WEBVTT 00:00:00.000 --> 00:00:01.520 It's kind of strange when you really think about 00:00:01.520 --> 00:00:05.160 it. What is? Well, how we treat modern AI. We 00:00:05.160 --> 00:00:07.400 have this incredible world -changing technology, 00:00:07.679 --> 00:00:11.599 but we treat it like an expensive autocomplete. 00:00:11.779 --> 00:00:13.720 Right. Yeah, we really do. You sit down at your 00:00:13.720 --> 00:00:16.500 desk, you open a totally blank tab, you type 00:00:16.500 --> 00:00:19.019 out the exact same background context you typed 00:00:19.019 --> 00:00:21.440 yesterday. Just hoping to coax out a decent answer 00:00:21.440 --> 00:00:23.679 this time. Exactly. You do all this manual labor. 00:00:23.899 --> 00:00:27.260 It feels completely backward. Oh, it really is 00:00:27.260 --> 00:00:29.579 backward. I mean, it's basically Groundhog Day 00:00:29.579 --> 00:00:32.600 for knowledge workers. You end up spending vastly 00:00:32.600 --> 00:00:35.579 more time prepping the tool than actually doing 00:00:35.579 --> 00:00:39.979 the work. Welcome to this deep dive. Today, we 00:00:39.979 --> 00:00:43.140 are moving far away from that kind of stateless 00:00:43.140 --> 00:00:45.259 chatting. Finally. We're exploring a definitive 00:00:45.259 --> 00:00:48.859 March 2026 framework. And the mission here is 00:00:48.859 --> 00:00:51.159 very clear. We're going to build a self -correcting 00:00:51.159 --> 00:00:54.280 AI business OS. Yeah, we're moving past the basic 00:00:54.280 --> 00:00:57.259 chat box. The top 1 % of power users have already 00:00:57.259 --> 00:00:59.780 made this shift. It fundamentally changes how 00:00:59.780 --> 00:01:02.600 you interact with information. And, well, I have 00:01:02.600 --> 00:01:04.319 to admit something right up front here. Yeah. 00:01:04.599 --> 00:01:06.859 I still wrestle with prompt drift myself. Oh, 00:01:06.859 --> 00:01:09.530 sure. Everyone does. Starting from scratch every 00:01:09.530 --> 00:01:12.269 single time is exhausting. It takes a real cognitive 00:01:12.269 --> 00:01:14.870 toll on you. It really drains your energy. And, 00:01:14.890 --> 00:01:16.849 you know, there's an actual industry term for 00:01:16.849 --> 00:01:18.950 that exact feeling. It's called context decay. 00:01:19.150 --> 00:01:21.969 Context decay. Let's unpack that a bit. Right. 00:01:22.109 --> 00:01:25.049 So it's that manual, repetitive effort. You constantly 00:01:25.049 --> 00:01:27.569 have to re -explain your business, your tone, 00:01:27.569 --> 00:01:30.730 or your goals in every new tab. Yeah. Most people 00:01:30.730 --> 00:01:33.689 use AI completely wrong right now. They treat 00:01:33.689 --> 00:01:36.510 it as a stateless, high -end search engine. Stateless. 00:01:37.609 --> 00:01:39.969 Meaning it holds absolutely no permanent memory 00:01:39.969 --> 00:01:42.290 from yesterday. Exactly. Every single session 00:01:42.290 --> 00:01:45.129 starts totally blank. Right. So the quality of 00:01:45.129 --> 00:01:48.129 your output relies entirely on your mood. It 00:01:48.129 --> 00:01:50.469 literally depends on how well you happen to explain 00:01:50.469 --> 00:01:52.670 things that specific morning. That makes total 00:01:52.670 --> 00:01:55.090 sense. It's kind of like hiring a brilliant world 00:01:55.090 --> 00:01:58.030 -class intern. Yeah. But this intern has zero 00:01:58.030 --> 00:02:00.230 short -term memory. You have to completely re 00:02:00.230 --> 00:02:02.170 -onboard them every single morning. Every single 00:02:02.170 --> 00:02:04.629 day. You have to explain what the company actually 00:02:04.629 --> 00:02:06.890 does every single time they sit down. That's 00:02:06.890 --> 00:02:08.250 the perfect way to look at it. I mean, it's an 00:02:08.250 --> 00:02:10.490 administrative nightmare, but there is a free 00:02:10.490 --> 00:02:13.729 Google ecosystem solution available now. It uses 00:02:13.729 --> 00:02:16.889 Gemini, Gems and Notebook LM working seamlessly 00:02:16.889 --> 00:02:19.750 together. Millions of people use these tools 00:02:19.750 --> 00:02:22.990 individually, but almost nobody uses them together 00:02:22.990 --> 00:02:27.159 correctly. So why is stateless? the default for 00:02:27.159 --> 00:02:30.199 these massive AI models. I mean, if it's so painful 00:02:30.199 --> 00:02:32.460 for the user, why build it that way? Well, because 00:02:32.460 --> 00:02:35.840 giving millions of users permanent, always active 00:02:35.840 --> 00:02:39.240 memory is incredibly expensive. Wiping the slate 00:02:39.240 --> 00:02:42.039 clean after every chat saves massive server costs. 00:02:42.120 --> 00:02:44.919 It's a hardware limitation, really. Ah, so standard 00:02:44.919 --> 00:02:48.819 chats prioritize speed and scale over long -term 00:02:48.819 --> 00:02:51.669 user memory. Precisely. But with this new framework, 00:02:51.870 --> 00:02:54.009 we can finally bypass that limitation. Okay. 00:02:54.210 --> 00:02:57.150 So we understand the amnesia problem. How do 00:02:57.150 --> 00:02:59.669 we actually build a permanent brain for the AI? 00:02:59.949 --> 00:03:01.909 You have to build it using three distinct layers. 00:03:01.990 --> 00:03:04.930 Each tool plays a very specific, non -overlapping 00:03:04.930 --> 00:03:07.030 role here. Okay. What's the first one? The first 00:03:07.030 --> 00:03:09.310 is the execution layer. That's Gemini. The main 00:03:09.310 --> 00:03:11.150 processing engine, the CPU, essentially. Right. 00:03:11.289 --> 00:03:14.509 It handles the fast, real -time reasoning. It 00:03:14.509 --> 00:03:17.150 generates the text. It synthesizes the ideas. 00:03:18.139 --> 00:03:21.180 But it has no prominent hard drive. Right. So 00:03:21.180 --> 00:03:23.759 the second piece is the consistency layer. That 00:03:23.759 --> 00:03:26.080 is where gems come into play. Those are the custom 00:03:26.080 --> 00:03:28.560 assistants you can build. Yes. But think of them 00:03:28.560 --> 00:03:31.620 as strict behavioral rules. You configure your 00:03:31.620 --> 00:03:34.979 rules exactly once. You set the specific rule. 00:03:35.159 --> 00:03:37.599 You set the exact formatting. And most importantly, 00:03:37.699 --> 00:03:40.240 you set the strict anti -hallucination rules. 00:03:40.520 --> 00:03:42.719 Which is critical. Okay. Because hallucination 00:03:42.719 --> 00:03:45.860 is when AI confidently invents false information 00:03:45.860 --> 00:03:49.250 that sounds real. Spot on. You explicitly tell 00:03:49.250 --> 00:03:51.750 the gem never to do that. It acts as a behavioral 00:03:51.750 --> 00:03:54.469 guardrail. Got it. Then the third layer is the 00:03:54.469 --> 00:03:57.490 knowledge layer. That's Notebook LM. It acts 00:03:57.490 --> 00:04:00.229 as a vector database. Which is a storage system 00:04:00.229 --> 00:04:02.469 organizing information by its underlying meaning. 00:04:02.650 --> 00:04:04.810 Exactly. It holds your source -controlled immutable 00:04:04.810 --> 00:04:06.789 facts. It's your permanent hard drive. Beat. 00:04:06.889 --> 00:04:08.889 This is where the big 2026 update really matters, 00:04:08.949 --> 00:04:11.770 doesn't it? Yeah. Gemini 3 .1 Pro now allows 00:04:11.770 --> 00:04:14.889 direct notebook embedding. Right. And that completely 00:04:14.889 --> 00:04:19.040 changes the game. Custom gems now live natively 00:04:19.040 --> 00:04:21.500 inside your proprietary knowledge base. Wow. 00:04:21.759 --> 00:04:24.160 You don't have to constantly reattach 20 different 00:04:24.160 --> 00:04:26.959 PDF files every single morning. The rules and 00:04:26.959 --> 00:04:29.879 the facts are fused together permanently. It 00:04:29.879 --> 00:04:31.620 feels a lot like running a high -end restaurant. 00:04:32.040 --> 00:04:34.660 Oh, how so? Well, Notebook LM is your locked 00:04:34.660 --> 00:04:37.279 pantry. It holds all the raw ingredients and 00:04:37.279 --> 00:04:39.800 the undisputed facts. Okay, yeah. The gems are 00:04:39.800 --> 00:04:42.680 your strict recipes. They are the unbending rules 00:04:42.680 --> 00:04:46.279 for how a dish must be prepared. And Gemini is 00:04:46.279 --> 00:04:49.120 just the line chef. It does the real -time execution 00:04:49.120 --> 00:04:51.579 based on the recipe and the ingredients it's 00:04:51.579 --> 00:04:53.980 given. I really love that analogy. A chef needs 00:04:53.980 --> 00:04:56.060 both the strict recipe and the right ingredients 00:04:56.060 --> 00:04:59.060 to cook a consistent meal. Right. One tool alone 00:04:59.060 --> 00:05:01.750 simply cannot create consistency. long -term 00:05:01.750 --> 00:05:04.750 memory, and factual accuracy all at once. You 00:05:04.750 --> 00:05:07.069 need all three layers working in harmony. What 00:05:07.069 --> 00:05:09.170 happens if someone tries to skip the consistency 00:05:09.170 --> 00:05:11.170 layer? Say they just plug the notebook directly 00:05:11.170 --> 00:05:14.629 into a standard Gemini chat. The AI just improvises 00:05:14.629 --> 00:05:17.569 the tone and the structure entirely. It'll find 00:05:17.569 --> 00:05:19.910 the right facts, but it ignores your professional 00:05:19.910 --> 00:05:22.370 standards completely. Without rules, you just 00:05:22.370 --> 00:05:25.720 get confident, well -formatted. But totally unpredictable 00:05:25.720 --> 00:05:28.879 guesses. Exactly. You end up with a structural 00:05:28.879 --> 00:05:31.379 mess that you have to rewrite anyway. Okay, but 00:05:31.379 --> 00:05:34.079 just having the tools clearly isn't enough. I 00:05:34.079 --> 00:05:35.939 imagine putting the wrong documents in the wrong 00:05:35.939 --> 00:05:37.839 places completely collapses the whole system. 00:05:38.019 --> 00:05:40.600 It really does. It's a garbage in, garbage out 00:05:40.600 --> 00:05:44.000 situation. There are two major design decisions 00:05:44.000 --> 00:05:46.040 you have to make here. What's the first one? 00:05:46.199 --> 00:05:48.779 Decision number one is strictly separating your 00:05:48.779 --> 00:05:51.040 stable knowledge from your dynamic knowledge. 00:05:51.579 --> 00:05:53.800 Wait, keeping stable and dynamic knowledge totally 00:05:53.800 --> 00:05:56.699 separated sounds great in theory. Yeah. But in 00:05:56.699 --> 00:05:59.040 practice, isn't managing two separate locations 00:05:59.040 --> 00:06:01.959 just recreating the exact administrative headache 00:06:01.959 --> 00:06:04.920 we're trying to automate away? Why not just dump 00:06:04.920 --> 00:06:07.180 everything into one giant notebook and let the 00:06:07.180 --> 00:06:10.769 AI sort it out? Because the AI's retrieval mechanism 00:06:10.769 --> 00:06:14.050 doesn't work like human intuition. Stable knowledge 00:06:14.050 --> 00:06:16.790 is permanent. It rarely ever changes. Like what? 00:06:16.910 --> 00:06:19.829 We're talking about core brand guidelines, master 00:06:19.829 --> 00:06:22.610 product briefs, foundational audience research. 00:06:22.949 --> 00:06:25.790 Those live permanently in the master truth notebook. 00:06:26.189 --> 00:06:28.910 And dynamic knowledge. What does that look like? 00:06:29.009 --> 00:06:32.379 That's your weekly performance reports. or the 00:06:32.379 --> 00:06:35.500 shifting data for specific short -term campaigns, 00:06:35.720 --> 00:06:38.639 they absolutely must stay out of the master notebook. 00:06:38.839 --> 00:06:40.860 You only attach them at the session level when 00:06:40.860 --> 00:06:43.779 needed for a specific task. So if I dump a weekly 00:06:43.779 --> 00:06:46.519 report into the permanent notebook, what actually 00:06:46.519 --> 00:06:49.339 happens mechanically? you create massive data 00:06:49.339 --> 00:06:52.180 pollution. If you have 50 weekly reports and 00:06:52.180 --> 00:06:54.860 one master brand guideline in the same database, 00:06:55.139 --> 00:06:58.259 the sheer volume of the weekly reports mathematically 00:06:58.259 --> 00:07:00.899 drowns out the core guidelines in the vector 00:07:00.899 --> 00:07:03.939 space. The AI gets confused about what is actually 00:07:03.939 --> 00:07:06.160 important. Mixing them really is data pollution. 00:07:06.399 --> 00:07:08.720 It's basically like printing yesterday's weather 00:07:08.720 --> 00:07:10.759 report in a permanent history textbook. Beat. 00:07:10.939 --> 00:07:13.339 It just deeply confuses the system's sense of 00:07:13.339 --> 00:07:15.459 priority. That's exactly what happens under the 00:07:15.459 --> 00:07:18.240 hood. The system starts quoting last week's failed 00:07:18.240 --> 00:07:21.240 ad spend as a permanent company mandate. Then 00:07:21.240 --> 00:07:23.740 you end up manually fact checking every single 00:07:23.740 --> 00:07:26.939 paragraph it writes. It defeats the entire purpose 00:07:26.939 --> 00:07:29.060 of automating the work in the first place. So 00:07:29.060 --> 00:07:31.949 stable stays in the notebook. Dynamic stays out 00:07:31.949 --> 00:07:34.750 and gets attached on the fly. What is the second 00:07:34.750 --> 00:07:38.110 major design decision? Decision two is locking 00:07:38.110 --> 00:07:40.550 the functional rules entirely inside the gem. 00:07:40.670 --> 00:07:43.189 You don't ever put your system rules in the daily 00:07:43.189 --> 00:07:45.689 prompt. So the gem defines exactly how the system 00:07:45.689 --> 00:07:48.189 works. Right. The gem dictates the formatting, 00:07:48.490 --> 00:07:50.889 the tone, and the constraints. And the daily 00:07:50.889 --> 00:07:52.990 prompt simply defines what you want done today. 00:07:53.149 --> 00:07:55.350 I see. The prompt handles the specific immediate 00:07:55.350 --> 00:07:58.610 goal or the target audience. This strict separation 00:07:58.610 --> 00:08:02.129 makes the entire system endlessly reusable. Is 00:08:02.129 --> 00:08:04.569 there a foolproof way to identify if a document 00:08:04.569 --> 00:08:07.250 is stable or dynamic before I upload it? Ask 00:08:07.250 --> 00:08:09.759 yourself one simple question. Will the information 00:08:09.759 --> 00:08:12.459 inside this document still be completely accurate 00:08:12.459 --> 00:08:15.180 and relevant next quarter? If it changes every 00:08:15.180 --> 00:08:17.339 quarter, keep it out of the main notebook entirely. 00:08:17.699 --> 00:08:21.220 Precisely. Keep the foundation relentlessly clean. 00:08:21.420 --> 00:08:24.019 Theory is great. But let's see this machine actually 00:08:24.019 --> 00:08:27.220 run in the wild. Let's look at the Anything LLM 00:08:27.220 --> 00:08:30.620 case study from the sources. Yes. Anything LLM 00:08:30.620 --> 00:08:34.659 is a private local desktop AI application. Their 00:08:34.659 --> 00:08:37.779 internal team has very heavy recurring tasks. 00:08:38.220 --> 00:08:41.450 Like what kind of tasks? Launch messaging. technical 00:08:41.450 --> 00:08:45.070 webinars, sales enablement materials. And without 00:08:45.070 --> 00:08:46.909 a layered system, they're basically starting 00:08:46.909 --> 00:08:49.070 from zero every single time they need to write 00:08:49.070 --> 00:08:51.870 an email. Exactly. But by using this three -layer 00:08:51.870 --> 00:08:54.110 system, they built a permanent truth notebook. 00:08:54.450 --> 00:08:57.629 It holds their core internal product brief and 00:08:57.629 --> 00:08:59.850 their master audience analysis. That's the stable 00:08:59.850 --> 00:09:01.789 foundation we just talked about. Right. Then 00:09:01.789 --> 00:09:04.629 they built a product messaging engine gem. And 00:09:04.629 --> 00:09:07.549 inside that gem, they used a very strict surface 00:09:07.549 --> 00:09:10.340 before generate rule. Let's explain that mechanically. 00:09:10.399 --> 00:09:12.299 How does surface before generate actually work 00:09:12.299 --> 00:09:14.919 under the hood? The gem basically acts like a 00:09:14.919 --> 00:09:17.580 bouncer at a club. Okay. Before Gemini is even 00:09:17.580 --> 00:09:19.519 allowed to start predicting the next word of 00:09:19.519 --> 00:09:22.080 your blog post, the gem physically forces it 00:09:22.080 --> 00:09:24.320 to run a mandatory search query against your 00:09:24.320 --> 00:09:26.840 notebook LM database. So it has to physically 00:09:26.840 --> 00:09:30.139 find the verified facts first. It must surface 00:09:30.139 --> 00:09:32.620 product truths before the system is allowed to 00:09:32.620 --> 00:09:34.960 write a single generated word. Exactly. It's 00:09:34.960 --> 00:09:38.360 an agentic loop. Step one. Query the database. 00:09:38.740 --> 00:09:42.159 Step two, read the retrieved facts. Step three, 00:09:42.320 --> 00:09:45.740 generate the text. And if it doesn't find anything? 00:09:45.940 --> 00:09:48.080 If step one fails to find anything, the system 00:09:48.080 --> 00:09:51.159 halts. Let's look at three real -world scenarios 00:09:51.159 --> 00:09:53.980 from their team. Scenario one is a simple daily 00:09:53.980 --> 00:09:57.179 task, generating launch angles for early adopters. 00:09:57.600 --> 00:09:59.899 That sounds pretty clean and fast. Incredibly 00:09:59.899 --> 00:10:02.159 fast. You use the Stablemaster notebook and the 00:10:02.159 --> 00:10:04.659 locked messaging gem. Your daily prompt just 00:10:04.659 --> 00:10:07.519 contains the specific task. Write three launch 00:10:07.519 --> 00:10:09.940 angles. Because the rules and facts are already 00:10:09.940 --> 00:10:12.860 loaded, the output is instantly consistent and 00:10:12.860 --> 00:10:15.600 completely reliable. Scenario two requires much 00:10:15.600 --> 00:10:17.940 deeper evidence. Say you want to target enterprise 00:10:17.940 --> 00:10:20.379 IT directors. Those are incredibly skeptical, 00:10:20.639 --> 00:10:23.559 compliance -focused professionals. A generic 00:10:23.559 --> 00:10:25.360 marketing pitch will instantly turn them off. 00:10:25.519 --> 00:10:27.379 Right. So this is where you bring in those session 00:10:27.379 --> 00:10:30.379 -level dynamic docs. Things like specific IT 00:10:30.379 --> 00:10:34.059 use cases or recent security pain point sheets. 00:10:34.320 --> 00:10:36.799 You drag and drop them into the chat just for 00:10:36.799 --> 00:10:40.210 this one specific prompt. Yes. These temporary 00:10:40.210 --> 00:10:42.629 docs tell the gem exactly what this specific 00:10:42.629 --> 00:10:45.470 audience cares about today. The system instantly 00:10:45.470 --> 00:10:47.710 connects your permanent product features from 00:10:47.710 --> 00:10:49.950 the notebook to the specific outcomes in the 00:10:49.950 --> 00:10:51.950 dynamic doc. That's powerful. Yeah, output gets 00:10:51.950 --> 00:10:54.730 highly detailed. It automatically includes narrative 00:10:54.730 --> 00:10:57.090 pillars and specific proof sections. Because 00:10:57.090 --> 00:10:59.690 the gem was explicitly told to surface hard truths, 00:10:59.850 --> 00:11:02.309 not just generate marketing fluff. Exactly. Now, 00:11:02.370 --> 00:11:04.549 scenario three is where it gets really advanced. 00:11:04.769 --> 00:11:07.210 It's about cross -referencing. You actually use 00:11:07.210 --> 00:11:09.389 two completely separate... notebooks at the same 00:11:09.389 --> 00:11:12.129 time oh wow one is your core project knowledge 00:11:12.129 --> 00:11:14.870 the other is your performance intelligence so 00:11:14.870 --> 00:11:17.389 comparing permanent expectations against shifting 00:11:17.389 --> 00:11:20.309 reality right you attach both notebooks to the 00:11:20.309 --> 00:11:23.029 same gemini session the first notebook provides 00:11:23.029 --> 00:11:25.429 the original marketing claims you made the second 00:11:25.429 --> 00:11:28.309 notebook provides the actual q1 revenue and engagement 00:11:28.309 --> 00:11:30.870 data so it's actively comparing the original 00:11:30.870 --> 00:11:34.519 claims against the hard reality Then it's proposing 00:11:34.519 --> 00:11:37.879 updates to the truth notebook based on what actually 00:11:37.879 --> 00:11:42.440 worked. Two sec silence. Whoa! Imagine a self 00:11:42.440 --> 00:11:45.000 -auditing loop. where the AI actually has to 00:11:45.000 --> 00:11:47.860 prove its claims using your exact documents before 00:11:47.860 --> 00:11:50.019 speaking. That changes everything. It really 00:11:50.019 --> 00:11:52.419 is a massive paradigm shift. It creates actual 00:11:52.419 --> 00:11:56.139 verifiable reliability. You can ask Gemini how 00:11:56.139 --> 00:11:58.820 each specific document contributed to its final 00:11:58.820 --> 00:12:01.559 answer. Yeah. It actively shows its exact work. 00:12:01.600 --> 00:12:03.960 It gives you footnotes. But looking closely at 00:12:03.960 --> 00:12:06.320 the sources, those citations usually point to 00:12:06.320 --> 00:12:08.320 the notebook overview. They don't always point 00:12:08.320 --> 00:12:10.659 to the specific lines within the massive documents. 00:12:10.899 --> 00:12:13.649 Can we really trust it blindly? Absolutely not 00:12:13.649 --> 00:12:15.750 blindly. It reliably points you to the right 00:12:15.750 --> 00:12:18.029 source document, which saves you hours of searching. 00:12:18.169 --> 00:12:20.649 But you still must verify the exact numbers or 00:12:20.649 --> 00:12:22.730 specific quotes manually. So it builds immense 00:12:22.730 --> 00:12:25.529 trust, but you still need manual checks for critical 00:12:25.529 --> 00:12:29.210 claims. You always need a human in the loop for 00:12:29.210 --> 00:12:32.149 the final ultimate sign -off. It's an assistant, 00:12:32.289 --> 00:12:36.389 not an autonomous CEO sponsor. This deep dive 00:12:36.389 --> 00:12:38.850 is brought to you by our partners. When you are 00:12:38.850 --> 00:12:41.350 building systems that scale, you need infrastructure 00:12:41.350 --> 00:12:44.639 you can trust. Check out our sponsors link in 00:12:44.639 --> 00:12:46.919 the description. Now let's get back to the deep 00:12:46.919 --> 00:12:50.019 dive. Okay, so this structural architecture is 00:12:50.019 --> 00:12:52.220 incredible for corporate teams, but it's not 00:12:52.220 --> 00:12:54.879 just for software product managers, right? How 00:12:54.879 --> 00:12:57.500 does the everyday listener, someone who is just 00:12:57.500 --> 00:13:00.059 trying to learn faster and manage their own life, 00:13:00.179 --> 00:13:04.070 use this OS today? It applies beautifully to 00:13:04.070 --> 00:13:07.129 absolutely any knowledge -heavy workflow in your 00:13:07.129 --> 00:13:09.129 personal life. Let's take a personal job search 00:13:09.129 --> 00:13:11.570 system as an example. Okay. How does that practically 00:13:11.570 --> 00:13:13.509 look when you're setting it up? Well, your master 00:13:13.509 --> 00:13:15.950 notebook is your permanent career vault. It holds 00:13:15.950 --> 00:13:18.470 your master CV, your personal career narrative, 00:13:18.590 --> 00:13:20.830 and your target industry research. That is your 00:13:20.830 --> 00:13:23.129 stable foundation. And the gem. What are the 00:13:23.129 --> 00:13:25.299 rules for a job search? The gem becomes your 00:13:25.299 --> 00:13:27.860 application engine. You program it to adapt your 00:13:27.860 --> 00:13:29.820 tone for cover letters. You strictly instruct 00:13:29.820 --> 00:13:32.139 it to never, ever invent credentials. Right. 00:13:32.240 --> 00:13:34.879 Obviously. And you tell it to aggressively flag 00:13:34.879 --> 00:13:37.940 any mismatches between your resume and the job 00:13:37.940 --> 00:13:40.740 requirements. And the dynamic session docs. Those 00:13:40.740 --> 00:13:42.899 are the specific job descriptions you're applying 00:13:42.899 --> 00:13:46.360 for that day. Or, you know, a target company's 00:13:46.360 --> 00:13:49.259 recent annual report. You bring them in just 00:13:49.259 --> 00:13:51.960 for that one specific application session so 00:13:51.960 --> 00:13:54.159 they don't pollute your master career. your notebook. 00:13:54.419 --> 00:13:57.320 That makes perfect sense. It saves you from rewriting 00:13:57.320 --> 00:13:59.120 your history every time you apply somewhere new. 00:13:59.360 --> 00:14:02.679 What about something more personal, like a health 00:14:02.679 --> 00:14:05.700 and household system? This is incredibly useful. 00:14:06.059 --> 00:14:08.779 The master notebook holds your stable family 00:14:08.779 --> 00:14:12.659 health history, routine GP letters, permanent 00:14:12.659 --> 00:14:15.179 dietary restrictions. All the unchangeable medical 00:14:15.179 --> 00:14:17.529 facts. Right. The gem is configured as a health 00:14:17.529 --> 00:14:20.269 decision injure. It structures advice clearly. 00:14:20.429 --> 00:14:23.809 It explicitly flags potential medical contradictions. 00:14:23.870 --> 00:14:25.769 It's super important. Most importantly, it's 00:14:25.769 --> 00:14:27.750 restricted to stay within your uploaded documents 00:14:27.750 --> 00:14:30.809 only. No searching the wild web for random medical 00:14:30.809 --> 00:14:33.490 advice. And then you just add your new temporary 00:14:33.490 --> 00:14:36.370 blood test results as dynamic session level docs. 00:14:36.840 --> 00:14:40.120 Exactly. Or maybe a highly specific new research 00:14:40.120 --> 00:14:42.500 paper you want summarized against your personal 00:14:42.500 --> 00:14:45.980 medical history. It evaluates the new data against 00:14:45.980 --> 00:14:48.669 your permanent baseline. I can see how that works 00:14:48.669 --> 00:14:51.190 for rigid data like medical records. Right. But 00:14:51.190 --> 00:14:53.370 what about something inherently subjective like 00:14:53.370 --> 00:14:56.509 creative writing or content strategy? Does the 00:14:56.509 --> 00:14:59.210 consistency layer suffocate the creativity? Not 00:14:59.210 --> 00:15:01.750 at all. Let's do a content strategy use case. 00:15:01.909 --> 00:15:04.990 It doesn't suffocate it. It focuses it. The notebook 00:15:04.990 --> 00:15:07.970 holds your stable brand growth guidelines, your 00:15:07.970 --> 00:15:10.889 deeper audience psychology research, the hard 00:15:10.889 --> 00:15:13.370 data from past campaign results. And the gem 00:15:13.370 --> 00:15:16.759 acts as a content strategy engine. Yes. It maintains 00:15:16.759 --> 00:15:19.279 those strict brand standards. It seamlessly adapts 00:15:19.279 --> 00:15:21.340 formats from blogs to tweets. And again, it never 00:15:21.340 --> 00:15:23.779 invents performance data. The dynamic session 00:15:23.779 --> 00:15:26.019 docs are just your trending daily topics or the 00:15:26.019 --> 00:15:28.440 specific brief for that week. Building this is 00:15:28.440 --> 00:15:30.879 exactly like stacking Lego blocks of data. You 00:15:30.879 --> 00:15:33.139 painstakingly snap the stable base together once. 00:15:33.610 --> 00:15:35.769 And then you just swap out the colorful top pieces, 00:15:35.789 --> 00:15:37.950 the dynamic docs, for different daily tasks. 00:15:38.289 --> 00:15:40.450 That's a really great visual. It saves so much 00:15:40.450 --> 00:15:43.090 cognitive overhead. You fundamentally stop explaining 00:15:43.090 --> 00:15:45.289 yourself to a machine, and you finally start 00:15:45.289 --> 00:15:47.850 executing your actual vision. But looking at 00:15:47.850 --> 00:15:50.250 all this, it feels like a lot of initial setup. 00:15:50.570 --> 00:15:53.289 What should the very first step be for a listener 00:15:53.289 --> 00:15:56.250 who is feeling completely overwhelmed by their 00:15:56.250 --> 00:15:59.169 open tabs right now? Don't try to build the whole 00:15:59.169 --> 00:16:02.970 OS at once. Pick one single highly repeating 00:16:02.970 --> 00:16:06.610 task that annoys you. Upload just two permanent 00:16:06.610 --> 00:16:09.409 stable documents to a brand new notebook. Set 00:16:09.409 --> 00:16:12.809 up one simple gem with three basic rules. Run 00:16:12.809 --> 00:16:15.669 it once. Pick one repetitive workflow. Build 00:16:15.669 --> 00:16:18.009 your truth notebook and never explain it again. 00:16:18.169 --> 00:16:21.129 That very first run is absolute magic. You suddenly 00:16:21.129 --> 00:16:23.649 realize that AI finally remembers exactly how 00:16:23.649 --> 00:16:25.649 you think and how you work. Let's pull back and 00:16:25.649 --> 00:16:27.730 synthesize the entire blueprint we've just drawn 00:16:27.730 --> 00:16:30.629 today. We are actively shifting away from casual, 00:16:30.750 --> 00:16:33.490 stateless AI use. We're moving to a highly intentional, 00:16:33.789 --> 00:16:36.230 highly structured system. It changes how you 00:16:36.230 --> 00:16:38.950 process information entirely. It turns AI from 00:16:38.950 --> 00:16:42.009 a slot machine into a reliable utility. Notebook 00:16:42.009 --> 00:16:45.929 LM is the unchangeable truth. Gems dictate the 00:16:45.929 --> 00:16:48.809 strict thinking rules. And Gemini is the engine 00:16:48.809 --> 00:16:50.909 that executes the vision. And the key takeaway 00:16:50.909 --> 00:16:53.789 is that separation creates clarity. You deliberately 00:16:53.789 --> 00:16:56.049 separate the raw data from the behavioral rules. 00:16:56.190 --> 00:16:58.769 You intentionally separate stable, permanent 00:16:58.769 --> 00:17:02.230 facts from dynamic, daily updates. And that precise 00:17:02.230 --> 00:17:05.109 combination is what creates actual, reliable 00:17:05.109 --> 00:17:07.609 insight. Exactly. When you bring them together 00:17:07.609 --> 00:17:10.650 correctly, it feels like magic. But it isn't 00:17:10.650 --> 00:17:12.970 magic. It's just exceptionally good, disciplined 00:17:12.970 --> 00:17:15.650 data architecture. Go build your very first truth 00:17:15.650 --> 00:17:18.410 notebook today. Seriously. Start with just two 00:17:18.410 --> 00:17:21.390 permanent documents. Test it yourself. Feel the 00:17:21.390 --> 00:17:24.009 difference between stateless chatting and a stateful 00:17:24.009 --> 00:17:26.089 system. It will completely change your entire 00:17:26.089 --> 00:17:28.250 perspective on what AI can actually do for you. 00:17:28.309 --> 00:17:30.130 You will absolutely never go back to stateless 00:17:30.130 --> 00:17:31.910 chatting again. I want to leave you with something 00:17:31.910 --> 00:17:34.690 to ponder as we wrap up. What if the true lasting 00:17:34.690 --> 00:17:38.269 value of AI in 2026 isn't its ability to endlessly 00:17:38.269 --> 00:17:41.049 generate brand new things? What if its real superpower 00:17:41.049 --> 00:17:43.650 is its ability to rigorously enforce your existing 00:17:43.650 --> 00:17:46.410 high quality standards against a massive sea 00:17:46.410 --> 00:17:49.539 of daily noise? Beat. Think about that. Outro 00:17:49.539 --> 00:17:49.759 music.