WEBVTT 00:00:00.000 --> 00:00:02.240 You know that feeling? Oh, I do. It's like 8 00:00:02.240 --> 00:00:05.379 AM on a Tuesday. You open up X, maybe LinkedIn, 00:00:05.540 --> 00:00:08.240 and your stomach just drops. The feed is just 00:00:08.240 --> 00:00:11.240 screaming at you. Exactly. Yesterday, everyone 00:00:11.240 --> 00:00:14.720 said you had to master chat GPT. Today, there's 00:00:14.720 --> 00:00:17.460 a new tool that supposedly makes that one completely 00:00:17.460 --> 00:00:20.320 obsolete. It feels like you're running a race 00:00:20.320 --> 00:00:23.920 where the finish line just keeps moving. We call 00:00:23.920 --> 00:00:27.420 it AI fatigue. It is exhausting. And it's totally 00:00:27.420 --> 00:00:29.620 counterproductive. It creates this paralysis 00:00:29.620 --> 00:00:32.500 where you feel like if you aren't using 50 different 00:00:32.500 --> 00:00:35.100 apps, you're already failing. But here's the 00:00:35.100 --> 00:00:37.159 secret we're going to unpack today. The people 00:00:37.159 --> 00:00:39.859 who are actually winning this game. They are 00:00:39.859 --> 00:00:42.460 not the ones chasing every shiny new toy. They 00:00:42.460 --> 00:00:44.740 are the ones mastering the fundamentals. Right. 00:00:44.759 --> 00:00:47.200 They are memorizing dashboards. They're learning 00:00:47.200 --> 00:00:50.520 how the machine actually thinks. Welcome to the 00:00:50.520 --> 00:00:52.840 Deep Dive. Today, we are slowing things down. 00:00:53.000 --> 00:00:55.759 We're taking a look at the 2026 AI Fundamentals 00:00:55.759 --> 00:00:58.100 Handbook. I love this. And think of this as the 00:00:58.100 --> 00:01:00.619 antidote to all that overwhelming noise. We're 00:01:00.619 --> 00:01:02.219 going to strip away the hype and just look at 00:01:02.219 --> 00:01:04.879 the underlying skills. The driving skills. The 00:01:04.879 --> 00:01:07.219 driving skills that apply no matter what vehicle 00:01:07.219 --> 00:01:09.319 you're in. I love that analogy. It's so good. 00:01:09.400 --> 00:01:11.560 If you learn to drive on a Toyota, you can hop 00:01:11.560 --> 00:01:14.319 into a Ford and you'll be just fine. Sure. But 00:01:14.319 --> 00:01:16.819 if you only memorize which button turns on the 00:01:16.819 --> 00:01:19.769 radio in that Toyota, You are lost the second 00:01:19.769 --> 00:01:23.450 you switch cars. Today is about learning to drive. 00:01:24.170 --> 00:01:26.560 We have a roadmap for this conversation. We're 00:01:26.560 --> 00:01:29.260 covering five pillars that the source material 00:01:29.260 --> 00:01:33.120 says are essential for 2026. OK. We're talking 00:01:33.120 --> 00:01:36.700 prompting, but the rigorous way. We're talking 00:01:36.700 --> 00:01:39.980 categorization, agents, open source, and something 00:01:39.980 --> 00:01:42.540 called vibe coding. Which is my absolute favorite 00:01:42.540 --> 00:01:45.200 part. It's wild. We will get there. But first, 00:01:45.340 --> 00:01:47.359 let's start with the most critical skill, just 00:01:47.359 --> 00:01:50.060 talking to the machine. The source makes a really 00:01:50.060 --> 00:01:52.319 interesting distinction here. It says most people 00:01:52.319 --> 00:01:55.450 fail because they treat AI like Google. This 00:01:55.450 --> 00:01:58.670 is the classic mistake. When we use Google, we've 00:01:58.670 --> 00:02:00.730 been trained for 20 years to think in keywords. 00:02:01.310 --> 00:02:04.890 Best pasta recipe. Weather London. Right. A search 00:02:04.890 --> 00:02:08.210 query. It's a search query. But AI isn't a search 00:02:08.210 --> 00:02:11.229 engine. It's a reasoning engine. You aren't searching. 00:02:11.650 --> 00:02:14.030 You're constructing. Constructing. That is a 00:02:14.030 --> 00:02:16.669 key word. It is. You're building a set of instructions. 00:02:17.490 --> 00:02:19.909 The source introduces a framework for this. It's 00:02:19.909 --> 00:02:24.439 called T -C -R -E -I. T -C -R -E -I. OK, let's 00:02:24.439 --> 00:02:28.080 unpack this acronym. So T stands for task. The 00:02:28.080 --> 00:02:30.139 task. And this is where people get lazy. They 00:02:30.139 --> 00:02:32.780 say, help me with a workout. That's way too vague. 00:02:33.039 --> 00:02:36.120 It's so weak. Yeah. The AI has to guess. Are 00:02:36.120 --> 00:02:38.620 you a bodybuilder? Are you training for a marathon? 00:02:39.199 --> 00:02:41.860 The framework demands strong verbs. So instead 00:02:41.860 --> 00:02:45.240 of help me, use create a detailed four week running 00:02:45.240 --> 00:02:47.020 schedule. It's the difference between asking 00:02:47.020 --> 00:02:50.379 a friend what should I eat versus order me a 00:02:50.379 --> 00:02:53.039 pepperoni pizza. Precisely. Clarity is king. 00:02:53.539 --> 00:02:56.039 Which brings us to C for context. Who, where, 00:02:56.159 --> 00:02:58.439 and why. Exactly. I love the example the handbook 00:02:58.439 --> 00:03:00.159 gave here. It really paints a clear picture. 00:03:00.319 --> 00:03:01.919 It does. So instead of just create a running 00:03:01.919 --> 00:03:04.039 schedule, you add all this detail. I am a 40 00:03:04.039 --> 00:03:06.120 -year -old male. I haven't exercised in five 00:03:06.120 --> 00:03:08.580 years. I have bad knees, and I can only run on 00:03:08.580 --> 00:03:10.759 Tuesdays, Thursdays, and Saturdays. Think about 00:03:10.759 --> 00:03:13.379 how much better the output becomes with that 00:03:13.379 --> 00:03:16.939 context. Without it, the AI gives you this generic 00:03:16.939 --> 00:03:19.500 plan that might just blow out your knees in week 00:03:19.500 --> 00:03:22.039 one. With context, it understands constraints. 00:03:22.259 --> 00:03:24.400 It becomes a personalized coach, not some generic 00:03:24.400 --> 00:03:26.500 article generator. Then we have our references. 00:03:26.620 --> 00:03:28.460 The source calls this one the secret weapon. 00:03:28.580 --> 00:03:31.620 And it is. This is the step almost everyone skips. 00:03:32.219 --> 00:03:35.479 It's incredibly difficult to describe a vibe 00:03:35.479 --> 00:03:39.449 or a tone with just adjectives. If I tell you 00:03:39.449 --> 00:03:42.590 to write something professional but witty, your 00:03:42.590 --> 00:03:44.389 definition of that is probably totally different 00:03:44.389 --> 00:03:46.270 from mine. So instead of trying to describe it, 00:03:46.270 --> 00:03:48.949 you just show it. Exactly. You paste in an example, 00:03:49.229 --> 00:03:50.870 a previous email you wrote that worked really 00:03:50.870 --> 00:03:53.169 well, a LinkedIn post you liked. You just tell 00:03:53.169 --> 00:03:56.169 the AI, use the writing style, sentence length, 00:03:56.530 --> 00:03:59.250 and tone of the text below. It mimics the pattern. 00:03:59.370 --> 00:04:02.710 And that's how you escape that robotic generic 00:04:02.710 --> 00:04:04.969 AI voice. It's the only way. Then finally, you 00:04:04.969 --> 00:04:08.069 have E and I. evaluate and iterate the reality 00:04:08.069 --> 00:04:10.129 check I have to be vulnerable here for a second 00:04:10.129 --> 00:04:12.830 this is where I struggle I still have this magic 00:04:12.830 --> 00:04:15.650 button expectation you know yeah everyone does 00:04:15.650 --> 00:04:18.509 I type in the prompt and if the result isn't 00:04:18.509 --> 00:04:21.149 perfect I get frustrated I call it prompt drift 00:04:21.149 --> 00:04:23.649 I just kind of give up you and everyone else 00:04:23.649 --> 00:04:26.430 but the source makes such a vital point here 00:04:26.430 --> 00:04:30.069 the AI gets you 80 % of the way there which is 00:04:30.069 --> 00:04:33.790 a miracle in itself it is but that last 20 % 00:04:34.269 --> 00:04:37.069 That is on you. You have to talk back to it. 00:04:37.129 --> 00:04:39.230 Make it more enthusiastic. You missed the part 00:04:39.230 --> 00:04:41.889 about my knees. Shorten that third paragraph. 00:04:42.110 --> 00:04:44.470 So it's a collaboration, not a transaction. Correct. 00:04:44.649 --> 00:04:47.290 Let me ask you this, though. If the task and 00:04:47.290 --> 00:04:49.949 context are clear, if I tell it what to do and 00:04:49.949 --> 00:04:52.910 who I am, why does the source say the reference 00:04:52.910 --> 00:04:55.629 step is the one people usually skip? Why is that 00:04:55.629 --> 00:04:57.829 the hurdle? Because it requires digging up old 00:04:57.829 --> 00:05:00.899 work. It requires friction. You have to go find 00:05:00.899 --> 00:05:03.139 a file, copy and paste it. We're lazy. We want 00:05:03.139 --> 00:05:06.040 to type one sentence and be done. But that friction 00:05:06.040 --> 00:05:08.420 is the only price you pay to escape mediocrity. 00:05:08.839 --> 00:05:10.579 It's the only way to sound like you and not some 00:05:10.579 --> 00:05:13.240 robot. That's a perfect segue into the tools 00:05:13.240 --> 00:05:15.439 themselves. Because once you know how to talk 00:05:15.439 --> 00:05:17.660 to them, you realize there are hundreds of them. 00:05:18.519 --> 00:05:20.779 The source calls this section the core four. 00:05:20.839 --> 00:05:22.800 And this is all about simplification. You do 00:05:22.800 --> 00:05:25.279 not need 50 apps. You just need to fill four 00:05:25.279 --> 00:05:28.600 buckets. Bucket one. General reasoning engines. 00:05:28.800 --> 00:05:32.129 The brain. These are your daily drivers. Your 00:05:32.129 --> 00:05:35.709 chat GPT, Claude, Gemini. They're for logic, 00:05:35.970 --> 00:05:38.149 for writing, for brainstorming. The source says 00:05:38.149 --> 00:05:40.930 just pick one based on vibes. Yeah, the author 00:05:40.930 --> 00:05:43.189 mentioned they like Gemini for writing, but chat 00:05:43.189 --> 00:05:45.750 GPT for logic. And that's fine. They're all racing 00:05:45.750 --> 00:05:48.310 neck and neck. Just pick one. But, and this is 00:05:48.310 --> 00:05:51.149 so critical, do not use them for facts. Which 00:05:51.149 --> 00:05:53.990 brings us to bucket two, research engines. This 00:05:53.990 --> 00:05:56.250 is maybe the most dangerous confusion in AI right 00:05:56.250 --> 00:05:59.180 now. General engines, like ChatGPT, are prediction 00:05:59.180 --> 00:06:01.720 machines. They just predict the next most likely 00:06:01.720 --> 00:06:04.240 word. They don't know facts. They hallucinate. 00:06:04.560 --> 00:06:07.519 If you ask ChatGPT who won the World Cup in 1998, 00:06:08.360 --> 00:06:10.680 it might be right, or I might just say a country 00:06:10.680 --> 00:06:12.839 that sounds plausible. So for facts, you need 00:06:12.839 --> 00:06:15.779 a different tool, a research engine. Right. Tools 00:06:15.779 --> 00:06:18.639 like Perplexity or Consensus. These actually 00:06:18.639 --> 00:06:20.699 connect to the live internet. They read real 00:06:20.699 --> 00:06:22.839 websites. They cite their sources. They don't 00:06:22.839 --> 00:06:25.139 just guess, they report. OK, so bucket three. 00:06:25.259 --> 00:06:28.720 Specialist tools, the artists. These are for 00:06:28.720 --> 00:06:31.180 when good enough isn't good enough. Mid -Journey 00:06:31.180 --> 00:06:34.399 for images, 11 Labs for audio, Sora for video. 00:06:34.579 --> 00:06:37.379 So a general tool like ChatGPT can make an image, 00:06:37.660 --> 00:06:40.160 sure. But a specialist tool like Mid -Journey 00:06:40.160 --> 00:06:43.720 will make art. And finally, bucket four, workflow 00:06:43.720 --> 00:06:48.279 automators, the glue. Xavier, make. These are 00:06:48.279 --> 00:06:49.680 the boring ones that will save your life. They 00:06:49.680 --> 00:06:51.579 just move data. When I get an email attachment 00:06:51.579 --> 00:06:53.600 and save it to Dropbox, that kind of thing. So 00:06:53.600 --> 00:06:55.779 looking at this whole landscape, the danger isn't 00:06:55.779 --> 00:06:58.459 missing out on the latest hot tool. The danger 00:06:58.459 --> 00:07:00.720 is using a brain tool when you actually need 00:07:00.720 --> 00:07:03.420 a research tool. Exactly. You're using a creative 00:07:03.420 --> 00:07:06.160 storyteller to check facts. That is just a recipe 00:07:06.160 --> 00:07:08.740 for disaster. Moving on to the third pillar. 00:07:08.939 --> 00:07:10.879 This one feels like a real glimpse into the future. 00:07:10.959 --> 00:07:14.000 The shift from chatbots to agents. This is the 00:07:14.000 --> 00:07:17.139 difference between an assistant and an intern. 00:07:17.310 --> 00:07:19.629 break that down for us. Okay think about the 00:07:19.629 --> 00:07:22.949 old way or I mean the current way for most people. 00:07:22.970 --> 00:07:25.589 Right. The chat bot. You treat it like a conversation 00:07:25.589 --> 00:07:27.790 partner. You want to book a flight to London. 00:07:27.910 --> 00:07:30.029 You ask what flights are available. It gives 00:07:30.029 --> 00:07:32.250 you a list. Then you have to go to the website. 00:07:32.449 --> 00:07:34.649 You type in your credit card. You hit book. You're 00:07:34.649 --> 00:07:36.329 the middleman. You're doing all the leg work. 00:07:36.449 --> 00:07:39.350 Right. An agent is different. With an agent you 00:07:39.350 --> 00:07:41.910 just say book the cheapest flight to London for 00:07:41.910 --> 00:07:45.220 Tuesday. Yeah. Then you walk away. Wow. The agent 00:07:45.220 --> 00:07:48.220 browses the web, it selects the flight, it inputs 00:07:48.220 --> 00:07:50.540 your payment info, and it emails you the ticket. 00:07:50.660 --> 00:07:53.579 That's a massive shift. It's delegating versus 00:07:53.579 --> 00:07:55.620 just asking for information. It changes your 00:07:55.620 --> 00:07:58.620 role entirely. The source uses the example of 00:07:58.620 --> 00:08:01.180 researching coffee history. In the chatbot era, 00:08:01.680 --> 00:08:03.860 you ask for articles, you read them, you synthesize 00:08:03.860 --> 00:08:06.000 them. Right. In the agent era, you tell the agent, 00:08:06.519 --> 00:08:08.839 read 10 articles on coffee history, filter out 00:08:08.839 --> 00:08:12.079 the bad info, and write me an outline. It's incredible. 00:08:12.180 --> 00:08:14.399 We're seeing this with tools like Perplexity's 00:08:14.399 --> 00:08:17.379 Agent Features or Claude Projects, where it acts 00:08:17.379 --> 00:08:20.220 as an expert on your specific data. It's awe 00:08:20.220 --> 00:08:22.779 -inspiring, really. We're moving from being the 00:08:22.779 --> 00:08:25.420 worker bees to being the managers of these digital 00:08:25.420 --> 00:08:28.420 interns. So does this mean the fundamental skill 00:08:28.420 --> 00:08:30.899 itself shifts? If I'm not doing the research, 00:08:31.060 --> 00:08:33.679 what am I doing? You're verifying. The skill 00:08:33.679 --> 00:08:36.929 shifts from gathering the news... to being the 00:08:36.929 --> 00:08:38.929 editor -in -chief. Right. We become editors -in 00:08:38.929 --> 00:08:41.629 -chief, not the beat reporter. Exactly. Your 00:08:41.629 --> 00:08:44.450 judgment becomes your most valuable asset. We're 00:08:44.450 --> 00:08:46.350 going to take a quick break to thank our partners 00:08:46.350 --> 00:08:49.860 who make this deep dive possible. And we are 00:08:49.860 --> 00:08:52.360 back. We've talked about prompting, tools, and 00:08:52.360 --> 00:08:54.500 agents. Now we need to talk about something that 00:08:54.500 --> 00:08:57.620 feels a bit more technical. But the source argues 00:08:57.620 --> 00:09:00.720 is actually about privacy and longevity. Open 00:09:00.720 --> 00:09:03.460 source. Owning the engine. Right. The concept 00:09:03.460 --> 00:09:06.740 here is renting versus owning. When we use chat 00:09:06.740 --> 00:09:09.620 GPT, we are renting intelligence from open AI. 00:09:09.740 --> 00:09:12.299 And that comes with baggage. First, privacy. 00:09:12.440 --> 00:09:14.840 They can see your data. If you're pasting in 00:09:14.840 --> 00:09:16.980 confidential business strategy or your personal 00:09:16.980 --> 00:09:19.240 journal, that's just sitting on someone else's 00:09:19.240 --> 00:09:21.820 server. Yeah, that's a big one. Second, cost. 00:09:22.340 --> 00:09:24.600 You're paying that monthly subscription. And 00:09:24.600 --> 00:09:28.360 third, control. If their servers go down, your 00:09:28.360 --> 00:09:31.080 brain goes offline. So the solution is open source 00:09:31.080 --> 00:09:34.120 models. Things like Llama 3 or DeepSeek. Exactly. 00:09:34.200 --> 00:09:36.120 These are models you can actually download. You 00:09:36.120 --> 00:09:38.500 download the brain directly to your laptop. It 00:09:38.500 --> 00:09:41.100 runs locally. So no data leaves your computer. 00:09:41.389 --> 00:09:43.970 None. It works offline. It costs zero dollars. 00:09:44.190 --> 00:09:47.389 The source mentions a tool called Olama. O -L 00:09:47.389 --> 00:09:50.470 -L -A -M -A. Olama has been a complete game changer. 00:09:50.710 --> 00:09:52.549 It used to be you needed to be a hacker to run 00:09:52.549 --> 00:09:55.470 these things. Now you download Olama, click a 00:09:55.470 --> 00:09:57.690 button, and you have a chatbot running on your 00:09:57.690 --> 00:10:00.509 hard drive that looks just like ChatGPT. Is this 00:10:00.509 --> 00:10:03.710 strictly for the privacy nets, the tinfoil hat 00:10:03.710 --> 00:10:06.809 brigade, or is there a practical longevity argument 00:10:06.809 --> 00:10:09.190 here? It is absolutely longevity play. It's future 00:10:09.190 --> 00:10:11.350 proofing. How so? Imagine if Internet prices 00:10:11.350 --> 00:10:13.929 spike or regulations change or a company goes 00:10:13.929 --> 00:10:16.029 out of business. If you rely entirely on the 00:10:16.029 --> 00:10:18.090 cloud, you're vulnerable. If you have the model 00:10:18.090 --> 00:10:20.389 on your drive, you own your own digital brain 00:10:20.389 --> 00:10:23.409 forever. No one can take it away from you. That 00:10:23.409 --> 00:10:26.190 brings us to our final pillar. And honestly, 00:10:26.450 --> 00:10:29.200 this is the one that blew my mind the most. No 00:10:29.200 --> 00:10:31.279 code building or as the cool kids are calling 00:10:31.279 --> 00:10:36.019 it vibe coding I Love the term it perfectly captures 00:10:36.019 --> 00:10:39.120 how the barrier to entry has just well It's dissolved 00:10:39.120 --> 00:10:41.419 for 30 years if you wanted to build software 00:10:41.419 --> 00:10:44.360 you had to learn a foreign language Python JavaScript 00:10:44.360 --> 00:10:48.200 C++ says now English is the programming language 00:10:48.200 --> 00:10:50.399 the source gives this example of a small business 00:10:50.399 --> 00:10:52.659 owner They want a calculator on their website 00:10:52.659 --> 00:10:56.120 in the old days That's a $500 job for a freelancer 00:10:56.120 --> 00:10:58.990 or a month of trying to learn code yourself Now, 00:10:59.669 --> 00:11:02.870 you open a tool like Replet or Cursor, and you 00:11:02.870 --> 00:11:04.509 just describe it. What would that sound like? 00:11:04.809 --> 00:11:07.309 You just type, create a webpage with sliders 00:11:07.309 --> 00:11:10.409 for hours in rate, multiply them for total cost, 00:11:10.730 --> 00:11:13.309 make the background blue. And it just appears. 00:11:13.409 --> 00:11:15.169 And if you don't like the blue... You don't rewrite 00:11:15.169 --> 00:11:17.009 the code. You just say, change the background 00:11:17.009 --> 00:11:19.289 to white. And the AI rewrites the code for you. 00:11:19.549 --> 00:11:21.450 This is empowering on a level I don't think we 00:11:21.450 --> 00:11:24.230 fully grasp yet. It means the only limit is your 00:11:24.230 --> 00:11:27.370 imagination. The tools listed like Replet for 00:11:27.370 --> 00:11:30.370 beginners with its agent feature or Google AI 00:11:30.370 --> 00:11:33.450 Studio for prototyping, they allow anyone to 00:11:33.450 --> 00:11:36.350 manifest an idea. But here's the catch. If the 00:11:36.350 --> 00:11:39.230 syntax barrier is gone, if I don't need to know 00:11:39.230 --> 00:11:42.269 where the semicolon goes, what becomes the new 00:11:42.269 --> 00:11:44.750 bottleneck for creation? Clarity of thought, 00:11:45.110 --> 00:11:48.210 logic. If you can't explain specifically what 00:11:48.210 --> 00:11:51.730 you want, if you have fuzzy thinking, The AI 00:11:51.730 --> 00:11:53.950 cannot build it. The computer will do exactly 00:11:53.950 --> 00:11:55.570 what you say. Which is often the problem if you 00:11:55.570 --> 00:11:57.129 don't know what you're saying. So we're back 00:11:57.129 --> 00:11:59.590 to prompting, in a way. Everything is prompting. 00:11:59.669 --> 00:12:01.269 Coding is just prompting with higher stakes. 00:12:01.470 --> 00:12:03.370 So let's zoom out. We've covered a lot of ground 00:12:03.370 --> 00:12:06.889 today. TCREI, the Core 4, agents, open source, 00:12:07.149 --> 00:12:10.029 vibe coding. It's a lot. It is. But remember 00:12:10.029 --> 00:12:12.490 the mission here. We're trying to stop the fatigue. 00:12:12.730 --> 00:12:14.710 It goes back to that driving lesson analogy. 00:12:15.070 --> 00:12:17.029 If you master these fundamentals, it doesn't 00:12:17.029 --> 00:12:19.649 really matter what tool comes out in 2027 or 00:12:19.649 --> 00:12:22.169 2030. Exactly. The shift we're describing is 00:12:22.169 --> 00:12:24.789 moving from being a consumer or a renter of these 00:12:24.789 --> 00:12:28.070 tools to being a pilot, an owner. The source 00:12:28.070 --> 00:12:30.049 ends with a challenge. And I want to pass this 00:12:30.049 --> 00:12:32.549 on to you, the listener. Don't try to do all 00:12:32.549 --> 00:12:33.970 of this at once. That's how you get burnout. 00:12:34.289 --> 00:12:37.700 Pick one thing. Just one. Maybe this week, you 00:12:37.700 --> 00:12:40.600 focus purely on that TCREI framework. Don't worry 00:12:40.600 --> 00:12:43.399 about agents or coding. Just get really, really 00:12:43.399 --> 00:12:46.399 good at writing structured prompts. Or do a spring 00:12:46.399 --> 00:12:48.519 cleaning. Look at your bookmarks bar. Do you 00:12:48.519 --> 00:12:51.679 have 20 AI tools saved? Delete them. Pick your 00:12:51.679 --> 00:12:54.179 core forebrain research specialist automator 00:12:54.179 --> 00:12:56.399 and commit to them. Simplicity is the ultimate 00:12:56.399 --> 00:12:58.360 sophistication here. You are no longer just a 00:12:58.360 --> 00:13:01.500 user. You are a pilot. So start your engine. 00:13:01.659 --> 00:13:04.379 And enjoy the ride. Thanks for diving duck with 00:13:04.379 --> 00:13:05.700 us. We'll catch you on the next one.