WEBVTT 00:00:00.000 --> 00:00:01.919 If you are constantly hunting for the perfect 00:00:01.919 --> 00:00:05.860 god prompt to fix your AI outputs, you are looking 00:00:05.860 --> 00:00:10.179 in the wrong place. Beat. The real bottleneck 00:00:10.179 --> 00:00:13.259 in your workflow is you. Yeah, it really is a 00:00:13.259 --> 00:00:15.400 soul -crushing cycle. You know, just copy -pasting 00:00:15.400 --> 00:00:18.000 text into a little box all day. It gets exhausting 00:00:18.000 --> 00:00:20.539 very quickly. Welcome to the Deep Dive. Today 00:00:20.539 --> 00:00:23.920 we are exploring a March 2026 framework by Max 00:00:23.920 --> 00:00:27.039 Ann. We are looking at how to externalize your 00:00:27.039 --> 00:00:29.579 memory. Right. And we are moving way beyond that 00:00:29.579 --> 00:00:31.879 standard chat window. It requires fundamentally 00:00:31.879 --> 00:00:34.280 changing your approach to the work itself. We're 00:00:34.280 --> 00:00:36.140 going to show you how to bring AI directly to 00:00:36.140 --> 00:00:38.420 your files. You will build a system that works 00:00:38.420 --> 00:00:40.759 without you constantly holding its hand. Which 00:00:40.759 --> 00:00:43.140 is the dream, right? But let us unpack an uncomfortable 00:00:43.140 --> 00:00:46.380 truth first. Beat. We carry so much context inside 00:00:46.380 --> 00:00:49.500 our own heads, the AI simply cannot see any of 00:00:49.500 --> 00:00:52.609 it. Exactly. And when the AI fails... We immediately 00:00:52.609 --> 00:00:54.409 blame the model. We treat it like a software 00:00:54.409 --> 00:00:57.250 bug. But blaming the AI there is completely missing 00:00:57.250 --> 00:00:59.250 the point. It really is. I mean, it is like blaming 00:00:59.250 --> 00:01:00.810 a hammer when you were trying to build a house 00:01:00.810 --> 00:01:03.469 one nail at a time. The tool is perfectly fine. 00:01:03.570 --> 00:01:05.810 The method itself is broken. I have to make a 00:01:05.810 --> 00:01:08.950 vulnerable admission here. I still wrestle with 00:01:08.950 --> 00:01:11.290 hoarding context in my head. I get frustrated 00:01:11.290 --> 00:01:13.709 when the AI misses the mark. Well, every single 00:01:13.709 --> 00:01:16.329 person does it initially. It feels safer to control 00:01:16.329 --> 00:01:19.810 the information directly. It is a very hard habit 00:01:19.810 --> 00:01:22.879 to break. It is. But Max -Anne breaks this down 00:01:22.879 --> 00:01:26.319 into three phases of AI maturity. This framework 00:01:26.319 --> 00:01:29.500 explains exactly why we get stuck. So phase one 00:01:29.500 --> 00:01:31.719 is where most people start, right? Right. Phase 00:01:31.719 --> 00:01:33.659 one is the adopt phase. This is your standard 00:01:33.659 --> 00:01:36.120 manual prompting. You learn the basic tools and 00:01:36.120 --> 00:01:39.620 experiment a bit. You apply the AI to small, 00:01:39.640 --> 00:01:43.060 isolated tasks, like drafting a quick email. 00:01:43.260 --> 00:01:46.000 Yeah, or summarizing a short article. But the 00:01:46.000 --> 00:01:48.159 impact there is strictly limited. Most of the 00:01:48.159 --> 00:01:50.359 heavy lifting remains entirely manual. You are 00:01:50.359 --> 00:01:52.319 still driving every single interaction. Exactly. 00:01:52.480 --> 00:01:54.879 We treat the AI exactly like a search engine. 00:01:55.000 --> 00:01:57.579 But phase two is the adapt phase. And this is 00:01:57.579 --> 00:01:59.920 where the real prize is hidden. The adapt phase. 00:02:00.239 --> 00:02:02.719 How does that differ from just adopting the tool? 00:02:03.000 --> 00:02:05.560 Instead of simply using the AI, you redesign 00:02:05.560 --> 00:02:08.240 how the work gets done. You intentionally design 00:02:08.240 --> 00:02:10.840 workflows for the AI to read. Oh, I see. So you 00:02:10.840 --> 00:02:13.710 change your own behavior first. And that paves 00:02:13.710 --> 00:02:16.509 the way for phase three, which is automate. That 00:02:16.509 --> 00:02:19.389 is where entire workflows run autonomously in 00:02:19.389 --> 00:02:21.849 the background. Leaving humans to just focus 00:02:21.849 --> 00:02:25.189 on high level direction. But phase two feels 00:02:25.189 --> 00:02:27.930 like the critical leap there. It absolutely is. 00:02:28.090 --> 00:02:30.590 The framework provides a brilliant example of 00:02:30.590 --> 00:02:33.310 this. Think about an executive who records every 00:02:33.310 --> 00:02:36.349 single meeting. Okay. They know their AI assistant 00:02:36.349 --> 00:02:40.009 will transcribe the audio, so they actively change 00:02:40.009 --> 00:02:42.189 their behavior in the room. They alter how they 00:02:42.189 --> 00:02:44.530 actually speak during the meeting. Precisely. 00:02:44.530 --> 00:02:47.030 During the meeting, they repeat key decisions 00:02:47.030 --> 00:02:49.610 out loud. They speak very clearly before moving 00:02:49.610 --> 00:02:51.650 to the next topic. So they aren't doing it for 00:02:51.650 --> 00:02:53.229 the people in the room. They are doing it strictly 00:02:53.229 --> 00:02:55.810 for the transcript itself. Right, because a clearer 00:02:55.810 --> 00:02:59.569 transcript naturally leads to far better AI summaries. 00:02:59.610 --> 00:03:02.770 The raw data being fed to the AI is upgraded. 00:03:02.949 --> 00:03:05.199 That is fa - Fascinating. Are there other practical 00:03:05.199 --> 00:03:07.159 examples of that? Yeah, there is another great 00:03:07.159 --> 00:03:10.060 one in the text. Someone stops saving their project 00:03:10.060 --> 00:03:12.960 files as PDFs. They switch to saving them as 00:03:12.960 --> 00:03:15.789 CSV or markdown files instead. Let us break down 00:03:15.789 --> 00:03:18.469 why that actually matters. A multi -column PDF 00:03:18.469 --> 00:03:22.229 with images is a nightmare for an AI. Oh, it 00:03:22.229 --> 00:03:24.830 is terrible. The text extraction gets completely 00:03:24.830 --> 00:03:28.090 jumbled and confused. It loses the logical reading 00:03:28.090 --> 00:03:31.069 order entirely. But Markdown structures data 00:03:31.069 --> 00:03:34.030 perfectly for language models. Exactly. It uses 00:03:34.030 --> 00:03:36.770 clear headers and simple text formatting. These 00:03:36.770 --> 00:03:39.189 are incredibly small changes to your daily routine, 00:03:39.409 --> 00:03:42.050 but they compound massively over time. Every 00:03:42.050 --> 00:03:44.830 interaction becomes clearer and far more productive 00:03:44.830 --> 00:03:47.710 yeah but moving from phase one to phase two seems 00:03:47.710 --> 00:03:50.050 overwhelming it can feel that way yeah how does 00:03:50.050 --> 00:03:52.490 someone force themselves to make this shift when 00:03:52.490 --> 00:03:55.289 they are already busy it really requires adjusting 00:03:55.289 --> 00:03:58.310 very small daily habits you do not overhaul your 00:03:58.310 --> 00:04:00.650 entire workflow overnight you just make sure 00:04:00.650 --> 00:04:03.090 that ai can participate from the very start so 00:04:03.090 --> 00:04:05.830 we adapt our habits so the ai can actually read 00:04:05.830 --> 00:04:08.270 our work that is the core philosophy and it leads 00:04:08.270 --> 00:04:11.069 us to the very first major shift in this entire 00:04:11.069 --> 00:04:13.840 framework Since we established we need to adapt 00:04:13.840 --> 00:04:16.199 our habits, we have to look at how we store context. 00:04:16.519 --> 00:04:20.100 Right. The first habit to change is how we hold 00:04:20.100 --> 00:04:23.399 on to information. We must externalize our memory 00:04:23.399 --> 00:04:25.459 completely. We have to stop repeating ourselves 00:04:25.459 --> 00:04:29.220 every single day. Two seconds, silence. Think 00:04:29.220 --> 00:04:31.579 about your brain right now. It is absolutely 00:04:31.579 --> 00:04:34.660 overflowing with hidden context. You know, a 00:04:34.660 --> 00:04:37.939 specific client. absolutely hates bullet points. 00:04:38.120 --> 00:04:40.980 Yeah. Or, you know, your company pricing changed 00:04:40.980 --> 00:04:43.459 three months ago. You know your manager wants 00:04:43.459 --> 00:04:46.240 reports kept strictly under one page. But the 00:04:46.240 --> 00:04:48.540 AI does not know any of that. It lives entirely 00:04:48.540 --> 00:04:51.399 isolated inside your head. So you start every 00:04:51.399 --> 00:04:54.139 single conversation from scratch. You re -explain 00:04:54.139 --> 00:04:56.899 the exact same preferences over and over. Or 00:04:56.899 --> 00:04:59.370 you forget. and the AI misses the mark entirely, 00:04:59.689 --> 00:05:02.350 you are acting as the memory drive for the machine. 00:05:02.569 --> 00:05:04.589 Which makes you the ultimate bottleneck in the 00:05:04.589 --> 00:05:06.970 process. Well, how do we actually fix this? The 00:05:06.970 --> 00:05:09.529 framework introduces three distinct layers of 00:05:09.529 --> 00:05:12.170 memory externalization. Layer one is your system 00:05:12.170 --> 00:05:13.990 instruction. This is your static memory, right? 00:05:14.069 --> 00:05:16.089 Let us define this clearly for everyone. Sure. 00:05:16.230 --> 00:05:18.310 Permanent background rules that guide how the 00:05:18.310 --> 00:05:20.790 AI should always behave. It helps to think of 00:05:20.790 --> 00:05:23.910 system instructions as a foundation. It is not 00:05:23.910 --> 00:05:26.250 just a prompt you type out. No, it covers your 00:05:26.250 --> 00:05:29.529 highly stable, unchanging preferences. Your specific 00:05:29.529 --> 00:05:32.470 brand voice or strict formatting rules go here. 00:05:32.649 --> 00:05:35.269 You write it down once and you never repeat it 00:05:35.269 --> 00:05:37.589 again. This completely eliminates what we call 00:05:37.589 --> 00:05:40.120 prompt drift. Right. Prompt shift is when your 00:05:40.120 --> 00:05:43.259 outputs slowly degrade over multiple interactions. 00:05:43.779 --> 00:05:46.720 Because the system instructions anchor the AI's 00:05:46.720 --> 00:05:49.120 behavior permanently. Exactly. And then we move 00:05:49.120 --> 00:05:51.300 to layer two. Layer two is the knowledge base. 00:05:51.540 --> 00:05:54.259 This is also a form of static memory. It is a 00:05:54.259 --> 00:05:56.819 dedicated document library the AI can reference 00:05:56.819 --> 00:05:59.100 anytime. So this would hold your standard contract 00:05:59.100 --> 00:06:02.990 templates. Or maybe your internal company style 00:06:02.990 --> 00:06:05.850 guides. Right. When the AI writes a new project 00:06:05.850 --> 00:06:08.829 proposal, it pulls from that exact template automatically. 00:06:09.189 --> 00:06:11.709 You no longer have to manually attach the same 00:06:11.709 --> 00:06:14.250 five reference files every single time. But layers 00:06:14.250 --> 00:06:16.850 one and two still have a major limitation. They 00:06:16.850 --> 00:06:19.290 are entirely static. They only change if you 00:06:19.290 --> 00:06:21.529 physically go in and manually update them yourself. 00:06:21.889 --> 00:06:24.810 Which brings us to the real breakthrough. Layer 00:06:24.810 --> 00:06:27.629 3 is where the entire system comes alive. Upbeat 00:06:27.629 --> 00:06:30.529 is the dynamic memory layer. Yes. Layer 3 is 00:06:30.529 --> 00:06:33.269 memory files. Let us define this one clearly, 00:06:33.410 --> 00:06:35.949 too. Text files, the AI updates itself to remember 00:06:35.949 --> 00:06:39.449 facts across sessions. Exactly. Tools like Cloud 00:06:39.449 --> 00:06:43.230 Code or OpenAI Codex use this specific layer. 00:06:43.569 --> 00:06:46.790 They read and write to files like claudy .md 00:06:46.790 --> 00:06:49.810 or agentsets .md. This is absolutely fascinating 00:06:49.810 --> 00:06:52.410 to me. We are leaving the browser behind. entirely 00:06:52.410 --> 00:06:54.930 totally these files leave locally right on your 00:06:54.930 --> 00:06:57.430 own machine they sit strictly inside your designated 00:06:57.430 --> 00:06:59.990 project folders and they update automatically 00:06:59.990 --> 00:07:02.649 in the background as the ai works mechanically 00:07:02.649 --> 00:07:05.290 how is this actually happening well the ai operates 00:07:05.290 --> 00:07:09.910 via the command line or an ide An IDE is an integrated 00:07:09.910 --> 00:07:12.389 development environment. So it is granted directory 00:07:12.389 --> 00:07:14.870 access to your local files. Right. And over time, 00:07:15.009 --> 00:07:17.689 the AI actively learns your nuanced preferences. 00:07:18.110 --> 00:07:20.610 It logs details about your ongoing projects. 00:07:20.850 --> 00:07:22.790 It maps out your unique working style without 00:07:22.790 --> 00:07:25.050 you typing a single thing. You stop being the 00:07:25.050 --> 00:07:27.290 memory bottleneck entirely. But there is a massive 00:07:27.290 --> 00:07:29.430 caveat we have to mention here. Standard chat 00:07:29.430 --> 00:07:32.629 apps reset their memory every single session. 00:07:33.029 --> 00:07:36.750 Ah, right. Regular chat GPT browser window will 00:07:36.750 --> 00:07:38.870 not do this. Gemini in your web browser will 00:07:38.870 --> 00:07:41.069 not do this either. No, they will not. You absolutely 00:07:41.069 --> 00:07:44.449 need specific tools for layer three to actually 00:07:44.449 --> 00:07:47.069 function. You need a system granted permission 00:07:47.069 --> 00:07:50.209 to read and write files locally. A simple web 00:07:50.209 --> 00:07:52.569 chat window is fundamentally incapable of doing 00:07:52.569 --> 00:07:55.649 this. But if the AI is autonomously writing its 00:07:55.649 --> 00:07:58.769 own memory files, I have a probing question here. 00:07:58.930 --> 00:08:01.430 Go for it. What happens if it learns a bad habit 00:08:01.430 --> 00:08:04.089 or completely misunderstands a core preference? 00:08:04.370 --> 00:08:06.670 That is the beauty of keeping these files locally 00:08:06.670 --> 00:08:10.089 on your machine. You retain absolute full control 00:08:10.089 --> 00:08:12.290 over the entire system. So you can easily open 00:08:12.290 --> 00:08:15.050 those markdown files in any simple text editor. 00:08:15.709 --> 00:08:18.310 It is completely transparent. There is no hidden 00:08:18.310 --> 00:08:21.149 black box of secret training data. Just open 00:08:21.149 --> 00:08:23.370 the text file, delete the bad rule, and the AI 00:08:23.370 --> 00:08:25.819 forgets it. It really is that simple. You see 00:08:25.819 --> 00:08:27.899 exactly what the AI believes to be true about 00:08:27.899 --> 00:08:30.139 your workflow. We are going to keep building 00:08:30.139 --> 00:08:33.759 on this framework. Once your AI has its own memory, 00:08:33.980 --> 00:08:35.940 we have to look at what it actually works on. 00:08:36.080 --> 00:08:38.240 We need to stop acting as a human file router. 00:08:38.539 --> 00:08:40.860 But first, a quick word from the folks who help 00:08:40.860 --> 00:08:44.860 keep this deep dive running. Sponsor. Okay, let 00:08:44.860 --> 00:08:47.960 us jump right back into Max Anne's framework. 00:08:48.379 --> 00:08:51.159 We just established how to externalize your memory. 00:08:51.669 --> 00:08:54.750 The AI finally remembers exactly how you like 00:08:54.750 --> 00:08:57.190 to work. But knowing how you work is only half 00:08:57.190 --> 00:08:59.690 the battle. The next big bottleneck is routing 00:08:59.690 --> 00:09:02.769 the actual work itself. Right. If you are still 00:09:02.769 --> 00:09:05.990 dragging and dropping 50 files into a browser 00:09:05.990 --> 00:09:08.289 chat box, you have a problem. You are functioning 00:09:08.289 --> 00:09:11.080 as a manual file router. That approach simply 00:09:11.080 --> 00:09:13.299 does not scale up. It creates massive friction. 00:09:13.440 --> 00:09:16.179 You inevitably hit a file upload limit. Then 00:09:16.179 --> 00:09:17.980 you have to strategically choose what context 00:09:17.980 --> 00:09:20.379 to remove. You lose vital information between 00:09:20.379 --> 00:09:22.460 different chat batches. You just cross your fingers 00:09:22.460 --> 00:09:24.360 and hope the outputs remain consistent. It is 00:09:24.360 --> 00:09:27.059 super frustrating. I love the vivid analogy the 00:09:27.059 --> 00:09:29.360 framework uses for this. It is like hiring a 00:09:29.360 --> 00:09:31.840 brilliant world -class assistant, but forcing 00:09:31.840 --> 00:09:34.690 them to read documents. through a tiny mail slot 00:09:34.690 --> 00:09:37.029 in the door. Yeah, you would not lock an MIT 00:09:37.029 --> 00:09:39.250 graduate in a hallway and slide pages under the 00:09:39.250 --> 00:09:41.590 door one by one. But that is exactly what we 00:09:41.590 --> 00:09:44.169 do with standard web interfaces. We need to let 00:09:44.169 --> 00:09:47.409 the AI fully into the filing room. We have to 00:09:47.409 --> 00:09:50.169 bring the AI directly to the files themselves. 00:09:50.610 --> 00:09:53.610 You use folder -based project tools for this 00:09:53.610 --> 00:09:56.970 shift. Tools like Cloud Code or Codex are built 00:09:56.970 --> 00:10:00.470 specifically for this exact purpose. You establish 00:10:00.470 --> 00:10:02.789 a simple project folder right on your desktop. 00:10:03.179 --> 00:10:05.659 The AI is granted permission to read everything 00:10:05.659 --> 00:10:08.139 inside that specific holder. It accesses the 00:10:08.139 --> 00:10:10.360 data directly, bypassing the clipboard entirely. 00:10:10.659 --> 00:10:12.700 You aren't copying or pasting a single line of 00:10:12.700 --> 00:10:15.519 text. Let us look at a concrete one -off use 00:10:15.519 --> 00:10:17.820 case to see how this feels. Imagine you create 00:10:17.820 --> 00:10:20.960 a brand new simple folder. Okay. You drop in 00:10:20.960 --> 00:10:23.519 50 raw transcripts from past client meetings. 00:10:23.720 --> 00:10:26.620 You ask your AI tool to analyze all of them at 00:10:26.620 --> 00:10:29.639 once. Whoa. Imagine dropping 50 meeting transcripts 00:10:29.639 --> 00:10:32.240 into a folder and the AI maps the entire project 00:10:32.240 --> 00:10:35.100 instantly. It extracts the big wins, the losses, 00:10:35.220 --> 00:10:37.960 and the hidden upsell opportunities all in one 00:10:37.960 --> 00:10:41.289 go. It produces a comprehensive summary without 00:10:41.289 --> 00:10:44.169 you managing a single batch of uploads. It just 00:10:44.169 --> 00:10:46.990 reads the directory, processes the text, and 00:10:46.990 --> 00:10:49.490 outputs the result. That is incredibly powerful 00:10:49.490 --> 00:10:52.529 for a one -off task. But the framework details 00:10:52.529 --> 00:10:55.389 ongoing use cases that are even more impressive. 00:10:55.669 --> 00:10:58.429 Like what? Think about keeping a dedicated folder 00:10:58.429 --> 00:11:01.909 for one very important client. After every single 00:11:01.909 --> 00:11:04.840 weekly meeting, You just drop the new audio transcript 00:11:04.840 --> 00:11:07.759 directly into that folder. Oh, and the AI spots 00:11:07.759 --> 00:11:10.059 the new file and reviews it automatically. Exactly. 00:11:10.279 --> 00:11:13.340 It updates the ongoing executive summaries. It 00:11:13.340 --> 00:11:15.860 identifies new strategic insights based on the 00:11:15.860 --> 00:11:18.159 historical context. It tracks complex behavioral 00:11:18.159 --> 00:11:20.559 patterns across dozens of conversations over 00:11:20.559 --> 00:11:22.960 a very long period of time. You just add one 00:11:22.960 --> 00:11:25.039 single file and the system handles the rest. 00:11:25.279 --> 00:11:28.399 But a folder is just a digital box. What actually 00:11:28.399 --> 00:11:31.120 makes this automation work without constant supervision? 00:11:31.659 --> 00:11:34.159 The secret engine is the instructions file. It 00:11:34.159 --> 00:11:36.759 sits quietly inside every single project folder 00:11:36.759 --> 00:11:40.259 you create. This is that clod .md or agents .md 00:11:40.259 --> 00:11:42.379 file we mentioned a moment ago. Right. The AI 00:11:42.379 --> 00:11:45.279 is programmed to read this file first every single 00:11:45.279 --> 00:11:47.580 time it boots up. A well -structured instructions 00:11:47.580 --> 00:11:51.279 file requires three very specific core elements, 00:11:51.419 --> 00:11:53.820 right? Yes. The first element is called what 00:11:53.820 --> 00:11:56.639 is here. This acts as a map of the entire folder 00:11:56.639 --> 00:11:59.600 structure. It shows the AI the subfolders, explains 00:11:59.600 --> 00:12:02.240 what each file type is for, and outlines the 00:12:02.240 --> 00:12:04.919 data formats. The second element is called tasks. 00:12:05.340 --> 00:12:08.580 This explicitly tells the AI what to do in various 00:12:08.580 --> 00:12:11.100 different scenarios. For example, it might say, 00:12:11.220 --> 00:12:14.139 when a new raw transcript is added, process it 00:12:14.139 --> 00:12:17.259 immediately. Then... Go update the main client 00:12:17.259 --> 00:12:19.139 summary document with the new findings. Right. 00:12:19.200 --> 00:12:21.500 And the third element is the most crucial part 00:12:21.500 --> 00:12:24.240 of the entire system. It is called the self -update 00:12:24.240 --> 00:12:27.519 rule. Beat. This blew my mind when I first read 00:12:27.519 --> 00:12:30.659 it. It is brilliant. If any completely new unmapped 00:12:30.659 --> 00:12:33.460 files are added to the folder, the AI updates 00:12:33.460 --> 00:12:36.269 the instructions file to include them. the instruction 00:12:36.269 --> 00:12:38.850 file literally maintains itself as the project 00:12:38.850 --> 00:12:41.809 naturally grows the text gives a fantastic example 00:12:41.809 --> 00:12:44.429 of a youtube creator utilizing this they build 00:12:44.429 --> 00:12:46.629 a master project folder for their entire channel 00:12:46.629 --> 00:12:49.549 history each time they finish a video they drop 00:12:49.549 --> 00:12:52.399 the new script directly into the folder And the 00:12:52.399 --> 00:12:54.720 AI immediately gets to work in the background. 00:12:55.000 --> 00:12:57.980 It checks the new script against all the previous 00:12:57.980 --> 00:13:00.779 videos in the archive. It detects any overlapping 00:13:00.779 --> 00:13:03.519 topics or repeated talking points. Right. It 00:13:03.519 --> 00:13:06.480 updates a master topic index and logs all the 00:13:06.480 --> 00:13:09.220 metadata automatically. It formats the descriptions. 00:13:09.980 --> 00:13:12.899 sets timeline markers, and generates tags without 00:13:12.899 --> 00:13:15.779 any human intervention. No complex, expensive 00:13:15.779 --> 00:13:18.519 automation platform like Zapier is required here. 00:13:18.600 --> 00:13:20.860 Just a folder, a markdown instructions file, 00:13:21.080 --> 00:13:23.440 and an AI tool that can read local directories. 00:13:23.559 --> 00:13:25.919 Your personal workflow suddenly scales exponentially 00:13:25.919 --> 00:13:28.720 without you acting as the manual middleman. It 00:13:28.720 --> 00:13:31.279 changes absolutely everything about how you manage 00:13:31.279 --> 00:13:33.860 large amounts of complex data. I do have a practical 00:13:33.860 --> 00:13:36.320 question about this, though. How does the AI 00:13:36.320 --> 00:13:39.100 not get hopelessly confused when completely new, 00:13:39.259 --> 00:13:42.220 totally unmapped files are randomly dumped into 00:13:42.220 --> 00:13:44.639 the folder? Well, that is exactly what that self 00:13:44.639 --> 00:13:47.240 -update rule is designed for. The AI automatically 00:13:47.240 --> 00:13:50.159 updates its own master instructions file to properly 00:13:50.159 --> 00:13:52.340 categorize those brand new additions. A self 00:13:52.340 --> 00:13:54.799 -updating master checklist tells the AI exactly 00:13:54.799 --> 00:13:57.659 how to handle new files. Precisely. It becomes 00:13:57.659 --> 00:13:59.399 a completely self -sustaining organizational 00:13:59.399 --> 00:14:03.200 system. Two second silence. So the AI now has 00:14:03.200 --> 00:14:06.600 its own dynamic memory. It has direct, frictionless 00:14:06.600 --> 00:14:09.200 access to your project files. But there is a 00:14:09.200 --> 00:14:12.259 third, final shift required to make this truly 00:14:12.259 --> 00:14:14.919 autonomous. Because if you still have to manually 00:14:14.919 --> 00:14:17.120 proofread every single summary it generates, 00:14:17.320 --> 00:14:19.320 you haven't really solved the problem. Exactly. 00:14:19.460 --> 00:14:21.779 You are still the ultimate bottleneck. You have 00:14:21.779 --> 00:14:24.059 just transitioned into being a human checkbox. 00:14:24.279 --> 00:14:27.360 And a human checkbox can and should be removed 00:14:27.360 --> 00:14:29.559 from the routine process entirely. Right. We 00:14:29.559 --> 00:14:31.500 usually review outputs to decide if they're... 00:14:31.500 --> 00:14:33.559 actually good enough to send for deep judgment 00:14:33.559 --> 00:14:36.360 heavy strategic work that human review still 00:14:36.360 --> 00:14:39.100 really matters yeah absolutely but for routine 00:14:39.100 --> 00:14:41.940 standardized tasks it is a massive waste of your 00:14:41.940 --> 00:14:45.279 time weekly summaries standard proposals or structured 00:14:45.279 --> 00:14:48.120 status updates do not need heavy human review 00:14:48.120 --> 00:14:51.019 that review step just adds time without adding 00:14:51.019 --> 00:14:53.700 any real tangible value to the final product 00:14:53.700 --> 00:14:56.879 so the third major shift is to strictly externalize 00:14:56.879 --> 00:14:59.320 your standards you have to define what good actually 00:14:59.320 --> 00:15:01.899 means and and you have to define it in strictly 00:15:01.899 --> 00:15:05.200 binary terms. Vague, subjective standards are 00:15:05.200 --> 00:15:08.080 the absolute enemy here. Asking the AI, does 00:15:08.080 --> 00:15:10.519 this sound professional, is completely useless. 00:15:10.879 --> 00:15:13.960 It is entirely subjective. You need rigid criteria 00:15:13.960 --> 00:15:17.159 with perfectly clear yes or no answers. Let us 00:15:17.159 --> 00:15:19.460 use executive meeting summaries as our practical 00:15:19.460 --> 00:15:21.899 example here. You know, a truly good summary... 00:15:22.000 --> 00:15:24.879 always follows very specific structural rules. 00:15:25.039 --> 00:15:27.200 But you usually just feel it intuitively. You 00:15:27.200 --> 00:15:29.440 need to write those rules down. Yeah. Rule one, 00:15:29.679 --> 00:15:32.019 does the summary open with a direct reference 00:15:32.019 --> 00:15:34.539 to the previous meeting? Yes or no? Rule two, 00:15:34.740 --> 00:15:37.720 is the entire document strictly under 200 words 00:15:37.720 --> 00:15:40.419 total? Yes or no? Rule three, is there a clear, 00:15:40.519 --> 00:15:43.360 bolded deadline next to every single action item? 00:15:43.500 --> 00:15:46.379 Yes or no? Rule four, does each paragraph lead 00:15:46.379 --> 00:15:48.990 with the main key point? Yes or no? Those are 00:15:48.990 --> 00:15:52.350 four very clear, highly binary criteria. Once 00:15:52.350 --> 00:15:54.909 the rules are mathematically clear, the AI drafts 00:15:54.909 --> 00:15:57.029 the initial summary. Then it tests its own draft 00:15:57.029 --> 00:15:59.389 against that exact checklist. It identifies exactly 00:15:59.389 --> 00:16:01.950 what failed, fixes the errors, and tests the 00:16:01.950 --> 00:16:04.570 document again. It does all of this internal 00:16:04.570 --> 00:16:07.889 looping before you ever even see the file. It 00:16:07.889 --> 00:16:09.870 is the fundamental difference between an AI that 00:16:09.870 --> 00:16:12.889 simply drafts and an AI that actually finishes 00:16:12.889 --> 00:16:17.120 the job. It drafts. checks, revises, and delivers 00:16:17.120 --> 00:16:19.519 a fully completed product. But most people will 00:16:19.519 --> 00:16:22.019 get stuck right here. It is a very common friction 00:16:22.019 --> 00:16:24.639 point. They know good work when they see it. 00:16:24.720 --> 00:16:27.759 But they cannot explicitly explain why it is 00:16:27.759 --> 00:16:30.919 good. How do you build a strict binary checklist 00:16:30.919 --> 00:16:33.120 if you don't even know your own internal rules? 00:16:33.610 --> 00:16:36.490 There is a brilliant, highly practical workaround 00:16:36.490 --> 00:16:39.309 for this exact problem. Step one is to collect 00:16:39.309 --> 00:16:42.649 five to ten examples of genuinely good past outputs. 00:16:42.889 --> 00:16:45.230 You grab your best proposals, your cleanest reports, 00:16:45.370 --> 00:16:47.629 or your most effective emails, whatever you specifically 00:16:47.629 --> 00:16:50.600 need the AI to replicate. Step two, you feed 00:16:50.600 --> 00:16:53.440 those perfect examples directly to the AI. You 00:16:53.440 --> 00:16:55.679 ask it to reverse engineer a binary checklist 00:16:55.679 --> 00:16:58.039 based solely on those good examples. You are 00:16:58.039 --> 00:16:59.980 asking the machine to figure out what makes your 00:16:59.980 --> 00:17:02.419 work good. Right. And then step three, you review 00:17:02.419 --> 00:17:05.319 that generated checklist carefully. You adjust 00:17:05.319 --> 00:17:07.299 or delete anything that doesn't actually matter 00:17:07.299 --> 00:17:10.200 to the final product. Then you paste that refined 00:17:10.200 --> 00:17:13.079 checklist directly into your permanent system 00:17:13.079 --> 00:17:15.900 instructions. And step four is the most crucial 00:17:15.900 --> 00:17:18.920 prompt addition of all. You explicitly tell the 00:17:18.920 --> 00:17:22.039 AI to run an internal evaluation loop. It must 00:17:22.039 --> 00:17:24.660 draft the document, self -evaluate against the 00:17:24.660 --> 00:17:27.720 binary checklist, revise the text, and evaluate 00:17:27.720 --> 00:17:30.380 again. It runs this tight loop repeatedly until 00:17:30.380 --> 00:17:32.920 it clears the checklist completely. It only delivers 00:17:32.920 --> 00:17:35.400 the final result to your desktop after it passes 00:17:35.400 --> 00:17:38.539 every single criterion. It forces the AI to be 00:17:38.539 --> 00:17:42.660 its own strict quality assurance manager. I have 00:17:42.660 --> 00:17:44.579 to push back a little here. If you are listening 00:17:44.579 --> 00:17:47.380 to this and thinking, I do not even trust my 00:17:47.380 --> 00:17:49.960 AI to write a two -sentence email without hallucinating. 00:17:50.079 --> 00:17:52.380 I get it. Sure. Are we really supposed to just 00:17:52.380 --> 00:17:55.099 blindly trust that the AI graded its own homework 00:17:55.099 --> 00:17:57.900 correctly? Not blindly, no. That would be incredibly 00:17:57.900 --> 00:18:00.180 reckless. The entire goal of this framework is 00:18:00.180 --> 00:18:03.079 earned trust. You still review the outputs heavily 00:18:03.079 --> 00:18:05.119 at the very start of the process. You are auditing 00:18:05.119 --> 00:18:07.680 the system's internal logic. You step back only 00:18:07.680 --> 00:18:10.000 when the system definitively proves its consistency 00:18:10.000 --> 00:18:12.539 based on hard historical evidence. You let it 00:18:12.539 --> 00:18:15.599 earn its autonomy. Exactly. We verify early on. 00:18:15.980 --> 00:18:18.339 So we can step back and trust the automated results. 00:18:18.619 --> 00:18:20.799 That is the ultimate goal. You are trying to 00:18:20.799 --> 00:18:23.839 stop wasting your limited attention on routine 00:18:23.839 --> 00:18:26.500 mechanical checks. You want to spend your human 00:18:26.500 --> 00:18:28.519 judgment where it actually matters. Two seconds 00:18:28.519 --> 00:18:31.779 silence. Let us synthesize this entire journey. 00:18:32.160 --> 00:18:34.460 We have covered a lot of deep technical ground 00:18:34.460 --> 00:18:37.619 today. The overarching theme of Max Anne's framework 00:18:37.619 --> 00:18:41.180 is very clear. The real magic isn't in finding 00:18:41.180 --> 00:18:44.579 some secret, perfectly phrased prompt. The magic 00:18:44.579 --> 00:18:47.119 is in building an architectural system that compounds 00:18:47.119 --> 00:18:49.539 over time. It all starts with a memory layer 00:18:49.539 --> 00:18:51.740 that actually learns and updates itself dynamically. 00:18:52.079 --> 00:18:54.059 Then you add dedicated file folders that the 00:18:54.059 --> 00:18:56.720 AI can access directly. You remove the manual 00:18:56.720 --> 00:18:59.579 copy pasting and file routing entirely from your 00:18:59.579 --> 00:19:02.569 day. Finally, you establish strict binary quality 00:19:02.569 --> 00:19:05.609 standards. This allows the AI to check its own 00:19:05.609 --> 00:19:07.809 routine work through an autonomous internal loop. 00:19:07.950 --> 00:19:10.049 Each individual shift strengthens the others. 00:19:10.230 --> 00:19:12.470 Stronger memory improves how your local files 00:19:12.470 --> 00:19:14.910 are handled. Better file handling creates much 00:19:14.910 --> 00:19:17.990 richer context for the evaluation loop. And better 00:19:17.990 --> 00:19:20.890 evaluation naturally leads to vastly cleaner 00:19:20.890 --> 00:19:23.349 outputs. You get significantly better results 00:19:23.349 --> 00:19:25.990 next week without adding a single ounce of extra 00:19:25.990 --> 00:19:29.779 effort. The system itself is just... quietly 00:19:29.779 --> 00:19:31.599 getting smarter in the background. We really 00:19:31.599 --> 00:19:33.720 encourage you to just pick one of these three 00:19:33.720 --> 00:19:36.019 shifts to implement today. Do not try to boil 00:19:36.019 --> 00:19:38.500 the ocean and do it all at once. Start by externalizing 00:19:38.500 --> 00:19:40.660 your memory. The rest of the system will naturally 00:19:40.660 --> 00:19:43.099 build from there. It really is about escaping 00:19:43.099 --> 00:19:46.259 that little browser chat box. You are building 00:19:46.259 --> 00:19:48.599 a permanent, compounding workflow that lives 00:19:48.599 --> 00:19:52.160 on your own machine. You are intentionally removing 00:19:52.160 --> 00:19:55.079 yourself as the central bottleneck. The system 00:19:55.079 --> 00:19:57.359 carries the heavy workload and keeps things constantly 00:19:57.359 --> 00:20:00.759 moving forward. Right. I want to leave you with 00:20:00.759 --> 00:20:03.420 a thought to mull over. If your AI system can 00:20:03.420 --> 00:20:06.359 now learn your highly nuanced preferences, if 00:20:06.359 --> 00:20:08.740 it can route your complex files effortlessly, 00:20:08.859 --> 00:20:11.279 if it can grade its own routine work reliably, 00:20:11.640 --> 00:20:15.059 what high -level, deeply human problems are you 00:20:15.059 --> 00:20:17.700 now freed up to actually solve? That is the real 00:20:17.700 --> 00:20:19.539 question we all need to answer. Thanks for joining 00:20:19.539 --> 00:20:21.640 us on this deep dive. Until next time.