WEBVTT 00:00:00.000 --> 00:00:02.200 The world of software is changing so fast that 00:00:02.200 --> 00:00:04.759 even the top developers are saying they feel 00:00:04.759 --> 00:00:08.160 completely lost. Absolutely. The game isn't about 00:00:08.160 --> 00:00:10.679 writing perfect code anymore. No. It's about 00:00:10.679 --> 00:00:15.919 orchestrating this unpredictable AI layer. So 00:00:15.919 --> 00:00:18.820 we have to ask, what does that new baseline for 00:00:18.820 --> 00:00:21.649 professional actually look like? Welcome to the 00:00:21.649 --> 00:00:23.929 Deep Dive. Today we're looking at sources that 00:00:23.929 --> 00:00:26.370 really map out the foundation of next -gen AI. 00:00:26.750 --> 00:00:30.010 We're talking about the literal energy that powers 00:00:30.010 --> 00:00:32.750 it, the automation flows that build it, and, 00:00:32.829 --> 00:00:35.369 well, the identity crisis hitting the people 00:00:35.369 --> 00:00:37.729 who make it all work. And our mission today is 00:00:37.729 --> 00:00:40.689 to unpack these three huge interconnected shifts. 00:00:41.289 --> 00:00:43.350 We're going to start with Google's massive bet 00:00:43.350 --> 00:00:46.210 on clean power using Corolla batteries. I mean, 00:00:46.229 --> 00:00:48.460 it's almost sci -fi. Then we'll drill down into 00:00:48.460 --> 00:00:51.079 the surprising secret behind building AI agents, 00:00:51.200 --> 00:00:53.299 which is really just focusing on a handful of 00:00:53.299 --> 00:00:55.179 core automation nodes. And finally, we're going 00:00:55.179 --> 00:00:57.820 to dissect the brutally honest truth that Andrzej 00:00:57.820 --> 00:01:00.200 Karpathy laid out about the new programming paradigm. 00:01:00.759 --> 00:01:04.140 It's forcing developers to become managers of 00:01:04.140 --> 00:01:06.260 chaos. Okay, let's unpack this. Let's do it. 00:01:06.340 --> 00:01:08.519 So our first source gets right to the heart of 00:01:08.519 --> 00:01:11.359 the problem. For any company trying to build 00:01:11.359 --> 00:01:15.659 AGI. Energy. Energy. You need massive nonstop. 00:01:16.109 --> 00:01:19.790 Carbon -free compute. 24 -7. And the old ways, 00:01:19.969 --> 00:01:23.909 you know, lithium, rare earth minerals, they 00:01:23.909 --> 00:01:26.569 just can't scale. Not cleanly enough, not fast 00:01:26.569 --> 00:01:28.569 enough. Yeah, and the supply chain for that stuff 00:01:28.569 --> 00:01:31.890 is a huge headache. It's volatile, geopolitically 00:01:31.890 --> 00:01:34.709 tricky, just a mess. Which brings us to Google's 00:01:34.709 --> 00:01:37.969 solution with this company, Energy Dome. They're 00:01:37.969 --> 00:01:40.829 developing these huge -scale co -euro batteries. 00:01:40.969 --> 00:01:43.109 And this is the brilliant part. The system uses 00:01:43.109 --> 00:01:45.750 only co -euro steel and water. Things that are 00:01:45.750 --> 00:01:47.810 cheap, abundant. Exactly. You can source them 00:01:47.810 --> 00:01:49.810 anywhere. It's a completely closed system, which 00:01:49.810 --> 00:01:51.650 is why it lasts so long. Okay, so how does it 00:01:51.650 --> 00:01:53.870 actually work? Well, when you have excess solar 00:01:53.870 --> 00:01:56.430 or wind power during the day, that electricity 00:01:56.430 --> 00:01:58.870 is used to compress carbon dioxide gas into a 00:01:58.870 --> 00:02:01.810 liquid. And that stores the energy. Like a giant 00:02:01.810 --> 00:02:04.810 battery. A giant silent battery. Then at night, 00:02:04.870 --> 00:02:07.049 or whenever the grid needs a boost, you just 00:02:07.049 --> 00:02:09.229 release the liquid to a yoyo. It turns back into 00:02:09.229 --> 00:02:12.870 a gas. Spins a turbine and generates power. It's 00:02:12.870 --> 00:02:15.430 basically a kinetic energy system. And the metrics 00:02:15.430 --> 00:02:17.430 here are why Google is jumping on this. This 00:02:17.430 --> 00:02:19.550 isn't just some research project. No, not at 00:02:19.550 --> 00:02:23.250 all. We're talking about a 75 % roundtrip efficiency. 00:02:23.550 --> 00:02:26.490 Very competitive. But the longevity is what gets 00:02:26.490 --> 00:02:29.849 me. Zero degradation. Over 30 years. And it's 00:02:29.849 --> 00:02:32.150 about half the cost of lithium -ion equivalent. 00:02:32.610 --> 00:02:35.090 So instead of waiting a decade to build, say, 00:02:35.250 --> 00:02:38.000 a nuclear plant. Right. They can just drop one 00:02:38.000 --> 00:02:39.860 of these massive domes. They honestly look like 00:02:39.860 --> 00:02:41.360 they're from Mars out in the desert and get going. 00:02:41.500 --> 00:02:44.240 It completely changes the strategy. It's modular. 00:02:44.439 --> 00:02:47.639 It's fast. And it's clean. Whoa. Yeah. Imagine 00:02:47.639 --> 00:02:50.400 scaling to a billion queries on energy that's 00:02:50.400 --> 00:02:53.500 just that clean and stable. It's transformative 00:02:53.500 --> 00:02:56.060 for AGI development. That kind of reliability 00:02:56.060 --> 00:02:58.759 is the dream. So these domes are obviously huge, 00:02:58.919 --> 00:03:01.280 though. We're talking football field sized. What's 00:03:01.280 --> 00:03:04.500 the main tradeoff for all this clean, scalable 00:03:04.500 --> 00:03:06.780 energy? The tradeoff against speed and efficiency 00:03:06.780 --> 00:03:10.280 is simply the sheer physical size, the land it 00:03:10.280 --> 00:03:13.020 requires. Okay, so we have this incredibly stable 00:03:13.020 --> 00:03:16.060 bedrock of power. Now let's move up the stack 00:03:16.060 --> 00:03:20.219 to the actual AI agents. Automation. Right. And 00:03:20.219 --> 00:03:22.900 if you've ever opened a platform like NANA, it 00:03:22.900 --> 00:03:26.960 can look like just chaos. It's a firehose of 00:03:26.960 --> 00:03:29.800 information, yeah. 100 plus nodes, arrows going 00:03:29.800 --> 00:03:32.509 everywhere. It's overwhelming. But the key insight 00:03:32.509 --> 00:03:35.370 from our sources is that the experts, the people 00:03:35.370 --> 00:03:38.830 building really complex agents, they aren't memorizing 00:03:38.830 --> 00:03:41.490 all 100 of those tools. Not even close. They're 00:03:41.490 --> 00:03:43.870 focusing on just a handful. Exactly. They rely 00:03:43.870 --> 00:03:47.150 on just 17 core nodes for about 80 % of all their 00:03:47.150 --> 00:03:49.810 automations. So the goal isn't to master every 00:03:49.810 --> 00:03:53.669 tool. It's to recognize the situation, the pattern, 00:03:53.669 --> 00:03:56.530 and the flow. Precisely. And a node here is just 00:03:56.530 --> 00:03:58.530 a functional block. Think of it like a digital 00:03:58.530 --> 00:04:01.139 Lego piece that does one single thing. Like checking 00:04:01.139 --> 00:04:04.259 some data or running a condition. Yeah, an IFELS 00:04:04.259 --> 00:04:06.539 decision, a loop to process a list, that kind 00:04:06.539 --> 00:04:08.800 of thing. Okay. Once you really know those 17 00:04:08.800 --> 00:04:11.479 core functions, building a workflow becomes almost 00:04:11.479 --> 00:04:14.159 plug and play. You stop staring at the huge menu. 00:04:14.379 --> 00:04:16.339 And you start thinking. You start thinking, okay, 00:04:16.399 --> 00:04:18.720 I need a loop right here, or this calls for an 00:04:18.720 --> 00:04:21.139 execute code block. So that's the real skill. 00:04:21.220 --> 00:04:23.660 Not the tool knowledge, but thinking in flows. 00:04:23.819 --> 00:04:26.459 That's the mindset shift. It defines the current 00:04:26.459 --> 00:04:29.529 AI skills gap. The person who can look at a business 00:04:29.529 --> 00:04:32.050 problem and map out that flow. They're the most 00:04:32.050 --> 00:04:33.949 valuable person in the room. Without a doubt. 00:04:34.050 --> 00:04:36.970 They bridge the logic and the execution. So why 00:04:36.970 --> 00:04:39.970 is recognizing that flow more valuable than just 00:04:39.970 --> 00:04:42.329 mastering every single tool you can find? It 00:04:42.329 --> 00:04:44.709 defines you as an orchestrator. And that lets 00:04:44.709 --> 00:04:47.569 you bridge the growing AI skills gap. Okay, so 00:04:47.569 --> 00:04:50.629 we have stable power, efficient workflows. But 00:04:50.629 --> 00:04:53.769 how are people in the real world actually reacting 00:04:53.769 --> 00:04:57.420 to all this? This is where it gets tricky. Public 00:04:57.420 --> 00:05:00.079 trust is a massive friction point, especially 00:05:00.079 --> 00:05:02.800 with something as visible as advertising. I can 00:05:02.800 --> 00:05:05.180 imagine. Yeah. The data is pretty stark. If you 00:05:05.180 --> 00:05:08.160 mark an ad as AI -generated, the number of clicks 00:05:08.160 --> 00:05:12.300 it gets drops by about 31%. Wow, 31%. That's 00:05:12.300 --> 00:05:14.660 huge. It's a commercial failure. Yeah. And it 00:05:14.660 --> 00:05:16.079 makes perfect sense when you remember all the 00:05:16.079 --> 00:05:18.360 AI ads that got roasted last year. Oh, yeah. 00:05:18.560 --> 00:05:20.980 Meta's AI granny that looked like she was melting. 00:05:21.040 --> 00:05:24.339 Right. Or the weird Coca -Cola trucks that defied 00:05:24.339 --> 00:05:26.779 physics. Yeah. And that McDonald's Christmas 00:05:26.779 --> 00:05:30.379 ad. A total mess. That uncanny valley effect 00:05:30.379 --> 00:05:32.600 is so powerful. If something feels off, people 00:05:32.600 --> 00:05:35.019 just pull back. Exactly. And that suspicion is 00:05:35.019 --> 00:05:37.540 reflected in the numbers. But, you know, while 00:05:37.540 --> 00:05:40.300 the public is cautious, AI deployment in other 00:05:40.300 --> 00:05:42.560 areas is still moving at lightning speed. Right. 00:05:42.620 --> 00:05:45.139 Like NVIDIA just dropped a new free guide on 00:05:45.139 --> 00:05:48.339 how to fine tune your own LLMs locally. And an 00:05:48.339 --> 00:05:50.730 LLM is just a computer model that... generates 00:05:50.730 --> 00:05:52.829 human -like text. So that's happening at the 00:05:52.829 --> 00:05:54.949 developer level. And then you have the really 00:05:54.949 --> 00:05:57.949 weird stuff. Like the Chinese Agibot? Yeah, renting 00:05:57.949 --> 00:06:00.329 out humanoid robots for weddings and concerts. 00:06:00.410 --> 00:06:02.449 The future is definitely here, ready or not. 00:06:02.569 --> 00:06:04.689 And the money is still there. We saw that in 00:06:04.689 --> 00:06:07.069 India, AI startups raised over $600 million. 00:06:07.509 --> 00:06:10.550 So the investment is flowing, but... Investors 00:06:10.550 --> 00:06:14.170 are getting pickier. They want to see practical... 00:06:14.430 --> 00:06:17.569 real world value, not just blue sky promises. 00:06:17.850 --> 00:06:20.110 Which circles back to the ad problem. They need 00:06:20.110 --> 00:06:22.990 agents that work, that are reliable, and that 00:06:22.990 --> 00:06:26.329 don't creep people out. So with this negative 00:06:26.329 --> 00:06:29.089 reaction to ads, does that mean consumer facing 00:06:29.089 --> 00:06:31.649 AI applications are going to slow down? I don't 00:06:31.649 --> 00:06:34.089 think they'll slow down, but it will force much 00:06:34.089 --> 00:06:36.629 higher standards for transparency and practical 00:06:36.629 --> 00:06:39.110 value. That pressure for reliability brings us 00:06:39.110 --> 00:06:42.220 to our last segment, the human element. This 00:06:42.220 --> 00:06:44.860 programming identity crisis. Yeah, we have to 00:06:44.860 --> 00:06:47.300 talk about Andrej Karpathy's post. For anyone 00:06:47.300 --> 00:06:49.079 who doesn't know, he's one of the most respected 00:06:49.079 --> 00:06:52.899 minds in AI. And he posted on X just being brutally 00:06:52.899 --> 00:06:55.220 honest, saying he's never felt so left behind 00:06:55.220 --> 00:06:57.579 as a programmer. And millions of developers immediately 00:06:57.579 --> 00:07:00.259 responded saying, yes, that's exactly how I feel. 00:07:00.379 --> 00:07:03.639 The issue is this quiet takeover of a new programming 00:07:03.639 --> 00:07:06.500 paradigm. The job has fundamentally changed. 00:07:06.819 --> 00:07:10.259 The old way was so predictable, so logical. Write 00:07:10.259 --> 00:07:13.779 code, test it. Debug the syntax, deploy. You 00:07:13.779 --> 00:07:16.240 are in complete control. The job is almost unrecognizable. 00:07:16.420 --> 00:07:18.680 You're prompting LLMs, orchestrating agents, 00:07:18.980 --> 00:07:21.160 wrangling all these different tools and APIs. 00:07:21.660 --> 00:07:23.920 The work isn't debugging your code syntax anymore. 00:07:24.120 --> 00:07:26.899 It's debugging the AI's behavior. And that's 00:07:26.899 --> 00:07:29.480 because of this new abstraction layer that tools 00:07:29.480 --> 00:07:32.180 like chat GPT agents have created. Right. And 00:07:32.180 --> 00:07:34.560 because of that layer, the whole vocabulary is 00:07:34.560 --> 00:07:37.000 changed overnight. If you don't know terms like 00:07:37.000 --> 00:07:40.240 MCP or LSP. You're already behind. Exactly. And 00:07:40.240 --> 00:07:42.120 MCP is just the model control plane which decides 00:07:42.120 --> 00:07:44.600 which tool the AI should use. Yeah. LSP is large 00:07:44.600 --> 00:07:46.839 scale prompting. And then there's context and 00:07:46.839 --> 00:07:49.360 memory. Context being the info for a single task 00:07:49.360 --> 00:07:51.779 and memory being the history over a whole conversation. 00:07:52.240 --> 00:07:55.060 But it's not just learning new words. It's the 00:07:55.060 --> 00:07:58.079 unpredictability of it all. That's the core challenge. 00:07:58.500 --> 00:08:01.319 You are no longer writing code that does exactly 00:08:01.319 --> 00:08:03.800 what you tell it to do. You're managing this 00:08:03.800 --> 00:08:06.740 living, unpredictable system. One that suffers 00:08:06.740 --> 00:08:08.959 from things like prompt drift, where the AI's 00:08:08.959 --> 00:08:11.600 output changes even if your input stays the same. 00:08:11.800 --> 00:08:14.180 Yeah. I still wrestle with prompt drift myself. 00:08:14.439 --> 00:08:16.860 It feels like the foundation just shifts every 00:08:16.860 --> 00:08:19.040 month. It adds this layer of anxiety to the whole 00:08:19.040 --> 00:08:21.480 process. That anxiety is the psychological toll. 00:08:22.120 --> 00:08:24.959 The honest truth is managing that unpredictable 00:08:24.959 --> 00:08:28.279 behavior is the job now. So if a top developer 00:08:28.279 --> 00:08:30.779 like Carpathy feels left behind by all this, 00:08:31.000 --> 00:08:33.840 what is the single most important skill for a 00:08:33.840 --> 00:08:36.659 new developer to build right now? It has to be 00:08:36.659 --> 00:08:39.779 critical thinking. Strong diagnostic skills to 00:08:39.779 --> 00:08:42.799 debug AI behavior and manage that unpredictability. 00:08:43.000 --> 00:08:44.899 So we've moved from the physical infrastructure 00:08:44.899 --> 00:08:46.860 all the way to the psychological impact today. 00:08:47.039 --> 00:08:49.580 Quite a journey. Let's recap the three big shifts 00:08:49.580 --> 00:08:52.590 we saw in the sources. Okay, first. AI infrastructure 00:08:52.590 --> 00:08:57.190 is moving to massive, reliable, low -carbon solutions 00:08:57.190 --> 00:08:59.750 like those Cocoa Your batteries. They're built 00:08:59.750 --> 00:09:02.649 for 30 years of stability to meet AGI's energy 00:09:02.649 --> 00:09:05.429 demands. Second, mastering automation isn't about 00:09:05.429 --> 00:09:07.590 the tools anymore. It's about recognizing flow 00:09:07.590 --> 00:09:10.610 patterns, those 17 core nodes, so you can be 00:09:10.610 --> 00:09:12.970 an orchestrator, not just a user. And third, 00:09:13.399 --> 00:09:15.740 The job of a programmer has fundamentally changed. 00:09:16.000 --> 00:09:18.820 It's now about orchestrating and debugging this 00:09:18.820 --> 00:09:21.860 unpredictable AI layer. It's a shift from being 00:09:21.860 --> 00:09:24.679 a craftsman to being a manager of chaos. Knowledge 00:09:24.679 --> 00:09:26.980 is power, but it seems like all the sources suggest 00:09:26.980 --> 00:09:29.899 success now comes from being that orchestrator. 00:09:30.019 --> 00:09:33.139 We saw that continuous, stable energy is critical 00:09:33.139 --> 00:09:36.519 for AGI. And yet... The developers building on 00:09:36.519 --> 00:09:39.419 top of it are struggling with this new, unpredictable 00:09:39.419 --> 00:09:42.399 software layer. Which brings us to a final thought 00:09:42.399 --> 00:09:44.720 for you to take with you. Yeah. If the physical 00:09:44.720 --> 00:09:47.080 infrastructure for AI is being solved with these 00:09:47.080 --> 00:09:50.879 massive, unchanging 30 -year batteries, why is 00:09:50.879 --> 00:09:53.019 the software layer still so incredibly volatile? 00:09:53.399 --> 00:09:55.659 That's a great question. What happens when those 00:09:55.659 --> 00:09:58.320 two radically different speeds, the rock -solid 00:09:58.320 --> 00:10:00.919 stability of physics and the utter chaos of code, 00:10:01.979 --> 00:10:04.240 Thanks for diving deep with us. We hope this 00:10:04.240 --> 00:10:05.759 knowledge gives you a competitive edge.