WEBVTT 00:00:00.000 --> 00:00:03.080 I still wrestle with AI news overwhelm myself. 00:00:03.259 --> 00:00:05.620 Oh yeah. That is incredibly common right now. 00:00:05.740 --> 00:00:07.860 I mean every single morning brings another brand 00:00:07.860 --> 00:00:10.300 new tool. Right. It feels completely exhausting 00:00:10.300 --> 00:00:13.179 to always be behind. You open your browser and 00:00:13.179 --> 00:00:17.280 just feel totally lost. The landscape moves way 00:00:17.280 --> 00:00:20.059 faster than humans can read. Exactly. We simply 00:00:20.059 --> 00:00:22.820 cannot process that volume of daily information. 00:00:23.019 --> 00:00:26.140 It's impossible. Welcome to the Deep Dive. Today 00:00:26.140 --> 00:00:28.960 our main mission is escaping that information 00:00:28.960 --> 00:00:31.719 mess. We are exploring a completely different 00:00:31.719 --> 00:00:34.619 approach to AI. We are turning Claude Cowork 00:00:34.619 --> 00:00:37.579 into your automated team. It researches, analyzes, 00:00:37.820 --> 00:00:39.759 and formats data while you sleep. Yeah, it's 00:00:39.759 --> 00:00:41.659 kind of like stacking Lego blocks of workflow 00:00:41.659 --> 00:00:44.200 together. OK, let's unpack this. So we are moving 00:00:44.200 --> 00:00:47.359 far past the simple chatbot interface. We're 00:00:47.359 --> 00:00:49.899 building a localized digital workforce for you. 00:00:50.020 --> 00:00:52.719 It lives directly on your local computer desktop, 00:00:52.799 --> 00:00:54.799 right? Exactly. Currently, it's available for 00:00:54.799 --> 00:00:57.460 pro and team paid plans. Before building, we 00:00:57.460 --> 00:00:59.619 have to deeply understand this paradigm shift. 00:00:59.939 --> 00:01:02.460 This system is fundamentally different from opening 00:01:02.460 --> 00:01:04.900 a web browser. That's a crucial mental shift 00:01:04.900 --> 00:01:07.590 you must make. It acts entirely differently than 00:01:07.590 --> 00:01:11.189 a normal website does. You access it via a special 00:01:11.189 --> 00:01:14.150 dedicated co -work tab. It isn't just another 00:01:14.150 --> 00:01:16.489 browser tab you forget about. It sits at the 00:01:16.489 --> 00:01:18.769 core of your operating system. Right. It actually 00:01:18.769 --> 00:01:21.969 sees your local files directly. It seamlessly 00:01:21.969 --> 00:01:24.549 talks to your other installed applications. Wow. 00:01:24.840 --> 00:01:27.060 So it can read documents right off your hard 00:01:27.060 --> 00:01:30.180 drive? Yeah. And it runs complex tasks in the 00:01:30.180 --> 00:01:32.859 background. It continues working while you are 00:01:32.859 --> 00:01:35.280 looking completely away. You do not have to babysit 00:01:35.280 --> 00:01:37.340 the application anymore. I think of the main 00:01:37.340 --> 00:01:40.400 difference like this. Regular Claude is a smart 00:01:40.400 --> 00:01:43.299 friend you can call. Okay. But Cowork is a smart 00:01:43.299 --> 00:01:45.819 assistant sitting directly beside you. It actually 00:01:45.819 --> 00:01:47.579 has his hands on your own keyboard. It knows 00:01:47.579 --> 00:01:49.719 exactly where you keep your Notion pages. Yes. 00:01:49.879 --> 00:01:51.939 It knows who your important clients actually 00:01:51.939 --> 00:01:54.299 are. And it absolutely never gets tired of doing 00:01:54.299 --> 00:01:57.260 research. There are five core blocks to this 00:01:57.260 --> 00:02:00.540 entire system. You really must understand each 00:02:00.540 --> 00:02:02.879 part for this to work. Let us walk through these 00:02:02.879 --> 00:02:05.159 blocks organically. You start by setting up your 00:02:05.159 --> 00:02:07.439 distinct digital rooms. We call these different 00:02:07.439 --> 00:02:09.900 spaces your folders. You put all related project 00:02:09.900 --> 00:02:12.599 files into one office. Right. Next, you have 00:02:12.599 --> 00:02:15.740 to give the system some rules. Skills act as 00:02:15.740 --> 00:02:18.719 your saved permanent operational playbooks. Then 00:02:18.719 --> 00:02:21.759 we get to the actual brain of the system. Memory 00:02:21.759 --> 00:02:24.099 is where Claude learns deeply about your work. 00:02:25.240 --> 00:02:27.860 From there, we need to export the finished work. 00:02:28.380 --> 00:02:30.400 Connectors act as the digital bridges to other 00:02:30.400 --> 00:02:33.439 apps. Finally, automation is the set and forget 00:02:33.439 --> 00:02:35.599 scheduling system. Right, and if you skip even 00:02:35.599 --> 00:02:38.270 one block... The illusion breaks. The system 00:02:38.270 --> 00:02:40.389 just goes back to acting like a generic chatbot. 00:02:40.590 --> 00:02:42.930 The memory feature is the most fascinating part 00:02:42.930 --> 00:02:46.569 to me. It fixes the generic robot problem completely. 00:02:46.849 --> 00:02:49.629 It really does. I mean, using AI without memory 00:02:49.629 --> 00:02:52.310 is like hiring a brilliant employee, but refusing 00:02:52.310 --> 00:02:54.370 to tell them what your company actually makes. 00:02:54.490 --> 00:02:56.789 That captures the core problem perfectly. Haha. 00:02:56.990 --> 00:02:59.810 Yeah. Memory uses global instructions to answer 00:02:59.810 --> 00:03:01.990 the question, who am I? You put your general 00:03:01.990 --> 00:03:03.909 preferences and business goals there. Right. 00:03:04.129 --> 00:03:07.310 But you also have specific localized folder instructions. 00:03:07.689 --> 00:03:11.189 Which answer? What is this specific project? 00:03:11.669 --> 00:03:14.889 Yes. It tells Claude exactly what to focus on 00:03:14.889 --> 00:03:17.810 today. How do we prevent the AI from giving us 00:03:17.810 --> 00:03:21.629 generic corporate jargon? Well, you must use 00:03:21.629 --> 00:03:25.990 I and B in global instructions. State your exact 00:03:25.990 --> 00:03:28.490 business goals from a first -person perspective. 00:03:29.010 --> 00:03:32.590 So first -person pronouns force the AI into your 00:03:32.590 --> 00:03:35.370 specific perspective. Exactly. It completely 00:03:35.370 --> 00:03:37.590 changes the tone of the final output. Now that 00:03:37.590 --> 00:03:39.810 Claude knows exactly who you are, it needs the 00:03:39.810 --> 00:03:41.830 permanent rulebook. Right. You do not want to 00:03:41.830 --> 00:03:44.169 repeat yourself every single morning. That completely 00:03:44.169 --> 00:03:46.289 defeats the underlying purpose of true automation. 00:03:46.469 --> 00:03:48.909 That is exactly what a skill is designed for. 00:03:49.069 --> 00:03:51.509 Yeah. A skill is basically just a saved, refined 00:03:51.509 --> 00:03:53.889 prompt. Instead of explaining your needs daily, 00:03:54.009 --> 00:03:56.889 you just call the skill. OK, let me play double's 00:03:56.889 --> 00:03:58.789 advocate here for a second. Why go through all 00:03:58.789 --> 00:04:00.610 this trouble? What do you mean? Why not just 00:04:00.610 --> 00:04:03.289 ask Claude for a summary? Why spend 20 minutes 00:04:03.289 --> 00:04:07.069 building a custom playbook? Because basic summaries 00:04:07.069 --> 00:04:09.150 completely miss the crucial business insights. 00:04:09.830 --> 00:04:12.349 A generic summary just adds to your daily information 00:04:12.349 --> 00:04:16.170 overload. Ah, I see. You actually need real targeted 00:04:16.170 --> 00:04:19.089 insights to make better decisions. We are building 00:04:19.089 --> 00:04:21.790 the research analyzer skill today. The instructions 00:04:21.790 --> 00:04:24.930 must be incredibly specific to work well. Exactly. 00:04:25.250 --> 00:04:27.629 The instructions must specify finding emerging 00:04:27.629 --> 00:04:30.689 trends in the industry. We want it extracting 00:04:30.689 --> 00:04:33.529 hard data with real sources. It must provide 00:04:33.529 --> 00:04:36.089 a strategic so -what for the business. Right. 00:04:36.230 --> 00:04:38.889 And it should format everything with clean, readable 00:04:38.889 --> 00:04:41.990 bullet points. absolutely no corporate jargon 00:04:41.990 --> 00:04:44.970 is ever allowed. Testing this skill is a crucial 00:04:44.970 --> 00:04:47.709 step in the process. You run a long, complex 00:04:47.709 --> 00:04:49.970 TechCrunch article through it first. Yeah, you 00:04:49.970 --> 00:04:52.290 really want to see if it being lazy. You need 00:04:52.290 --> 00:04:54.290 to ensure it follows the complex instructions. 00:04:54.649 --> 00:04:57.350 There are also powerful pre -built plugins available 00:04:57.350 --> 00:05:00.050 to use, like the productivity plugin for your 00:05:00.050 --> 00:05:03.410 tasks and calendar. and web search plugins can 00:05:03.410 --> 00:05:05.730 combine with your custom skills. What do we do 00:05:05.730 --> 00:05:07.930 if the output feels too vague during a trial 00:05:07.930 --> 00:05:10.509 run? Just go back into the skill settings and 00:05:10.509 --> 00:05:13.129 add specifics. You might add a line like, always 00:05:13.129 --> 00:05:15.730 look for funding rounds. Treat vague outputs 00:05:15.730 --> 00:05:19.189 as a missing rule and update the playbook. Precisely. 00:05:19.529 --> 00:05:22.449 It learns exactly what you want over time. Skills 00:05:22.449 --> 00:05:26.490 are undeniably powerful tools, but manually triggering 00:05:26.490 --> 00:05:29.810 them is still annoying manual labor. Yeah. Let 00:05:29.810 --> 00:05:32.230 us make the machine work while you are fast asleep. 00:05:32.490 --> 00:05:34.389 This is the real magic of the automation phase. 00:05:34.750 --> 00:05:36.769 You use the schedule command to set automatic 00:05:36.769 --> 00:05:39.730 tasks. You just toggle a few switches for folders, 00:05:40.050 --> 00:05:42.290 connectors, and instructions. You could schedule 00:05:42.290 --> 00:05:45.310 your entire research system for 6 a .m. But there's 00:05:45.310 --> 00:05:47.930 one major operational catch you must remember. 00:05:48.610 --> 00:05:50.990 Because it runs locally, the computer cannot 00:05:50.990 --> 00:05:53.639 be asleep. Here's where it gets really interesting. 00:05:53.740 --> 00:05:57.100 Yeah. The laptop lid absolutely cannot be closed 00:05:57.100 --> 00:05:59.019 during this time. Your physical machine must 00:05:59.019 --> 00:06:01.240 stay active during the scheduled window. Right. 00:06:01.579 --> 00:06:04.379 Once it is awake, you can run parallel multitasking. 00:06:05.100 --> 00:06:08.040 Unlike older sequential AI, Cowork runs multiple 00:06:08.040 --> 00:06:10.699 tasks simultaneously. It can run three to five 00:06:10.699 --> 00:06:13.120 tasks at the exact same time. It is exactly like 00:06:13.120 --> 00:06:15.680 having a whole team of assistants. It can check 00:06:15.680 --> 00:06:18.040 Google algorithm updates in one distinct stream. 00:06:18.160 --> 00:06:20.759 It can find new YouTube productivity videos in 00:06:20.759 --> 00:06:22.959 another. and analyze competitor newsletters all 00:06:22.959 --> 00:06:26.779 at the very same time. But there is a physical 00:06:26.779 --> 00:06:28.939 hardware limit here. It uses your actual computer 00:06:28.939 --> 00:06:31.459 processing power to function. Yeah. Running 20 00:06:31.459 --> 00:06:34.120 tasks might make your laptop fans scream loudly. 00:06:34.800 --> 00:06:36.939 Three to five is the ideal operational sweet 00:06:36.939 --> 00:06:39.600 spot. It is fascinating that the limitation isn't 00:06:39.600 --> 00:06:43.160 the AI's brain. The real bottleneck is your physical 00:06:43.160 --> 00:06:46.899 laptop's cooling fan. It really grounds the abstract 00:06:46.899 --> 00:06:49.920 technology in physical reality. What is the absolute 00:06:49.920 --> 00:06:52.560 biggest mistake people make on their first automated 00:06:52.560 --> 00:06:54.759 schedule? Oh, they close their laptop lid when 00:06:54.759 --> 00:06:57.639 they go to bed, the computer sleeps, and the 00:06:57.639 --> 00:07:00.600 entire AI automation dies instantly. A closed 00:07:00.600 --> 00:07:03.240 laptop lid instantly kills your entire automated 00:07:03.240 --> 00:07:05.399 morning routine. This is a very frustrating lesson 00:07:05.399 --> 00:07:07.720 to learn the hard way. Sponsor. So the massive 00:07:07.720 --> 00:07:09.959 amount of data is processed perfectly at 6am. 00:07:10.240 --> 00:07:13.180 But where does it actually go after the AI finishes? 00:07:13.720 --> 00:07:15.959 Data trapped inside a chat window is just more 00:07:15.959 --> 00:07:18.879 noise. That is exactly why we deeply need connectors. 00:07:19.139 --> 00:07:21.680 They are digital bridges to your real daily workspaces. 00:07:21.930 --> 00:07:24.329 You can connect Claw directly to your personal 00:07:24.329 --> 00:07:27.310 Notion workspace. You create a professional research 00:07:27.310 --> 00:07:29.970 database inside Notion very easily. It tracks 00:07:29.970 --> 00:07:32.790 the source, key insight, and date incredibly 00:07:32.790 --> 00:07:35.250 cleanly. Everything stays beautifully organized 00:07:35.250 --> 00:07:37.430 and fully automated. You can also connect it 00:07:37.430 --> 00:07:40.649 directly to your Slack workspace. This is a total 00:07:40.649 --> 00:07:43.430 game changer. It sends daily bullet point briefings 00:07:43.430 --> 00:07:45.850 to a private channel automatically. You just 00:07:45.850 --> 00:07:47.509 read them on your phone over morning breakfast. 00:07:47.730 --> 00:07:51.050 Then there is the very powerful Zapier MCP protocol. 00:07:51.259 --> 00:07:53.740 Can you define what that actually is for us? 00:07:54.079 --> 00:07:56.959 A digital skeleton key letting AI talk to thousands 00:07:56.959 --> 00:07:59.899 of apps. A digital skeleton key letting AI talk 00:07:59.899 --> 00:08:03.079 to thousands of apps. That is a perfect, simple 00:08:03.079 --> 00:08:05.660 way to clearly describe it. It safely connects 00:08:05.660 --> 00:08:09.379 to over 8 ,000 different digital tools. But you 00:08:09.379 --> 00:08:12.139 should absolutely start slow with just notion 00:08:12.139 --> 00:08:14.480 first. Now we have to talk about the power of 00:08:14.480 --> 00:08:18.060 subagents. Yes. There are four very smart distinct 00:08:18.060 --> 00:08:20.759 patterns they use naturally. First is research 00:08:20.759 --> 00:08:23.199 parallelism for running multiple tasks at once. 00:08:23.560 --> 00:08:26.379 Second is asset parallelism for creating several 00:08:26.379 --> 00:08:29.160 different files at once. Third is sequential 00:08:29.160 --> 00:08:32.360 chaining. This works exactly like a digital factory 00:08:32.360 --> 00:08:35.179 assembly line. The output of one agent becomes 00:08:35.179 --> 00:08:38.840 the input for the next. And fourth is batch processing 00:08:38.840 --> 00:08:41.899 for massive, demanding industry -level work. 00:08:42.179 --> 00:08:44.379 You track all this complex activity on a detailed 00:08:44.379 --> 00:08:47.200 visual dashboard. It constantly shows the exact 00:08:47.200 --> 00:08:50.740 status, progress, and activity. Two sec silence. 00:08:51.840 --> 00:08:55.230 Wow. Imagine scaling to a billion queries. or 00:08:55.230 --> 00:08:58.210 just watching it crush 10 to 20 posts in 90 short 00:08:58.210 --> 00:09:01.009 seconds. It truly saves at least an hour of tedious 00:09:01.009 --> 00:09:03.610 manual labor. With 8 ,000 apps available, what 00:09:03.610 --> 00:09:05.409 happens if we connect everything at once? The 00:09:05.409 --> 00:09:07.970 entire system becomes incredibly messy very quickly. 00:09:08.549 --> 00:09:11.090 AI hallucinations start triggering wrong automated 00:09:11.090 --> 00:09:13.370 actions everywhere across your apps. Connecting 00:09:13.370 --> 00:09:15.769 too many apps creates a tangled web of automated 00:09:15.769 --> 00:09:18.149 mistakes. Exactly. You will completely break 00:09:18.149 --> 00:09:20.509 your own daily workflow. Let us bring all these 00:09:20.509 --> 00:09:22.509 Lego blocks together into the daily routine. 00:09:22.909 --> 00:09:25.149 We need to look closely at the ultimate master 00:09:25.149 --> 00:09:27.490 workflow. It is a beautiful five -step morning 00:09:27.490 --> 00:09:30.490 routine that changes everything. First, memory 00:09:30.490 --> 00:09:33.470 knows the overriding goal is practical AI. Second, 00:09:34.009 --> 00:09:36.450 the schedule triggers the entire system at 6 00:09:36.450 --> 00:09:40.059 a .m. shoply. Third, Sub -agents aggressively 00:09:40.059 --> 00:09:43.580 search the web for five big news stories. Fourth, 00:09:43.779 --> 00:09:47.000 the research analyzer skill processes those complicated 00:09:47.000 --> 00:09:50.500 stories deeply. Finally, connectors save the 00:09:50.500 --> 00:09:52.960 detailed formatted report straight to Notion, 00:09:53.279 --> 00:09:55.440 and they send a quick readable summary directly 00:09:55.440 --> 00:09:58.100 to your Slack. You wake up and simply open Slack 00:09:58.100 --> 00:10:00.679 on your phone. You read the number one key insight 00:10:00.679 --> 00:10:03.519 immediately while drinking coffee. Then you click 00:10:03.519 --> 00:10:06.220 the provided Notion link. for the full report. 00:10:06.720 --> 00:10:09.740 You did absolutely zero manual searching or tedious 00:10:09.740 --> 00:10:11.980 summarizing yourself. You only did the high -level 00:10:11.980 --> 00:10:14.379 strategic thinking and the final writing. That 00:10:14.379 --> 00:10:16.639 is the massive difference between working too 00:10:16.639 --> 00:10:19.179 hard and having a smart team. So what does this 00:10:19.179 --> 00:10:22.019 all mean? You are no longer a manual data worker. 00:10:22.379 --> 00:10:24.700 You are now the strategic manager of a virtual 00:10:24.700 --> 00:10:27.000 team. You direct the complex flow of high -level 00:10:27.000 --> 00:10:30.080 information instead. For a listener feeling completely 00:10:30.080 --> 00:10:32.399 overwhelmed right now, what is the very first 00:10:32.399 --> 00:10:35.159 tiny step they should take today? Build one single 00:10:35.159 --> 00:10:37.360 research skill and link it directly to your Notion 00:10:37.360 --> 00:10:40.220 database. Do not try anything else yet. Start 00:10:40.220 --> 00:10:43.059 small by building one specific research skill 00:10:43.059 --> 00:10:46.460 and connecting it. Yes. Keep it incredibly simple 00:10:46.460 --> 00:10:48.860 until you fully trust the system. Well, let's 00:10:48.860 --> 00:10:51.840 wrap this up. Sure. The big idea of this deep 00:10:51.840 --> 00:10:54.240 dive is that Claude Cowork isn't a web tool. 00:10:54.620 --> 00:10:57.940 It is a localized parallel processing team. Built 00:10:57.940 --> 00:11:00.620 solidly on skills, memory, and connectors. And 00:11:00.620 --> 00:11:02.759 it works tirelessly while you are fast asleep. 00:11:03.250 --> 00:11:05.769 If your localized AI is doing all the collecting, 00:11:05.950 --> 00:11:07.809 summarizing, and formatting before you even pour 00:11:07.809 --> 00:11:11.049 your coffee, what uniquely human skills should 00:11:11.049 --> 00:11:13.669 you be spending your newly freed up time on today? 00:11:14.169 --> 00:11:16.389 Beat. Thank you for joining us on this deep dive. 00:11:16.450 --> 00:11:18.710 Hey Curious, see you next time. Out. T -Row music.