WEBVTT 00:00:00.000 --> 00:00:03.020 Imagine asking a machine, you know, just using 00:00:03.020 --> 00:00:05.639 plain English to build a whole website and it 00:00:05.639 --> 00:00:09.380 just hands it over. Flawless. Seconds later. 00:00:09.599 --> 00:00:13.000 Right. No debugging. Nothing. Or maybe think 00:00:13.000 --> 00:00:15.339 about an AI assistant with a company credit cards, 00:00:15.560 --> 00:00:17.420 just booking your flights, managing expenses 00:00:17.420 --> 00:00:20.879 all on its own. Welcome to the deep dive. Yeah, 00:00:21.100 --> 00:00:23.600 that shift you're talking about from AI is just 00:00:23.600 --> 00:00:26.379 a tool we kind of poke at to AI is this agent, 00:00:26.559 --> 00:00:28.600 something that acts for us. That's really the 00:00:28.600 --> 00:00:29.920 core of what we're seeing today. It feels like 00:00:29.920 --> 00:00:32.340 it's accelerating every week. It really is. The 00:00:32.340 --> 00:00:35.119 number of new tools, it's a lot. So our mission 00:00:35.119 --> 00:00:38.060 here is simple, cut through all that noise, and 00:00:38.060 --> 00:00:40.179 track where AI is getting genuinely autonomous. 00:00:40.899 --> 00:00:42.619 So we've got a plan for you today focused right 00:00:42.619 --> 00:00:44.840 on that self -sufficiency idea. First, we'll 00:00:44.840 --> 00:00:47.520 hit the coding battleground, speed, precision, 00:00:47.920 --> 00:00:50.340 critical stuff there. Second, the creative side, 00:00:50.880 --> 00:00:54.140 visuals, audio, how AI is becoming more like 00:00:54.140 --> 00:00:56.340 a collaborator. Interesting. And finally, the 00:00:56.340 --> 00:00:59.340 really deep integration stuff, AI moving into, 00:00:59.340 --> 00:01:03.390 well, global governance. and commerce. Big picture. 00:01:03.689 --> 00:01:05.469 OK, let's unpack that, starting with coding. 00:01:05.849 --> 00:01:08.890 GPT -5 codex. They're calling it the coder's 00:01:08.890 --> 00:01:11.909 new best friend. Why is this command line thing 00:01:11.909 --> 00:01:15.230 such a big deal compared to, say, a web chat, 00:01:15.750 --> 00:01:18.269 especially for pros? Yeah, good question. It 00:01:18.269 --> 00:01:20.750 really boils down to professional velocity, speed. 00:01:21.430 --> 00:01:24.530 Codex is a command line tool, a CLI. Now, that 00:01:24.530 --> 00:01:27.250 might just sound like typing words into a black 00:01:27.250 --> 00:01:29.780 window, but for developers. People in the trenches, 00:01:29.840 --> 00:01:32.159 that's super useful. It means faster iteration, 00:01:32.579 --> 00:01:34.819 much easier integration into their existing setups, 00:01:35.019 --> 00:01:37.319 like build scripts, automated workflows. Oh, 00:01:37.439 --> 00:01:39.659 OK. Just tell it what you want, like codecs create, 00:01:39.879 --> 00:01:42.459 website weather Ho Chi Minh City, and boom, it 00:01:42.459 --> 00:01:45.079 spits out the code instantly. So it's about speed 00:01:45.079 --> 00:01:46.780 and fitting right into how they already work. 00:01:47.040 --> 00:01:49.459 Exactly. And the real kicker, the differentiator, 00:01:49.620 --> 00:01:51.459 it's precision. We saw this play out in a head 00:01:51.459 --> 00:01:53.819 -to -head test, building a simple to -do list 00:01:53.819 --> 00:01:57.060 website. Input box, add button, cross out tasks. 00:01:57.680 --> 00:02:00.459 basic stuff. Okay. And how did it stack up against 00:02:00.459 --> 00:02:03.980 Claude code and PropX tool? Well, Codex just 00:02:03.980 --> 00:02:06.900 did it. Delivered exactly what was asked. No 00:02:06.900 --> 00:02:09.479 extra fluff, no back and forth. When we asked 00:02:09.479 --> 00:02:11.939 it to add a clear all button, it just slotted 00:02:11.939 --> 00:02:14.379 it in neatly and the thing worked perfectly. 00:02:14.479 --> 00:02:17.439 First try. Wow. Okay. And Claude? Claude code. 00:02:18.120 --> 00:02:20.620 It was a bit different. It showed this what you 00:02:20.620 --> 00:02:23.060 might call feature creep. It started asking questions 00:02:23.060 --> 00:02:25.300 like, do you want to save the list? Which was 00:02:25.300 --> 00:02:28.659 an ask for. And then it decided on its own to 00:02:28.659 --> 00:02:31.259 add columns like priority and due date. Made 00:02:31.259 --> 00:02:33.039 the whole thing more complicated than the simple 00:02:33.039 --> 00:02:35.659 request. And worse, the main thing, the cross 00:02:35.659 --> 00:02:38.770 out function. Sometimes it was buggy. That's 00:02:38.770 --> 00:02:41.430 interesting. So Claude was maybe trying to be 00:02:41.430 --> 00:02:44.689 smarter, anticipate needs, but Codex 1 by just 00:02:44.689 --> 00:02:47.949 sticking to the script, literally. So is that 00:02:47.949 --> 00:02:50.189 ability to just follow orders precisely, not 00:02:50.189 --> 00:02:52.030 try to guess, is that the most critical thing 00:02:52.030 --> 00:02:54.590 for these tools right now? Precision and avoiding 00:02:54.590 --> 00:02:56.969 that feature creep. Yeah, absolutely key for 00:02:56.969 --> 00:02:59.729 keeping projects simple and reliable. OK, so 00:02:59.729 --> 00:03:02.789 if Codex is like the efficient code writer, then 00:03:02.789 --> 00:03:05.310 Repled Agent 3, well, that's more like the full 00:03:05.310 --> 00:03:07.819 application engineer. This one takes us a big 00:03:07.819 --> 00:03:10.360 step further. It lets you build an entire working 00:03:10.360 --> 00:03:13.860 app just by describing the idea. That feels like 00:03:13.860 --> 00:03:16.039 a huge jump from just getting code snippets. 00:03:16.099 --> 00:03:18.960 What's the right analogy here? It's really honestly 00:03:18.960 --> 00:03:21.340 like having an expert software engineer sitting 00:03:21.340 --> 00:03:24.020 right there with you. You describe the whole 00:03:24.020 --> 00:03:26.639 app idea, let's say, a simple recipe blog. You 00:03:26.639 --> 00:03:28.960 want photo galleries, separate ingredient pages, 00:03:29.060 --> 00:03:31.439 maybe a search bar, and the agent just handles 00:03:31.439 --> 00:03:33.560 it. All the coding, compiling, even hosting, 00:03:33.860 --> 00:03:35.659 it all happens in the background. Okay, now here's 00:03:35.659 --> 00:03:37.639 the part that really caught my eye, where it 00:03:37.639 --> 00:03:40.439 feels less like a tool and more like an agent. 00:03:40.759 --> 00:03:43.919 Agent 3 can apparently think about its own work. 00:03:44.300 --> 00:03:48.599 If the code gets too messy, it simplifies itself. 00:03:48.699 --> 00:03:51.620 Yeah, that self -correction thing is... It's 00:03:51.620 --> 00:03:53.879 pushing boundaries. It means the tool is managing 00:03:53.879 --> 00:03:56.120 its own complexity, its own technical debt, you 00:03:56.120 --> 00:03:58.439 could say. I mean, I still wrestle with prompt 00:03:58.439 --> 00:04:02.319 drift myself sometimes, where the AI kind of 00:04:02.319 --> 00:04:04.500 forgets the original goal over time. So that 00:04:04.500 --> 00:04:08.000 self -correction, wow, that sounds amazing, especially 00:04:08.000 --> 00:04:10.340 for bigger projects. It's revolutionary, definitely. 00:04:10.680 --> 00:04:12.939 But there are trade -offs, right? It's slower 00:04:12.939 --> 00:04:15.139 than the command line tools, can take 10 minutes, 00:04:15.300 --> 00:04:18.769 maybe more, for complex apps. And crucially, 00:04:18.949 --> 00:04:21.009 because it is auto -simplifying things behind 00:04:21.009 --> 00:04:23.589 the scenes, you give up some control. You don't 00:04:23.589 --> 00:04:25.569 see every little coding step. It's really built 00:04:25.569 --> 00:04:28.410 for automation, less for detailed human tweaking. 00:04:28.629 --> 00:04:31.470 But wait, if the AI is simplifying the code, 00:04:32.410 --> 00:04:35.189 how does a human engineer debug it later if they 00:04:35.189 --> 00:04:37.550 don't know what choices the AI made, what corners 00:04:37.550 --> 00:04:39.269 it might have cut? Doesn't that add friction 00:04:39.269 --> 00:04:41.550 back in? That's the balance, isn't it? It's about 00:04:41.550 --> 00:04:43.689 choosing the right tool for the job. Codecs for 00:04:43.689 --> 00:04:46.639 control, Agent 3 for speed. and that hands -off 00:04:46.639 --> 00:04:49.560 automation. OK, let's pivot then. Visuals, audio, 00:04:49.899 --> 00:04:52.819 start with Lumere 3. Big focus on cinematic quality, 00:04:53.100 --> 00:04:56.379 specifically high dynamic range HDR video. Can 00:04:56.379 --> 00:04:58.959 you break down HDR simply for us? Yeah, sure 00:04:58.959 --> 00:05:03.430 thing. So HDR. basically fixes the biggest problem 00:05:03.430 --> 00:05:06.410 with standard video, SDR. You know how in normal 00:05:06.410 --> 00:05:09.829 videos, dark spots become just black blobs and 00:05:09.829 --> 00:05:11.949 bright spots get totally blown out just white? 00:05:12.170 --> 00:05:15.170 Yeah, definitely. SDR fixes that. It captures 00:05:15.170 --> 00:05:17.449 way more detail in both the really dark shadows 00:05:17.449 --> 00:05:20.009 and the super bright highlights all at the same 00:05:20.009 --> 00:05:23.709 time. Makes the image feel incredibly real deep, 00:05:24.269 --> 00:05:25.889 kind of like how our eyes actually see light. 00:05:26.029 --> 00:05:28.329 So if you're filming something high contrast, 00:05:28.569 --> 00:05:31.269 like I don't know, an astronaut on a shiny planet 00:05:31.269 --> 00:05:34.209 with two suns blazing. HDR means you see the 00:05:34.209 --> 00:05:36.370 detail and the astronaut shatters, and maybe 00:05:36.370 --> 00:05:38.709 the faint nebula in the background sky. That's 00:05:38.709 --> 00:05:40.709 the idea, yeah. That's the high concept use. 00:05:41.089 --> 00:05:43.149 But the practical feature may be even more impressive. 00:05:43.389 --> 00:05:45.509 Ray 3 lets you upload a normal video from your 00:05:45.509 --> 00:05:48.050 phone, standard SDR, and it can upgrade it to 00:05:48.050 --> 00:05:51.290 HDR quality, just with a prompt. Whoa. Yeah, 00:05:51.550 --> 00:05:53.350 whoa. I mean, just imagine the potential there 00:05:53.350 --> 00:05:55.089 for independent filmmakers, right? Getting that 00:05:55.089 --> 00:05:57.449 really complex cinematic lighting look from just 00:05:57.449 --> 00:06:00.290 a prompt, no fancy gear needed. It's wild. That 00:06:00.290 --> 00:06:02.970 upgrade capability is huge. OK, shifting to design 00:06:02.970 --> 00:06:06.329 tools. Rave AI. They're doing something new, 00:06:06.550 --> 00:06:08.670 mimicking pro software with layered editing. 00:06:09.050 --> 00:06:11.529 Yes. And this is a massive efficiency boost for 00:06:11.529 --> 00:06:13.600 designers. All right. Instead of trying to tweak 00:06:13.600 --> 00:06:15.459 images with text prompts, which can be super 00:06:15.459 --> 00:06:18.259 frustrating, like, move the text a tiny bit left, 00:06:19.100 --> 00:06:22.360 no more. It's been there. Rave AI automatically 00:06:22.360 --> 00:06:24.759 breaks the image down into separate layers, just 00:06:24.759 --> 00:06:26.779 like you get in Photoshop or other pro tools. 00:06:26.899 --> 00:06:29.600 So if I generate, say, a poster for orange juice, 00:06:30.199 --> 00:06:32.639 the AI sees the glass, the straw, the juice itself, 00:06:32.819 --> 00:06:36.110 the text that says sunshine in a glass. puts 00:06:36.110 --> 00:06:38.509 them on different layers. Exactly. It turns generative 00:06:38.509 --> 00:06:41.569 AI into non -destructive editing. You can literally 00:06:41.569 --> 00:06:44.170 click the text layer, resize it, change the font, 00:06:44.550 --> 00:06:47.269 click the orange layer, make it bigger, all without 00:06:47.269 --> 00:06:49.310 messing up anything else around it. OK, that 00:06:49.310 --> 00:06:51.610 sounds way more intuitive. It's huge for day 00:06:51.610 --> 00:06:53.689 -to -day marketing design work. Anywhere you 00:06:53.689 --> 00:06:56.449 need to constantly adjust layouts or text, much 00:06:56.449 --> 00:06:58.709 faster. And sticking with images for a sexy dream 00:06:58.709 --> 00:07:01.939 4 .0, its strength is apparently Perfect text 00:07:01.939 --> 00:07:04.300 and images. That's been a notorious weak spot 00:07:04.300 --> 00:07:07.560 for AI, right? Oh, absolutely. No more weird, 00:07:07.959 --> 00:07:10.699 melty letters or bizarre spellings baked right 00:07:10.699 --> 00:07:13.560 into the picture. And beyond text, it's also 00:07:13.560 --> 00:07:15.819 really good at sticking to specific details. 00:07:16.240 --> 00:07:18.779 You ask for a 16 .9 aspect ratio, you get 16 00:07:18.779 --> 00:07:22.079 .9. You specify exact clothing colors it delivers. 00:07:22.480 --> 00:07:24.779 It cuts down on all that tedious fixing of small 00:07:24.779 --> 00:07:27.360 errors. OK, so we have this visual fidelity jump 00:07:27.360 --> 00:07:30.139 with HDR and this structural control with layers. 00:07:30.250 --> 00:07:33.029 Which one do you think hits creative workflows 00:07:33.029 --> 00:07:36.189 forced makes the biggest immediate impact? Layered 00:07:36.189 --> 00:07:38.589 editing. That'll immediately boost efficiency 00:07:38.589 --> 00:07:41.410 for daily marketing and design tasks. No question. 00:07:41.750 --> 00:07:44.709 Mid -roll sponsor read. All right, we're back. 00:07:44.829 --> 00:07:47.209 Let's talk about the workplace. Gamma 3 .0, the 00:07:47.209 --> 00:07:49.430 presentation tool. It's moving beyond just making 00:07:49.430 --> 00:07:51.810 slides, integrating more deeply. Yeah, big time. 00:07:52.110 --> 00:07:54.430 First off, the interface is way easier. It's 00:07:54.430 --> 00:07:56.490 chat based now. You just make the text bigger 00:07:56.490 --> 00:07:59.370 and boom, it does it instantly. Nice. But the 00:07:59.370 --> 00:08:01.029 real breakthrough, and this ties back to that 00:08:01.029 --> 00:08:03.930 agency theme, is the new API feature for businesses. 00:08:04.089 --> 00:08:05.829 OK, tell us about that. What does it do? So the 00:08:05.829 --> 00:08:08.230 API lets other company systems automatically 00:08:08.230 --> 00:08:11.329 trigger Gamma to create a custom sales presentation. 00:08:11.930 --> 00:08:14.829 Imagine this. A salesperson finishes a call. 00:08:15.509 --> 00:08:17.569 The system automatically generates a tailored 00:08:17.569 --> 00:08:20.069 deck for that specific client based on the call. 00:08:20.110 --> 00:08:23.410 Wow, OK. The AI could, in theory, listen to the 00:08:23.410 --> 00:08:25.730 call, grab the key points, the client's priorities, 00:08:25.810 --> 00:08:28.300 and then build the presentation. all without 00:08:28.300 --> 00:08:31.339 the salesperson having to manually prep it. That 00:08:31.339 --> 00:08:33.879 drastically cuts down sales cycle time. Huge 00:08:33.879 --> 00:08:36.620 competitive edge. That's a massive time saver. 00:08:36.840 --> 00:08:39.080 OK, and also on the creation front, 11 Labs. 00:08:39.620 --> 00:08:41.960 They started as just a voice generator, but now 00:08:41.960 --> 00:08:44.720 with Studio 3 .0, they're aiming to be a full 00:08:44.720 --> 00:08:46.480 audio studio. Yeah, it's a complete audio hub 00:08:46.480 --> 00:08:49.000 now. They've added complex stuff like voice remixing, 00:08:49.100 --> 00:08:51.759 an AI script generator, and a proper professional 00:08:51.759 --> 00:08:53.740 timeline editor, all in one place. So you could 00:08:53.740 --> 00:08:56.539 basically create a whole short audio story. Exactly. 00:08:56.720 --> 00:08:58.440 have different characters, each with a unique 00:08:58.440 --> 00:09:00.799 AI voice, add sound effects, footsteps, doors 00:09:00.799 --> 00:09:03.299 closing, layer in background music or ambiance, 00:09:03.519 --> 00:09:05.799 all in one screen. The convenience factor seems 00:09:05.799 --> 00:09:07.820 key there. You're not jumping between tools anymore. 00:09:08.240 --> 00:09:11.379 Scripting, voicing, mixing, it's all integrated. 00:09:11.539 --> 00:09:13.580 Like you could build a whole forest, soundscape, 00:09:13.740 --> 00:09:18.720 wind, birds, a stream right there. So does putting 00:09:18.720 --> 00:09:20.919 everything onto one screen like that actually 00:09:20.919 --> 00:09:24.179 boost creativity more than using separate specialized 00:09:24.179 --> 00:09:27.190 tools? What do you think? Yeah, I think so. Reducing 00:09:27.190 --> 00:09:29.090 that friction, it keeps you in the creative flow, 00:09:29.289 --> 00:09:31.250 stops you from context switching all the time. 00:09:32.070 --> 00:09:33.570 All right, let's shift to the highest level now. 00:09:33.730 --> 00:09:36.789 The public sphere. AI moving from assistant to, 00:09:36.789 --> 00:09:39.669 well, official agent. This news out of Albania 00:09:39.669 --> 00:09:42.639 is... Pretty stunning. First country to appoint 00:09:42.639 --> 00:09:45.220 an AI official for public purchasing. Yeah, headline 00:09:45.220 --> 00:09:47.379 grabbing for sure. And they're describing this 00:09:47.379 --> 00:09:50.159 AI as 100 % incorruptible, which makes it this 00:09:50.159 --> 00:09:52.340 huge experiment in governance. But we got to 00:09:52.340 --> 00:09:54.779 clarify right for you listening. Is this AI just 00:09:54.779 --> 00:09:57.720 like auditing purchase records? Or is it actually 00:09:57.720 --> 00:10:00.059 choosing vendors, approving huge payments on 00:10:00.059 --> 00:10:02.240 its own? The reporting implies it's managing 00:10:02.240 --> 00:10:05.820 public purchasing, which suggests... Yeah, a 00:10:05.820 --> 00:10:07.940 high level of decision -making. And if that's 00:10:07.940 --> 00:10:10.080 right, it throws up some critical questions immediately, 00:10:10.320 --> 00:10:13.269 like if the AI messes up. Makes a bad call that 00:10:13.269 --> 00:10:15.929 costs millions. Who's responsible exactly? Is 00:10:15.929 --> 00:10:18.230 it the government minister overseeing it the 00:10:18.230 --> 00:10:20.889 people who coded the algorithm? Where does the 00:10:20.889 --> 00:10:23.870 buck stop that accountability piece is just central 00:10:23.870 --> 00:10:26.350 and also just because it's incorruptible That 00:10:26.350 --> 00:10:28.549 doesn't automatically mean it's fair. Yeah, is 00:10:28.549 --> 00:10:31.309 it just introducing a new kind of bias? Maybe 00:10:31.309 --> 00:10:34.149 one that's harder to see into how public money 00:10:34.149 --> 00:10:38.230 gets spent. Mm -hmm Meanwhile to even enable 00:10:38.230 --> 00:10:42.049 this kind of autonomous action in commerce, Google's 00:10:42.049 --> 00:10:44.690 leading a big project. They're working with MasterCard, 00:10:44.870 --> 00:10:48.070 Visa, Amex, the giants to create a common standard 00:10:48.070 --> 00:10:51.429 so AI agents can securely use credit cards. And 00:10:51.429 --> 00:10:54.350 this is foundational, really critical. You need 00:10:54.350 --> 00:10:56.409 this standard if you want your AI assistant to, 00:10:56.409 --> 00:10:58.769 say, book your flight someday or order your groceries 00:10:58.769 --> 00:11:01.029 or pay bills without you having to click approve 00:11:01.029 --> 00:11:04.269 on every single tiny transaction. A shared secure 00:11:04.269 --> 00:11:07.450 standard. That's basically the green light for 00:11:07.450 --> 00:11:10.269 this whole future of AI agents doing stuff in 00:11:10.269 --> 00:11:12.389 the real world commercially. And one last quick 00:11:12.389 --> 00:11:15.370 note, this one on privacy. OpenAI mentioned plans 00:11:15.370 --> 00:11:18.330 to use AI to guess a user's age. Based on how 00:11:18.330 --> 00:11:20.990 they chat, the topics they discuss, the stated 00:11:20.990 --> 00:11:24.730 reason is parental controls for under -18s. But 00:11:24.730 --> 00:11:27.649 the immediate worry is, well, what private data 00:11:27.649 --> 00:11:29.690 is it looking at to make that guess? It's that 00:11:29.690 --> 00:11:31.889 constant push -pull, isn't it? Safety versus 00:11:31.889 --> 00:11:34.169 surveillance. Often these safety features need 00:11:34.169 --> 00:11:36.529 more data, which bumps up against privacy concerns. 00:11:37.549 --> 00:11:40.129 So back to the AI minister idea. If you have 00:11:40.129 --> 00:11:42.570 an incorruptible AI making purchasing decisions, 00:11:43.129 --> 00:11:45.210 does its moral authority rely entirely on it 00:11:45.210 --> 00:11:47.009 being totally secure and totally accountable? 00:11:47.409 --> 00:11:49.429 Accountability is crucial, absolutely. But these 00:11:49.429 --> 00:11:51.549 standards seem necessary if we're moving towards 00:11:51.549 --> 00:11:54.789 more autonomous AI. OK, so let's wrap up this 00:11:54.789 --> 00:11:57.250 deep dive. The big trend we've seen today, across 00:11:57.250 --> 00:11:59.870 coding, creative work, even governance, is clear. 00:12:00.159 --> 00:12:02.799 AI is shifting from being just a helpful tool 00:12:02.799 --> 00:12:05.200 that reacts to becoming a powerful agent that 00:12:05.200 --> 00:12:07.519 acts, more autonomous. Yeah, the tools are getting 00:12:07.519 --> 00:12:10.059 simpler to use, definitely more powerful, and 00:12:10.059 --> 00:12:12.159 they're weaving themselves right into these really 00:12:12.159 --> 00:12:15.179 specialized, high -stakes workflows. And I'd 00:12:15.179 --> 00:12:17.100 really encourage you, if you're curious, try 00:12:17.100 --> 00:12:20.679 this out yourself. Use Codex. Maybe build a simple 00:12:20.679 --> 00:12:23.440 portfolio site. Or play with RevAI's layers. 00:12:23.600 --> 00:12:26.120 Design a quick graphic. Feel that control. The 00:12:26.120 --> 00:12:28.120 power is getting really accessible. The agent 00:12:28.120 --> 00:12:30.460 is kind of waiting. And maybe a final thought 00:12:30.460 --> 00:12:32.580 to leave you with connecting a couple of these 00:12:32.580 --> 00:12:36.519 threads. What actually happens when an incorruptible 00:12:36.519 --> 00:12:40.340 AI minister like Albania's model gets the power 00:12:40.340 --> 00:12:44.519 to spend money autonomously using that new secure 00:12:44.519 --> 00:12:47.519 Google Visa payment standard? What are the real 00:12:47.519 --> 00:12:50.720 limits, practical, ethical of a fully automated 00:12:50.720 --> 00:12:54.110 state purchasing system? It creates this closed 00:12:54.110 --> 00:12:55.970 loop of digital authority. It's something we're 00:12:55.970 --> 00:12:57.289 really just starting to wrap our heads around. 00:12:57.389 --> 00:12:59.450 Lots to think about. We hope you'll keep exploring 00:12:59.450 --> 00:13:01.330 these ideas and join us for the next deep dive.