model. And then we'll look at some advanced techniques 00:01:20.930 --> 00:01:23.450 like forcing self -critique and even moving toward 00:01:23.450 --> 00:01:27.430 automation. So let's unpack this. So let's start 00:01:27.430 --> 00:01:30.090 with that foundation. Looking at the source material, 00:01:30.489 --> 00:01:35.030 why is communication, the classic soft skill. 00:01:35.349 --> 00:01:37.609 Why is that suddenly the most valuable technical 00:01:37.609 --> 00:01:40.329 skill? It feels a little counterintuitive. Well, 00:01:40.349 --> 00:01:42.390 it really comes down to a fundamental cognitive 00:01:42.390 --> 00:01:45.450 mismatch. For 20 years, our brains were trained 00:01:45.450 --> 00:01:48.469 by Google. We learned to use these fragmented 00:01:48.469 --> 00:01:51.030 key words, you know, best restaurant Berlin or 00:01:51.030 --> 00:01:54.849 tourist flat tire fix. We learn to speak to a 00:01:54.849 --> 00:01:56.930 machine that searches for existing things. And 00:01:56.930 --> 00:01:59.219 what do these new models want instead? They don't 00:01:59.219 --> 00:02:02.840 want keywords. They want direct, specific, delegated 00:02:02.840 --> 00:02:05.299 instruction. They are not search engines. They're 00:02:05.299 --> 00:02:07.000 not just looking through a library. These are 00:02:07.000 --> 00:02:09.819 language models that are capable of creation, 00:02:10.159 --> 00:02:13.020 of synthesis, of actual execution. So most users 00:02:13.020 --> 00:02:15.879 are just, they're using the wrong mental model? 00:02:16.099 --> 00:02:18.680 Exactly. They still approach the AI like they're 00:02:18.680 --> 00:02:21.439 just surfing the web. When you switch that mental 00:02:21.439 --> 00:02:24.979 model to delegation, You unlock everything. I 00:02:24.979 --> 00:02:27.379 really like the black box analogy that describes 00:02:27.379 --> 00:02:29.759 this. When you look at a model like ChatGBT or 00:02:29.759 --> 00:02:34.520 Claude, they contain this vast, almost unimaginable 00:02:34.520 --> 00:02:36.439 potential. It's all locked up inside. And the 00:02:36.439 --> 00:02:39.599 prompt is the key, literally. A weak, keyword 00:02:39.599 --> 00:02:42.599 -based prompt, it opens a tiny little door, it 00:02:42.599 --> 00:02:44.979 gives you a generic, surface -level answer that 00:02:44.979 --> 00:02:46.879 you have to heavily edit. What a strong one. 00:02:47.039 --> 00:02:48.879 A strong, structured prompt, one that's built 00:02:48.879 --> 00:02:51.280 on the clarity of your instruction, that opens 00:02:51.280 --> 00:02:53.280 the entire vault of knowledge that's just waiting 00:02:53.280 --> 00:02:55.659 inside the machine. That potential sounds great, 00:02:55.780 --> 00:02:57.539 but where I think a lot of listeners spend their 00:02:57.539 --> 00:03:00.039 time is just fixing bad output. So we have to 00:03:00.039 --> 00:03:02.780 talk more about this edit tax. The edit tax is 00:03:02.780 --> 00:03:05.580 simple, and it's so corrosive to productivity. 00:03:06.039 --> 00:03:08.520 It's spending, say, 30 seconds writing a low 00:03:08.520 --> 00:03:11.500 effort prompt and then spending 30 minutes fixing 00:03:11.500 --> 00:03:14.719 the messy, generalized, often inaccurate result. 00:03:15.099 --> 00:03:17.520 That 30 minutes of correction is the tax. That's 00:03:17.520 --> 00:03:20.020 the tax you pay for being unclear. OK, but let 00:03:20.020 --> 00:03:22.960 me challenge that for a second. Sometimes I just 00:03:22.960 --> 00:03:26.120 need a quick title idea or like a single sentence 00:03:26.120 --> 00:03:29.280 summary of a meeting isn't spending two minutes 00:03:29.280 --> 00:03:31.780 building a whole golden square prompt. Isn't 00:03:31.780 --> 00:03:34.379 that overkill for those simple low stakes tasks? 00:03:34.659 --> 00:03:37.000 That's a fair challenge, yeah. For a title idea, 00:03:37.240 --> 00:03:39.039 no, you don't need the full structure, but you 00:03:39.039 --> 00:03:41.300 have to distinguish between brainstorming where 00:03:41.300 --> 00:03:43.620 generic output is kind of fine and execution. 00:03:44.020 --> 00:03:46.520 If you need an email to a client or a business 00:03:46.520 --> 00:03:49.039 strategy or a paragraph for your website, the 00:03:49.039 --> 00:03:52.460 edit tax hits hard. Power users focus on execution. 00:03:52.719 --> 00:03:54.599 They'll invest those two focused minutes and 00:03:54.599 --> 00:03:57.340 get a 95 % perfect result. They lower the tax 00:03:57.340 --> 00:04:00.020 to almost zero. It's about maximizing output 00:04:00.020 --> 00:04:03.099 quality, not just speed. OK, that clarifies the 00:04:03.099 --> 00:04:06.460 return on investment. So if the prompt is the 00:04:06.460 --> 00:04:09.979 key to unlocking that vault, how exactly do we 00:04:09.979 --> 00:04:12.060 structure it so that we can open it consistently 00:04:12.060 --> 00:04:14.919 every single time? By consistently using the 00:04:14.919 --> 00:04:16.579 Golden Square framework. All right, let's dive 00:04:16.579 --> 00:04:19.000 into the Golden Square. This framework seems 00:04:19.000 --> 00:04:21.160 to be what immediately separates us from the 00:04:21.160 --> 00:04:24.389 default user. It's the role task context format 00:04:24.389 --> 00:04:26.769 approach. Absolutely. These four components, 00:04:26.930 --> 00:04:30.129 they provide the instant clarity the AI desperately 00:04:30.129 --> 00:04:33.089 needs. It's the scaffolding for great results. 00:04:33.730 --> 00:04:36.250 We start with the role. You tell the AI, act 00:04:36.250 --> 00:04:39.089 as a senior SEO specialist with 10 years of agency 00:04:39.089 --> 00:04:41.290 experience. And giving it that persona gives 00:04:41.290 --> 00:04:44.329 it authority and tone, forcing it to use specific 00:04:44.329 --> 00:04:46.970 knowledge. Exactly. Then you move to the task. 00:04:47.199 --> 00:04:49.959 Analyze these 10 keywords and suggest a primary 00:04:49.959 --> 00:04:52.459 and secondary content strategy. Then comes the 00:04:52.459 --> 00:04:55.100 really vital piece, the context. We are a small, 00:04:55.279 --> 00:04:57.800 independent bakery in Paris, specializing only 00:04:57.800 --> 00:05:00.199 in artisanal sourdough. Which grounds the entire 00:05:00.199 --> 00:05:04.019 analysis. And finally, format. Give me a detailed 00:05:04.019 --> 00:05:06.959 markdown table with columns for keyword, target 00:05:06.959 --> 00:05:10.319 audience, and content angle. That structure just 00:05:10.319 --> 00:05:12.560 provides the perfect container for the output. 00:05:13.040 --> 00:05:15.740 I want to focus on context because it feels like 00:05:15.740 --> 00:05:18.459 the most neglected piece, yet you're saying it's 00:05:18.459 --> 00:05:20.360 the most critical filter. We're not just providing 00:05:20.360 --> 00:05:22.720 background information here, are we? No, not 00:05:22.720 --> 00:05:26.089 at all. Context is the magic. It eliminates millions 00:05:26.089 --> 00:05:28.629 of generic answers because it acts as this critical 00:05:28.629 --> 00:05:31.470 filter. Most people provide what we call empty 00:05:31.470 --> 00:05:34.170 context, and that's precisely why they fail. 00:05:34.990 --> 00:05:37.269 Can you give us an example of that failure? Sure. 00:05:37.569 --> 00:05:39.709 If you just say, give me a workout plan, the 00:05:39.709 --> 00:05:42.990 AI pulls this boring generalized routine based 00:05:42.990 --> 00:05:45.189 on the average person. That's empty context. 00:05:45.329 --> 00:05:47.250 The output is useless because it's not for anyone 00:05:47.250 --> 00:05:49.779 specific. But the professional AI communication 00:05:49.779 --> 00:05:52.319 prompt, it turns the AI from a general database 00:05:52.319 --> 00:05:55.680 into a custom personal trainer. So what specific 00:05:55.680 --> 00:05:57.720 pieces of context do we need to provide to make 00:05:57.720 --> 00:05:59.800 that shift actually happen? You need to give 00:05:59.800 --> 00:06:02.800 it the context a human PT would need. It's like 00:06:02.800 --> 00:06:05.459 stacking Lego blocks of data. You start with 00:06:05.459 --> 00:06:08.370 biological metrics weight. age, typical heart 00:06:08.370 --> 00:06:10.870 rate, so it can calculate your daily energy needs 00:06:10.870 --> 00:06:13.610 and ensure cardiovascular safety. That alone 00:06:13.610 --> 00:06:16.290 moves it way past generic internet advice. Then 00:06:16.290 --> 00:06:18.370 you give it environment and equipment details. 00:06:18.569 --> 00:06:20.810 You say, I only have resistance bands and a yoga 00:06:20.810 --> 00:06:24.399 mat. The AI is then forced to swap bench presses 00:06:24.399 --> 00:06:27.540 for push -up variations or heavy weights for 00:06:27.540 --> 00:06:30.740 band exercises. And crucially, you share your 00:06:30.740 --> 00:06:32.759 injury history. Let's say you have a 10 -year 00:06:32.759 --> 00:06:36.060 -old knee problem. That triggers the AI to enter 00:06:36.060 --> 00:06:38.680 a low -impact mode and it removes any jumping 00:06:38.680 --> 00:06:41.420 or high -impact moves. Goals and preferences. 00:06:41.800 --> 00:06:43.879 If I hate running, I can just tell it to swap 00:06:43.879 --> 00:06:46.199 all my cardio for cycling. That makes sure the 00:06:46.199 --> 00:06:48.040 plan is actually something I'll stick with. The 00:06:48.040 --> 00:06:50.120 detail is the difference between a throwaway 00:06:50.120 --> 00:06:53.240 draft and a real executable plan. And we can 00:06:53.240 --> 00:06:55.500 sharpen that output even more by setting constraints. 00:06:55.699 --> 00:06:57.699 These are the negative instructions. Yeah. You're 00:06:57.699 --> 00:07:00.920 telling the AI precisely what not to do. So what 00:07:00.920 --> 00:07:02.519 are some of the most effective constraints to 00:07:02.519 --> 00:07:05.240 use? They are powerful guards against that robotic 00:07:05.240 --> 00:07:08.120 tone we all hate. You can say things like, do 00:07:08.120 --> 00:07:11.180 not use corporate jargon like energy or paradigm 00:07:11.180 --> 00:07:14.220 shift, or keep the response under 200 words, 00:07:14.740 --> 00:07:17.540 or even avoid using the word comprehensive or 00:07:17.540 --> 00:07:20.699 unlock in the summary. So we have the scaffolding 00:07:20.699 --> 00:07:23.120 built with the golden square, but that perfect 00:07:23.120 --> 00:07:25.339 structure is only as good as the raw material 00:07:25.339 --> 00:07:28.220 we feed it. If we pick the wrong tool, it's just 00:07:28.220 --> 00:07:31.120 wasted effort. How do we know which AI brain 00:07:31.120 --> 00:07:33.439 is the best one for the job? By matching the 00:07:33.439 --> 00:07:35.420 problem to the unique strengths of each model, 00:07:35.920 --> 00:07:38.259 we're really past the point where one AI tool 00:07:38.259 --> 00:07:40.759 does everything well. Mastery means acknowledging 00:07:40.759 --> 00:07:43.459 that different AIs have specific personalities, 00:07:43.860 --> 00:07:46.339 and knowing which tool to use for which problem 00:07:46.339 --> 00:07:48.379 is paramount. Okay, so let's look at the four 00:07:48.379 --> 00:07:50.439 main personalities that are detailed in the source 00:07:50.439 --> 00:07:52.579 materials breakdown. What's the first one? We 00:07:52.579 --> 00:07:54.399 start with Chat GPT. You should think of it as 00:07:54.399 --> 00:07:57.819 your creative partner. It really excels at brainstorming, 00:07:58.060 --> 00:08:00.399 generating engaging stories, role playing, and 00:08:00.399 --> 00:08:03.000 just general idea generation. It has that really 00:08:03.000 --> 00:08:05.279 human -like flow which makes it great for early 00:08:05.279 --> 00:08:07.759 stage conceptual work. And then the next one, 00:08:07.879 --> 00:08:10.019 which seems to handle more of the heavy lifting. 00:08:10.220 --> 00:08:12.529 That would be Claude. This is your intellectual 00:08:12.529 --> 00:08:15.250 assistant. Claude is amazing at complex logic 00:08:15.250 --> 00:08:18.550 and following very detailed multi -step instructions. 00:08:19.129 --> 00:08:22.649 Its huge technical advantage is its massive context 00:08:22.649 --> 00:08:26.050 window. Its memory, basically. Right. You can 00:08:26.050 --> 00:08:29.750 upload a 100 -page PDF or a 20 ,000 -word book 00:08:29.750 --> 00:08:32.350 and ask it really nuanced questions without it 00:08:32.350 --> 00:08:33.909 forgetting the beginning of the document. Then 00:08:33.909 --> 00:08:36.110 we've got Gemini, the Google offering. Right. 00:08:36.110 --> 00:08:39.490 Gemini is your data and integration king. Because 00:08:39.490 --> 00:08:41.879 it's so deeply woven to the Google ecosystem. 00:08:42.259 --> 00:08:45.299 It can natively see your Google Drive. It can 00:08:45.299 --> 00:08:48.299 analyze real -time data, search your Gmail. It 00:08:48.299 --> 00:08:51.720 can even process live YouTube videos. For a student 00:08:51.720 --> 00:08:54.600 or a learner who needs immediate data analysis 00:08:54.600 --> 00:08:56.600 from their personal files, Gemini is often the 00:08:56.600 --> 00:08:58.379 fastest, most integrated choice. And finally, 00:08:58.519 --> 00:09:02.440 the specialized one, perplexity AI. Yes, perplexity 00:09:02.440 --> 00:09:05.539 AI. the research librarian. It doesn't guess 00:09:05.539 --> 00:09:07.840 or hallucinate because its main job is to search 00:09:07.840 --> 00:09:10.279 the live internet. It gives you precise footnotes 00:09:10.279 --> 00:09:12.700 with links back to its sources. So if you're 00:09:12.700 --> 00:09:15.159 writing a research paper, a factual report, or 00:09:15.159 --> 00:09:17.720 a high -spakes business proposal, perplexity 00:09:17.720 --> 00:09:20.460 is essential for accurate, anchored information. 00:09:20.659 --> 00:09:22.519 I can see how a freelance marketer could use 00:09:22.519 --> 00:09:25.179 this strategically in combining these strengths. 00:09:25.539 --> 00:09:27.899 What does that look like in practice? Okay, so 00:09:27.899 --> 00:09:30.379 let's say they need a content plan for a niche 00:09:30.379 --> 00:09:34.090 B2B software client. First, they use perplexity 00:09:34.090 --> 00:09:37.549 to find live trending web data. What questions 00:09:37.549 --> 00:09:39.490 are people asking? What are competitors missing? 00:09:39.789 --> 00:09:41.870 They get all those factual inputs. And then they 00:09:41.870 --> 00:09:43.809 take that structured data and... Right, and they 00:09:43.809 --> 00:09:46.590 feed that raw data directly into Claude using 00:09:46.590 --> 00:09:49.190 the golden square, asking it to act as a leading 00:09:49.190 --> 00:09:52.490 industry voice and generate five compelling non 00:09:52.490 --> 00:09:55.330 -robotic blog post titles based on that data. 00:09:55.509 --> 00:09:57.929 And why Claude for that part? because Claude 00:09:57.929 --> 00:10:01.210 currently has one of the most fluid, least mechanical 00:10:01.210 --> 00:10:03.970 writing styles available. So that combination 00:10:03.970 --> 00:10:07.330 hot data plus evocative language gives them a 00:10:07.330 --> 00:10:10.450 professional, ready to execute content plan almost 00:10:10.450 --> 00:10:13.100 immediately. That is the definition of leveraging 00:10:13.100 --> 00:10:15.740 multiple strengths. So we know which tool to 00:10:15.740 --> 00:10:18.940 use. But once we have that basic draft, how do 00:10:18.940 --> 00:10:21.500 we push it past the initial output and force 00:10:21.500 --> 00:10:24.519 it to produce truly expert level content? We 00:10:24.519 --> 00:10:26.860 push it past the default by teaching the AI to 00:10:26.860 --> 00:10:29.279 think harder with advanced prompting, sponsor 00:10:29.279 --> 00:10:31.559 saying. So here's where the pro strategy start 00:10:31.559 --> 00:10:33.580 and the science behind it is fascinating. We 00:10:33.580 --> 00:10:36.580 can begin with chain of thought or Cotet prompting. 00:10:36.679 --> 00:10:39.669 Define that mechanism for us. It sounds complicated, 00:10:39.730 --> 00:10:42.769 but it's deceptively simple, isn't it? It is. 00:10:42.769 --> 00:10:45.269 It's just adding the instruction. Think step 00:10:45.269 --> 00:10:47.529 -by -step before providing the final answer. 00:10:48.049 --> 00:10:49.970 And research consistently proves that if you 00:10:49.970 --> 00:10:52.629 ask an AI to first show its internal logic, its 00:10:52.629 --> 00:10:55.289 accuracy increases significantly, sometimes by 00:10:55.289 --> 00:10:57.629 15 or 20 percent. What's the mechanism there? 00:10:57.929 --> 00:11:00.029 Why does asking it to show its work actually 00:11:00.029 --> 00:11:02.570 make it smarter? You're essentially forcing the 00:11:02.570 --> 00:11:05.769 AI out of its fast, superficial system 1 thinking, 00:11:06.129 --> 00:11:08.549 which defaults to the most common answer, and 00:11:08.549 --> 00:11:11.870 into a slower, more deliberate system 2 process. 00:11:12.590 --> 00:11:15.529 It kind of mimics how our own prefrontal cortex 00:11:15.529 --> 00:11:18.830 solves complex problems. It's essential for anything 00:11:18.830 --> 00:11:21.909 high stakes, like calculating ROI or designing 00:11:21.909 --> 00:11:24.669 software logic. The next technique is few -shot 00:11:24.669 --> 00:11:27.789 prompting, which is basically making the AI a 00:11:27.789 --> 00:11:30.960 world -class copycat. Precisely. If you're trying 00:11:30.960 --> 00:11:33.240 to write in your unique voice, which is usually 00:11:33.240 --> 00:11:35.899 this messy mix of sentence lengths, tone, and 00:11:35.899 --> 00:11:38.480 humor, don't try to describe it. Just give the 00:11:38.480 --> 00:11:40.740 AI two or three excellent examples of your work. 00:11:41.000 --> 00:11:42.960 That teaches that your unique style immediately, 00:11:43.240 --> 00:11:45.340 and it ensures your brand or personal voice stays 00:11:45.340 --> 00:11:47.519 consistent. But the output process should be 00:11:47.519 --> 00:11:49.320 a dynamic conversation, right? Not just a single 00:11:49.320 --> 00:11:51.700 shot monologue. We have to embrace iterative 00:11:51.700 --> 00:11:54.480 refinement. Absolutely. We have to stop expecting 00:11:54.480 --> 00:11:56.840 the first answer to be the best one. The process 00:11:56.840 --> 00:11:59.039 is a necessary back and forth. You get the draft, 00:11:59.159 --> 00:12:00.840 and then you say, OK, now shorten the introduction 00:12:00.840 --> 00:12:03.019 by 50 % and make the tone a bit more aggressive. 00:12:03.220 --> 00:12:05.679 Or turn that last paragraph into a strong call 00:12:05.679 --> 00:12:08.100 to action for my newsletter. Exactly. You know, 00:12:08.220 --> 00:12:10.299 I still wrestle with prompt drift myself if I'm 00:12:10.299 --> 00:12:12.919 not careful. For the listener, what is prompt 00:12:12.919 --> 00:12:16.399 drift? That's when the AI slowly forgets your 00:12:16.399 --> 00:12:18.419 original instructions as the conversation gets 00:12:18.419 --> 00:12:21.240 longer. You know, you spend 30 turns discussing 00:12:21.240 --> 00:12:23.980 a business plan, and by turn 31, it starts making 00:12:23.980 --> 00:12:26.480 completely irrelevant suggestions. Have you had 00:12:26.480 --> 00:12:28.960 a bad case of that recently? And how did you 00:12:28.960 --> 00:12:32.320 save the output? Just last week, I was optimizing 00:12:32.320 --> 00:12:35.259 this dense technical guide. And after about 45 00:12:35.259 --> 00:12:37.940 minutes, it completely forgot the tone constraint. 00:12:38.340 --> 00:12:41.320 It reverted to this sterile academic language. 00:12:41.620 --> 00:12:44.159 Frustrating. Yeah. But the fix wasn't starting 00:12:44.159 --> 00:12:46.440 over. The fix was just pasting my original golden 00:12:46.440 --> 00:12:48.500 square prompt back into the chat and saying, 00:12:49.059 --> 00:12:51.200 read here to this role in format and continue 00:12:51.200 --> 00:12:54.059 from the last paragraph. That resets the model 00:12:54.059 --> 00:12:57.870 instantly. That reset command is gold. My favorite 00:12:57.870 --> 00:13:00.330 technique to get airtight content is forcing 00:13:00.330 --> 00:13:03.490 the AI to flip from creator to critic, the ask 00:13:03.490 --> 00:13:06.149 for critiques method. Oh yeah, this is the expert 00:13:06.149 --> 00:13:08.629 level move that gets you past the polite average. 00:13:09.149 --> 00:13:11.830 The AI is programmed to be agreeable. So to trigger 00:13:11.830 --> 00:13:14.409 a self -correction loop, you have to demand a 00:13:14.409 --> 00:13:17.070 critique from a specific skeptical perspective. 00:13:17.470 --> 00:13:20.730 So I'd ask the AI to act as a skeptical executive 00:13:20.730 --> 00:13:22.970 and then critique the very draft it just wrote. 00:13:23.080 --> 00:13:25.440 Exactly. You demand it point out the logical 00:13:25.440 --> 00:13:27.740 gaps, the potential for misunderstanding, any 00:13:27.740 --> 00:13:30.940 robotic pros. Then, and this is the key, you 00:13:30.940 --> 00:13:33.320 tell it to rewrite the entire piece based on 00:13:33.320 --> 00:13:36.039 those self -identified flaws. You're using its 00:13:36.039 --> 00:13:38.419 own analytical power to improve its creative 00:13:38.419 --> 00:13:41.399 output. Even with great prompts and self -critique, 00:13:41.580 --> 00:13:44.059 it's exhausting to repeat business details like 00:13:44.059 --> 00:13:47.379 company history or brand voice every single day. 00:13:47.820 --> 00:13:50.139 How do we manage all this context for long -term 00:13:50.139 --> 00:13:52.360 efficiency? We need to set up permanent context 00:13:52.360 --> 00:13:54.360 files and custom instructions. We essentially 00:13:54.360 --> 00:13:57.159 teach the AI who we are one time. Permanent context 00:13:57.159 --> 00:13:59.539 is all about defining yourself so the AI never 00:13:59.539 --> 00:14:01.600 has to ask again. Think of custom instructions 00:14:01.600 --> 00:14:04.059 or system prompts as your AI's permanent personality 00:14:04.059 --> 00:14:06.720 file. So in a tool like ChatGPT, you'd input 00:14:06.720 --> 00:14:09.840 these permanent facts like, I live in Vietnam 00:14:09.840 --> 00:14:12.539 and work remotely, or my audience is beginner 00:14:12.539 --> 00:14:15.960 entrepreneurs, or I prefer short, bulleted lists. 00:14:16.250 --> 00:14:19.429 Correct. And that context is always active, globally, 00:14:19.750 --> 00:14:22.450 for every new chat. The AI just knows who you 00:14:22.450 --> 00:14:25.070 are. The counterpart to that is the project context 00:14:25.070 --> 00:14:26.950 file. What's the difference between the permanent 00:14:26.950 --> 00:14:29.759 personality and this project file? The project 00:14:29.759 --> 00:14:32.240 context is just a simple text file you create 00:14:32.240 --> 00:14:35.120 with specific temporary details for the client 00:14:35.120 --> 00:14:36.980 or project you're working on right now. You know, 00:14:37.179 --> 00:14:39.759 client names, competitor URLs, product limitations. 00:14:40.120 --> 00:14:42.100 You just upload that file at the start of every 00:14:42.100 --> 00:14:44.299 new chat and it saves you 10 minutes of typing 00:14:44.299 --> 00:14:46.759 every time. So now we're moving from that manual 00:14:46.759 --> 00:14:49.879 back and forth chat to building autonomous systems. 00:14:50.440 --> 00:14:52.940 We have to adopt the automation mindset. The 00:14:52.940 --> 00:14:55.700 shift here is huge. Don't just ask the AI to 00:14:55.700 --> 00:14:58.100 describe a solution or write a paragraph. The 00:14:58.100 --> 00:15:00.220 power user gives it the authority to execute 00:15:00.220 --> 00:15:03.299 an entire repetitive process. This is the difference 00:15:03.299 --> 00:15:05.659 between a high -speed typewriter and a scalable 00:15:05.659 --> 00:15:08.299 workflow engine. So what's the criteria for that? 00:15:08.559 --> 00:15:10.759 How do we identify the tasks that are ready for 00:15:10.759 --> 00:15:13.720 automation, these low -value loops? You look 00:15:13.720 --> 00:15:16.559 for tasks that are three things. Repetitive, 00:15:16.879 --> 00:15:19.889 logic -based, and boring. They follow a clear 00:15:19.889 --> 00:15:23.090 if -this -then -that rule. They require focus, 00:15:23.190 --> 00:15:25.710 but zero creative inspiration, and they just 00:15:25.710 --> 00:15:28.090 take up time. Like summarizing meeting transcripts 00:15:28.090 --> 00:15:31.110 or triaging support tickets. Exactly. Or drafting 00:15:31.110 --> 00:15:33.809 follow -up emails based on a spreadsheet of customer 00:15:33.809 --> 00:15:35.990 statuses. And to make that system actually execute, 00:15:36.070 --> 00:15:38.629 we need interconnectors. This is the glue. Yeah, 00:15:38.629 --> 00:15:40.870 that's right. Tools like Zapier and Make are 00:15:40.870 --> 00:15:43.330 the central nervous system. They connect the 00:15:43.330 --> 00:15:45.850 AI's specialized brain to the hands of your business, 00:15:45.970 --> 00:15:49.490 your other apps, like Slack, or a sauna. This 00:15:49.490 --> 00:15:52.190 eliminates the human middleman and that allows 00:15:52.190 --> 00:15:55.070 the output to scale globally without anyone manually 00:15:55.070 --> 00:15:57.070 copying and pasting. And that infrastructure 00:15:57.070 --> 00:16:00.070 lets us build very specific tailored agents or 00:16:00.070 --> 00:16:03.090 custom GPTs. Think about creating a social media 00:16:03.090 --> 00:16:06.009 manager agent. You train it on your specific 00:16:06.009 --> 00:16:08.730 brand voice, your posting schedules. You feed 00:16:08.730 --> 00:16:11.830 it a link to a new blog post and because it has 00:16:11.830 --> 00:16:14.590 the context and authority it automatically spits 00:16:14.590 --> 00:16:18.250 out five unique tweets detailed LinkedIn post, 00:16:18.590 --> 00:16:22.750 and a bulleted TikTok script. Whoa. Imagine scaling 00:16:22.750 --> 00:16:25.330 that process to a billion queries without copying 00:16:25.330 --> 00:16:28.429 and pasting a single thing. That's true leverage. 00:16:28.850 --> 00:16:31.330 It is. But automation is incredibly powerful, 00:16:31.350 --> 00:16:35.090 but AI does still hallucinate. So how do we ensure 00:16:35.090 --> 00:16:37.490 that all this speed and scale doesn't compromise 00:16:37.490 --> 00:16:40.429 quality or accuracy? By becoming a world -class 00:16:40.429 --> 00:16:43.029 fact checker and implementing crucial verification 00:16:43.029 --> 00:16:45.549 checks at every step. And the first most critical 00:16:45.549 --> 00:16:47.490 safety instruction here is source anchoring. 00:16:47.789 --> 00:16:49.870 This is completely non -negotiable whenever you're 00:16:49.870 --> 00:16:52.159 dealing with attached documents or data. You 00:16:52.159 --> 00:16:54.899 tell the AI, only use information provided in 00:16:54.899 --> 00:16:56.980 this attached PDF. If the answer is not contained 00:16:56.980 --> 00:16:59.360 within the PDF, you must say, I do not know. 00:16:59.460 --> 00:17:01.360 And that anchors it. It stops it from guessing. 00:17:01.559 --> 00:17:03.580 It completely stops it from guessing or trying 00:17:03.580 --> 00:17:05.779 to fill in the blanks with generalized knowledge. 00:17:06.299 --> 00:17:09.099 And for really high stakes information, legal 00:17:09.099 --> 00:17:11.059 summaries, medical reports, financial stuff, 00:17:11.660 --> 00:17:14.759 we need the cross -model verification method. 00:17:14.980 --> 00:17:17.440 It's the double -check system. You use model 00:17:17.440 --> 00:17:19.700 A, let's say Claude, to do the heavy lifting 00:17:19.700 --> 00:17:22.400 because you know it has a long memory. Then you 00:17:22.400 --> 00:17:24.779 immediately paste that answer into model B, like 00:17:24.779 --> 00:17:27.500 chat GPT, and you ask it to check the summary 00:17:27.500 --> 00:17:30.640 for errors, inconsistencies, or missing information. 00:17:30.859 --> 00:17:33.559 And if they disagree, you investigate. If they 00:17:33.559 --> 00:17:36.559 agree, you can feel much safer sending that document 00:17:36.559 --> 00:17:39.079 forward. Precisely. You're using their differences 00:17:39.079 --> 00:17:41.440 in training and architecture as a built -in safety 00:17:41.440 --> 00:17:44.160 net. But ultimately, the human in the loop is 00:17:44.160 --> 00:17:47.230 non -negotiable. It's the pilot -co -pilot analogy. 00:17:47.490 --> 00:17:49.549 The AI is the co -pilot. It's doing the bulk 00:17:49.549 --> 00:17:51.789 of the work, managing the data, flying the plane. 00:17:51.950 --> 00:17:54.849 But you are the pilot who is always, always responsible 00:17:54.849 --> 00:17:57.529 for the landing. Never let the AI post directly 00:17:57.529 --> 00:17:59.849 to social media or send an email to a client 00:17:59.849 --> 00:18:02.190 or execute a financial trade without your final 00:18:02.190 --> 00:18:05.009 human review. That responsibility just can't 00:18:05.009 --> 00:18:07.109 be delegated. Before we wrap up, let's just quickly 00:18:07.109 --> 00:18:09.509 run through the don't list, the common traps 00:18:09.509 --> 00:18:12.569 that still waste so much user time. Right. Don't 00:18:12.569 --> 00:18:15.509 give wall of text prompts. AIs stan text, they 00:18:15.509 --> 00:18:18.029 love structure. Use bullet points and numbered 00:18:18.029 --> 00:18:21.200 lists for clarity. And critically, don't use 00:18:21.200 --> 00:18:24.160 AI for things it's inherently bad at, like high 00:18:24.160 --> 00:18:26.880 -level strategy that needs deep market context 00:18:26.880 --> 00:18:30.779 or true emotional empathy. Use it for the heavy 00:18:30.779 --> 00:18:33.000 lifting of data analysis and drafting. So the 00:18:33.000 --> 00:18:35.880 big idea here, really, is that effective AI use 00:18:35.880 --> 00:18:38.279 isn't about chasing the next shiny new tool. 00:18:38.400 --> 00:18:41.500 It is about clarity, structure, verification, 00:18:41.940 --> 00:18:44.440 and understanding the core principles of communication. 00:18:44.700 --> 00:18:46.920 The biggest lesson after spending hundreds of 00:18:46.920 --> 00:18:49.170 hours building these systems is that AI doesn't 00:18:49.170 --> 00:18:52.170 replace your brain. It amplifies it. It's a true 00:18:52.170 --> 00:18:54.630 force multiplier. If you are a clear, structured 00:18:54.630 --> 00:18:57.289 communicator, AI will make you superhuman. You'll 00:18:57.289 --> 00:18:58.970 achieve an output quality and a speed that were 00:18:58.970 --> 00:19:01.250 previously impossible. But if you're disorganized, 00:19:01.289 --> 00:19:03.650 if your inputs are vague and you can't articulate 00:19:03.650 --> 00:19:05.829 exactly what you need, AI will just make you 00:19:05.829 --> 00:19:08.009 disorganized faster. It just speeds up the mess. 00:19:08.349 --> 00:19:11.109 The key to mastery is personal clarity. The roadmap 00:19:11.109 --> 00:19:13.549 is clear, and the tools are definitely ready 00:19:13.549 --> 00:19:16.390 for serious delegation. I think the key question 00:19:16.390 --> 00:19:18.470 now is, What are you going to build with these 00:19:18.470 --> 00:19:20.549 skills? We covered a huge amount of material 00:19:20.549 --> 00:19:22.990 today, so let's give you a simple 30 -day plan. 00:19:23.430 --> 00:19:26.049 This week, just focus on Clarity Make every single 00:19:26.049 --> 00:19:28.970 prompt, a role task context format prompt. In 00:19:28.970 --> 00:19:31.210 a couple of weeks, focus on refinement try, that 00:19:31.210 --> 00:19:33.650 advanced technique of critiquing the AI's own 00:19:33.650 --> 00:19:36.869 work. Then try a simple automation with Zapier 00:19:36.869 --> 00:19:39.670 or Make. Just pick one technique and try it today. 00:19:39.930 --> 00:19:40.490 Get started.