WEBVTT 00:00:00.000 --> 00:00:02.600 Every single hour you spend, you know, clicking 00:00:02.600 --> 00:00:05.519 around and manually gathering research, well, 00:00:05.799 --> 00:00:07.719 it is an hour your competition is getting ahead. 00:00:07.820 --> 00:00:10.119 Oh, absolutely. And not because they are smarter 00:00:10.119 --> 00:00:12.480 than you, not because they're working harder. 00:00:12.740 --> 00:00:15.080 They just automated the boring stuff. Right. 00:00:15.199 --> 00:00:17.620 It is kind of the harsh reality of modern knowledge 00:00:17.620 --> 00:00:20.480 work. Honestly, we are wasting so much of our 00:00:20.480 --> 00:00:23.960 cognitive energy on just like repetitive mechanical 00:00:23.960 --> 00:00:26.440 tasks. Yeah, just trying to get the information 00:00:26.440 --> 00:00:28.620 in front of us. Exactly. So welcome to the deep 00:00:28.620 --> 00:00:31.719 dive. I want to set our tone right away today, 00:00:32.119 --> 00:00:35.320 because we are unpacking a really fundamental 00:00:35.320 --> 00:00:37.560 shift here. For sure. This is not just, you know, 00:00:37.640 --> 00:00:39.780 another quick productivity hack to save you five 00:00:39.780 --> 00:00:42.460 minutes. We are looking at how you interface 00:00:42.460 --> 00:00:45.119 with human knowledge itself. It is a big one. 00:00:45.240 --> 00:00:47.979 It really is. So here is the roadmap for today. 00:00:48.380 --> 00:00:51.200 First, we will look at why people use Google's 00:00:51.200 --> 00:00:54.359 Notebook LM completely the wrong way. Oh, they 00:00:54.359 --> 00:00:57.880 definitely do. Yeah. Then we will bring in Anthropics 00:00:57.880 --> 00:01:01.039 Claude. to act as a sort of conductor to fully 00:01:01.039 --> 00:01:04.439 automate it. A conductor? I like that. And then 00:01:04.439 --> 00:01:06.819 we will look at the actual step -by -step setup. 00:01:07.519 --> 00:01:10.159 And finally, we will set it loose to build these 00:01:10.159 --> 00:01:13.140 autonomous daily research machines. It genuinely 00:01:13.140 --> 00:01:15.079 sounds like magic when you first see it work. 00:01:15.299 --> 00:01:17.560 But I mean, it is really just very smart engineering. 00:01:18.000 --> 00:01:20.920 Exactly. So let's approach this calmly. Stay 00:01:20.920 --> 00:01:23.829 inquisitive today. Don't let the technical automation 00:01:23.829 --> 00:01:26.310 overwhelm you because the underlying philosophy 00:01:26.310 --> 00:01:29.390 here is what actually matters. Let's look at 00:01:29.390 --> 00:01:32.189 the first piece of this puzzle, Notebook LM. 00:01:33.150 --> 00:01:36.400 It is a brilliant tool. But I think almost everyone 00:01:36.400 --> 00:01:39.099 is using it totally wrong. They absolutely are. 00:01:39.200 --> 00:01:41.560 And, you know, to understand why, we kind of 00:01:41.560 --> 00:01:43.439 have to break down what the tool actually is. 00:01:44.180 --> 00:01:46.640 Because Notebook LM is built by Google. It is 00:01:46.640 --> 00:01:49.099 powered by their Gemini model. But it operates 00:01:49.099 --> 00:01:51.640 completely differently than, say, chat GPT or 00:01:51.640 --> 00:01:53.959 a standard cloud interface. It is much more constrained. 00:01:54.060 --> 00:01:56.780 Radically constrained. You feed it specific sources, 00:01:56.780 --> 00:01:59.180 and it uses only those sources to answer your 00:01:59.180 --> 00:02:01.099 question. Right. So it is entirely closed off. 00:02:01.159 --> 00:02:03.099 It is not just pulling random facts from the 00:02:03.099 --> 00:02:05.980 broader internet at all. Exactly. It works exclusively 00:02:05.980 --> 00:02:08.560 with what you give it. So you can drop in a dense 00:02:08.560 --> 00:02:12.759 PDF file. You can add web URLs or paste YouTube 00:02:12.759 --> 00:02:14.759 video links. Or even just raw meeting notes, 00:02:14.780 --> 00:02:16.919 right? Yeah, exactly. Raw text works perfectly. 00:02:17.060 --> 00:02:19.280 And then it essentially acts as an expert on 00:02:19.280 --> 00:02:21.379 that specific pile of data. Well, more than just 00:02:21.379 --> 00:02:24.599 an expert, it generates these massive professional 00:02:24.599 --> 00:02:27.520 outputs based entirely on your inputs. Oh, wow. 00:02:27.699 --> 00:02:30.280 Yeah, it can create incredibly clean slide decks. 00:02:30.400 --> 00:02:32.960 It generates steady flashcards. It builds complex 00:02:32.960 --> 00:02:36.009 mind maps. It even creates those two host audio 00:02:36.009 --> 00:02:38.889 summaries, which is essentially a completely 00:02:38.889 --> 00:02:41.409 custom podcast analyzing your documents. It is 00:02:41.409 --> 00:02:44.770 an incredible suite of tools, but As you look 00:02:44.770 --> 00:02:47.810 closer at the actual daily workflow, there is 00:02:47.810 --> 00:02:50.330 a glaring issue. Yeah, the manual labor. Exactly. 00:02:50.469 --> 00:02:53.150 The manual way people interact with this is deeply 00:02:53.150 --> 00:02:55.710 flawed. It is totally exhausting, because most 00:02:55.710 --> 00:02:58.590 people, they open Notebook LM, they click to 00:02:58.590 --> 00:03:00.229 create a new notebook, and then they have to 00:03:00.229 --> 00:03:02.590 go hunt for sources somewhere else. Wow, tabbing 00:03:02.590 --> 00:03:05.289 back and forth. Right. They find a YouTube video, 00:03:05.490 --> 00:03:07.990 copy the link, paste it into the tool, and then 00:03:07.990 --> 00:03:09.530 they just sit there and wait for it to load. 00:03:09.810 --> 00:03:12.139 Which takes a minute. Then they find an article, 00:03:12.319 --> 00:03:14.120 copy it, paste it, wait again. Then they finally 00:03:14.120 --> 00:03:16.319 click through some menu to generate a slide deck. 00:03:16.580 --> 00:03:19.780 They are doing it one by one. Every single time. 00:03:20.240 --> 00:03:22.080 Every single day. You know what reminds me of 00:03:22.080 --> 00:03:24.479 the early days of automobiles? You had to stand 00:03:24.479 --> 00:03:27.180 out in the mud and hand crank the engine just 00:03:27.180 --> 00:03:29.719 to get the car started. That is a great analogy. 00:03:30.000 --> 00:03:32.060 The human is the bottleneck in the loop here. 00:03:32.099 --> 00:03:35.300 You are doing all the mechanical labor of moving 00:03:35.300 --> 00:03:38.860 data from point A to point B. And that loop simply 00:03:38.860 --> 00:03:41.150 does not scale. I mean, you have to be physically 00:03:41.150 --> 00:03:43.310 sitting at your desk. You have to remember to 00:03:43.310 --> 00:03:45.870 actually do the research. And you have to manually 00:03:45.870 --> 00:03:48.849 click through every single action yourself. Your 00:03:48.849 --> 00:03:51.229 brain is basically tired before you even start 00:03:51.229 --> 00:03:53.750 reading the output. Which brings up a really 00:03:53.750 --> 00:03:57.530 interesting tension here. We want the AI to be 00:03:57.530 --> 00:04:00.669 autonomous to go out and get the data for us. 00:04:00.689 --> 00:04:02.830 But we also want it to be perfectly accurate, 00:04:03.389 --> 00:04:05.349 which is why we use Notebook LM in the first 00:04:05.349 --> 00:04:10.569 place. So why is restricting the AI to only your 00:04:10.569 --> 00:04:13.330 uploaded sources actually its biggest superpower? 00:04:13.830 --> 00:04:16.750 It eliminates hallucinations by citing the exact 00:04:16.750 --> 00:04:19.430 source document for every single claim. That 00:04:19.430 --> 00:04:22.529 is profound. There is no guessing. None. There 00:04:22.529 --> 00:04:25.430 are no vague, well, I think this is true kind 00:04:25.430 --> 00:04:28.310 of responses. Everything is entirely grounded 00:04:28.310 --> 00:04:31.430 in reality. Exactly. But that grounding is useless. 00:04:31.689 --> 00:04:35.149 If we are still hand cranking the engine, we 00:04:35.149 --> 00:04:37.629 need to remove the human bottleneck entirely. 00:04:37.810 --> 00:04:40.069 We need a conductor to orchestrate this. And 00:04:40.069 --> 00:04:42.870 that conductor is Anthropix Claude. But we need 00:04:42.870 --> 00:04:45.220 to be clear here, not... the standard Claude 00:04:45.220 --> 00:04:47.839 you use in your web browser. Claude is an AI 00:04:47.839 --> 00:04:49.639 assistant, sure, but it has different forms. 00:04:49.660 --> 00:04:51.860 It goes way beyond just answering questions. 00:04:52.180 --> 00:04:54.060 When you give it the right environment, it can 00:04:54.060 --> 00:04:56.660 take real action. It can search the web, write 00:04:56.660 --> 00:04:59.540 files, and crucially, it can communicate with 00:04:59.540 --> 00:05:02.060 other software. The source material is very specific 00:05:02.060 --> 00:05:03.939 about this, actually. There are three different 00:05:03.939 --> 00:05:05.439 versions of Claude, and we should clarify that 00:05:05.439 --> 00:05:08.399 so nobody gets lost. Yeah, that is important. 00:05:08.860 --> 00:05:11.079 So if you are a Claude user, you generally see 00:05:11.079 --> 00:05:14.360 three main modes. First is chat. That is your 00:05:14.360 --> 00:05:17.160 regular conversation. Best for just asking questions. 00:05:17.240 --> 00:05:20.439 Pretty standard. Right. Second is projects, where 00:05:20.439 --> 00:05:22.939 Claude works alongside you with a specific set 00:05:22.939 --> 00:05:25.699 of files you upload. But to build an autonomous 00:05:25.699 --> 00:05:28.079 machine, we actually need the third mode. Yeah. 00:05:28.319 --> 00:05:31.769 Claude code. Exactly. Cloud Code is a completely 00:05:31.769 --> 00:05:33.970 different beast. It is a developer tool that 00:05:33.970 --> 00:05:36.769 runs directly in your computer's terminal. It 00:05:36.769 --> 00:05:39.449 executes code, reads your local files, and connects 00:05:39.449 --> 00:05:42.649 to external tools. It operates as a fully autonomous 00:05:42.649 --> 00:05:44.709 agent that can actually do things right on your 00:05:44.709 --> 00:05:46.970 machine. And this is where the magic really happens. 00:05:47.149 --> 00:05:50.589 We are connecting the action -taker, Cloud Code, 00:05:50.610 --> 00:05:53.310 to the synthesizer, Notebook LM. Yes. But there 00:05:53.310 --> 00:05:55.970 is a massive technical hurdle here. Notebook 00:05:55.970 --> 00:05:59.139 LM does not have an official public API. Google 00:05:59.139 --> 00:06:01.540 has not built a back door for other apps to talk 00:06:01.540 --> 00:06:03.920 to it. Right. And without a public API, software 00:06:03.920 --> 00:06:06.420 tools cannot easily shake hands. They cannot 00:06:06.420 --> 00:06:08.639 exchange data in the background. Which is a problem. 00:06:08.860 --> 00:06:11.980 A huge problem. And that is exactly where a new 00:06:11.980 --> 00:06:14.560 piece of technology called MCP comes in to save 00:06:14.560 --> 00:06:16.680 the day. I see that acronym everywhere right 00:06:16.680 --> 00:06:18.740 now. And I will be honest, my eyes usually kind 00:06:18.740 --> 00:06:22.139 of glaze over. Model context protocol. It sounds 00:06:22.139 --> 00:06:25.310 incredibly dense. Let's define that jargon simply. 00:06:25.730 --> 00:06:28.410 What is MCP actually doing for us? A digital 00:06:28.410 --> 00:06:31.790 bridge letting AI operate other apps like a human. 00:06:32.009 --> 00:06:35.990 A digital bridge. I like that. So MCP bridges 00:06:35.990 --> 00:06:39.170 the gap. And a developer named Jacob Ben -Babitt 00:06:39.170 --> 00:06:41.750 actually built a specific package for this bridge 00:06:41.750 --> 00:06:45.189 called the Notebook MCP package. It is a brilliant 00:06:45.189 --> 00:06:47.620 piece of engineering, honestly. Cloud Code sends 00:06:47.620 --> 00:06:49.980 natural language instructions to this MCP server 00:06:49.980 --> 00:06:52.660 running right on your machine. The server translates 00:06:52.660 --> 00:06:54.500 those instructions and handles all the heavy 00:06:54.500 --> 00:06:56.879 lifting. It creates the notebooks, loads the 00:06:56.879 --> 00:06:59.360 sources, and triggers the audio or slide generation. 00:06:59.399 --> 00:07:01.459 Well, I have a vulnerable admission to make here. 00:07:01.839 --> 00:07:04.660 Even knowing how powerful all of this is, I still 00:07:04.660 --> 00:07:06.600 get intimidated whenever I have to open a command 00:07:06.600 --> 00:07:09.100 terminal. Oh, totally. Yeah, staring at a black 00:07:09.100 --> 00:07:11.500 screen with blinking text, it feels like stepping 00:07:11.500 --> 00:07:14.019 into the matrix. Are we doing actual coding here? 00:07:14.189 --> 00:07:16.689 That feeling is completely normal. The terminal 00:07:16.689 --> 00:07:19.529 is super intimidating, but I promise you, no 00:07:19.529 --> 00:07:21.629 coding experience is required for this at all. 00:07:21.910 --> 00:07:24.209 Really? You are not writing software. You are 00:07:24.209 --> 00:07:26.250 just talking to Claude and telling it to run 00:07:26.250 --> 00:07:28.910 a few installation commands for you. OK, but 00:07:28.910 --> 00:07:31.029 let's clarify the mechanism of how this bridge 00:07:31.029 --> 00:07:34.209 actually works. Since there is no official Google 00:07:34.209 --> 00:07:37.050 API quietly passing data back and forth, how 00:07:37.050 --> 00:07:39.170 is Claude actually interacting with Notebook 00:07:39.170 --> 00:07:43.180 LM? The MCP package quietly opens Chrome in the 00:07:43.180 --> 00:07:45.939 background and clicks buttons for you. It is 00:07:45.939 --> 00:07:48.379 literally acting as a ghost in the machine. Exactly. 00:07:48.560 --> 00:07:51.459 It mimics human clicking. It opens a hidden browser, 00:07:52.040 --> 00:07:54.480 navigates to the website, and clicks the upload 00:07:54.480 --> 00:07:57.769 button just like you would. Yeah. That is incredibly 00:07:57.769 --> 00:08:00.350 clever, but it also raises some real security 00:08:00.350 --> 00:08:02.769 concerns, which makes the actual step -by -step 00:08:02.769 --> 00:08:06.050 setup critical to get right. Yes. So let's walk 00:08:06.050 --> 00:08:07.829 through the actual build. It is surprisingly 00:08:07.829 --> 00:08:09.850 straightforward. You do not even need to download 00:08:09.850 --> 00:08:12.850 a separate scary terminal application. Oh, nice. 00:08:12.910 --> 00:08:16.050 You just open your standard Claw desktop app, 00:08:16.129 --> 00:08:18.589 and you will see a little code tab right at the 00:08:18.589 --> 00:08:20.529 top. You just click that. And then you establish 00:08:20.529 --> 00:08:23.370 a workspace. You click Select Folder at the bottom 00:08:23.370 --> 00:08:26.019 left. and you point it to just an empty folder 00:08:26.019 --> 00:08:28.980 on your computer, that becomes Claude's home 00:08:28.980 --> 00:08:32.419 base for this specific project. Next, you need 00:08:32.419 --> 00:08:34.940 to install Jacob's bridge package. Because Claude 00:08:34.940 --> 00:08:37.100 Code acts like a developer inside your terminal, 00:08:37.399 --> 00:08:39.919 you do not go downloading zip files from random 00:08:39.919 --> 00:08:42.019 websites. You just type it out. Exactly. You 00:08:42.019 --> 00:08:45.120 just type a plain English prompt. You tell Claude 00:08:45.120 --> 00:08:48.240 to reach out to the internet, grab the specific 00:08:48.240 --> 00:08:51.639 notebook MCP package, and install it globally 00:08:51.639 --> 00:08:54.159 on your machine. Then you tell Claude to run 00:08:54.159 --> 00:08:56.879 a setup command to link that newly installed 00:08:56.879 --> 00:08:59.460 package directly to your Claude Code environment. 00:08:59.610 --> 00:09:02.289 Claude executes both of those requests automatically. 00:09:02.649 --> 00:09:04.690 It downloads the files, configures the bridge, 00:09:04.870 --> 00:09:07.389 and reports back when it is done. There are no 00:09:07.389 --> 00:09:09.269 confusing configuration files you have to edit 00:09:09.269 --> 00:09:11.990 manually. It's remarkably smooth. But then we 00:09:11.990 --> 00:09:14.769 hit the authentication step. You have to connect 00:09:14.769 --> 00:09:16.830 this whole system to your Google account so it 00:09:16.830 --> 00:09:19.529 can actually access Notebook LM. Right. So you 00:09:19.529 --> 00:09:22.149 ask Claude to run the authentication setup. And 00:09:22.149 --> 00:09:25.309 at that point, a visible Chrome browser window 00:09:25.309 --> 00:09:28.500 pops open on your screen. You can actually see 00:09:28.500 --> 00:09:31.080 it this time. Yes, you simply log in to your 00:09:31.080 --> 00:09:33.519 Google account just like normal. And you just 00:09:33.519 --> 00:09:36.440 let it load. Once the login is completely finished, 00:09:37.059 --> 00:09:39.539 you tap back over to Claude code, press enter, 00:09:39.919 --> 00:09:42.700 and it saves a secure cookie. From that point 00:09:42.700 --> 00:09:45.679 on, Claude can access Notebook LM without ever 00:09:45.679 --> 00:09:47.919 asking for your password again. But this is exactly 00:09:47.919 --> 00:09:50.509 where we need to pause. And talk about the risks. 00:09:50.769 --> 00:09:53.190 Yes. I want to push back on this step heavily 00:09:53.190 --> 00:09:55.909 because the source material highlights a crucial 00:09:55.909 --> 00:09:58.809 warning that we simply cannot skip. No, we cannot. 00:09:59.129 --> 00:10:01.850 This entire process relies on browser automation. 00:10:02.190 --> 00:10:04.529 It is literally controlling a Chrome browser 00:10:04.529 --> 00:10:07.929 to click buttons. And Google has very sophisticated 00:10:07.929 --> 00:10:10.990 detection systems designed to catch automated 00:10:10.990 --> 00:10:14.110 robotic behavior. It is a very real risk. Google 00:10:14.110 --> 00:10:16.690 systems might see a browser clicking things perfectly 00:10:16.690 --> 00:10:20.090 every 0 .1 seconds and flag the account for unusual 00:10:20.090 --> 00:10:22.029 activity. So the implication here is that you 00:10:22.029 --> 00:10:24.769 could lose access to your account. I strongly 00:10:24.769 --> 00:10:28.190 advise you, do not use your primary personal 00:10:28.190 --> 00:10:30.990 Gmail or your main work account for this setup. 00:10:31.110 --> 00:10:34.250 Never. Create a dedicated separate Google account 00:10:34.250 --> 00:10:37.950 just for this automated research. Keep it entirely 00:10:37.950 --> 00:10:40.789 isolated. It completely removes the risk of your 00:10:40.789 --> 00:10:43.789 main emails or photos getting locked up in a 00:10:43.789 --> 00:10:46.669 Google security sweep. That is a very smart boundary 00:10:46.669 --> 00:10:49.090 to set. Just treat it like a burner account for 00:10:49.090 --> 00:10:51.789 your AI assistant. Exactly. So once you have 00:10:51.789 --> 00:10:54.409 authenticated that dedicated account, your next 00:10:54.409 --> 00:10:57.529 step isn't to build a massive project. It is 00:10:57.529 --> 00:11:00.649 to test the plumbing. You can have CROD run a 00:11:00.649 --> 00:11:02.750 system health check. You just tell it to run 00:11:02.750 --> 00:11:04.669 its doctor protocol and report back if any of 00:11:04.669 --> 00:11:06.289 the connections are broken. But you also need 00:11:06.289 --> 00:11:08.850 a real -world test before you try to automate 00:11:08.850 --> 00:11:11.549 your entire life. Verify that the ghost in the 00:11:11.549 --> 00:11:14.409 machine can actually see your notebook LM account. 00:11:14.889 --> 00:11:16.970 What is the single biggest mistake people make 00:11:16.970 --> 00:11:19.230 right after installing this? Assuming it worked 00:11:19.230 --> 00:11:21.070 without running the quick connection test to 00:11:21.070 --> 00:11:23.570 list recent notebooks. Exactly. They assume the 00:11:23.570 --> 00:11:25.429 installation was flawless. They try to build 00:11:25.429 --> 00:11:28.299 a complex workflow and it completely fails. you 00:11:28.299 --> 00:11:31.120 must test it first. Just type into Claude, access 00:11:31.120 --> 00:11:33.340 my Notebook LM account, and list the three most 00:11:33.340 --> 00:11:35.980 recent notebooks I have. If Claude types back 00:11:35.980 --> 00:11:38.220 the exact names of your notebooks, you know the 00:11:38.220 --> 00:11:40.480 bridge is solid. A simple two -minute check saves 00:11:40.480 --> 00:11:44.279 you hours of debugging. Right. So we have laid 00:11:44.279 --> 00:11:47.379 the foundation here. We have successfully bridged 00:11:47.379 --> 00:11:50.360 Claude code's action -taking ability with Notebook 00:11:50.360 --> 00:11:53.470 LM's analytical engine. We are no longer hand 00:11:53.470 --> 00:11:55.850 -cranking the Model T. Now we build the autonomous 00:11:55.850 --> 00:11:58.330 machine. This is where the friction of learning 00:11:58.330 --> 00:12:01.830 just completely evaporates. We have got three 00:12:01.830 --> 00:12:05.269 core real -world use cases from our sources to 00:12:05.269 --> 00:12:06.850 break down. Let's start with the most common 00:12:06.850 --> 00:12:09.990 one. Researching a totally new topic from scratch. 00:12:10.409 --> 00:12:12.470 Imagine there is a subject you need to get smart 00:12:12.470 --> 00:12:14.980 on fast. But you do not have the time to hunt 00:12:14.980 --> 00:12:17.379 down the articles yourself. You can give Claude 00:12:17.379 --> 00:12:20.259 one highly specific directive. You open Claude 00:12:20.259 --> 00:12:22.580 code and type, search the web for recent YouTube 00:12:22.580 --> 00:12:24.919 videos and highly rated articles about AI agents 00:12:24.919 --> 00:12:27.179 published in the last 30 days. So you are giving 00:12:27.179 --> 00:12:30.059 it the boundaries of the research. But then you 00:12:30.059 --> 00:12:32.659 instruct it on what to do with that data, evaluate 00:12:32.659 --> 00:12:35.960 them, and load the five best sources into a new 00:12:35.960 --> 00:12:39.059 notebook, LM notebook called AI Agents Research, 00:12:39.539 --> 00:12:44.019 April 2026. And finally, you demand the exact 00:12:44.019 --> 00:12:46.679 output you want to see. You say, once those specific 00:12:46.679 --> 00:12:50.059 sources are loaded, generate a slide deck. I 00:12:50.059 --> 00:12:52.980 want a clean, dark background, exactly eight 00:12:52.980 --> 00:12:55.720 slides written in a professional corporate tone. 00:12:56.059 --> 00:12:58.639 It is entirely hands -off. Claude takes that 00:12:58.639 --> 00:13:00.700 prompt, goes out to the web, curates the sources, 00:13:01.200 --> 00:13:03.600 silently opens that ghost browser, creates the 00:13:03.600 --> 00:13:06.639 notebook, uploads every link, and triggers the 00:13:06.639 --> 00:13:08.600 slide generation. Wow. You just sit back and 00:13:08.600 --> 00:13:10.440 review the final presentation. The second use 00:13:10.440 --> 00:13:12.940 case brings us a bit closer to home. You do not 00:13:12.940 --> 00:13:15.120 always need to search the web. You can turn your 00:13:15.120 --> 00:13:17.960 own messy local files into polished presentations. 00:13:18.059 --> 00:13:20.139 Oh, that is huge. Let's say you just finished 00:13:20.139 --> 00:13:22.960 a chaotic brainstorm. You have a raw markdown 00:13:22.960 --> 00:13:25.220 file on your desktop full of unorganized meeting 00:13:25.220 --> 00:13:27.519 notes. You do not want to manually copy and paste 00:13:27.519 --> 00:13:29.379 all that text. You just point Claude to the local 00:13:29.379 --> 00:13:31.700 file path on your computer. Right. You prompt 00:13:31.700 --> 00:13:34.100 it simply. You say, take the markdown file, sitting 00:13:34.100 --> 00:13:37.379 at this specific document's path, silently upload 00:13:37.379 --> 00:13:39.419 it as a source to a brand new notebook called 00:13:39.419 --> 00:13:42.000 Content Strategy Deck. Then, generate a seven 00:13:42.000 --> 00:13:44.299 slide presenter deck from it using a minimal 00:13:44.299 --> 00:13:46.480 light design. Claude reaches into your local 00:13:46.480 --> 00:13:49.779 folder, reads the draft, uploads it, and builds 00:13:49.779 --> 00:13:53.360 the visual carousel. It bridges your local hard 00:13:53.360 --> 00:13:55.740 drive directly to Google's AI without you clicking 00:13:55.740 --> 00:13:58.059 a single thing. It is absolutely brilliant for 00:13:58.059 --> 00:14:01.600 summarizing dense internal reports or turning 00:14:01.600 --> 00:14:04.200 draft scripts into visuals. But the third use 00:14:04.200 --> 00:14:06.950 case. This is the one that really shifts how 00:14:06.950 --> 00:14:09.649 we absorb information. I agree. It is focused 00:14:09.649 --> 00:14:12.610 entirely on deep accelerated learning. Learning 00:14:12.610 --> 00:14:15.250 complex new frameworks or really technical skills. 00:14:15.350 --> 00:14:17.690 Right. You can feed it incredibly dense study 00:14:17.690 --> 00:14:21.070 materials. Let's say you have three heavy technical 00:14:21.070 --> 00:14:23.629 PDFs on Python programming sitting in your downloads 00:14:23.629 --> 00:14:26.049 folder. You tell Claude to create a notebook 00:14:26.049 --> 00:14:29.179 from those exact files. Then you prompt it. Generate 00:14:29.179 --> 00:14:32.399 a set of 30 study flashcards, focus purely on 00:14:32.399 --> 00:14:35.320 definitions and syntax. And while you are at 00:14:35.320 --> 00:14:38.139 it, generate a mind map showing how all these 00:14:38.139 --> 00:14:40.620 Python concepts connect to each other. The true 00:14:40.620 --> 00:14:43.240 genius here isn't just that it makes flashcards. 00:14:43.679 --> 00:14:46.580 It is how Notebook LM grounds the information. 00:14:47.220 --> 00:14:49.559 Every single flashcard it generates contains 00:14:49.559 --> 00:14:52.200 a specific citation. Right. It links directly 00:14:52.200 --> 00:14:54.340 back to the exact page and the exact PDF you 00:14:54.340 --> 00:14:56.820 uploaded. Yeah. Think about the feedback loop 00:14:56.820 --> 00:15:00.039 there. If you are studying... and you get a complex 00:15:00.039 --> 00:15:02.820 syntax question wrong, you just click the citation 00:15:02.820 --> 00:15:05.039 link on the flashcard. And boom, you are there. 00:15:05.240 --> 00:15:07.679 You are instantly looking at the original source 00:15:07.679 --> 00:15:10.679 context. It is a perfectly closed loop study 00:15:10.679 --> 00:15:12.320 system. You never have to go digging through 00:15:12.320 --> 00:15:14.559 a 400 page textbook to figure out why you were 00:15:14.559 --> 00:15:17.000 wrong. Whoa. I mean, imagine scaling that up. 00:15:17.279 --> 00:15:19.679 Imagine waking up every single morning to a custom 00:15:19.679 --> 00:15:22.539 10 minute audio briefing on decentralized finance 00:15:22.539 --> 00:15:24.779 generated entirely from the deepest research 00:15:24.779 --> 00:15:27.320 papers while you were sleeping. A daily? highly 00:15:27.320 --> 00:15:29.700 personalized podcast built just for you from 00:15:29.700 --> 00:15:32.220 the exact sources you trust. It is incredible. 00:15:32.220 --> 00:15:34.259 It really is. But that brings up a practical 00:15:34.259 --> 00:15:36.720 issue. If we are automating this to run every 00:15:36.720 --> 00:15:39.740 day, we want the outputs to look and sound right 00:15:39.740 --> 00:15:42.799 consistently. True. We do not want a dark corporate 00:15:42.799 --> 00:15:46.159 slide deck on Monday and a neon pink messy deck 00:15:46.159 --> 00:15:49.899 on Tuesday. How do you prevent Claude from generating 00:15:49.899 --> 00:15:52.840 totally random slide designs every time? Tell 00:15:52.840 --> 00:15:55.639 Claude to save a specific slide design as a name 00:15:55.639 --> 00:15:58.299 style for later. That is a phenomenal pro tip 00:15:58.299 --> 00:16:00.820 right there. If Claude accidentally generates 00:16:00.820 --> 00:16:03.320 a slide design you absolutely love, you can just 00:16:03.320 --> 00:16:05.639 tell it to save that aesthetic as a name style 00:16:05.639 --> 00:16:08.200 in its memory. Right, exactly. Then in your daily 00:16:08.200 --> 00:16:10.299 prompts, you just say, use my standard corporate 00:16:10.299 --> 00:16:12.740 style. It makes the system significantly faster 00:16:12.740 --> 00:16:15.139 and totally consistent. Which perfectly sets 00:16:15.139 --> 00:16:17.620 up. the final piece of this whole puzzle. We 00:16:17.620 --> 00:16:19.500 do not just want to run these commands manually 00:16:19.500 --> 00:16:21.759 every morning. We need to talk about scheduling, 00:16:22.139 --> 00:16:24.580 system limits, and making this run on an invisible 00:16:24.580 --> 00:16:28.259 timer. We want true autonomy. And Claude Code 00:16:28.259 --> 00:16:30.360 actually has a built -in scheduling system to 00:16:30.360 --> 00:16:32.480 handle this using something called cron jobs. 00:16:32.759 --> 00:16:35.580 Cron jobs are basically just the standard computing 00:16:35.580 --> 00:16:37.980 method for running tasks on a precise timer. 00:16:38.720 --> 00:16:40.919 The phrase sounds highly technical, but Claude 00:16:40.919 --> 00:16:43.440 abstracts all the complexity away. You literally 00:16:43.440 --> 00:16:45.440 just type natural language into the terminal. 00:16:46.080 --> 00:16:48.940 You say, create a scheduled task that runs every 00:16:48.940 --> 00:16:52.629 single morning at 8 a .m. And you add, the task 00:16:52.629 --> 00:16:55.450 should search for the top five AI news stories 00:16:55.450 --> 00:16:58.330 from the past 24 hours, load them into a new 00:16:58.330 --> 00:17:01.350 notebook named after today's date, then generate 00:17:01.350 --> 00:17:04.410 both a slide deck summary and a two -host audio 00:17:04.410 --> 00:17:06.849 overview. It is programming through conversation. 00:17:07.109 --> 00:17:09.970 But as we look at this, there is a massive physical 00:17:09.970 --> 00:17:12.690 limitation to the system that we absolutely must 00:17:12.690 --> 00:17:15.130 discuss. Oh, definitely. A local scheduled task 00:17:15.130 --> 00:17:17.289 only runs when your computer is physically awake. 00:17:17.529 --> 00:17:20.450 Right. If you set that task for 8am, but your 00:17:20.450 --> 00:17:23.369 laptop is closed and asleep in your bag, the 00:17:23.369 --> 00:17:26.170 task simply will not run. It is helpful to think 00:17:26.170 --> 00:17:28.710 of a local task, like an eager physical assistant 00:17:28.710 --> 00:17:30.970 sitting in your office. They are fully prepped 00:17:30.970 --> 00:17:33.990 and ready to work. Yeah. But they cannot do the 00:17:33.990 --> 00:17:36.170 morning research if you lock them inside a dark 00:17:36.170 --> 00:17:38.250 room over the weekend. They need the machine 00:17:38.250 --> 00:17:40.410 to be awake. You essentially have two options 00:17:40.410 --> 00:17:43.329 to solve this. The first is brute force. You 00:17:43.329 --> 00:17:45.430 can just adjust your computer's power settings 00:17:45.430 --> 00:17:47.930 to ensure it never goes to sleep overnight. Which 00:17:47.930 --> 00:17:51.750 works. It is easy, but it is not exactly elegant 00:17:51.750 --> 00:17:54.460 or practical for a laptop. The second option 00:17:54.460 --> 00:17:57.740 is to use a remote task. Cloud Code supports 00:17:57.740 --> 00:18:00.599 pushing these scheduled tasks to the cloud using 00:18:00.599 --> 00:18:03.279 a GitHub integration. Oh, that is smart. Basically, 00:18:03.619 --> 00:18:05.559 your workspace and instructions are pushed to 00:18:05.559 --> 00:18:08.660 a remote server. Anthropic systems wake up, run 00:18:08.660 --> 00:18:10.940 the automation remotely, and drop the results 00:18:10.940 --> 00:18:13.319 right into your account. It takes a few extra 00:18:13.319 --> 00:18:15.140 minutes to configure the GitHub connection, but 00:18:15.140 --> 00:18:18.579 once you do, it is entirely reliable. Your actual 00:18:18.579 --> 00:18:20.980 physical laptop can be completely powered down 00:18:20.980 --> 00:18:23.299 at the bottom of your backpack, and your research 00:18:23.299 --> 00:18:26.240 machine will still run flawlessly at 8 a .m. 00:18:26.539 --> 00:18:28.759 Now, let's talk about the system limits, because 00:18:28.759 --> 00:18:30.759 eventually you are going to want to feed this 00:18:30.759 --> 00:18:33.180 thing a massive amount of data. Oh, for sure. 00:18:33.539 --> 00:18:35.680 Notebook LM currently operates on a free tier 00:18:35.680 --> 00:18:38.579 and a newly introduced paid tier. The free tier 00:18:38.579 --> 00:18:42.119 is actually Remarkably generous. It allows up 00:18:42.119 --> 00:18:45.380 to 50 individual sources per notebook. If you 00:18:45.380 --> 00:18:48.500 hit that limit, the paid plans, which run around 00:18:48.500 --> 00:18:53.019 $14 to $20 a month, allow up to 300 sources per 00:18:53.019 --> 00:18:55.839 single notebook. But the real shocker to me is 00:18:55.839 --> 00:18:59.140 the sheer size allowed for each source. Each 00:18:59.140 --> 00:19:01.799 individual document you upload can be up to 500 00:19:01.799 --> 00:19:05.299 ,000 words. That translates to roughly 1 million 00:19:05.299 --> 00:19:08.019 tokens per document. To put that in perspective, 00:19:08.339 --> 00:19:11.059 that comfortably covers almost any massive academic 00:19:11.059 --> 00:19:13.460 research paper, a dense legal filing, or even 00:19:13.460 --> 00:19:15.740 a full -length textbook. You could feed it entire 00:19:15.740 --> 00:19:19.359 libraries of data day after day. But as with 00:19:19.359 --> 00:19:21.720 any complex system, things will inevitably break. 00:19:21.980 --> 00:19:23.859 They always do. Websites change their layouts, 00:19:24.180 --> 00:19:26.420 networks drop, automation always encounters some 00:19:26.420 --> 00:19:28.880 kind of friction. If I set a task for eight in 00:19:28.880 --> 00:19:31.079 the morning and it fails, how do I find out what 00:19:31.079 --> 00:19:33.599 broke? Just ask Claude in the terminal what happened 00:19:33.599 --> 00:19:35.980 and it explains the error log. It is entirely 00:19:35.980 --> 00:19:37.759 self -diagnosing. You do not have to go digging 00:19:37.759 --> 00:19:39.640 through lines of error code. No, thankfully. 00:19:40.099 --> 00:19:42.900 Claude code automatically logs any failures in 00:19:42.900 --> 00:19:45.559 your terminal history. You just type, what happened 00:19:45.559 --> 00:19:48.000 with my scheduled task from this morning? It 00:19:48.000 --> 00:19:50.740 reads its own failure log and explains the issue 00:19:50.740 --> 00:19:53.380 back to you in plain English. It is a remarkably 00:19:53.380 --> 00:19:56.480 resilient setup once you get it dialed in. We 00:19:56.480 --> 00:19:58.740 have covered a massive amount of ground today. 00:19:58.990 --> 00:20:01.390 Let's slow down for a moment here. I want to 00:20:01.390 --> 00:20:03.549 synthesize the philosophy behind all of this 00:20:03.549 --> 00:20:06.029 because I think it is really easy to get lost 00:20:06.029 --> 00:20:08.930 in the technical mechanics of packages and background 00:20:08.930 --> 00:20:11.140 browsers. Very easy. We are not just sharing 00:20:11.140 --> 00:20:13.779 a neat software trick today. We are looking at 00:20:13.779 --> 00:20:16.559 a true paradigm shift in how we handle human 00:20:16.559 --> 00:20:19.339 knowledge. The old way was entirely defined by 00:20:19.339 --> 00:20:22.059 manual labor. Your own two hands were the bottleneck 00:20:22.059 --> 00:20:24.740 to your understanding. Exactly. By bridging a 00:20:24.740 --> 00:20:27.500 relentless action taker like Claude with an analytical 00:20:27.500 --> 00:20:30.299 engine like Notebook LM, you evolve. You shift 00:20:30.299 --> 00:20:32.440 from being the doer, the person clicking, copying, 00:20:32.579 --> 00:20:35.079 and pasting, to being the director. You simply 00:20:35.079 --> 00:20:37.519 set the strategic goal, you define the boundaries, 00:20:37.779 --> 00:20:39.829 and then you just review the synthesized knowledge 00:20:39.829 --> 00:20:41.769 that is handed back to you. You are managing 00:20:41.769 --> 00:20:44.210 the AI. You are not operating the software anymore. 00:20:44.309 --> 00:20:47.809 You free up your cognitive load entirely. Your 00:20:47.809 --> 00:20:50.849 mind is no longer exhausted by the mere act of 00:20:50.849 --> 00:20:53.289 gathering the information. You actually have 00:20:53.289 --> 00:20:55.829 the energy and the clarity to apply that information 00:20:55.829 --> 00:20:58.650 to your life or your business. Don't just listen 00:20:58.650 --> 00:21:00.829 to this deep dive and nod along. I have a challenge 00:21:00.829 --> 00:21:05.890 for you this week. Pick one topic, just one topic 00:21:05.890 --> 00:21:08.859 that you are intensely curious about right now. 00:21:09.160 --> 00:21:11.299 Something that genuinely excites you. Run the 00:21:11.299 --> 00:21:14.119 quick connection test, give Claude one highly 00:21:14.119 --> 00:21:17.279 specific prompt to research that topic, and watch 00:21:17.279 --> 00:21:20.539 it build your notebook automatically. The absolute 00:21:20.539 --> 00:21:22.900 best way to understand the power of the system 00:21:22.900 --> 00:21:25.460 is to sit back and watch the ghost in the machine 00:21:25.460 --> 00:21:27.519 work right in front of you. It fundamentally 00:21:27.519 --> 00:21:29.619 changes your perspective on what is possible 00:21:29.619 --> 00:21:32.420 with a computer, but also leaves me with a lingering 00:21:32.420 --> 00:21:34.339 thought. I want to pass directly to you before 00:21:34.339 --> 00:21:36.900 we go. If the friction of gathering, reading, 00:21:37.140 --> 00:21:39.440 and organizing knowledge completely disappears 00:21:39.440 --> 00:21:42.539 from our lives, what happens to our human curiosity? 00:21:43.579 --> 00:21:46.220 Does this frictionless access make us more deeply 00:21:46.220 --> 00:21:48.980 inquisitive, allowing us to ask bigger, harder 00:21:48.980 --> 00:21:52.160 questions? Or do we risk outsourcing our fundamental 00:21:52.160 --> 00:21:54.319 sense of wonder to the machines doing the reading 00:21:54.319 --> 00:21:56.640 for us? It's something to ponder as you build 00:21:56.640 --> 00:21:58.819 your own machine. Fascinating question. Until 00:21:58.819 --> 00:21:59.380 next time.