ReadMultiplex.com: The Last Program: How AI Is Replacing Every App You Use.의 전체 녹취록

your intense curiosity, you really need 00:01:47.079 --> 00:01:49.799 to support Brian. You really do. Please become 00:01:49.799 --> 00:01:52.099 a member of his community or at the very least 00:01:52.099 --> 00:01:54.459 just buy him a coffee. Your support is the engine 00:01:54.459 --> 00:01:56.579 that keeps these paradigm shifting explorations 00:01:56.579 --> 00:01:59.099 going. So go support Brian Romilly and his amazing 00:01:59.099 --> 00:02:01.980 work. It's definitely worth it because the depth 00:02:01.980 --> 00:02:04.900 of analysis in this document warrants that support. 00:02:05.469 --> 00:02:08.509 I mean, it doesn't just predict a trend. It outlines 00:02:08.509 --> 00:02:11.150 a complete structural collapse and rebirth of 00:02:11.150 --> 00:02:14.050 the software industry. And it grounds it in very 00:02:14.050 --> 00:02:16.610 specific technological mechanisms, right? Exactly. 00:02:16.750 --> 00:02:19.669 It upends 50 years of software engineering philosophy. 00:02:20.550 --> 00:02:23.669 We are charting a course from a highly specific 00:02:23.669 --> 00:02:27.729 explosive catalyst event in October 2026, mapping 00:02:27.729 --> 00:02:31.050 the economic and legal fallout, and finally arriving 00:02:31.050 --> 00:02:33.270 at that post -app reality you just described 00:02:33.270 --> 00:02:36.090 in 2036. So let's look at that mechanism that 00:02:36.090 --> 00:02:39.069 sets this entire collapse in motion. The document 00:02:39.069 --> 00:02:42.430 pinpoints early October 2026. A developer named 00:02:42.430 --> 00:02:44.909 Brandon Thomas, working at a company called Artcraft, 00:02:45.129 --> 00:02:48.229 basically initiates what we can only call a shockwave. 00:02:48.409 --> 00:02:49.930 A total shockwave in the developer community. 00:02:50.129 --> 00:02:54.240 Right. He directs a suite of Claude Opus 5 .5 00:02:54.240 --> 00:02:57.900 AI agents to create clean room re -implementations 00:02:57.900 --> 00:03:00.800 of seven flagship Adobe applications. And he 00:03:00.800 --> 00:03:02.960 tasks the AI with writing these new applications 00:03:02.960 --> 00:03:05.979 entirely in the Rust programming language. And 00:03:05.979 --> 00:03:08.520 the timeline here is the first major indicator 00:03:08.520 --> 00:03:11.159 of the paradigm shift. I mean, the repositories 00:03:11.159 --> 00:03:13.860 for these clones appeared on GitHub between September 00:03:13.860 --> 00:03:18.349 30th and October 1st. Two days. Two days. Functionality 00:03:18.349 --> 00:03:20.349 that originally required decades of iteration. 00:03:20.889 --> 00:03:23.490 Thousands of specialized human engineers. It 00:03:23.490 --> 00:03:26.569 was replicated over a single weekend. It's just 00:03:26.569 --> 00:03:29.270 wild to think about. And the developer community's 00:03:29.270 --> 00:03:31.949 reaction was instantaneous. The document notes 00:03:31.949 --> 00:03:33.909 that the Photoshop equivalent, which they dubbed 00:03:33.909 --> 00:03:38.129 Photocraft, hit 45 ,000 GitHub stars in 10 days. 00:03:38.389 --> 00:03:40.949 Which is astronomical for a new repo. Oh, totally. 00:03:41.129 --> 00:03:44.210 And the entire suite exceeded 86 ,000 stars by 00:03:44.210 --> 00:03:47.110 October 11th. But the detail that really signals 00:03:47.110 --> 00:03:48.629 the inflection point, the thing that blew my 00:03:48.629 --> 00:03:51.409 mind. is that the majority of the commits, the 00:03:51.409 --> 00:03:53.629 actual lines of code being submitted to the project, 00:03:53.810 --> 00:03:56.870 carried a co -author line crediting the AI model. 00:03:57.110 --> 00:03:59.729 Yeah, the human was just the director. The AI 00:03:59.729 --> 00:04:02.990 was the primary builder. Which forces us to really 00:04:02.990 --> 00:04:05.530 examine the actual process used here. Because 00:04:05.530 --> 00:04:07.990 they didn't just steal the code, did they? Nope, 00:04:08.050 --> 00:04:11.229 not at all. The project explicitly stated they 00:04:11.229 --> 00:04:14.150 did not copy proprietary code. They utilized 00:04:14.150 --> 00:04:16.490 a method of reverse engineering by observation. 00:04:17.259 --> 00:04:19.980 Okay, let's pause and draw a hard line here between 00:04:19.980 --> 00:04:23.100 observation and decompilation. Because this is 00:04:23.100 --> 00:04:25.000 where the legal and technical arguments usually 00:04:25.000 --> 00:04:27.600 get, you know, really tangled up. Right. It's 00:04:27.600 --> 00:04:30.319 an important distinction. Historically, if you 00:04:30.319 --> 00:04:32.420 wanted to clone a piece of software, you might 00:04:32.420 --> 00:04:35.100 try decompilation. You take the compiled binary 00:04:35.100 --> 00:04:37.620 file, that executable black box that runs on 00:04:37.620 --> 00:04:40.040 your machine, and you run it through a decompiler 00:04:40.040 --> 00:04:42.379 to spit out something resembling the original 00:04:42.379 --> 00:04:45.180 source code. But decompilation is famously flawed. 00:04:45.750 --> 00:04:48.509 And legally hazardous, too. Right. Because the 00:04:48.509 --> 00:04:51.089 compiler strips out all the human intent. Exactly. 00:04:51.129 --> 00:04:54.310 The variable names, the comments, the elegant 00:04:54.310 --> 00:04:56.649 logic structures, all of that is optimized away 00:04:56.649 --> 00:04:59.810 into raw machine code. It's just ones and zeros 00:04:59.810 --> 00:05:02.250 at that point. It's not just unbaking a cake. 00:05:02.879 --> 00:05:05.279 It's more like trying to reconstruct a highly 00:05:05.279 --> 00:05:08.319 complex multilingual novel by only looking at 00:05:08.319 --> 00:05:10.819 a spreadsheet that details how many times each 00:05:10.819 --> 00:05:13.379 letter of the alphabet was used. That is a great 00:05:13.379 --> 00:05:15.730 analogy. You know, you have the raw materials, 00:05:15.790 --> 00:05:18.790 but the narrative flow, the grammar, the human 00:05:18.790 --> 00:05:21.569 intent behind the story was

completely shredded 00:05:21.569 --> 00:05:24.649 by the compiler. It's a mess. And looking at 00:05:24.649 --> 00:05:27.350 that mess often violates end -user license agreements 00:05:27.350 --> 00:05:30.170 anyway. And that is precisely why the methodology 00:05:30.170 --> 00:05:32.769 outlined in the last program bypasses the binary 00:05:32.769 --> 00:05:36.410 entirely. The AI agents didn't even attempt to 00:05:36.410 --> 00:05:39.329 unpack Adobe's compiled code. They just observed 00:05:39.329 --> 00:05:42.009 the software as a user would, but at the speed 00:05:42.009 --> 00:05:44.660 and scale of computation. Let's get into the 00:05:44.660 --> 00:05:47.519 technical weeds here of how a text -based or 00:05:47.519 --> 00:05:50.939 even a multimodal LLM actually observes complex 00:05:50.939 --> 00:05:53.120 visual software like a photo editor. I mean, 00:05:53.120 --> 00:05:57.180 how does an AI look at a screen and write Rust 00:05:57.180 --> 00:06:00.660 code to match it? Well, it requires a synthesis 00:06:00.660 --> 00:06:03.939 of a few different capabilities. First, the AI 00:06:03.939 --> 00:06:06.360 leverages computer vision to analyze the graphical 00:06:06.360 --> 00:06:09.300 user interface. It literally maps the visual 00:06:09.300 --> 00:06:11.720 state of the application. Like it sees where 00:06:11.720 --> 00:06:13.939 the buttons and sliders are. Right. Then it interacts 00:06:13.939 --> 00:06:16.279 with the program. So it feeds it a specific image 00:06:16.279 --> 00:06:19.040 file, applies, let's say, a Gaussian blur filter 00:06:19.040 --> 00:06:21.860 with a radius of five pixels. And then it observes 00:06:21.860 --> 00:06:25.040 the new visual state. But crucially, it doesn't 00:06:25.040 --> 00:06:27.339 just look at the screen. It analyzes the underlying 00:06:27.339 --> 00:06:30.959 file formats. Ah, right. Because the specifications 00:06:30.959 --> 00:06:34.600 for formats like PSD or PDF, those are publicly 00:06:34.600 --> 00:06:37.379 documented. Anyone can read how a PSD file is 00:06:37.379 --> 00:06:40.699 structured. Exactly. So the AI agent. generates 00:06:40.699 --> 00:06:43.519 an input file, applies a function in the target 00:06:43.519 --> 00:06:45.980 software, and then mathematically compares the 00:06:45.980 --> 00:06:48.300 output file against that public specification. 00:06:48.779 --> 00:06:52.459 So it's looking at the math? Yes. Once it understands 00:06:52.459 --> 00:06:55.079 the exact mathematical transformation that occurred, 00:06:55.259 --> 00:06:58.620 say, a specific matrix multiplication applied 00:06:58.620 --> 00:07:02.040 to the RGB values of a pixel array, it proposes 00:07:02.040 --> 00:07:04.199 a piece of Rust code to replicate that exact 00:07:04.199 --> 00:07:06.639 math. And then it checks its own work, right? 00:07:06.720 --> 00:07:09.319 It compiles its proposed Rust code, runs the 00:07:09.319 --> 00:07:11.139 exact same input file through it, and compares 00:07:11.139 --> 00:07:14.060 its output to Adobe's output. Perfectly said. 00:07:14.160 --> 00:07:16.639 And if the hexadecimal values of the resulting 00:07:16.639 --> 00:07:19.759 files don't match, it iterates. It tweaks the 00:07:19.759 --> 00:07:22.160 Rust code and tries again. Millions of times 00:07:22.160 --> 00:07:25.860 over. Millions of times. It is an automated high 00:07:25.860 --> 00:07:29.459 -speed loop of hypothesis, execution, evaluation, 00:07:29.980 --> 00:07:32.759 and refinement. And the choice of Rust here. 00:07:33.079 --> 00:07:34.959 as highlighted in the document, is not accidental 00:07:34.959 --> 00:07:39.079 at all. Why Rust specifically? Well, Rust offers 00:07:39.079 --> 00:07:41.800 memory safety guarantees and fearless concurrency. 00:07:42.120 --> 00:07:44.459 That means the resulting application is incredibly 00:07:44.459 --> 00:07:47.620 fast and less prone to the kind of legacy memory 00:07:47.620 --> 00:07:51.019 leaks that plague older C++ code bases. So it's 00:07:51.019 --> 00:07:52.959 actually building a better foundation. Exactly. 00:07:53.180 --> 00:07:56.240 The AI isn't just copying the function. It's 00:07:56.240 --> 00:07:58.480 often writing structurally superior, more modern 00:07:58.480 --> 00:08:01.240 code to execute that function. Which naturally 00:08:01.240 --> 00:08:03.699 leads to the question of viability because, you 00:08:03.699 --> 00:08:06.379 know, 45 ,000 GitHub stars usually indicates 00:08:06.379 --> 00:08:09.480 massive hype, but hype doesn't equal a production 00:08:09.480 --> 00:08:12.079 ready tool. Very true. If I am a professional 00:08:12.079 --> 00:08:14.500 art director working at an agency and I drop 00:08:14.500 --> 00:08:18.540 a hundred layer PSD file onto my desk, is Photocraft 00:08:18.540 --> 00:08:20.990 actually going to handle that on day 10? The 00:08:20.990 --> 00:08:22.850 source document is actually highly objective 00:08:22.850 --> 00:08:25.389 about this limitation. It clearly states that 00:08:25.389 --> 00:08:27.829 while Photocraft achieved high coverage of standard 00:08:27.829 --> 00:08:30.550 menu commands, and it demonstrated solid round 00:08:30.550 --> 00:08:33.970 -trip success on test PSD files, meaning it could 00:08:33.970 --> 00:08:36.210 open the file, modify it, and save it without 00:08:36.210 --> 00:08:38.870 corrupting the data structure, its internal readiness 00:08:38.870 --> 00:08:42.049 for heavy, professional daily work was still 00:08:42.049 --> 00:08:44.309 lower than the original. Right, the edge cases. 00:08:45.080 --> 00:08:47.659 The undocumented legacy features, the highly 00:08:47.659 --> 00:08:50.379 specific color profile management behaviors that 00:08:50.379 --> 00:08:52.539 a professional just relies on without even thinking 00:08:52.539 --> 00:08:55.120 about them. The clone is going to stumble there, 00:08:55.159 --> 00:08:58.720 isn't it? It will stumble initially, yes. But 00:08:58.720 --> 00:09:00.740 the

document argues that focusing on the immediate 00:09:00.740 --> 00:09:03.600 gap misses the historical magnitude of the event. 00:09:03.759 --> 00:09:05.799 The gap between the clone and the original is 00:09:05.799 --> 00:09:07.860 just a matter of time and iteration. It's not 00:09:07.860 --> 00:09:11.039 a fundamental barrier. The dam has broken. Exactly. 00:09:11.590 --> 00:09:13.889 The critical takeaway is that the scarcity of 00:09:13.889 --> 00:09:16.330 the binary has been broken. The scarcity of the 00:09:16.330 --> 00:09:19.730 binary. Let's dig into that, because for 50 years, 00:09:19.929 --> 00:09:22.309 the software industry's entire economic model 00:09:22.309 --> 00:09:24.289 has been predicated on the assumption that writing 00:09:24.289 --> 00:09:27.490 complex software is impossibly hard, highly expensive, 00:09:27.690 --> 00:09:31.009 and requires a massive organization. A huge moat. 00:09:31.190 --> 00:09:35.059 Right. The compiled binary was this scarce, highly 00:09:35.059 --> 00:09:37.639 valuable artifact. If you wanted the capability 00:09:37.639 --> 00:09:40.120 to edit layers in a photograph, you had to rent 00:09:40.120 --> 00:09:42.639 the binary from the organization that spent $100 00:09:42.639 --> 00:09:45.720 million building it. And the October 2026 event 00:09:45.720 --> 00:09:48.539 demonstrates that the capability to perform those 00:09:48.539 --> 00:09:51.480 complex operations is no longer tied to the proprietary 00:09:51.480 --> 00:09:54.639 binary. The operations can be systematically 00:09:54.639 --> 00:09:58.019 extracted and perfectly replicated by an automated 00:09:58.019 --> 00:10:01.519 observer in a matter of hours. The moat has completely 00:10:01.519 --> 00:10:04.379 evaporated. So if the ability to composite images 00:10:04.379 --> 00:10:07.480 or sort massive databases or compile code is 00:10:07.480 --> 00:10:10.259 no longer a scarce resource, what happens to 00:10:10.259 --> 00:10:13.200 the value of software? I mean, the text introduces 00:10:13.200 --> 00:10:16.419 this concept of the surviving scarcity. If code 00:10:16.419 --> 00:10:19.200 is infinitely and instantly reproducible, what 00:10:19.200 --> 00:10:22.059 is the bottleneck? The bottleneck reverts to 00:10:22.059 --> 00:10:25.399 human judgment. Human judgment. Yes. The document 00:10:25.399 --> 00:10:27.779 articulates this beautifully by examining what 00:10:27.779 --> 00:10:31.539 an AI generated clone actually lacks. We have 00:10:31.539 --> 00:10:33.580 an open source clone that can perfectly composite 00:10:33.580 --> 00:10:36.820 20 layers of high resolution imagery. It understands 00:10:36.820 --> 00:10:38.919 the math of a drop shadow and the physics of 00:10:38.919 --> 00:10:41.120 light rendering perfectly. Better than a human, 00:10:41.200 --> 00:10:44.379 probably. Right. But the AI agent possesses no 00:10:44.379 --> 00:10:46.620 concept of why the composite is being constructed. 00:10:48.129 --> 00:10:50.070 It

doesn't know if the resulting image is meant 00:10:50.070 --> 00:10:52.529 to evoke nostalgia for a marketing campaign or 00:10:52.529 --> 00:10:55.210 if it's meant to accurately document a journalistic 00:10:55.210 --> 00:10:57.909 event without misleading the public. It cannot 00:10:57.909 --> 00:11:00.350 determine if the output advances your underlying 00:11:00.350 --> 00:11:04.360 purpose. The AI can provide the mechanism, but 00:11:04.360 --> 00:11:07.000 only the human can provide the intent and the 00:11:07.000 --> 00:11:09.559 final evaluation of success. That's a profound 00:11:09.559 --> 00:11:12.940 shift. It is. And this brings us to a profound 00:11:12.940 --> 00:11:16.559 philosophical reframing in the text. It posits 00:11:16.559 --> 00:11:19.120 that software... as we have historically understood 00:11:19.120 --> 00:11:22.820 it, was merely a translation layer. A translator 00:11:22.820 --> 00:11:25.039 between human intent and machine limitation. 00:11:25.620 --> 00:11:27.600 Think about the evolution of human -computer 00:11:27.600 --> 00:11:31.159 interaction. The machine only understands microscopic 00:11:31.159 --> 00:11:34.490 electrical state changes, zeros and ones. To 00:11:34.490 --> 00:11:36.470 get the machine to do work, humans had to learn 00:11:36.470 --> 00:11:39.009 its language. Right, we started with punch cards. 00:11:39.389 --> 00:11:42.149 Then we moved to command -line interfaces, learning 00:11:42.149 --> 00:11:45.059 highly rigid syntax. Then we invented graphical 00:11:45.059 --> 00:11:47.580 user interfaces, replacing typed commands with 00:11:47.580 --> 00:11:50.480 drop -down menus and icons. But even a drop -down 00:11:50.480 --> 00:11:52.960 menu is basically just a dictionary. When you 00:11:52.960 --> 00:11:54.960 launch a massive spreadsheet application, you 00:11:54.960 --> 00:11:57.240 have a core intent in your mind. You're thinking, 00:11:57.340 --> 00:12:00.799 tell me if our Q3 marketing spend actually resulted 00:12:00.799 --> 00:12:03.360 in a higher customer retention rate, and flag 00:12:03.360 --> 00:12:05.799 any regions that deviated from the trend. That 00:12:05.799 --> 00:12:08.039 is your intent. But the computer couldn't accept 00:12:08.039 --> 00:12:11.200 that intent. It refused ordinary language. So 00:12:11.200 --> 00:12:13.419 you, the human, had to act as the translator. 00:12:13.720 --> 00:12:16.460 Exactly. You had to launch the spreadsheet. You 00:12:16.460 --> 00:12:18.019 had to know which buttons to click to import 00:12:18.019 --> 00:12:21.039 the CSV data. You had to manually write the syntax 00:12:21.039 --> 00:12:24.159 for VLOOKUPs and pivot tables. Yeah. You had 00:12:24.159 --> 00:12:26.360 to map the cells, apply conditional formatting, 00:12:26.539 --> 00:12:28.700 and manually construct the graph. You had to 00:12:28.700 --> 00:12:32.259 break down your high -level intent into 50 micro 00:12:32.259 --> 00:12:34.399 -operations that the software's interface

could 00:12:34.399 --> 00:12:37.210 actually understand. And we became so accustomed 00:12:37.210 --> 00:12:40.289 to this translation work that we started to conflate 00:12:40.289 --> 00:12:42.809 it with the actual job. We put proficient in 00:12:42.809 --> 00:12:45.690 Excel or Adobe Creative Suite Master on our resumes, 00:12:45.789 --> 00:12:48.669 as if navigating the labyrinth of a software 00:12:48.669 --> 00:12:51.429 interface was the actual valuable skill. Wow. 00:12:51.590 --> 00:12:54.309 We took pride in being mechanics when we really 00:12:54.309 --> 00:12:56.990 just wanted to be drivers. We memorized the transmission 00:12:56.990 --> 00:12:59.049 ratios just so we could go to the grocery store. 00:12:59.309 --> 00:13:01.850 That's it exactly. The source document argues 00:13:01.850 --> 00:13:04.009 that navigating the software was never the actual 00:13:04.009 --> 00:13:07.399 work. The work was extracting the anomalous data 00:13:07.399 --> 00:13:10.679 to make the business decision. Right. The work 00:13:10.679 --> 00:13:12.440 was adjusting the contrast so the photograph 00:13:12.440 --> 00:13:15.340 communicated the right emotion. The software 00:13:15.340 --> 00:13:17.259 was just the hurdle you had to clear to execute 00:13:17.259 --> 00:13:19.799 your judgment. So when the AI agents demonstrate 00:13:19.799 --> 00:13:22.039 they can reconstruct the software's capabilities 00:13:22.039 --> 00:13:25.139 on demand, they are essentially taking over the 00:13:25.139 --> 00:13:28.049 job of translation. You just state the intent 00:13:28.049 --> 00:13:30.750 in ordinary language and the AI agent automatically 00:13:30.750 --> 00:13:33.529 maps that intent to the necessary computational 00:13:33.529 --> 00:13:36.669 operations. Which completely eliminates the need 00:13:36.669 --> 00:13:39.029 for the permanent translator. You no longer need 00:13:39.029 --> 00:13:41.269 the massive bloated software application sitting 00:13:41.269 --> 00:13:43.250 on your hard drive waiting for you to click its 00:13:43.250 --> 00:13:46.230 icons. Okay, this moves us into the timeline 00:13:46.230 --> 00:13:49.750 the document sets for 2028. This is the era of 00:13:49.750 --> 00:13:53.389 the program as a temporary object. A massive 00:13:53.389 --> 00:13:56.110 conceptual leap. The text suggests that within 00:13:56.110 --> 00:13:58.350 two years of the Catalyst event, most commercial 00:13:58.350 --> 00:14:01.190 software whose behavior can be observed will 00:14:01.190 --> 00:14:04.730 exist as open implementations. But it goes a 00:14:04.730 --> 00:14:07.070 step further, doesn't it? It says we won't just 00:14:07.070 --> 00:14:09.590 replace proprietary apps with open source apps. 00:14:09.750 --> 00:14:14.049 We will stop using apps entirely. Yes. The concept 00:14:14.049 --> 00:14:16.429 of the installed application begins to dissolve. 00:14:16.909 --> 00:14:19.289 When you state a complex job to your system, 00:14:19.350 --> 00:14:21.789 let's say, produce a layout fitting these images 00:14:21.789 --> 00:14:25.309 with space for a caption. Or... Match the lighting 00:14:25.309 --> 00:14:27.409 of this reference video clip to these three raw 00:14:27.409 --> 00:14:30.610 clips. The system does not launch a named program. 00:14:31.289 --> 00:14:33.629 So walk me through the actual computer science 00:14:33.629 --> 00:14:36.029 of this transient environment, because I'm trying 00:14:36.029 --> 00:14:38.330 to picture this on my machine. How does an operating 00:14:38.330 --> 00:14:41.690 system handle on -the -fly computation without 00:14:41.690 --> 00:14:44.309 a permanent application structure? It relies 00:14:44.309 --> 00:14:46.970 on an evolution of containerization and modular 00:14:46.970 --> 00:14:50.129 architecture. It's likely utilizing technologies 00:14:50.129 --> 00:14:54.669 akin to WebAssembly. or highly optimized sandboxed 00:14:54.669 --> 00:14:57.750 environments. Okay. So when you state your intent, 00:14:58.070 --> 00:15:00.809 a local AI routing model translates that intent 00:15:00.809 --> 00:15:03.629 into a directed acyclic graph of mathematical 00:15:03.629 --> 00:15:07.210 operations. It pulls the necessary open source 00:15:07.210 --> 00:15:09.909 operational libraries, say an image parsing library 00:15:09.909 --> 00:15:13.070 in a matrix transformation library. And then 00:15:13.070 --> 00:15:15.750 it compiles a highly specific lightweight executable 00:15:15.750 --> 00:15:19.090 just for this single request. It compiles it 00:15:19.090 --> 00:15:22.169 right then and there. Yes. It executes the code, 00:15:22.549 --> 00:15:25.110 renders the video frame, hands the result back 00:15:25.110 --> 00:15:27.309 to the display layer, and then the executable 00:15:27.309 --> 00:15:30.269 is immediately purged for memory. The computation 00:15:30.269 --> 00:15:33.190 exists exclusively for the duration of the request. 00:15:33.409 --> 00:15:36.029 Exactly. No background process is left running, 00:15:36.110 --> 00:15:39.629 no registry key is modified, no massive application 00:15:39.629 --> 00:15:41.830 suite consuming your system's resources while 00:15:41.830 --> 00:15:43.809 it's just sitting idle. Okay, I have to push 00:15:43.809 --> 00:15:46.009 back hard on the physics of this, though, because... 00:15:46.309 --> 00:15:49.070 The latency of compiling a custom program from 00:15:49.070 --> 00:15:51.429 scratch every single time I want to crop an image 00:15:51.429 --> 00:15:53.750 or sort a list, it just sounds absurd. I know, 00:15:53.769 --> 00:15:56.049 it sounds counterintuitive. We are talking about 00:15:56.049 --> 00:15:59.529 replacing highly optimized, pre -compiled C++ 00:15:59.529 --> 00:16:02.629 binaries with AI -generated, just -in -time code 00:16:02.629 --> 00:16:05.590 execution. Isn't this going to be painfully slow? 00:16:06.190 --> 00:16:08.950 And won't it require massive, expensive cloud 00:16:08.950 --> 00:16:11.250

computing power? I mean, if I have to wait five 00:16:11.250 --> 00:16:13.669 seconds for an AI to write a script just to turn 00:16:13.669 --> 00:16:16.220 my text bold, the entire paradigm collapses. 00:16:16.580 --> 00:16:19.720 That is the natural skepticism, absolutely. But 00:16:19.720 --> 00:16:21.299 it stems from a misunderstanding of what the 00:16:21.299 --> 00:16:24.620 AI is actually generating. The AI is not writing 00:16:24.620 --> 00:16:26.879 a new image rendering engine from scratch every 00:16:26.879 --> 00:16:30.019 time you ask for a crop. Ah, okay. Clarify that 00:16:30.019 --> 00:16:32.399 distinction because it's critical. The heavy 00:16:32.399 --> 00:16:34.860 lifting, the actual processing of pixels, or 00:16:34.860 --> 00:16:38.460 the cryptographic hashing of a file, or the complex 00:16:38.460 --> 00:16:41.299 sorting algorithms, those remain pre -compiled. 00:16:41.559 --> 00:16:44.100 highly optimized mathematical libraries stored 00:16:44.100 --> 00:16:46.200 locally on your machine. So the foundation is 00:16:46.200 --> 00:16:48.379 already there. Exactly. What the AI generates 00:16:48.379 --> 00:16:51.299 on the fly is merely the glue code. It writes 00:16:51.299 --> 00:16:53.419 the tiny script that routes your specific data 00:16:53.419 --> 00:16:55.799 through those pre -existing hyper -fast libraries 00:16:55.799 --> 00:16:58.860 based on your ordinary language request. Oh, 00:16:59.100 --> 00:17:02.159 so it's like a highly intelligent API caller. 00:17:02.519 --> 00:17:05.019 It's not reinventing the wheel. It's just instantly 00:17:05.019 --> 00:17:06.980 writing the code that connects the steering wheel 00:17:06.980 --> 00:17:09.940 to the axle based on where I want to turn. That 00:17:09.940 --> 00:17:12.849 is the... perfect way to look at it. Generating 00:17:12.849 --> 00:17:15.450 50 lines of glue code in Rust or Python takes 00:17:15.450 --> 00:17:18.420 milliseconds for a local model. The execution 00:17:18.420 --> 00:17:20.700 of the underlying mathematical library is just 00:17:20.700 --> 00:17:22.819 as fast as it would be if Adobe had called it. 00:17:22.980 --> 00:17:26.059 Okay, that makes sense. Which brings us to arguably 00:17:26.059 --> 00:17:28.319 the most controversial and fascinating claim 00:17:28.319 --> 00:17:31.240 in the entire document, the hardware renaissance. 00:17:31.559 --> 00:17:33.960 This is where the economics really get turned 00:17:33.960 --> 00:17:36.519 upside down. You mentioned local models just 00:17:36.519 --> 00:17:40.000 now. The text insists that this revolution doesn't 00:17:40.000 --> 00:17:42.819 require a massive server farm. It claims this 00:17:42.819 --> 00:17:45.119 transient computation happens locally, right 00:17:45.119 --> 00:17:47.880 on the user's desk, using surprisingly old hardware. 00:17:48.160 --> 00:17:50.660 The document paints a very specific scenario. 00:17:50.920 --> 00:17:53.480 It argues that a standard Linux machine from 00:17:53.480 --> 00:17:57.319 2015 or 2018, something sitting under a desk 00:17:57.319 --> 00:17:59.400 collecting dust because it can no longer run 00:17:59.400 --> 00:18:02.880 the latest bloated commercial software, suddenly 00:18:02.880 --> 00:18:05.960 becomes a highly capable primary computer again. 00:18:06.420 --> 00:18:08.579 I need to scrutinize this. Because the hardware 00:18:08.579 --> 00:18:10.640 requirements for AI right now are astronomical. 00:18:10.880 --> 00:18:14.099 The text explicitly claims that 8 to 16 gigabytes 00:18:14.099 --> 00:18:16.420 of RAM is enough to run these models. Yes, it 00:18:16.420 --> 00:18:19.720 does. But anyone who has tried to run a modern 00:18:19.720 --> 00:18:22.180 large language model knows that you need massive 00:18:22.180 --> 00:18:25.259 amounts of VRAM. A 70 billion parameter model 00:18:25.259 --> 00:18:28.180 cannot physically fit into 8 gigs of system RAM. 00:18:28.799 --> 00:18:32.000 Furthermore, an old 2015 motherboard is running 00:18:32.000 --> 00:18:36.710 PCIe Gen 3, maybe DDR3 memory. The memory bandwidth 00:18:36.710 --> 00:18:39.509 is just a fraction of what a modern unified architecture 00:18:39.509 --> 00:18:41.970 provides. You're not wrong about the specs. You'd 00:18:41.970 --> 00:18:44.470 be waiting minutes for a massive model just to 00:18:44.470 --> 00:18:46.630 load into memory, let alone generate tokens. 00:18:47.130 --> 00:18:49.730 So how does the document justify this claim? 00:18:50.190 --> 00:18:52.650 The justification relies on the rapid divergence 00:18:52.650 --> 00:18:55.730 between generalized LLMs, large language models, 00:18:55.930 --> 00:18:58.569 and specialized SLMs, which are small language 00:18:58.569 --> 00:19:01.869 models. Okay, small language models. You're absolutely 00:19:01.869 --> 00:19:04.410 right. that an omniscient chatbot that has memorized 00:19:04.410 --> 00:19:07.670 the entire internet requires massive VRAM and 00:19:07.670 --> 00:19:10.369 memory bandwidth. But the agent running on this 00:19:10.369 --> 00:19:13.390 2015 Linux machine is not a generalized chatbot. 00:19:13.569 --> 00:19:15.789 It doesn't need to know the history of the Roman 00:19:15.789 --> 00:19:18.329 Empire or how to write a sonnet in the style 00:19:18.329 --> 00:19:20.690 of Shakespeare. It only needs to know software 00:19:20.690 --> 00:19:23.670 syntax and file structures. Precisely. We are 00:19:23.670 --> 00:19:26.170 talking about highly quantized, task -specific 00:19:26.170 --> 00:19:29.619 models. A 4 -bit quantized model trained exclusively 00:19:29.619 --> 00:19:32.819 on routing ordinary language to standard programming 00:19:32.819 --> 00:19:35.779 libraries might only be 2 or 3 billion parameters. 00:19:36.160 --> 00:19:39.740 It requires perhaps 1 .5 to 2 gigabytes of RAM. 00:19:40.079 --> 00:19:42.839 Wow. So it fits easily into the memory footprint 00:19:42.839 --> 00:19:46.279 of an old ThinkPad? Easily. And because its output

00:19:46.279 --> 00:19:48.259 is constrained to short bursts of structural 00:19:48.259 --> 00:19:51.339 glue code, the token generation speed, even on 00:19:51.339 --> 00:19:54.900 a slower CPU with DDR3 memory, is measured in 00:19:54.900 --> 00:19:57.740 milliseconds. The text argues that the operating 00:19:57.740 --> 00:20:01.180 system only needs to provide file access, process 00:20:01.180 --> 00:20:04.779 management, and network exposure. A stable Linux 00:20:04.779 --> 00:20:06.900 kernel from a decade ago does this flawlessly. 00:20:07.339 --> 00:20:10.140 Let's ground this in reality then. What does 00:20:10.140 --> 00:20:13.319 this 2015 machine actually do for the user in 00:20:13.319 --> 00:20:16.259 this new paradigm? The source provides some great, 00:20:16.380 --> 00:20:19.960 highly specific daily use cases. Consider the... 00:20:20.230 --> 00:20:22.670 tedious administrative work that consumes hours 00:20:22.670 --> 00:20:25.109 of human capital right now. The document gives 00:20:25.109 --> 00:20:27.289 the example of saying to the system, extract 00:20:27.289 --> 00:20:29.829 the invoice numbers from this folder of PDFs 00:20:29.829 --> 00:20:32.130 and place them in a table that matches the format 00:20:32.130 --> 00:20:35.029 we used last month. Historically, that requires 00:20:35.029 --> 00:20:37.970 opening a bloated PDF reader, manually copying 00:20:37.970 --> 00:20:40.490 data, opening a massive spreadsheet application, 00:20:40.930 --> 00:20:43.410 pasting, formatting. It takes forever. With a 00:20:43.410 --> 00:20:45.809 local agent, the model instantly generates a 00:20:45.809 --> 00:20:48.670 script utilizing an open source PDF parsing library 00:20:48.670 --> 00:20:52.490 and a CSV writing library. It executes the extraction, 00:20:52.730 --> 00:20:55.089 formats the data based on its observation of 00:20:55.089 --> 00:20:57.289 your historical files, and presents the table. 00:20:57.490 --> 00:21:00.430 Boom. Or a hardware monitoring scenario, right? 00:21:00.509 --> 00:21:03.809 The text mentions saying, monitor the temperature 00:21:03.809 --> 00:21:06.470 of these attached external sensors and send me 00:21:06.470 --> 00:21:09.210 an alert only if the reading stays outside the 00:21:09.210 --> 00:21:12.039 70 degree range for more than 10 minutes. Instead 00:21:12.039 --> 00:21:14.299 of buying and installing an expensive proprietary 00:21:14.299 --> 00:21:16.900 dashboard application that consumes four gigabytes 00:21:16.900 --> 00:21:19.799 of RAM just to idle in the background, the agent 00:21:19.799 --> 00:21:22.839 writes a tiny, highly efficient background daemon 00:21:22.839 --> 00:21:26.559 in Rust. It compiles it, sets it to run, and 00:21:26.559 --> 00:21:28.799 the system load is practically zero. The machine 00:21:28.799 --> 00:21:31.859 is doing the actual calculation locally again, 00:21:32.000 --> 00:21:34.940 rather than acting as a dumb, thin client requesting 00:21:34.940 --> 00:21:37.460 computation from a cloud service that charges 00:21:37.460 --> 00:21:42.859 a $30 monthly subscription. absolute catastrophe 00:21:42.859 --> 00:21:45.200 for the current hardware and software economic 00:21:45.200 --> 00:21:47.500 models. I mean, the entire revenue structure 00:21:47.500 --> 00:21:50.160 of the hardware industry relies on planned obsolescence 00:21:50.160 --> 00:21:53.180 driven by software bloat. The upgrade treadmill. 00:21:53.500 --> 00:21:55.980 We don't buy a new laptop every three years because 00:21:55.980 --> 00:21:58.440 the silicon degraded. We buy it because the new 00:21:58.440 --> 00:22:00.720 operating system update and the new version of 00:22:00.720 --> 00:22:03.220 our daily software suite requires twice as much 00:22:03.220 --> 00:22:05.900 RAM just to maintain the same perceived speed 00:22:05.900 --> 00:22:08.460 we had three years ago. But if the software is 00:22:08.460 --> 00:22:10.890 generated locally. Perfectly tailored to the 00:22:10.890 --> 00:22:13.069 constraints of the hardware's existing specifications, 00:22:13.509 --> 00:22:17.329 using unbloated, open implementations, that forced 00:22:17.329 --> 00:22:20.150 replacement cycle shatters. A machine that continues 00:22:20.150 --> 00:22:22.869 to run reliably becomes an enduring asset, not 00:22:22.869 --> 00:22:25.609 a depreciating liability. You might add a cheap 00:22:25.609 --> 00:22:28.609 neural processing unit via USB or something, 00:22:28.750 --> 00:22:30.490 but you don't throw away the motherboard and 00:22:30.490 --> 00:22:32.809 the chassis. Which brings us to the inevitable 00:22:32.809 --> 00:22:36.190 counterattack. Right. Because the multi -trillion 00:22:36.190 --> 00:22:38.650 dollar market caps of companies like Apple, Microsoft, 00:22:38.809 --> 00:22:41.309 and Adobe are predicated on you continuously 00:22:41.309 --> 00:22:43.470 buying hardware or paying monthly subscriptions. 00:22:43.789 --> 00:22:46.890 They are not going to passively watch as local 00:22:46.890 --> 00:22:49.230 open implementations render their products obsolete. 00:22:49.529 --> 00:22:52.990 The transition from proprietary binaries to transient 00:22:52.990 --> 00:22:55.269 computation is going to trigger the most brutal 00:22:55.269 --> 00:22:58.519 legal war in the history of technology. And the 00:22:58.519 --> 00:23:00.440 source document pinpoints the opening maneuver 00:23:00.440 --> 00:23:03.240 of this war. It specifically highlights a change 00:23:03.240 --> 00:23:05.720 in Adobe's general terms of use, citing Section 00:23:05.720 --> 00:23:09.859 17 going into effect in October 2025. Section 00:23:09.859 --> 00:23:14.000 17. Yes. Section 17 represents a desperate attempt 00:23:14.000 --> 00:23:17.099 to plug the dam. For decades, software licenses 00:23:17.099 --> 00:23:19.859 have explicitly banned decompilation and reverse 00:23:19.859 --> 00:23:22.539 engineering of the source code. But Section 17 00:23:22.539 --> 00:23:25.759 extends

the ban to the act of observation itself. 00:23:26.259 --> 00:23:28.160 Wait, observation. Exactly. I'm just looking 00:23:28.160 --> 00:23:31.619 at it. Exactly. It explicitly prohibits tracking, 00:23:31.819 --> 00:23:34.279 monitoring, or observing the behavior, inputs, 00:23:34.420 --> 00:23:36.880 or outputs of the running application in order 00:23:36.880 --> 00:23:39.559 to rebuild its functions or train a machine learning 00:23:39.559 --> 00:23:42.359 model. They saw the art craft scenario coming. 00:23:43.000 --> 00:23:45.380 They realized that if an AI can just watch the 00:23:45.380 --> 00:23:47.559 screen and figure out the math, the binary is 00:23:47.559 --> 00:23:50.299 indefensible. So they tried to make the act of 00:23:50.299 --> 00:23:52.920 watching illegal. And the corporate legal arsenal 00:23:52.920 --> 00:23:55.329 is immense. They will enforce these subscription 00:23:55.329 --> 00:23:57.670 contracts against any user or developer they 00:23:57.670 --> 00:24:01.009 suspect of utilizing observation agents. They 00:24:01.009 --> 00:24:03.809 will deploy thousands of unexpired utility patents 00:24:03.809 --> 00:24:06.529 covering specific image processing algorithms 00:24:06.529 --> 00:24:10.029 or user interface paradigms. They'll throw everything 00:24:10.029 --> 00:24:12.990 at it. They will leverage trade dress infringement, 00:24:13.009 --> 00:24:15.910 arguing that if an open source clone visually 00:24:15.910 --> 00:24:18.849 mimics the layout of their tools, it causes market 00:24:18.849 --> 00:24:22.799 confusion. But the critical question is... Will 00:24:22.799 --> 00:24:25.180 any of this actually stop the paradigm shift? 00:24:25.619 --> 00:24:28.119 The document argues forcefully that the corporate 00:24:28.119 --> 00:24:31.019 defense will fail. It roots this failure in a 00:24:31.019 --> 00:24:33.039 foundational principle of intellectual property 00:24:33.039 --> 00:24:36.299 law. Let's unpack the legal realities here. Why 00:24:36.299 --> 00:24:38.440 does a ban on observation ultimately collapse? 00:24:38.779 --> 00:24:40.759 Because it collides directly with the fundamental 00:24:40.759 --> 00:24:43.599 distinction in copyright law between expression 00:24:43.599 --> 00:24:47.640 and operation. Expression versus operation. Yes. 00:24:48.339 --> 00:24:50.740 Copyright law protects the specific expression 00:24:50.740 --> 00:24:53.980 of an idea, but it explicitly does not protect 00:24:53.980 --> 00:24:57.299 the idea itself, nor does it protect the operation, 00:24:57.579 --> 00:25:00.180 procedure, or system used to achieve a result. 00:25:00.700 --> 00:25:03.079 Let's use an analogy here to make that concrete, 00:25:03.240 --> 00:25:06.400 because legal theory can get dry. Consider a 00:25:06.400 --> 00:25:10.240 famous cookbook. A perfect example. A chef can 00:25:10.240 --> 00:25:12.539 copyright the specific way they wrote the story 00:25:12.539 --> 00:25:15.750 about their grandmother's pasta. They

can copyright 00:25:15.750 --> 00:25:18.349 the photography, the layout of the page, the 00:25:18.349 --> 00:25:20.509 specific adjectives used in the instructions. 00:25:20.730 --> 00:25:24.329 That is their expression. But they cannot copyright 00:25:24.329 --> 00:25:27.470 the operation of boiling water for 10 minutes 00:25:27.470 --> 00:25:30.910 to cook dried pasta. Exactly. If another chef 00:25:30.910 --> 00:25:33.609 reads the book, realizes that boiling water cooks 00:25:33.609 --> 00:25:36.170 pasta, and writes their own completely unique 00:25:36.170 --> 00:25:39.029 instructions detailing how to boil water, there 00:25:39.029 --> 00:25:41.710 is no copyright infringement. The first chef 00:25:41.710 --> 00:25:44.089 doesn't own the chemical reaction of heat and 00:25:44.089 --> 00:25:46.930 starch. So applying this to software, Adobe owns 00:25:46.930 --> 00:25:49.710 the copyright to the specific millions of lines 00:25:49.710 --> 00:25:52.890 of C++ code they wrote to build their photo editor. 00:25:53.069 --> 00:25:55.529 That is their expression. Right. But they do 00:25:55.529 --> 00:25:58.049 not own the mathematical concept of applying 00:25:58.049 --> 00:26:00.869 a Gaussian blur to a grid of pixels. They do 00:26:00.869 --> 00:26:03.349 not own the operation of, you know, making an 00:26:03.349 --> 00:26:06.269 image darker when a slider is moved left. And 00:26:06.269 --> 00:26:07.930 courts have historically defended this distinction 00:26:07.930 --> 00:26:10.950 rigorously, particularly regarding interoperability. 00:26:11.640 --> 00:26:14.740 There is a famous precedent here that the document's 00:26:14.740 --> 00:26:18.140 logic echoes perfectly. The story of Compaq cloning 00:26:18.140 --> 00:26:22.500 the IBM PC BIOS in 1982. Yes, the clean room 00:26:22.500 --> 00:26:26.400 design precedent. This is crucial history. IBM 00:26:26.400 --> 00:26:28.920 had a monopoly on the PC market because they 00:26:28.920 --> 00:26:31.819 owned the proprietary BIOS, the basic input output 00:26:31.819 --> 00:26:34.099 system that allowed the hardware to talk to the 00:26:34.099 --> 00:26:36.579 software. And if you wanted to build a computer 00:26:36.579 --> 00:26:39.299 that could run IBM software, you had to have 00:26:39.299 --> 00:26:41.380 their BIOS. But they wouldn't sell it to you. 00:26:41.640 --> 00:26:44.720 So Compact used a cleanroom approach. They set 00:26:44.720 --> 00:26:47.839 up two teams of engineers. Team A read the copyrighted 00:26:47.839 --> 00:26:50.420 IBM source code and documented exactly what it 00:26:50.420 --> 00:26:53.880 did, its inputs, its outputs, its timing. They 00:26:53.880 --> 00:26:56.400 wrote a purely functional specification manual. 00:26:56.720 --> 00:27:00.059 But Team A was legally contaminated. They had 00:27:00.059 --> 00:27:02.880 seen the copyrighted expression. Right. So Team 00:27:02.880 --> 00:27:06.960 B enters the picture. Exactly. Team B had never 00:27:06.960 --> 00:27:09.599

seen the IBM code. They were locked in a room 00:27:09.599 --> 00:27:11.700 with only the functional specification manual 00:27:11.700 --> 00:27:14.759 written by Team A. Team B wrote a brand new BIOS 00:27:14.759 --> 00:27:17.599 from scratch that performed the exact same operations. 00:27:17.980 --> 00:27:20.880 When IBM sued, the courts ruled in favor of Compaq. 00:27:21.160 --> 00:27:24.359 Team B only copied the function, the operation, 00:27:24.380 --> 00:27:27.460 which is not copyrightable. Fast forward to 2026. 00:27:27.819 --> 00:27:31.259 The AI agent acts as Team B. Precisely. The public 00:27:31.259 --> 00:27:33.420 documentation and the visual observation of the 00:27:33.420 --> 00:27:36.299 software act as the specification manual. The 00:27:36.299 --> 00:27:38.920 AI has never seen Adobe source code. It only 00:27:38.920 --> 00:27:41.279 observes the function and writes a completely 00:27:41.279 --> 00:27:43.759 new expression in Rust to achieve the same operation. 00:27:44.059 --> 00:27:47.380 The legal precedent strongly protects this. And 00:27:47.380 --> 00:27:50.279 regarding Adobe's new contractual ban on watching 00:27:50.279 --> 00:27:53.180 the software, the document points out the utter 00:27:53.180 --> 00:27:55.900 practical impossibility of enforcing it. It's 00:27:55.900 --> 00:27:58.349 totally unenforceable. I mean... How does a court 00:27:58.349 --> 00:28:00.970 distinguish between an AI agent rapidly observing 00:28:00.970 --> 00:28:04.190 a program's behavior and a human user rapidly 00:28:04.190 --> 00:28:06.190 clicking through a program while recording their 00:28:06.190 --> 00:28:09.390 screen for a YouTube tutorial? It cannot. The 00:28:09.390 --> 00:28:11.289 distinction collapses entirely when you consider 00:28:11.289 --> 00:28:14.369 that the AI can simply ingest the thousands of 00:28:14.369 --> 00:28:17.430 public tutorials, help manuals, and public API 00:28:17.430 --> 00:28:20.069 documentations that the vendor themselves published. 00:28:20.289 --> 00:28:22.569 If the operation can be described in ordinary 00:28:22.569 --> 00:28:25.289 human language, That description can be handed 00:28:25.289 --> 00:28:27.809 to a generative model that has never touched 00:28:27.809 --> 00:28:30.309 the proprietary software, and that model can 00:28:30.309 --> 00:28:32.829 write the code. It's unenforceable. It's like 00:28:32.829 --> 00:28:34.849 a restaurant banning you from remembering what 00:28:34.849 --> 00:28:36.710 the food tasted like so you don't try to cook 00:28:36.710 --> 00:28:39.869 it at home. That's a great way to put it. And 00:28:39.869 --> 00:28:42.529 this brings us to the Napster parallel. The document 00:28:42.529 --> 00:28:44.769 uses the music industry's crisis at the turn 00:28:44.769 --> 00:28:47.390 of the century to crystallize the software industry's 00:28:47.390 --> 00:28:50.130 impending doom. Let's really dig into the mechanics 00:28:50.130 --> 00:28:53.069 of that analogy. For the entire 20th century, 00:28:53.420 --> 00:28:55.359 The music industry's business model was tied 00:28:55.359 --> 00:28:58.099 to physical scarcity. They believed they were 00:28:58.099 --> 00:29:00.200 in the business of manufacturing and distributing 00:29:00.200 --> 00:29:05.839 plastic discs, vinyl, cassettes, CDs. The physical 00:29:05.839 --> 00:29:08.160 object was the scarce container for the value, 00:29:08.259 --> 00:29:10.660 which was the audio recording itself. Right. 00:29:11.019 --> 00:29:13.900 Before widespread broadband internet, if you 00:29:13.900 --> 00:29:16.079 wanted to listen to a specific album on demand, 00:29:16.420 --> 00:29:19.200 buying the physical CD was the only practical 00:29:19.200 --> 00:29:22.160 method. But then digital compression formats 00:29:22.160 --> 00:29:25.019 like MP3 and peer -to -peer networks like Napster 00:29:25.019 --> 00:29:27.599 emerged. Suddenly, the audio recording could 00:29:27.599 --> 00:29:30.539 be perfectly replicated and instantly transmitted 00:29:30.539 --> 00:29:33.259 without the need for the plastic disc at all. 00:29:33.440 --> 00:29:35.519 The value is completely decoupled from the physical 00:29:35.519 --> 00:29:37.960 container. And the music industry's reaction 00:29:37.960 --> 00:29:40.059 was identical to what the tech industry will 00:29:40.059 --> 00:29:43.259 do. Total legal panic. Look at the A &M Records 00:29:43.259 --> 00:29:45.940 v. Napster case. The record label sued the platform. 00:29:46.200 --> 00:29:48.980 They sued individual college students. They spent 00:29:48.980 --> 00:29:51.079 millions developing digital rights management, 00:29:51.160 --> 00:29:52.960 trying to create CDs that couldn't be ripped 00:29:52.960 --> 00:29:55.400 to a computer. And it was entirely futile. They 00:29:55.400 --> 00:29:57.619 successfully shut down Napster, but the peer 00:29:57.619 --> 00:29:59.700 -to -peer genie was out of the bottle. BitTorrent, 00:29:59.740 --> 00:30:03.220 LimeWire, Kazaa immediately replaced it. The 00:30:03.220 --> 00:30:05.200 fundamental technological reality had shifted. 00:30:05.339 --> 00:30:07.980 Data routing was vastly more efficient than physical 00:30:07.980 --> 00:30:11.539 distribution. And the author of the last program 00:30:11.539 --> 00:30:14.119 argues that the software industry is making the 00:30:14.119 --> 00:30:17.900 exact same category error today. Tech conglomerates 00:30:17.900 --> 00:30:20.200 believe that the compiled binary file is the 00:30:20.200 --> 00:30:22.920 indispensable container of value. They believe 00:30:22.920 --> 00:30:25.500 they are selling access to that specific heavy 00:30:25.500 --> 00:30:29.069 proprietary file. But the 2026 ArtCraft event 00:30:29.069 --> 00:30:32.390 proves that the operations, the actual mathematical 00:30:32.390 --> 00:30:35.549 transformations the user desires, can be perfectly 00:30:35.549 --> 00:30:37.910 extracted and transmitted without the proprietary 00:30:37.910

--> 00:30:41.829 binary container. Observation and AI generation 00:30:41.829 --> 00:30:45.009 are the MP3 compression of the software world. 00:30:45.369 --> 00:30:48.289 While the legal nuances differ, you know, music 00:30:48.289 --> 00:30:50.769 copyright protects the fixed audio performance, 00:30:50.910 --> 00:30:53.250 whereas software copyright protects the written 00:30:53.250 --> 00:30:57.369 code, the practical outcome is identical. Adobe 00:30:57.369 --> 00:30:59.309 or Microsoft might win a lawsuit against one 00:30:59.309 --> 00:31:02.490 specific AI developer. They might force a GitHub 00:31:02.490 --> 00:31:05.390 repository to be taken down using a DMCA notice. 00:31:05.750 --> 00:31:08.450 But they cannot reinstate the fundamental condition 00:31:08.450 --> 00:31:11.569 of scarcity. Because once the knowledge of how 00:31:11.569 --> 00:31:14.210 to efficiently sequence an operation is generated 00:31:14.210 --> 00:31:16.650 and shared as open source code, it propagates 00:31:16.650 --> 00:31:20.339 instantly. You cannot sue math. out of existence. 00:31:20.619 --> 00:31:23.039 You cannot sue math out of existence. I love 00:31:23.039 --> 00:31:25.819 that. And once that open source code is out in 00:31:25.819 --> 00:31:28.099 the wild, the document predicts an acceleration 00:31:28.099 --> 00:31:30.920 that traditional software vendors simply cannot 00:31:30.920 --> 00:31:34.519 match. Let's move into section six, the acceleration 00:31:34.519 --> 00:31:38.119 of open implementations. Historically, maintaining 00:31:38.119 --> 00:31:40.559 an open source project was bottlenecked by human 00:31:40.559 --> 00:31:43.519 labor. If you found a bug in an open source photo 00:31:43.519 --> 00:31:46.200 editor, you had to be a skilled C++ developer 00:31:46.200 --> 00:31:49.289 to understand the code base, write a fix, and 00:31:49.289 --> 00:31:51.730 submit a pull request. Then the maintainers had 00:31:51.730 --> 00:31:54.230 to manually review your code, test it, and approve 00:31:54.230 --> 00:31:56.950 it. It was a system filled with friction. But 00:31:56.950 --> 00:31:59.329 when AI agents enter the workflow, that friction 00:31:59.329 --> 00:32:02.730 drops to zero. The text describes a scenario 00:32:02.730 --> 00:32:05.849 where any user, regardless of their coding ability, 00:32:06.190 --> 00:32:09.210 can drive the evolution of the software. Like 00:32:09.210 --> 00:32:11.470 if an accountant using an open source data sorting 00:32:11.470 --> 00:32:13.809 tool notices a calculation error in a specific 00:32:13.809 --> 00:32:16.609 edge case, they don't need to write code. Not 00:32:16.609 --> 00:32:19.779 at all. They simply describe the error in ordinary 00:32:19.779 --> 00:32:22.900 language to their local AI agent. The tool is 00:32:22.900 --> 00:32:24.839 failing to account for leap years when calculating 00:32:24.839 --> 00:32:27.660 the quarterly amortization schedule. Exactly. 00:32:28.220 --> 00:32:31.079 And the local agent immediately analyzes the 00:32:31.079 --> 00:32:33.680 open source code base, identifies the logical 00:32:33.680 --> 00:32:36.440 flaw, rewrites the specific function in Rust, 00:32:36.700 --> 00:32:39.180 runs automated tests to ensure it didn't break 00:32:39.180 --> 00:32:41.740 anything else, and submits the patch to the main 00:32:41.740 --> 00:32:45.359 repository. The repository becomes an autopoietic 00:32:45.359 --> 00:32:49.970 system. It heals and evolves constantly. It completely 00:32:49.970 --> 00:32:52.130 destroys the traditional vendor release schedule. 00:32:52.309 --> 00:32:54.150 You aren't waiting for a massive corporation 00:32:54.150 --> 00:32:56.869 to prioritize your bug fix in their Q4 service 00:32:56.869 --> 00:33:00.130 pack. The community agent fixes it in three minutes. 00:33:00.410 --> 00:33:03.509 This extreme adaptability also leads to the proliferation 00:33:03.509 --> 00:33:06.809 of forks. Because the underlying operations are 00:33:06.809 --> 00:33:09.569 transparent and modular, communities can instantly 00:33:09.569 --> 00:33:12.309 branch the software to serve highly niche requirements. 00:33:13.150 --> 00:33:15.490 The document mentions legacy integration, which 00:33:15.490 --> 00:33:17.809 is a massive headache for the enterprise world. 00:33:18.210 --> 00:33:20.990 Currently, migrating a Fortune 500 company from 00:33:20.990 --> 00:33:23.509 an old proprietary database system to a new one 00:33:23.509 --> 00:33:26.690 is a multi -year, multi -million dollar nightmare 00:33:26.690 --> 00:33:30.289 fraught with data loss. But in a world of AI 00:33:30.289 --> 00:33:32.910 -generated open implementations, migration becomes 00:33:32.910 --> 00:33:36.890 a trivial mathematical proof. The AI agent can 00:33:36.890 --> 00:33:39.009 ingest the public file formats of the legacy 00:33:39.009 --> 00:33:42.380 software. It generates an open clone. And to 00:33:42.380 --> 00:33:44.779 prove it works, the enterprise tasks the AI with 00:33:44.779 --> 00:33:47.079 running 10 ,000 of their most complex legacy 00:33:47.079 --> 00:33:49.200 files through both the old system and the new 00:33:49.200 --> 00:33:51.500 clone, confirming byte -for -byte identical outputs. 00:33:51.799 --> 00:33:54.700 The risk is eliminated. If the outputs match 00:33:54.700 --> 00:33:57.079 perfectly, the enterprise drops the multi -million 00:33:57.079 --> 00:33:59.720 dollar licensing contract immediately and switches 00:33:59.720 --> 00:34:02.460 to the open implementation. They might keep a 00:34:02.460 --> 00:34:04.759 single license copy on an air -gapped machine 00:34:04.759 --> 00:34:07.180 for legal compliance, but the daily workflow 00:34:07.180 --> 00:34:09.619 shifts entirely. There will also be forks driven 00:34:09.619 --> 00:34:12.760 by security or hardware constraints. The Department 00:34:12.760 --> 00:34:15.079 of Defense will utilize agents to strip every 00:34:15.079 --> 00:34:17.699 line of telemetry or

networking code out of an 00:34:17.699 --> 00:34:20.119 open implementation, creating a mathematically 00:34:20.119 --> 00:34:23.920 provable air -gapped version. Or a hobbyist will 00:34:23.920 --> 00:34:26.079 instruct their agent to recompile the implementation 00:34:26.079 --> 00:34:29.079 to run exclusively on the architecture of a $10 00:34:29.079 --> 00:34:32.300 Raspberry Pi. The software bends perfectly to 00:34:32.300 --> 00:34:34.300 the user's constraints, rather than forcing the 00:34:34.300 --> 00:34:36.840 user to upgrade their hardware to meet the software's 00:34:36.840 --> 00:34:40.440 bloat. It is a total inversion of power, and 00:34:40.440 --> 00:34:43.260 this acceleration pushes us inevitably toward 00:34:43.260 --> 00:34:45.420 the final horizon described in the source document, 00:34:45.719 --> 00:34:49.800 the years 2032 to 2036. This is where we reach 00:34:49.800 --> 00:34:53.239 the concept that truly breaks the brain, the 00:34:53.239 --> 00:34:57.440 end of code as a human -facing artifact. Up until 00:34:57.440 --> 00:35:00.760 this point in the timeline, 2026 to 2030, the 00:35:00.760 --> 00:35:02.960 AI agents were still writing human -readable 00:35:02.960 --> 00:35:06.119 programming languages like Rust or Python. They 00:35:06.119 --> 00:35:08.400 were writing scripts that a developer could theoretically 00:35:08.400 --> 00:35:12.000 open in a text editor and read. But the document 00:35:12.000 --> 00:35:14.920 states that by 2032, this intermediate step is 00:35:14.920 --> 00:35:17.260 abandoned. Because why generate human -readable 00:35:17.260 --> 00:35:20.239 code if a human is never going to read it? Exactly. 00:35:20.519 --> 00:35:23.099 Translating pure mathematical operations into 00:35:23.099 --> 00:35:25.579 the syntax of a human programming language requires 00:35:25.579 --> 00:35:29.420 computational overhead. By 2032, the system stops 00:35:29.420 --> 00:35:32.750 emitting inspectable code. It transitions to 00:35:32.750 --> 00:35:34.909 maintaining internal, multidimensional representations 00:35:34.909 --> 00:35:38.530 of operations. So it understands data transformation 00:35:38.530 --> 00:35:41.610 natively. It understands sorting, filtering, 00:35:41.949 --> 00:35:45.329 matrix multiplication, and compositing, not as 00:35:45.329 --> 00:35:48.269 lines of Python, but as abstract mathematical 00:35:48.269 --> 00:35:51.510 concepts. When you supply the ordinary language 00:35:51.510 --> 00:35:54.070 intent and the constraints, the system directly 00:35:54.070 --> 00:35:56.349 applies the necessary mathematical transformations 00:35:56.349 --> 00:35:59.389 to the data structure. The distinction between 00:35:59.389 --> 00:36:01.630 the act of writing software... And the act of 00:36:01.630 --> 00:36:04.800 using software evaporates completely. Neither 00:36:04.800 --> 00:36:07.480 activity exists anymore. Which sets the stage 00:36:07.480 --> 00:36:11.099 for the vivid picture of the year 2036,

the post 00:36:11.099 --> 00:36:14.139 -app workspace. We touched on this in the introduction, 00:36:14.300 --> 00:36:16.980 but let's explore the granular reality of working 00:36:16.980 --> 00:36:19.400 in this environment. The physical devices we 00:36:19.400 --> 00:36:21.460 use will transition from being hosts for heavy 00:36:21.460 --> 00:36:24.119 computation to serving purely as interfaces for 00:36:24.119 --> 00:36:27.059 intent. The screen on your desk or the augmented 00:36:27.059 --> 00:36:29.659 reality glasses you wear will not display a grid 00:36:29.659 --> 00:36:32.989 of app icons. There is no start menu. Think about 00:36:32.989 --> 00:36:35.389 a highly specific, frustrating task we deal with 00:36:35.389 --> 00:36:37.710 today. You receive a massive data dump from a 00:36:37.710 --> 00:36:41.329 client in a poorly formatted CSV file. You need 00:36:41.329 --> 00:36:43.630 to convert it into a nested JSON structure to 00:36:43.630 --> 00:36:46.500 feed it into a web visualization. Currently, 00:36:46.579 --> 00:36:48.760 you have to find a sketchy website that offers 00:36:48.760 --> 00:36:52.559 CSV to JSON conversion, worry about data privacy, 00:36:52.840 --> 00:36:55.079 upload it, configure the delimiter settings, 00:36:55.320 --> 00:36:57.539 download the result, and hope it parsed correctly. 00:36:58.159 --> 00:37:01.260 In the 2036 workspace, you simply drag the raw 00:37:01.260 --> 00:37:03.760 data file onto your workspace surface and say 00:37:03.760 --> 00:37:06.059 to the ambient system, convert this data into 00:37:06.059 --> 00:37:08.800 a nested JSON format. Group the entries by the 00:37:08.800 --> 00:37:10.800 region column and strip out any rows where the 00:37:10.800 --> 00:37:13.280 revenue value is null. And the system simply 00:37:13.280 --> 00:37:15.440 performs the mathematical transformation instantly. 00:37:15.960 --> 00:37:18.039 There is no user interface for the conversion 00:37:18.039 --> 00:37:20.639 tool. There is no application to launch. The 00:37:20.639 --> 00:37:24.000 new file just appears. Or consider creative work. 00:37:24.340 --> 00:37:26.960 You have a raw video file of an interview. You 00:37:26.960 --> 00:37:29.170 state, Remove the background hum from the audio, 00:37:29.389 --> 00:37:31.250 color grade the subject to match the lighting 00:37:31.250 --> 00:37:33.590 tone of this cinematic reference frame, and render 00:37:33.590 --> 00:37:35.769 a cut that isolates the three moments the subject 00:37:35.769 --> 00:37:39.119 mentioned the word scarcity. The result is rendered 00:37:39.119 --> 00:37:42.230 and presented. You didn't open Premiere Pro. 00:37:42.369 --> 00:37:44.750 You didn't manually tweak a parametric equalizer. 00:37:44.809 --> 00:37:47.610 You provided the intent and the aesthetic judgment. 00:37:47.750 --> 00:37:50.420 The system provided the execution. The document 00:37:50.420 --> 00:37:53.099 uses a phenomenal historical analogy to frame 00:37:53.099 --> 00:37:56.119 how bizarre our current reliance on applications 00:37:56.119 --> 00:37:59.280 will look from the vantage point of 2036. It 00:37:59.280 --> 00:38:01.539 compares our use of drop -down menus to the use 00:38:01.539 --> 00:38:05.079 of punch cards in the 1960s. It is the perfect 00:38:05.079 --> 00:38:09.239 metaphor. In 1965, if you wanted a mainframe 00:38:09.239 --> 00:38:11.539 to calculate payroll, you couldn't just tell 00:38:11.539 --> 00:38:14.300 it to do so. You had to sit at a key punch machine 00:38:14.300 --> 00:38:17.039 and punch physical holes into cardboard cards 00:38:17.039 --> 00:38:20.119 representing machine code. perfectly sequenced 00:38:20.119 --> 00:38:22.900 without a single error. You were adapting your 00:38:22.900 --> 00:38:25.380 human intent to the rigid physical constraints 00:38:25.380 --> 00:38:28.159 of the machine's intake mechanism. And today, 00:38:28.320 --> 00:38:31.079 when we want to remove a blemish from a photograph, 00:38:31.280 --> 00:38:34.480 we launch a massive application. Navigate to 00:38:34.480 --> 00:38:37.059 the healing brush tool in a sidebar, adjust the 00:38:37.059 --> 00:38:39.659 pixel radius slider, and click the mouse on specific 00:38:39.659 --> 00:38:42.179 coordinates. We are still adapting our human 00:38:42.179 --> 00:38:44.719 intent to the rigid graphical constraints of 00:38:44.719 --> 00:38:47.559 the software's intake mechanism. Both punch cards 00:38:47.559 --> 00:38:50.219 and graphical user interfaces are merely intermediate 00:38:50.219 --> 00:38:53.380 artifacts. They are necessary compromises required 00:38:53.380 --> 00:38:55.559 by a system incapable of interpreting natural 00:38:55.559 --> 00:38:58.940 language. But once the system can effortlessly 00:38:58.940 --> 00:39:01.579 parse, route, and execute ordinary language, 00:39:01.800 --> 00:39:04.500 the intermediate artifact loses its reason to 00:39:04.500 --> 00:39:08.389 exist. The application vanishes. And this leads 00:39:08.389 --> 00:39:11.530 us to the absolute defining quote of the last 00:39:11.530 --> 00:39:14.610 program. It is the thesis statement that synthesizes 00:39:14.610 --> 00:39:16.969 this entire technological arc. It's a powerful 00:39:16.969 --> 00:39:20.309 one. The author writes, the binary was never 00:39:20.309 --> 00:39:23.349 the work. It was the tax the machine charged 00:39:23.349 --> 00:39:27.010 for refusing ordinary language. It is a devastatingly 00:39:27.010 --> 00:39:29.500 accurate assessment. For decades, we believed 00:39:29.500 --> 00:39:32.260 that buying, installing, learning, and maintaining 00:39:32.260 --> 00:39:34.860 software applications was an inherent part of 00:39:34.860 --> 00:39:38.000 digital labor. We spent billions of hours and 00:39:38.000 --> 00:39:40.539 trillions of dollars servicing the needs of the 00:39:40.539 --> 00:39:43.219 binary. We paid the tax. We paid it with our 00:39:43.219 --> 00:39:45.960 time, our cognitive load, and our

subscription 00:39:45.960 --> 00:39:50.320 fees. So if the application is just a tax, and 00:39:50.320 --> 00:39:53.780 the tax is abolished by AI observation and transient 00:39:53.780 --> 00:39:56.840 computation, Where does all that human energy 00:39:56.840 --> 00:39:59.179 and attention go? Well, the attention shifts 00:39:59.179 --> 00:40:01.940 entirely back to the refinement of intent and 00:40:01.940 --> 00:40:04.739 the exercise of judgment. The text explains that 00:40:04.739 --> 00:40:06.559 the professional of the future is not someone 00:40:06.559 --> 00:40:09.219 who has memorized the obscure submenus of a design 00:40:09.219 --> 00:40:12.239 program or the syntax of a database query language. 00:40:12.400 --> 00:40:14.480 Right. The professional of the future is the 00:40:14.480 --> 00:40:16.699 person who can most precisely articulate the 00:40:16.699 --> 00:40:19.500 desired outcome and most rigorously judge the 00:40:19.500 --> 00:40:22.510 final result. The skill is no longer operating 00:40:22.510 --> 00:40:24.909 the instrument. The instrument computation becomes 00:40:24.909 --> 00:40:27.250 invisible and ubiquitous like basic arithmetic. 00:40:27.510 --> 00:40:30.190 You don't marvel at a calculator's ability to 00:40:30.190 --> 00:40:32.449 divide numbers. You simply use the result to 00:40:32.449 --> 00:40:34.969 balance your budget. Exactly. The human work 00:40:34.969 --> 00:40:38.130 remains. A financial auditor still needs to identify 00:40:38.130 --> 00:40:41.630 fraud. An editor still needs to pace a film to 00:40:41.630 --> 00:40:44.469 evoke tension. An architect still needs to design 00:40:44.469 --> 00:40:47.079 a structure that inspires awe. The fundamental 00:40:47.079 --> 00:40:50.059 need for a result in the physical or cultural 00:40:50.059 --> 00:40:53.860 world hasn't changed. What has changed is that 00:40:53.860 --> 00:40:56.480 the rigid, proprietary mediator sitting between 00:40:56.480 --> 00:40:58.719 the human mind and the computational processor 00:40:58.719 --> 00:41:02.639 has been removed. The judgments were always the 00:41:02.639 --> 00:41:05.789 work. That is an incredibly empowering thought 00:41:05.789 --> 00:41:08.690 to anchor on. Let's zoom out and recap the immense 00:41:08.690 --> 00:41:11.190 territory we've covered in this deep dive. We 00:41:11.190 --> 00:41:14.409 began with a catalyst in October 2026. Brandon 00:41:14.409 --> 00:41:17.289 Thomas and R -Craft demonstrating that AI agents 00:41:17.289 --> 00:41:19.610 could perfectly reverse engineer complex software 00:41:19.610 --> 00:41:23.210 capabilities simply by observing inputs and outputs, 00:41:23.469 --> 00:41:25.989 bypassing the proprietary source code entirely. 00:41:26.309 --> 00:41:28.909 We analyzed how that breakthrough shatters the 00:41:28.909 --> 00:41:31.730 historical scarcity of the compiled binary, shifting 00:41:31.730 --> 00:41:34.389 the remaining value entirely to human judgment 00:41:34.389 --> 00:41:37.269 and intent. We explored the mechanics of transient 00:41:37.269 --> 00:41:40.750 computation. a near future where software does 00:41:40.750 --> 00:41:43.590 not exist as permanent installations, but as 00:41:43.590 --> 00:41:46.309 temporary on -demand chains of mathematical operations 00:41:46.309 --> 00:41:49.849 generated to fulfill a specific request and immediately 00:41:49.849 --> 00:41:52.809 purged. We debated the profound hardware implications, 00:41:53.230 --> 00:41:57.230 the realization that highly quantized, localized 00:41:57.230 --> 00:41:59.809 AI models can run efficiently on older machines, 00:42:00.010 --> 00:42:02.309 effectively breaking the forced upgrade cycle 00:42:02.309 --> 00:42:05.030 and transforming aging hardware from a liability 00:42:05.030 --> 00:42:07.539 into a durable asset. We examine the inevitable 00:42:07.539 --> 00:42:10.139 legal warfare, the desperate attempts by tech 00:42:10.139 --> 00:42:12.539 conglomerates to ban the act of observation, 00:42:12.960 --> 00:42:15.639 and why those bans will collapse against the 00:42:15.639 --> 00:42:18.159 foundational copyright distinction between protected 00:42:18.159 --> 00:42:21.579 expression and unprotected operations, echoing 00:42:21.579 --> 00:42:23.800 the destruction of the music industry's physical 00:42:23.800 --> 00:42:26.480 monopoly during the Napster era. We trace the 00:42:26.480 --> 00:42:28.760 hyper -acceleration of open implementations, 00:42:28.760 --> 00:42:32.239 where AI -driven continuous integration bypasses 00:42:32.239 --> 00:42:34.679 traditional vendor release schedules. allowing 00:42:34.679 --> 00:42:36.860 massive enterprises to frictionlessly migrate 00:42:36.860 --> 00:42:39.579 away from legacy systems. And finally, we arrived 00:42:39.579 --> 00:42:43.500 at the post -app workspace of 2036, a reality 00:42:43.500 --> 00:42:45.679 where the concept of code is abstracted away 00:42:45.679 --> 00:42:48.639 entirely, and the human user interacts directly 00:42:48.639 --> 00:42:51.840 with computation through ordinary language, rendering 00:42:51.840 --> 00:42:54.440 the traditional application as obsolete as the 00:42:54.440 --> 00:42:57.159 punch card. It is a rigorous, deeply researched, 00:42:57.340 --> 00:42:59.579 and frankly thrilling vision of what is to come. 00:43:00.219 --> 00:43:02.199 And it leaves me with a final provocative thought 00:43:02.199 --> 00:43:04.320 that extends beyond the immediate scope of the 00:43:04.320 --> 00:43:07.360 document. Oh, what's that? Well, the author brilliantly 00:43:07.360 --> 00:43:10.599 categorized software binaries as a tax we paid 00:43:10.599 --> 00:43:12.940 because machines couldn't understand our language. 00:43:13.239 --> 00:43:16.039 If that framework holds true as we enter the 00:43:16.039 --> 00:43:19.000 era of ordinary language computation, what other 00:43:19.000 --> 00:43:21.849 massive... billion dollar structures in our society, 00:43:21.969 -->

00:43:24.710 in our bureaucracies, in our daily lives, are 00:43:24.710 --> 00:43:27.570 actually just temporary translation taxes waiting 00:43:27.570 --> 00:43:29.909 to be identified and abolished by observation. 00:43:30.349 --> 00:43:33.849 Wow. That is the exact critical lens we must 00:43:33.849 --> 00:43:36.590 apply as we evaluate every digital service we 00:43:36.590 --> 00:43:39.809 currently believe is indispensable. It absolutely 00:43:39.809 --> 00:43:42.309 is. I want to thank you for joining us on this 00:43:42.309 --> 00:43:45.030 incredible journey today. Before we close, a 00:43:45.030 --> 00:43:47.349 final reminder, you are listening to the Read 00:43:47.349 --> 00:43:50.250 Multiplex podcast. The ability to dissect these 00:43:50.250 --> 00:43:53.389 pivotal shifts, to look 10 years into the future 00:43:53.389 --> 00:43:55.730 with this level of clarity, is made possible 00:43:55.730 --> 00:43:59.409 by the pioneering work of Brian Romilly. Do not 00:43:59.409 --> 00:44:01.650 let this kind of analysis disappear behind a 00:44:01.650 --> 00:44:04.329 corporate wall. Go to his platform, become a 00:44:04.329 --> 00:44:06.829 member, or simply buy him a coffee. Your support 00:44:06.829 --> 00:44:09.130 ensures that these vital, paradigm -shifting 00:44:09.130 --> 00:44:12.630 deep dives continue to exist. The future is arriving 00:44:12.630 --> 00:44:15.840 at an accelerating pace. Understanding the mechanics 00:44:15.840 --> 00:44:18.719 of that arrival is the only reliable way to navigate 00:44:18.719 --> 00:44:21.480 it. We started this conversation talking about 00:44:21.480 --> 00:44:23.760 how much we love the permanence of our digital 00:44:23.760 --> 00:44:26.739 tools. We like knowing exactly where our apps 00:44:26.739 --> 00:44:29.400 are installed. But perhaps the greatest technological 00:44:29.400 --> 00:44:32.179 leap of our lifetime won't be a shiny new application. 00:44:32.800 --> 00:44:35.199 Perhaps the greatest leap will be the liberating 00:44:35.199 --> 00:44:37.360 realization that we never needed the application 00:44:37.360 --> 00:44:40.079 at all. Thanks for joining us on the Deep Dive. 00:44:40.139 --> 00:44:44.250 We'll catch you next time. Welcome to the multiplex, 00:44:44.329 --> 00:44:48.030 where the future comes alive. Past, present, 00:44:48.329 --> 00:44:53.389 future tech, open up your eyes. Brian Romella 00:44:53.389 --> 00:44:57.920 leads the way with research deep and true. Technology 00:44:57.920 --> 00:45:02.780 and history, he connects them all for you. Brian 00:45:02.780 --> 00:45:04.800 Romella leads the way with research deep and 00:45:04.800 --> 00:45:07.059 true. Technology and history, he connects them 00:45:07.059 --> 00:45:09.360 all for you. Ancient sparks to modern code, he 00:45:09.360 --> 00:45:43.940 traces every threat, go exploration. But humanity 00:45:43.940 --> 00:45:50.599 moves on. Love equation for AI. Zero human companies 00:45:50.599 --> 00:45:55.840 building from the start. Thermodynamic wages 00:45:55.840 --> 00:45:59.940 earning the key. Secrets we will read. Experiment 00:45:59.940 --> 00:46:03.920 in publishing real researches and spending every...