WEBVTT 00:00:00.000 --> 00:00:02.220 If you look at how software creation has changed, 00:00:02.359 --> 00:00:04.660 I mean, even in just the last few years, it's 00:00:04.660 --> 00:00:07.940 already dramatic. But I don't think anything 00:00:07.940 --> 00:00:10.560 really prepared me for the actual demonstrated 00:00:10.560 --> 00:00:14.080 reality that you could build a functional real 00:00:14.080 --> 00:00:18.120 -time 3D ray tracing simulator just by typing 00:00:18.120 --> 00:00:20.719 a single paragraph of text. Welcome to the deep 00:00:20.719 --> 00:00:25.399 dive. That ability to turn conversation, or what 00:00:25.399 --> 00:00:28.059 some people are calling vibe, into a complex 00:00:28.059 --> 00:00:31.199 tool. That's the new reality check, isn't it? 00:00:31.239 --> 00:00:33.920 It really is. And today we are cracking open 00:00:33.920 --> 00:00:36.719 the documentation, the analysis of 12 different 00:00:36.719 --> 00:00:39.100 applications that were built completely from 00:00:39.100 --> 00:00:41.299 scratch using just natural language problems. 00:00:41.380 --> 00:00:43.600 We'll stack apps in minutes. Exactly. Our mission 00:00:43.600 --> 00:00:45.539 is pretty straightforward. We need to pull out 00:00:45.539 --> 00:00:48.399 the most important lessons here about rapid prototyping, 00:00:48.500 --> 00:00:52.820 handling, extreme complexity, and critically 00:00:52.820 --> 00:00:54.979 understanding the real world limits. We're moving 00:00:54.979 --> 00:00:57.399 past the hype. Right into the details. The short 00:00:57.399 --> 00:01:02.500 version, the TLDR. is this. Gemini 3 .0 automates 00:01:02.500 --> 00:01:06.579 these really specialized tasks that before required 00:01:06.579 --> 00:01:09.400 such deep knowledge of coding frameworks. The 00:01:09.400 --> 00:01:13.180 real value is just. It's drastically lowering 00:01:13.180 --> 00:01:16.519 the barrier to entry for complex logic. And here's 00:01:16.519 --> 00:01:18.280 your immediate actionable tip from the source 00:01:18.280 --> 00:01:20.340 material. If you want to try and replicate this. 00:01:20.920 --> 00:01:24.540 Always, always request standalone HTML files 00:01:24.540 --> 00:01:27.239 in your prompt. Why is that? Because that guarantees 00:01:27.239 --> 00:01:29.560 the app is immediately testable. It's portable, 00:01:29.799 --> 00:01:32.420 ready to share or, you know, ready to debug. 00:01:32.680 --> 00:01:34.640 That's such a critical constraint to put on the 00:01:34.640 --> 00:01:37.629 AI. Because even just a year ago, if you suggested 00:01:37.629 --> 00:01:39.750 we'd be creating a physics engine or a custom 00:01:39.750 --> 00:01:41.670 board game just by typing. Yeah, you'd have been 00:01:41.670 --> 00:01:43.269 left out of the room. It would have been dismissed 00:01:43.269 --> 00:01:45.370 as science fiction. But now that shortcut is 00:01:45.370 --> 00:01:47.769 undeniably real. The tests we reviewed, they 00:01:47.769 --> 00:01:49.769 showed some shocking successes, you know, like 00:01:49.769 --> 00:01:51.969 those physics demos. But we're also going to 00:01:51.969 --> 00:01:54.109 look honestly at the limits. For sure. Especially 00:01:54.109 --> 00:01:57.409 around browser lag and some of the visual artifacts. 00:01:57.790 --> 00:02:00.290 It's an honest look at where vibe coding is today. 00:02:00.549 --> 00:02:02.489 Okay, let's start with procedural generation. 00:02:03.200 --> 00:02:05.260 This is essentially code writing code. We're 00:02:05.260 --> 00:02:07.519 talking geometry and a little bit of fun with 00:02:07.519 --> 00:02:09.500 this thing called the voxel creatures generator. 00:02:09.800 --> 00:02:12.400 Right. It started with a really simple high level 00:02:12.400 --> 00:02:15.639 prompt just asking for procedurally generated 00:02:15.639 --> 00:02:18.840 voxel creatures and biomechanical vehicles. And 00:02:18.840 --> 00:02:21.360 what's immediately fascinating here is that the 00:02:21.360 --> 00:02:23.939 A .I. didn't just spit out spatic images. It 00:02:23.939 --> 00:02:27.060 generated interactive physics. You can create 00:02:27.060 --> 00:02:28.860 a creature and then you click a button to scrap 00:02:28.860 --> 00:02:32.340 it and you watch the pieces just shatter. with 00:02:32.340 --> 00:02:35.500 really realistic, tumbling physics and shadows. 00:02:35.520 --> 00:02:37.800 It's like digital Lego blocks. And the system 00:02:37.800 --> 00:02:40.159 retains conversational memory, which is key. 00:02:40.620 --> 00:02:44.180 The source asked for a specific item, a retro 00:02:44.180 --> 00:02:47.270 Polaroid camera. And Gemini just generated a 00:02:47.270 --> 00:02:50.409 fully recognizable voxel camera in seconds. So 00:02:50.409 --> 00:02:52.830 it's handling both random generation and specific 00:02:52.830 --> 00:02:54.930 instruction at the same time. Yeah, inside a 00:02:54.930 --> 00:02:57.250 3D space. It's pretty amazing. We also saw the 00:02:57.250 --> 00:03:00.050 process of converting a flat 2D image. I think 00:03:00.050 --> 00:03:01.969 a Michael Jackson photo was used in the test. 00:03:02.150 --> 00:03:04.729 That's the one. And turning it into 3D voxel 00:03:04.729 --> 00:03:07.650 assets. So the AI has to analyze the composition, 00:03:07.949 --> 00:03:10.969 identify elements, figure out depth, and then 00:03:10.969 --> 00:03:14.979 output three distinct 3D scenes. Setup. mid -event 00:03:14.979 --> 00:03:18.120 and aftermath it's complex scene interpretation 00:03:18.120 --> 00:03:21.240 so what was the quality like well the assessment 00:03:21.240 --> 00:03:23.740 was interesting the results were described as 00:03:23.740 --> 00:03:26.860 simplistic but actually really cool they weren't 00:03:26.860 --> 00:03:29.560 photorealistic but they absolutely captured the 00:03:29.560 --> 00:03:31.639 essence of the image and its movement Okay, so 00:03:31.639 --> 00:03:33.419 let's unpack this a bit. When we look at that 00:03:33.419 --> 00:03:35.759 voxel generator, the source noted a small sort 00:03:35.759 --> 00:03:38.520 of rough edge. The app keeps the same color palette 00:03:38.520 --> 00:03:40.759 from the last design until you do a hard refresh. 00:03:41.319 --> 00:03:43.460 What does that color palette stickiness tell 00:03:43.460 --> 00:03:45.840 us about where these generated apps are right 00:03:45.840 --> 00:03:48.240 now? It tells us that generated apps often need 00:03:48.240 --> 00:03:50.419 iterative refinement. The foundation is there, 00:03:50.479 --> 00:03:52.659 but state management still needs a human review. 00:03:52.879 --> 00:03:55.530 And now we have to move into the... uh the really 00:03:55.530 --> 00:03:58.110 deep technical stuff we asked the ai to handle 00:03:58.110 --> 00:04:00.590 a task that is just notoriously difficult and 00:04:00.590 --> 00:04:03.050 computationally expensive real -time ray tracing 00:04:03.050 --> 00:04:06.789 and ray tracing for anyone unfamiliar is the 00:04:06.789 --> 00:04:09.270 rendering technique that simulates exactly how 00:04:09.270 --> 00:04:11.930 light bounces through a scene it's what gives 00:04:11.930 --> 00:04:13.949 you those hyper realistic reflections usually 00:04:13.949 --> 00:04:16.329 requires high -end gaming hardware oh yeah and 00:04:16.329 --> 00:04:19.509 historically weeks of highly specialized manual 00:04:19.509 --> 00:04:22.149 coding to set up the geometry and the physics 00:04:22.920 --> 00:04:25.779 The prompt was intense. It specifically requested 00:04:25.779 --> 00:04:30.160 a 3D ray tracing app in a mirror maze using PBR 00:04:30.160 --> 00:04:32.699 materials and high -end frameworks like 3JS. 00:04:32.779 --> 00:04:34.759 Okay, let's quickly define that jargon. PBR. 00:04:34.959 --> 00:04:37.360 That's physically based rendering. Right. So 00:04:37.360 --> 00:04:39.699 digital materials that react accurately to light, 00:04:39.779 --> 00:04:43.899 like real metal or rough plastic. And 3JS is 00:04:43.899 --> 00:04:46.540 just the standard library for creating 3D graphics 00:04:46.540 --> 00:04:49.360 in a web browser. The AI had to seamlessly integrate 00:04:49.360 --> 00:04:51.500 all of that. And what it delivered was, well, 00:04:51.579 --> 00:04:53.959 a fully navigable 3D environment. You have these 00:04:53.959 --> 00:04:56.220 floating metallic pieces where the reflections 00:04:56.220 --> 00:04:58.800 update dynamically. Light bounces correctly between 00:04:58.800 --> 00:05:01.160 three different light sources. That is a massive 00:05:01.160 --> 00:05:03.720 computational load. Achieved via conversation. 00:05:04.060 --> 00:05:06.339 Now, the source was honest. There were visible 00:05:06.339 --> 00:05:08.519 artifacts, some smart alignment issues, a bit 00:05:08.519 --> 00:05:11.180 of visual glitching. But for something generated 00:05:11.180 --> 00:05:15.379 conversationally in minutes, the quality is truly 00:05:15.379 --> 00:05:17.339 remarkable. I mean, that's the moment of wonder 00:05:17.339 --> 00:05:19.720 right there. I genuinely did a double take when 00:05:19.720 --> 00:05:21.720 I saw the metallic reflections. It looks like 00:05:21.720 --> 00:05:25.000 a high -end tech demo from maybe five years ago, 00:05:25.160 --> 00:05:27.319 but it was generated from a paragraph of text. 00:05:27.639 --> 00:05:30.939 That's a huge leap. It is. Alongside this, we 00:05:30.939 --> 00:05:33.540 saw the dot matrix image converter. You upload 00:05:33.540 --> 00:05:35.920 a photo and the app turns it into thousands of 00:05:35.920 --> 00:05:37.980 individual dots. And the interactivity is key 00:05:37.980 --> 00:05:40.480 there. It is. You can apply effects like data 00:05:40.480 --> 00:05:42.560 glitch or implosion and you watch the dots move 00:05:42.560 --> 00:05:44.740 around in 3D space and then perfectly reassemble. 00:05:45.389 --> 00:05:48.529 But if the output is visually glitchy, as the 00:05:48.529 --> 00:05:51.790 source said, how much time does that really save 00:05:51.790 --> 00:05:55.170 a developer? Is vibe coding just creating complex 00:05:55.170 --> 00:05:58.069 but, you know, broken starting points? Well, 00:05:58.110 --> 00:06:00.569 it showed AI can handle the advanced real -time 00:06:00.569 --> 00:06:03.529 rendering physics. That complex structure, which 00:06:03.529 --> 00:06:05.769 used to take weeks to set up, is now instantly 00:06:05.769 --> 00:06:08.990 accessible. Next up, we tested complex multi 00:06:08.990 --> 00:06:12.069 -body simulations. This is where you really stress 00:06:12.069 --> 00:06:14.170 the math capabilities of the generated code. 00:06:14.730 --> 00:06:17.129 We started with the particle collider simulator. 00:06:17.470 --> 00:06:19.709 Right. The goal was an educational visualization. 00:06:20.250 --> 00:06:23.930 A particle collider and a tokamak fusion reactor. 00:06:24.310 --> 00:06:27.269 And Gemini created two separate simulators showing 00:06:27.269 --> 00:06:30.370 particle acceleration, interaction, and detailed 00:06:30.370 --> 00:06:33.790 collision paths. The physics here has to be mathematically 00:06:33.790 --> 00:06:37.000 sound for it to be accurate. But this is where 00:06:37.000 --> 00:06:38.920 we hit the performance reality check. The source 00:06:38.920 --> 00:06:40.560 material really focused on this. It said the 00:06:40.560 --> 00:06:43.800 simulation runs very slowly in my browser. But 00:06:43.800 --> 00:06:45.180 that's because it's doing a lot of calculation. 00:06:45.540 --> 00:06:47.639 That performance lag is so important. The source 00:06:47.639 --> 00:06:51.000 noted that expecting a flawlessly optimized production 00:06:51.000 --> 00:06:53.899 -ready app right away is the big mistake users 00:06:53.899 --> 00:06:56.540 make. I still wrestle with this myself, actually. 00:06:56.560 --> 00:06:58.699 You see this incredible functional output, the 00:06:58.699 --> 00:07:00.899 whole collider visualized, and you kind of forget 00:07:00.899 --> 00:07:03.819 that optimization is still a manual, time -consuming 00:07:03.819 --> 00:07:06.660 step. It's easy to... expect immediate perfection. 00:07:07.100 --> 00:07:09.579 Totally. And the interactive end -body gravity 00:07:09.579 --> 00:07:12.459 simulator pushed this idea of complicity even 00:07:12.459 --> 00:07:15.420 further. This challenges the AI to accurately 00:07:15.420 --> 00:07:18.560 simulate Newtonian gravity with multiple bodies 00:07:18.560 --> 00:07:20.879 at once. Meaning every planet is affecting every 00:07:20.879 --> 00:07:23.180 other planet all the time. Constantly. And what 00:07:23.180 --> 00:07:25.120 you can do with it is incredible. You can add 00:07:25.120 --> 00:07:28.899 planets, moons, change their mass, and the simulation 00:07:28.899 --> 00:07:31.360 just immediately recalculates all the gravitational 00:07:31.360 --> 00:07:34.740 forces. Which is how you get those emergent chaotic... 00:07:34.860 --> 00:07:37.600 behaviors like orbital decay or a gravitational 00:07:37.600 --> 00:07:40.500 slingshot. And the AI responded perfectly to 00:07:40.500 --> 00:07:43.560 requests for cosmic events. You could instantly 00:07:43.560 --> 00:07:46.339 add a rogue star, a black hole, or even trigger 00:07:46.339 --> 00:07:48.620 a supernova. And it all affected the physics 00:07:48.620 --> 00:07:50.560 in real time. It proved the underlying engine 00:07:50.560 --> 00:07:53.769 was robust, not just a visual trick. So when 00:07:53.769 --> 00:07:56.269 dealing with these high computation tasks, like 00:07:56.269 --> 00:07:59.509 simulating 10 bodies interacting, is that lag 00:07:59.509 --> 00:08:02.389 a failure or is it the ultimate proof of concept? 00:08:02.589 --> 00:08:04.889 It's proof of concept. The complex foundation 00:08:04.889 --> 00:08:07.810 is sound, but performance optimization has to 00:08:07.810 --> 00:08:10.129 follow the AI generation, not come before it. 00:08:10.410 --> 00:08:11.949 All right. So moving away from visual physics, 00:08:12.170 --> 00:08:14.310 we looked at how this generative capability applies 00:08:14.310 --> 00:08:17.290 to pure data analysis and real world utility. 00:08:18.050 --> 00:08:21.600 The AI bubble research storyboard is. It's a 00:08:21.600 --> 00:08:24.160 fascinating case of macroeconomics meeting interactive 00:08:24.160 --> 00:08:27.339 visualization. Yeah, this one was cool. The prompt 00:08:27.339 --> 00:08:30.779 basically told the AI to act as a macroeconomist, 00:08:30.819 --> 00:08:33.779 gather historical data on financial bubbles like 00:08:33.779 --> 00:08:36.440 the dot -com era, the Great Depression, and then 00:08:36.440 --> 00:08:39.559 build an animated public dashboard to analyze 00:08:39.559 --> 00:08:42.789 the current AI market. And what it created was 00:08:42.789 --> 00:08:45.750 an interactive macroeconomic analysis tool. It 00:08:45.750 --> 00:08:49.129 had current AI market data, I think CapEx projections 00:08:49.129 --> 00:08:53.110 of $405 billion in 2025. Yeah, and Minsky cycle 00:08:53.110 --> 00:08:55.590 visualizations. Let's clarify that. CapEx, capital 00:08:55.590 --> 00:08:58.029 expenditure. That's the money corporations are 00:08:58.029 --> 00:08:59.889 spending on physical assets, right? Exactly. 00:08:59.889 --> 00:09:02.090 Like infrastructure and chips. That number is 00:09:02.090 --> 00:09:04.169 a critical proxy for massive corporate confidence. 00:09:04.620 --> 00:09:06.840 And the Minsky cycle just explains the five standard 00:09:06.840 --> 00:09:09.059 stages of speculative bubble, from displacement 00:09:09.059 --> 00:09:11.580 all the way to panic. So the AI didn't just plot 00:09:11.580 --> 00:09:14.320 data. It applied complex financial theory. It 00:09:14.320 --> 00:09:17.100 did. And the highlight was the bubble simulator 00:09:17.100 --> 00:09:20.279 game that was built into the dashboard. Users 00:09:20.279 --> 00:09:22.600 could make business decisions, raise funding, 00:09:22.840 --> 00:09:25.720 hire people, and the chart would update tracking 00:09:25.720 --> 00:09:28.679 risk until the demo day bubble popped. And the 00:09:28.679 --> 00:09:31.120 tool even provided its own strategic assessment. 00:09:31.299 --> 00:09:33.659 It did. It said something like, productive buttle, 00:09:33.679 --> 00:09:37.360 infrastructure survives, speculators die, banks 00:09:37.360 --> 00:09:40.500 safe, stocks at risk. That's predictive application, 00:09:40.740 --> 00:09:43.039 not just visualization. So moving from macro 00:09:43.039 --> 00:09:45.860 trends to micro analysis, we have the basketball 00:09:45.860 --> 00:09:48.779 shot analyzer. This is a total game changer for 00:09:48.779 --> 00:09:50.759 coaching and skill development. The workflow 00:09:50.759 --> 00:09:52.919 seems simple. You just upload a short clip of 00:09:52.919 --> 00:09:55.320 a jump shot and the system processes it frame 00:09:55.320 --> 00:09:57.639 by frame. Right, using computer vision. And the 00:09:57.639 --> 00:10:00.440 technical advantage is just so clear. It delivers 00:10:00.440 --> 00:10:03.440 this granular biomechanics, breakdown joint angles, 00:10:03.559 --> 00:10:06.019 wrist snap, foot placement. It analyzes the entire 00:10:06.019 --> 00:10:08.759 kinetic chain and gives you a shot score and 00:10:08.759 --> 00:10:11.480 even a pro comparison mode with skeleton overlays. 00:10:11.580 --> 00:10:13.700 So how does that economic bubble game really 00:10:13.700 --> 00:10:15.639 show the difference between just raw analysis 00:10:15.639 --> 00:10:18.700 and genuine application? it turns static data 00:10:18.700 --> 00:10:21.440 and historical frameworks into a dynamic educational 00:10:21.440 --> 00:10:24.059 and most importantly a playable scenario for 00:10:24.059 --> 00:10:26.919 the user to learn from our final segment looks 00:10:26.919 --> 00:10:29.539 at the ai's ability to model large -scale systems 00:10:29.539 --> 00:10:32.480 and integrate real -time inputs we're starting 00:10:32.480 --> 00:10:34.840 with the earth simulator with weather control 00:10:34.840 --> 00:10:38.740 the prompt asked for a perfect 3d earth globe 00:10:38.740 --> 00:10:42.259 realistic interactive cloud layers high quality 00:10:42.259 --> 00:10:46.039 rendering and it delivered a rotatable 3d earth 00:10:46.039 --> 00:10:48.730 where you can actually adjust environmental parameters, 00:10:48.970 --> 00:10:51.710 CO2 levels, temperature. The satellite feature 00:10:51.710 --> 00:10:53.730 was particularly impressive, I thought. It really 00:10:53.730 --> 00:10:56.169 was. The system calculated and displayed the 00:10:56.169 --> 00:10:58.649 correct orbital paths for communication and weather 00:10:58.649 --> 00:11:00.970 satellites, all moving in their actual trajectories. 00:11:01.090 --> 00:11:03.389 Then that's hard. That requires accurate geospatial 00:11:03.389 --> 00:11:05.389 calculation with real -time physics. It's not 00:11:05.389 --> 00:11:07.370 just a visual. And you could use natural language 00:11:07.370 --> 00:11:09.909 commands. You can type trigger a Category 5 hurricane, 00:11:10.070 --> 00:11:12.350 and the system updates the visual state. Amazing. 00:11:12.590 --> 00:11:15.110 Then we have the hand -tracking fluid dynamics 00:11:15.110 --> 00:11:18.149 simulator. This one is simpler. It just tracks 00:11:18.149 --> 00:11:20.409 your hands with a webcam to create these swirling 00:11:20.409 --> 00:11:23.940 fluid dynamic effects. appeal there is just demonstrating 00:11:23.940 --> 00:11:26.419 that real -time computer vision integration with 00:11:26.419 --> 00:11:29.080 generative effects it's the basis for future 00:11:29.080 --> 00:11:32.159 gesture based interfaces and finally the power 00:11:32.159 --> 00:11:35.360 of 10 animator this tested the scaling challenge 00:11:36.250 --> 00:11:38.970 generating an animation zooming all the way from 00:11:38.970 --> 00:11:42.350 a cosmic scale down to an atomic scale. And Gemini 00:11:42.350 --> 00:11:44.529 did this by coordinating with another model, 00:11:44.690 --> 00:11:47.769 NanoBanana, to blend the imagery across those 00:11:47.769 --> 00:11:50.169 different magnitudes. And that ability to coordinate 00:11:50.169 --> 00:11:53.570 specialized subtasks, telling one AI to handle 00:11:53.570 --> 00:11:55.809 image generation and another to handle the animation, 00:11:56.070 --> 00:11:58.509 that's the next huge leap. Though we do have 00:11:58.509 --> 00:12:00.909 to acknowledge the honest check, the content 00:12:00.909 --> 00:12:03.879 accuracy. It varied. There was an odd, unexplained 00:12:03.879 --> 00:12:06.740 hand artifact that appeared at the cellular level. 00:12:06.879 --> 00:12:09.419 Right. So the concept of AI directing other AI 00:12:09.419 --> 00:12:12.100 worked, but the final polish still needed a human 00:12:12.100 --> 00:12:15.000 eye. So what's the overarching lesson from combining 00:12:15.000 --> 00:12:17.919 vision, physics, and image generation across 00:12:17.919 --> 00:12:21.080 all these different scales? That the AI can successfully 00:12:21.080 --> 00:12:23.740 coordinate specialized subtasks and separate 00:12:23.740 --> 00:12:26.980 models to achieve these really complex multimodal 00:12:26.980 --> 00:12:30.120 results on a grand scale. When you look back 00:12:30.120 --> 00:12:33.539 at All 12 of these applications, I mean, from 00:12:33.539 --> 00:12:35.500 the Monopoly board generator, which we didn't 00:12:35.500 --> 00:12:37.580 even have time to get into, to that high -end 00:12:37.580 --> 00:12:40.419 ray tracing demo, the impressive part isn't any 00:12:40.419 --> 00:12:43.360 single one of them. No. It's the sheer totality 00:12:43.360 --> 00:12:47.009 of it, the diversity and the complexity. 12 functional 00:12:47.009 --> 00:12:49.149 applications built through conversation in a 00:12:49.149 --> 00:12:51.090 matter of minutes. That amount of work would 00:12:51.090 --> 00:12:53.330 have required significant development time and 00:12:53.330 --> 00:12:55.450 specialized expertise just a few months ago. 00:12:55.590 --> 00:12:57.710 Multiple teams, probably. The ultimate unlock 00:12:57.710 --> 00:12:59.950 here isn't just speed. It's the democratization 00:12:59.950 --> 00:13:03.110 of high -end tooling. The code is still complex, 00:13:03.309 --> 00:13:05.870 but the barrier to accessing that complexity 00:13:05.870 --> 00:13:08.429 has been, you know, just vaporized by conversation. 00:13:09.129 --> 00:13:12.129 So the real question then isn't whether AI can 00:13:12.129 --> 00:13:14.830 build perfect applications today. It's whether 00:13:14.830 --> 00:13:17.889 AI can make building useful, structurally sound 00:13:17.889 --> 00:13:21.009 things accessible to significantly more people. 00:13:21.190 --> 00:13:23.970 And based on these results, the answer is, it's 00:13:23.970 --> 00:13:27.309 an undeniable yes. So if the AI handles the complexity, 00:13:27.649 --> 00:13:30.409 does the value of the human developer now shift 00:13:30.409 --> 00:13:33.629 entirely to optimization? Or does this finally 00:13:33.629 --> 00:13:35.649 mean we can move beyond the screen and start 00:13:35.649 --> 00:13:37.649 building truly custom interfaces applications 00:13:37.649 --> 00:13:39.809 that just weren't worth the coding effort before? 00:13:40.190 --> 00:13:42.850 The tools are here today. What kind of complex 00:13:42.850 --> 00:13:45.529 logic or creative idea that you previously thought 00:13:45.529 --> 00:13:47.950 was impossible to code are you now ready to prototype? 00:13:48.490 --> 00:13:50.909 That's something fascinating to consider as we 00:13:50.909 --> 00:13:51.990 wrap up this deep dive.