WEBVTT 00:00:00.000 --> 00:00:02.500 You know, if you try to search for AI news today, 00:00:03.080 --> 00:00:05.660 you are just, you're immediately overwhelmed. 00:00:05.900 --> 00:00:08.060 Oh, it's not just information overload. It's 00:00:08.060 --> 00:00:10.900 a mountain of noise. It's a mountain of completely 00:00:10.900 --> 00:00:13.099 conflicting predictions. One source says total 00:00:13.099 --> 00:00:15.160 job loss. And the next says it's all a bubble 00:00:15.160 --> 00:00:17.699 that's about to pop. Yeah. It makes it so hard 00:00:17.699 --> 00:00:21.190 to know. who to trust, or what to actually do. 00:00:21.289 --> 00:00:24.010 And that noise is what freezes people. It's why 00:00:24.010 --> 00:00:26.850 we're doing this. Our mission here is to just 00:00:26.850 --> 00:00:29.589 cut through all of that hype. We went straight 00:00:29.589 --> 00:00:32.469 to the data. We're talking full reports, real 00:00:32.469 --> 00:00:34.729 industry tests from McKinsey, from Stanford, 00:00:35.009 --> 00:00:38.350 OpenAI, and EPOC AI. Welcome back to the deep 00:00:38.350 --> 00:00:40.750 dive. This isn't about guessing what's going 00:00:40.750 --> 00:00:45.310 to happen in 10 years. No. It's about six definitive 00:00:45.310 --> 00:00:48.479 data back changes happening right now. this year 00:00:48.479 --> 00:00:51.420 that affect your job. So, okay, let's unpack 00:00:51.420 --> 00:00:53.340 this. We've got a roadmap for you. We're going 00:00:53.340 --> 00:00:55.780 to start with why that race for the smartest 00:00:55.780 --> 00:00:58.219 model is pretty much over. Then we'll get into 00:00:58.219 --> 00:01:00.640 why you need to build workflows, not wait for 00:01:00.640 --> 00:01:03.679 some magic agent. And then my favorite part, 00:01:04.239 --> 00:01:06.180 how non -technical people are suddenly getting 00:01:06.180 --> 00:01:09.439 these incredible new superpowers. For the last 00:01:09.439 --> 00:01:12.099 couple of years, it felt like the entire conversation 00:01:12.099 --> 00:01:15.140 was just about benchmark scores. Oh, completely. 00:01:15.319 --> 00:01:17.739 Everyone was arguing about, you know, is it GPT 00:01:17.739 --> 00:01:20.920 -5? Is it Claude 3? Gemini 1 .5? Who's the best 00:01:20.920 --> 00:01:23.120 this week? Yeah, who's the best at geometry or 00:01:23.120 --> 00:01:26.959 writing a poem? It was this really intense race. 00:01:27.379 --> 00:01:30.840 But moving into 2026, the data is showing that 00:01:30.840 --> 00:01:33.540 that whole competition has, well, it's hit a 00:01:33.540 --> 00:01:36.319 plateau. It really has. And what's so fascinating 00:01:36.319 --> 00:01:39.200 is how the gap is just disappearing. We've got 00:01:39.200 --> 00:01:41.760 data from artificial analysis that confirms that 00:01:41.760 --> 00:01:44.920 all the top -tier models, their performance is 00:01:44.920 --> 00:01:46.739 clustering. They're all becoming good at the 00:01:46.739 --> 00:01:48.599 same things. We're all becoming good enough at 00:01:48.599 --> 00:01:50.540 pretty much everything. Math, writing, you name 00:01:50.540 --> 00:01:53.719 it. And it's not just the big expensive frontier 00:01:53.719 --> 00:01:55.599 models we're talking about. No, absolutely not. 00:01:55.719 --> 00:01:57.500 And this is where the whole landscape just fundamentally 00:01:57.500 --> 00:01:59.700 changes. You've got Stanford research confirming 00:01:59.700 --> 00:02:01.799 that open weight models. So these are the free 00:02:01.799 --> 00:02:05.019 to use ones, like Llama. Exactly. Llama DeepSeq. 00:02:05.140 --> 00:02:07.280 They're performing nearly as well as the incredibly 00:02:07.280 --> 00:02:10.259 expensive models from the big labs. Wait. OK, 00:02:10.300 --> 00:02:12.500 so if the free models are almost as good, why 00:02:12.500 --> 00:02:14.659 is anyone paying billions for the others? What 00:02:14.659 --> 00:02:16.759 are we missing? That's the critical question. 00:02:17.020 --> 00:02:21.039 And for, I'd say, 99 % of tasks, you're right, 00:02:21.139 --> 00:02:24.219 the difference is basically zero. For that last 00:02:24.219 --> 00:02:27.699 1 % super high precision complex reasoning, the 00:02:27.699 --> 00:02:30.219 big models still have a slight edge. But the 00:02:30.219 --> 00:02:32.199 bigger factor is cost, right? The bigger factor 00:02:32.199 --> 00:02:35.099 is efficiency and cost. It's transformative. 00:02:35.680 --> 00:02:38.400 Epoch AI has data showing hardware efficiency 00:02:38.400 --> 00:02:40.120 is just, it's skyrocketing. What does that mean 00:02:40.120 --> 00:02:42.699 in real terms? It means NVIDIA's newest chips 00:02:42.699 --> 00:02:46.840 use 105 ,000 times less energy to create text 00:02:46.840 --> 00:02:49.719 than chips from just a decade ago. Wow. So think 00:02:49.719 --> 00:02:51.860 about what that does. When the quality is basically 00:02:51.860 --> 00:02:54.219 the same across the board and the cost to run 00:02:54.219 --> 00:02:57.439 it plummets, AI becomes a commodity. Like water 00:02:57.439 --> 00:02:59.520 or electricity. Exactly like water. The source 00:02:59.520 --> 00:03:01.240 doesn't matter as much as the tap it comes out 00:03:01.240 --> 00:03:04.259 of. Precisely. So your choice isn't about some 00:03:04.259 --> 00:03:06.460 tiny difference in a benchmark score anymore. 00:03:06.939 --> 00:03:09.020 It's all about friction. Where you already spend 00:03:09.020 --> 00:03:12.300 your time. Yes. Where you live digitally. If 00:03:12.300 --> 00:03:14.319 you're in the Microsoft world with Excel and 00:03:14.319 --> 00:03:17.259 Word, reuse Copilot. It's right there. And if 00:03:17.259 --> 00:03:20.340 you're all in on Google Docs and Gmail, you use 00:03:20.340 --> 00:03:22.960 Gemini. Right. because it can see your stuff. 00:03:23.259 --> 00:03:25.460 I mean, I realized I was losing my mind switching 00:03:25.460 --> 00:03:27.419 tabs all day trying to find the quote unquote 00:03:27.419 --> 00:03:29.699 smartest model. Right. The mental fatigue is 00:03:29.699 --> 00:03:31.919 real. Sticking to one platform, the one that's 00:03:31.919 --> 00:03:35.360 already integrated, it honestly saves me probably 00:03:35.360 --> 00:03:38.580 30 minutes a day of just lost focus. So given 00:03:38.580 --> 00:03:40.860 that the models are basically commodities now, 00:03:41.139 --> 00:03:43.139 how much time should we really spend comparing 00:03:43.139 --> 00:03:46.250 those tiny benchmark? score differences. I'd 00:03:46.250 --> 00:03:48.330 say just focus on seamless integration. Forget 00:03:48.330 --> 00:03:50.590 about the tiny performance gains. That shift 00:03:50.590 --> 00:03:52.550 away from the models themselves brings us right 00:03:52.550 --> 00:03:55.650 to the next big thing, which is this pivot from 00:03:55.650 --> 00:03:59.409 chasing AI agents to building actual workflows. 00:04:00.090 --> 00:04:02.169 The hype around agents was just deafening, wasn't 00:04:02.169 --> 00:04:04.330 it? Oh, yeah. The idea that you could just say, 00:04:04.729 --> 00:04:07.449 hey, manage this project for me and do 50 steps 00:04:07.449 --> 00:04:09.689 while you went for coffee. A compelling vision, 00:04:09.770 --> 00:04:13.599 for sure. But the data offers a pretty big reality 00:04:13.599 --> 00:04:16.379 check. McKinsey found that less than 10 % of 00:04:16.379 --> 00:04:18.720 companies have actually scaled autonomous AI 00:04:18.720 --> 00:04:21.300 agents. So why are they failing? What's the roadblock? 00:04:21.500 --> 00:04:24.420 It's the risk. The risk is just too high. When 00:04:24.420 --> 00:04:26.720 you give it total control, mistakes are harder 00:04:26.720 --> 00:04:29.420 to find. It's almost impossible to monitor. And 00:04:29.420 --> 00:04:31.759 the security questions, especially with company 00:04:31.759 --> 00:04:35.699 data, are huge. If an agent makes a really bad 00:04:35.699 --> 00:04:39.180 mistake deep inside a process, trying to undo 00:04:39.180 --> 00:04:41.379 that damage is a nightmare. So you're saying 00:04:41.379 --> 00:04:43.579 the need for human oversight is the thing that's 00:04:43.579 --> 00:04:45.339 holding back full automation. It's the critical 00:04:45.339 --> 00:04:47.980 constraint, absolutely. So what's actually working? 00:04:48.100 --> 00:04:50.560 Workflows. The practical solution. This is what's 00:04:50.560 --> 00:04:53.470 happening right now. 20 % of all enterprise AI 00:04:53.470 --> 00:04:57.470 use is through these specific step -by -step 00:04:57.470 --> 00:04:59.529 processes where the human is still in the driver's 00:04:59.529 --> 00:05:01.670 seat. And you can see this in the real world 00:05:01.670 --> 00:05:04.689 in customer service, for example. The AI can 00:05:04.689 --> 00:05:07.610 verify an ID. It can summarize the call history. 00:05:07.649 --> 00:05:12.220 Right. But a human makes that final Tricky refund 00:05:12.220 --> 00:05:14.759 decision or in content creation. You don't just 00:05:14.759 --> 00:05:17.360 tell the AI. Hey run my social media Yeah, you 00:05:17.360 --> 00:05:19.879 build a workflow you give it a transcript the 00:05:19.879 --> 00:05:22.779 AI pulls five key points It drafts five posts 00:05:22.779 --> 00:05:25.439 and then you the human you edit for tone and 00:05:25.439 --> 00:05:27.740 you hit publish It's like stacking Lego blocks 00:05:27.740 --> 00:05:30.300 and you can start building these right now You 00:05:30.300 --> 00:05:33.720 can use a simple prompt to turn any messy task 00:05:33.720 --> 00:05:36.620 into a clean workflow Just ask the AI to map 00:05:36.620 --> 00:05:38.939 out the steps for you So if these autonomous 00:05:38.939 --> 00:05:41.220 agents are still struggling so much, what's that 00:05:41.220 --> 00:05:43.800 one key thing that's really preventing full automation? 00:05:44.100 --> 00:05:46.980 It's the absolute need for human monitoring to 00:05:46.980 --> 00:05:51.360 catch mistakes and to manage risk. Okay. So that 00:05:51.360 --> 00:05:53.199 success with workflows leads us to maybe the 00:05:53.199 --> 00:05:56.480 most exciting trend, especially for most of the 00:05:56.480 --> 00:05:58.300 people listening. It's the end of the technical 00:05:58.300 --> 00:06:00.540 wall. Yeah, that wall was high. In the past, 00:06:00.560 --> 00:06:02.459 if you wanted to build even a simple business 00:06:02.459 --> 00:06:04.680 tool like a dashboard, you had to know how to 00:06:04.680 --> 00:06:07.860 code. You needed a specialist. That whole requirement 00:06:07.860 --> 00:06:11.019 is just evaporating. There was this fascinating 00:06:11.019 --> 00:06:13.680 MIT study on what they call the equalizer effect. 00:06:14.040 --> 00:06:16.740 The equalizer effect. It shows that AI helps 00:06:16.740 --> 00:06:20.319 people with lower technical skills way more than 00:06:20.319 --> 00:06:23.519 it helps experts. A world -class coder gets a 00:06:23.519 --> 00:06:25.800 little bit faster. But someone with zero skill. 00:06:26.019 --> 00:06:28.500 They get a superpower instantly. And you can 00:06:28.500 --> 00:06:31.500 see that in the data right away. OpenAI said 00:06:31.500 --> 00:06:34.259 that coding -related messages from non -technical 00:06:34.259 --> 00:06:39.459 staff talking marketers. HR grew by 36 % in six 00:06:39.459 --> 00:06:41.040 months. They're building their own stuff. They're 00:06:41.040 --> 00:06:43.639 not waiting for IT anymore. And this is the insight 00:06:43.639 --> 00:06:46.060 for you, the listener. Your value isn't knowing 00:06:46.060 --> 00:06:49.319 where the semicolons go anymore. It's about understanding 00:06:49.319 --> 00:06:51.579 the business problem that needs to be solved. 00:06:52.180 --> 00:06:55.519 If you know what you need, the AI handles the 00:06:55.519 --> 00:06:58.470 how. It handles the technical side. Right. The 00:06:58.470 --> 00:07:00.689 person who actually understands the budget process 00:07:00.689 --> 00:07:02.949 can now build the tool for it without having 00:07:02.949 --> 00:07:05.370 to translate it all for a programmer. It changes 00:07:05.370 --> 00:07:07.970 who gets to be an innovator. You should really 00:07:07.970 --> 00:07:09.850 try this for yourself. Think of some annoying 00:07:09.850 --> 00:07:12.529 little task like cleaning up a messy spreadsheet 00:07:12.529 --> 00:07:15.529 of names. You can just ask the AI, give me a 00:07:15.529 --> 00:07:17.689 Google Apps script to fix this. I don't code. 00:07:17.970 --> 00:07:19.769 So give me the code and tell me exactly where 00:07:19.769 --> 00:07:21.769 to clay to make it work. And it will. It gives 00:07:21.769 --> 00:07:24.620 you the full solution. Whoa. Just take a second 00:07:24.620 --> 00:07:27.500 and imagine scaling that accessibility. A billion 00:07:27.500 --> 00:07:29.540 curious people who aren't developers building 00:07:29.540 --> 00:07:31.519 their own custom tools for their own specific 00:07:31.519 --> 00:07:34.220 problems. That's a massive shift in how companies 00:07:34.220 --> 00:07:36.279 work. So does this mean every non -technical 00:07:36.279 --> 00:07:39.379 person now needs to become a master of prompt 00:07:39.379 --> 00:07:42.060 engineering for code? No, not at all. You just 00:07:42.060 --> 00:07:43.839 have to focus on defining the problem clearly. 00:07:44.480 --> 00:07:47.810 The AI handles the technical execution. Now we 00:07:47.810 --> 00:07:50.329 have to shift gears a little bit away from what 00:07:50.329 --> 00:07:54.170 was the obsession of 2023 and 2024, which was 00:07:54.170 --> 00:07:56.689 prompt engineering. Ugh, the search for the magic 00:07:56.689 --> 00:07:58.990 words. That secret formula that would unlock 00:07:58.990 --> 00:08:01.769 everything. Right. Prompting just isn't the bottleneck 00:08:01.769 --> 00:08:03.509 anymore. The models are smart enough now. They 00:08:03.509 --> 00:08:05.430 can understand imperfect grammar. They get your 00:08:05.430 --> 00:08:08.110 intent. The real bottleneck, the competitive 00:08:08.110 --> 00:08:11.889 edge for 2026, is context. The fact gap. The 00:08:11.889 --> 00:08:14.449 fact gap. The AI knows everything on the public 00:08:14.449 --> 00:08:17.050 internet, but it knows nothing about you. And 00:08:17.050 --> 00:08:19.490 that's the key vulnerability, isn't it? It has 00:08:19.490 --> 00:08:22.930 no idea about your project deadlines, your company's 00:08:22.930 --> 00:08:24.750 tone of voice, that conversation you had with 00:08:24.750 --> 00:08:28.019 your boss yesterday. Without that context, the 00:08:28.019 --> 00:08:31.060 output is just generic. It sounds like AI. I'll 00:08:31.060 --> 00:08:33.340 admit, I still wrestle with prompt drift myself 00:08:33.340 --> 00:08:36.500 sometimes. I'll forget to feed it the critical 00:08:36.500 --> 00:08:38.799 background document before I ask a complex question. 00:08:38.899 --> 00:08:41.360 Then the answer is useless. Technically perfect, 00:08:41.519 --> 00:08:43.840 but completely useless for what I'm doing. That 00:08:43.840 --> 00:08:46.679 was my aha moment on context. And if you don't 00:08:46.679 --> 00:08:49.559 provide that context, the AI is just flying blind. 00:08:49.940 --> 00:08:52.139 The winners of 2026 are going to be the people 00:08:52.139 --> 00:08:54.600 and the companies who have their own data organized 00:08:54.600 --> 00:08:57.019 and ready for the AI. Because if your files are 00:08:57.019 --> 00:08:59.399 scattered everywhere, desktop, Dropbox, Google 00:08:59.399 --> 00:09:02.519 Drive, the AI can't help you. It can't. But the 00:09:02.519 --> 00:09:04.559 solutions are here. Things like Claude projects 00:09:04.559 --> 00:09:07.519 or custom GPTs let you upload your specific context. 00:09:08.100 --> 00:09:10.799 PDFs, brand guides, old reports. And it remembers 00:09:10.799 --> 00:09:13.600 that context for every conversation. Exactly. 00:09:13.799 --> 00:09:16.440 You can use it to enforce your style. Upload 00:09:16.440 --> 00:09:19.259 five of your old newsletters and say, study my 00:09:19.259 --> 00:09:22.440 tone, my sentence length. Now, draft a post in 00:09:22.440 --> 00:09:25.440 this exact style and do not use words like transform 00:09:25.440 --> 00:09:27.539 or harness. You're basically outsourcing your 00:09:27.539 --> 00:09:30.179 own quality control. So given how critical that 00:09:30.179 --> 00:09:33.440 context is, how important is it to maintain a 00:09:33.440 --> 00:09:36.179 single clean source of data over time? Well, 00:09:36.299 --> 00:09:38.820 your output quality depends entirely on having 00:09:38.820 --> 00:09:42.100 organized accessible source data. Let's talk 00:09:42.100 --> 00:09:45.059 about the elephant in the room, ads. It's now 00:09:45.059 --> 00:09:47.879 officially confirmed. Platforms like ChatGPT 00:09:47.879 --> 00:09:49.440 are going to start showing ads. And everyone's 00:09:49.440 --> 00:09:51.899 going to groan. Everyone will groan. But we have 00:09:51.899 --> 00:09:55.259 to admit, this is an unavoidable and probably 00:09:55.259 --> 00:09:57.440 necessary step for the industry. It absolutely 00:09:57.440 --> 00:09:59.539 is. It all comes down to the staggering cost. 00:09:59.639 --> 00:10:02.120 I mean, it costs millions of dollars every single 00:10:02.120 --> 00:10:04.960 day to run these models if they only relied on 00:10:04.960 --> 00:10:08.259 subscriptions. you'd create a massive AI divide. 00:10:08.519 --> 00:10:10.139 Explain what you mean by that divide. It means 00:10:10.139 --> 00:10:12.000 only the biggest companies, the richest people, 00:10:12.059 --> 00:10:14.559 could afford the best tools. It would kill innovation 00:10:14.559 --> 00:10:16.860 for everyone else. So ads are the trade -off. 00:10:17.000 --> 00:10:19.580 They're what allows for universal access. Exactly. 00:10:19.600 --> 00:10:22.620 It's what gives students, non -profits, researchers, 00:10:22.700 --> 00:10:25.100 and developing countries access to high -quality 00:10:25.100 --> 00:10:27.820 AI for free. It's the YouTube model, really. 00:10:27.820 --> 00:10:29.879 You can pay for premium to get rid of the ads. 00:10:29.960 --> 00:10:31.759 Or you watch a few ads and you still get the 00:10:31.759 --> 00:10:33.919 same world -class tools. But the big question 00:10:33.919 --> 00:10:37.830 here is trust. What will these ads look like? 00:10:38.169 --> 00:10:41.830 How do you keep the AI's advice pure? Everyone 00:10:41.830 --> 00:10:44.190 is very clear on this. The integrity of the model 00:10:44.190 --> 00:10:46.490 has to be protected. We'll probably see things 00:10:46.490 --> 00:10:48.610 like banner ads on the side of the screen totally 00:10:48.610 --> 00:10:50.990 separate from the conversation. So the AI won't 00:10:50.990 --> 00:10:53.350 suddenly start pitching products? It can't. If 00:10:53.350 --> 00:10:55.909 you ask how to fix a leaky pipe, it must not 00:10:55.909 --> 00:10:58.830 say, you should buy Brand X wrench. That would 00:10:58.830 --> 00:11:01.870 destroy trust instantly. So if ads are funding 00:11:01.870 --> 00:11:04.799 this access, What is the single biggest risk 00:11:04.799 --> 00:11:07.220 platforms face when it comes to influencing the 00:11:07.220 --> 00:11:09.360 model? Ruining user trust is the greatest threat 00:11:09.360 --> 00:11:12.000 to adoption. The line between advice and ads 00:11:12.000 --> 00:11:14.419 has to be crystal clear. So far, we've only been 00:11:14.419 --> 00:11:17.340 talking about AI on a screen, in documents. But 00:11:17.340 --> 00:11:20.960 the last huge trend that's quickly becoming real 00:11:20.960 --> 00:11:25.259 is embodied AI. AI with a physical body, moving 00:11:25.259 --> 00:11:27.259 around in the real world. And this isn't science 00:11:27.259 --> 00:11:29.509 fiction. It's already happening in very specific 00:11:29.509 --> 00:11:31.909 industries. Look at logistics and autonomous 00:11:31.909 --> 00:11:35.730 driving. Waymo has driven over a hundred million 00:11:35.730 --> 00:11:39.409 autonomous miles. And the data is just staggering. 00:11:39.850 --> 00:11:43.309 They're 96 % safer than a human driver. Autonomous 00:11:43.309 --> 00:11:45.549 taxis aren't a fantasy anymore. They're becoming 00:11:45.549 --> 00:11:48.309 normal in some cities. Normalizing fast. You 00:11:48.309 --> 00:11:50.350 see the same thing in warehouses with robots. 00:11:50.909 --> 00:11:53.909 Amazon's robots use AI to navigate and handle 00:11:53.909 --> 00:11:56.250 things they've never seen before. They've cut 00:11:56.250 --> 00:11:58.950 the order to shift time by almost 80 percent 00:11:58.950 --> 00:12:01.129 in some places. Which brings us to this really 00:12:01.129 --> 00:12:02.970 big idea that explains why this is happening 00:12:02.970 --> 00:12:07.100 so fast. Capital assets as software. That is 00:12:07.100 --> 00:12:09.860 the key frame for all of this. A traditional 00:12:09.860 --> 00:12:12.659 machine, like a car, just got worse over time, 00:12:12.840 --> 00:12:15.019 wear and tear. Right. The new model is that the 00:12:15.019 --> 00:12:16.879 machine gets smarter over time through software 00:12:16.879 --> 00:12:19.419 updates. The physical hardware is the same, but 00:12:19.419 --> 00:12:22.000 its brain gets better at driving, better at picking 00:12:22.000 --> 00:12:24.159 up boxes. So it appreciates an intelligence. 00:12:24.340 --> 00:12:26.539 Exactly. But we do need to keep expectations 00:12:26.539 --> 00:12:28.899 in check, especially around the house. Yeah, 00:12:29.000 --> 00:12:30.899 we shouldn't expect a robot to be folding our 00:12:30.899 --> 00:12:34.269 laundry next week. No. Rodney Brooks at MIT says 00:12:34.269 --> 00:12:37.049 we're still probably 15 years away from a general 00:12:37.049 --> 00:12:40.110 purpose robot that can do that reliably. The 00:12:40.110 --> 00:12:43.549 focus right now is on those specialized industrial 00:12:43.549 --> 00:12:46.250 machines. So which segment of the economy is 00:12:46.250 --> 00:12:49.470 really going to see the fastest adoption of this 00:12:49.470 --> 00:12:52.070 physical automation? It's those specialized industrial 00:12:52.070 --> 00:12:55.309 roles, logistics, high -volume manufacturing, 00:12:55.909 --> 00:12:57.769 and transportation. We've covered a lot of ground 00:12:57.769 --> 00:13:01.210 today. And the clear message seems to be that 00:13:01.210 --> 00:13:04.409 AI is moving away from that initial kind of theoretical 00:13:04.409 --> 00:13:07.070 model obsession. And toward practical integrated 00:13:07.070 --> 00:13:09.789 application, the shift is here. It is. And it's 00:13:09.789 --> 00:13:11.789 actionable. So let's run through that quick checklist 00:13:11.789 --> 00:13:14.590 for you, the learner. First, stop comparing models. 00:13:14.850 --> 00:13:16.850 Just pick the one that fits best into the tools 00:13:16.850 --> 00:13:19.470 you already use. Second, stop waiting for magic 00:13:19.470 --> 00:13:22.110 agents. Start thinking in terms of step -by -step 00:13:22.110 --> 00:13:24.889 workflows that you control. Third, don't wait 00:13:24.889 --> 00:13:28.269 for IT. Use AI to write code for you. Focus your 00:13:28.269 --> 00:13:30.309 energy on just defining the business problem. 00:13:30.690 --> 00:13:33.470 And fourth, organize your context files. Scattered 00:13:33.470 --> 00:13:36.889 data makes the smartest AI totally useless. Fifth, 00:13:37.289 --> 00:13:40.029 get ready for ads. Accept them as the price for 00:13:40.029 --> 00:13:43.029 universal access, but watch for that clear separation 00:13:43.029 --> 00:13:46.330 to maintain trust. And finally, keep an eye on 00:13:46.330 --> 00:13:48.590 physical automation, especially in logistics 00:13:48.590 --> 00:13:50.570 and industry. So your assignment for this week 00:13:50.570 --> 00:13:53.750 is simple. Just identify one weekly task, only 00:13:53.750 --> 00:13:56.210 one, that feels like a total chore. And then 00:13:56.210 --> 00:13:58.639 challenge yourself. Use that workflow prompt 00:13:58.639 --> 00:14:01.200 ID we talked about. Ask your favorite AI to help 00:14:01.200 --> 00:14:03.519 you break that chore down into steps. Figure 00:14:03.519 --> 00:14:06.019 out what the AI can do and what you need to keep 00:14:06.019 --> 00:14:09.299 for a final quality check. Start small, one workflow 00:14:09.299 --> 00:14:11.679 at a time, mastering that practical application. 00:14:12.039 --> 00:14:13.600 That's how you win in 2026.