WEBVTT 00:00:00.000 --> 00:00:02.220 Imagine waking up to find your morning emails 00:00:02.220 --> 00:00:05.019 already read. Your Slack messages are beautifully 00:00:05.019 --> 00:00:07.559 summarized. Yeah, that sounds like a total dream. 00:00:07.879 --> 00:00:10.240 Right. And your calendar is perfectly sorted. 00:00:10.900 --> 00:00:13.820 But not by a human assistant. Yeah. This was 00:00:13.820 --> 00:00:16.399 done by your laptop. It was working silently 00:00:16.399 --> 00:00:19.539 in the dark while you slept. It is wild to even 00:00:19.539 --> 00:00:22.359 think about. It really is. Today we are looking 00:00:22.359 --> 00:00:26.609 at the April 2026 Claude Cowork update. It completely 00:00:26.609 --> 00:00:30.370 turned a simple chatbot into a 24 -7 autonomous 00:00:30.370 --> 00:00:33.750 desktop OS. Welcome to the Deep Dive. Thanks 00:00:33.750 --> 00:00:35.649 for having me. We are definitely shifting gears 00:00:35.649 --> 00:00:37.950 here today. We are moving from just chatting 00:00:37.950 --> 00:00:40.450 with AI to actually scheduling a digital labor 00:00:40.450 --> 00:00:43.149 force. That is the true narrative through line 00:00:43.149 --> 00:00:45.030 today. Exactly. We are moving away from that 00:00:45.030 --> 00:00:47.670 simple preview window. You type a question, you 00:00:47.670 --> 00:00:50.210 get an answer. But these new updates completely 00:00:50.210 --> 00:00:53.009 flip that dynamic. It is no longer just a passive 00:00:53.009 --> 00:00:55.509 tool waiting for your input. Right. It is managing 00:00:55.509 --> 00:00:58.530 deep context now. It follows highly structured 00:00:58.530 --> 00:01:01.109 instructions. It executes ongoing workflows in 00:01:01.109 --> 00:01:03.890 the background. It is a massive behavioral shift. 00:01:04.170 --> 00:01:05.989 Let's talk about the foundational problem this 00:01:05.989 --> 00:01:09.629 solves first. The biggest hurdle with AI up until 00:01:09.629 --> 00:01:13.329 now has been amnesia. Oh, absolutely. The blank 00:01:13.329 --> 00:01:16.230 chat problem. Yeah. Every time you open a new 00:01:16.230 --> 00:01:19.010 chat window. The system just forgets everything. 00:01:19.370 --> 00:01:22.349 It resets the context completely. It is the most 00:01:22.349 --> 00:01:25.769 common complaint we see. You sit down to work 00:01:25.769 --> 00:01:27.890 and you waste half your morning just setting 00:01:27.890 --> 00:01:30.090 the stage. Right. You have to explain your business 00:01:30.090 --> 00:01:32.510 model all over again. Exactly. You explain your 00:01:32.510 --> 00:01:34.569 target audience. You have to dictate your brand's 00:01:34.569 --> 00:01:38.010 tone of voice. It is completely exhausting. It 00:01:38.010 --> 00:01:40.290 is kind of like hiring a brand new assistant 00:01:40.290 --> 00:01:42.290 every single morning. You spend two hours training 00:01:42.290 --> 00:01:44.090 them. They do an hour of work and then you fire 00:01:44.090 --> 00:01:46.510 them. Yeah, and then you repeat the entire process 00:01:46.510 --> 00:01:51.030 the very next day. The April 2026 update fixes 00:01:51.030 --> 00:01:53.590 this through a feature called Projects. And this 00:01:53.590 --> 00:01:55.989 relies entirely on scoped memory, right? Yes. 00:01:56.269 --> 00:01:58.969 Scoped memory is basically memory limited to 00:01:58.969 --> 00:02:02.250 a specific folder or context. It anchors the 00:02:02.250 --> 00:02:04.590 AI directly to your local file system. Okay, 00:02:04.629 --> 00:02:07.430 so it builds local vector embeddings. Let's define 00:02:07.430 --> 00:02:09.969 that AI jargon really quickly. Sure. It means 00:02:09.969 --> 00:02:13.210 breaking documents down into semantic mathematical 00:02:13.210 --> 00:02:16.289 chunks. So how does it actually remember things? 00:02:16.469 --> 00:02:19.650 I mean, without reading a massive PDF every single 00:02:19.650 --> 00:02:22.729 time I ask a question. That is the magic of local 00:02:22.729 --> 00:02:25.830 vector indexing. When you point the AI at a folder, 00:02:25.969 --> 00:02:28.610 it doesn't just read the text linearly. Right. 00:02:28.710 --> 00:02:31.669 It assigns mathematical values to those chunks 00:02:31.669 --> 00:02:34.469 based on their meaning. So when you ask a question 00:02:34.469 --> 00:02:37.270 later, it just pulls the exact mathematical vectors 00:02:37.270 --> 00:02:39.849 that match your... It is basically creating a 00:02:39.849 --> 00:02:42.750 highly efficient, invisible index card system. 00:02:42.990 --> 00:02:45.770 It remembers your past decisions and your style 00:02:45.770 --> 00:02:47.729 preferences. Yeah, creating a project is like 00:02:47.729 --> 00:02:50.430 giving the AI a dedicated physical office. You 00:02:50.430 --> 00:02:52.810 give it a desk, labeled folders, and a filing 00:02:52.810 --> 00:02:54.810 cabinet. And the alternative is just terrible. 00:02:55.229 --> 00:02:57.110 It is like shoving your entire professional life 00:02:57.110 --> 00:02:59.710 into a messy backpack every morning. Exactly. 00:02:59.710 --> 00:03:01.669 You throw your tax spreadsheets in with your 00:03:01.669 --> 00:03:03.889 creative writing. You can never find what you 00:03:03.889 --> 00:03:05.969 actually need. So you can start these projects 00:03:05.969 --> 00:03:09.110 from scratch or import an older setup. Or you 00:03:09.110 --> 00:03:10.770 can just point co -work at an existing folder 00:03:10.770 --> 00:03:13.139 on your computer. That third option is the game 00:03:13.139 --> 00:03:15.759 changer. Most people already have meticulously 00:03:15.759 --> 00:03:17.419 organized folders. You don't need to rebuild 00:03:17.419 --> 00:03:19.919 anything. You just point the software at your 00:03:19.919 --> 00:03:23.460 Q3 marketing folder. It instantly indexes the 00:03:23.460 --> 00:03:25.939 templates and past campaign documents. You do 00:03:25.939 --> 00:03:29.639 sex silence. I will be honest. My immediate instinct 00:03:29.639 --> 00:03:32.599 is to just create a massive master folder called 00:03:32.599 --> 00:03:37.050 my brain. Oh, no. Yeah. I kind of just want to 00:03:37.050 --> 00:03:39.250 dump my entire hard drive into it and let Claude 00:03:39.250 --> 00:03:40.830 sort it out. Why shouldn't I do that? Because 00:03:40.830 --> 00:03:42.930 you will totally destroy the vector indexing 00:03:42.930 --> 00:03:45.550 we just talked about. Really? Yeah. If you dump 00:03:45.550 --> 00:03:47.990 sales spreadsheets in with your YouTube scripts 00:03:47.990 --> 00:03:51.330 and grocery lists, the AI loses its semantic 00:03:51.330 --> 00:03:53.909 focus. Ah, I see. It is going to try to draw 00:03:53.909 --> 00:03:56.550 connections between your Q3 revenue and your 00:03:56.550 --> 00:03:58.949 recipe for banana bread. The context gets muddied. 00:03:59.240 --> 00:04:02.419 So mixing unrelated tasks makes the AI's output 00:04:02.419 --> 00:04:04.819 highly inconsistent and confused. You really 00:04:04.819 --> 00:04:06.719 have to keep those boundaries strict. Okay, so 00:04:06.719 --> 00:04:08.560 I have this perfectly organized digital office. 00:04:09.020 --> 00:04:12.939 But how does the AI actually do the work without 00:04:12.939 --> 00:04:15.860 me constantly nudging it? That brings us to schedule 00:04:15.860 --> 00:04:18.240 tasks and skills. This is where we leave the 00:04:18.240 --> 00:04:20.439 old world of repeating prompts behind. Right. 00:04:20.540 --> 00:04:22.839 We are building automated playbooks now. Skills 00:04:22.839 --> 00:04:25.540 are essentially step -by -step reusable instructions. 00:04:26.040 --> 00:04:28.879 Yes, but they are much more rigid than standard 00:04:28.879 --> 00:04:31.759 chat prompts. You write them almost like code. 00:04:32.079 --> 00:04:34.920 You dictate the inputs, the processing steps, 00:04:35.139 --> 00:04:38.620 and the exact output format. Then you schedule 00:04:38.620 --> 00:04:41.300 them. You can run them hourly or daily. Let's 00:04:41.300 --> 00:04:43.040 walk through a practical example. The morning 00:04:43.040 --> 00:04:45.519 briefing skill. This is incredibly practical. 00:04:45.779 --> 00:04:48.579 You schedule it to run at 8 in the morning. It 00:04:48.579 --> 00:04:51.500 literally wakes up before you do. It scans your 00:04:51.500 --> 00:04:54.259 Gmail inbox securely. Yeah, it reads your unread 00:04:54.259 --> 00:04:56.120 Slack channels and reviews your calendar for 00:04:56.120 --> 00:04:58.379 the day. Then it drafts a complete summary. It 00:04:58.379 --> 00:05:01.360 flags urgent messages. It even drafts proposed 00:05:01.360 --> 00:05:04.019 replies. It is just sitting there waiting for 00:05:04.019 --> 00:05:06.019 you when you finally open your laptop. That is 00:05:06.019 --> 00:05:09.040 amazing. Another fantastic use case is the YouTube 00:05:09.040 --> 00:05:11.959 Data Tracker. Creators absolutely love this one. 00:05:12.019 --> 00:05:14.100 It checks a specific YouTube channel automatically 00:05:14.100 --> 00:05:16.860 every night. Right. It pulls the new video titles, 00:05:17.000 --> 00:05:20.399 grabs the URLs, and checks the view counts. Then 00:05:20.399 --> 00:05:22.579 it automatically appends that fresh data right 00:05:22.579 --> 00:05:25.220 into a local spreadsheet. It acts exactly like 00:05:25.220 --> 00:05:28.259 a junior employee. One who never sleeps, never 00:05:28.259 --> 00:05:30.959 complains, and never misses a formatting deadline. 00:05:31.300 --> 00:05:33.860 Two -sec silence. Exactly. Actually, I have to 00:05:33.860 --> 00:05:36.019 make a vulnerable admission here. Oh, yeah. I 00:05:36.019 --> 00:05:38.259 still wrestle with starting from scratch every 00:05:38.259 --> 00:05:40.689 morning instead of automating. Really? Yeah, 00:05:40.730 --> 00:05:42.769 I think to myself, well, it will only take me 00:05:42.769 --> 00:05:45.250 five minutes to check these stats. It is a really 00:05:45.250 --> 00:05:47.689 hard habit to break. That is entirely normal. 00:05:47.889 --> 00:05:50.769 We are so conditioned to just do the manual labor 00:05:50.769 --> 00:05:53.290 ourselves. We think setting up the automation 00:05:53.290 --> 00:05:56.750 takes longer than the task, but it is about the 00:05:56.750 --> 00:05:59.089 compound interest of your time. That makes a 00:05:59.089 --> 00:06:01.310 lot of sense. Once you see a skill run flawlessly 00:06:01.310 --> 00:06:04.050 for a week, you never go back. But let's get 00:06:04.050 --> 00:06:06.569 into the mechanics of building these. Yeah. Why 00:06:06.569 --> 00:06:09.170 do standard prompts fail when you try to automate 00:06:09.170 --> 00:06:11.610 them? How specific did these skill instructions 00:06:11.610 --> 00:06:13.910 actually need to be? They need to be airtight. 00:06:14.209 --> 00:06:16.850 Large language models inherently suffer from 00:06:16.850 --> 00:06:19.230 prompt drift. Right. They introduce randomness. 00:06:19.250 --> 00:06:22.290 Yes. If you just write, summarize my emails, 00:06:22.629 --> 00:06:25.490 Monday's output might be two paragraphs. Tuesday's 00:06:25.490 --> 00:06:28.509 might be a haiku. Wow. Yeah. You have to dictate 00:06:28.509 --> 00:06:31.990 a strict JSON or markdown structure. You leave 00:06:31.990 --> 00:06:35.509 zero room for creative interpretation. So vague 00:06:35.509 --> 00:06:39.319 inputs create. messy outputs. You most strictly 00:06:39.319 --> 00:06:42.120 define the structure. Right. And that rigid structure 00:06:42.120 --> 00:06:44.360 is what makes the whole system actually reliable 00:06:44.360 --> 00:06:46.920 over time. Okay. So I have this incredible automated 00:06:46.920 --> 00:06:49.680 assistant running on my desktop. It is indexing 00:06:49.680 --> 00:06:52.319 my files and running scheduled skills. Yep. But 00:06:52.319 --> 00:06:54.220 what happens when I close my laptop and walk 00:06:54.220 --> 00:06:56.670 out the door? Am I suddenly cut off from all 00:06:56.670 --> 00:06:58.949 this power? Not anymore. That is where dispatch 00:06:58.949 --> 00:07:01.529 comes in. It completely changes your daily dynamic. 00:07:01.930 --> 00:07:03.910 Dispatch is essentially the remote control for 00:07:03.910 --> 00:07:05.910 your desktop ecosystem. Exactly. It lets you 00:07:05.910 --> 00:07:08.269 send tasks from the mobile app, but the actual 00:07:08.269 --> 00:07:10.769 heavy lifting runs on your desktop co -work setup 00:07:10.769 --> 00:07:13.230 back home. It is a fascinating piece of architecture. 00:07:13.569 --> 00:07:15.930 Your phone isn't doing the processing. No, it 00:07:15.930 --> 00:07:18.529 is sending an encrypted ping through Ampropix 00:07:18.529 --> 00:07:21.250 servers, which routes to a local daemon running 00:07:21.250 --> 00:07:24.430 quietly on your Mac or PC. But to make that connection 00:07:24.430 --> 00:07:26.949 work, you have to configure the machine correctly. 00:07:27.430 --> 00:07:30.230 Two settings are absolutely crucial. Right. First, 00:07:30.449 --> 00:07:33.509 you need to enable keep awake. And second, you 00:07:33.509 --> 00:07:37.170 need to toggle allow browser actions. Exactly. 00:07:37.410 --> 00:07:40.410 Keep awake ensures the local daemon stays active. 00:07:40.629 --> 00:07:43.089 It prevents your hard drive and network card 00:07:43.089 --> 00:07:46.410 from going to sleep. Allow browser actions lets 00:07:46.410 --> 00:07:49.110 the AI spin up an invisible browser instance 00:07:49.110 --> 00:07:52.819 to navigate web pages while you are gone. Let's 00:07:52.819 --> 00:07:54.800 paint a picture of how this actually plays out 00:07:54.800 --> 00:07:58.060 in the real world. Say you were at the gym. Okay. 00:07:58.199 --> 00:08:00.199 You are halfway through a run on the treadmill. 00:08:00.720 --> 00:08:03.560 You suddenly remember you need a specific vendor 00:08:03.560 --> 00:08:06.750 receipt for your accountant by noon. Normally, 00:08:06.769 --> 00:08:08.529 that ruins your workout. You just stress about 00:08:08.529 --> 00:08:10.310 it until you get home. Or you try to frantically 00:08:10.310 --> 00:08:12.189 search your email on your phone screen. Right. 00:08:12.350 --> 00:08:14.689 But with Dispatch, you just send a quick text 00:08:14.689 --> 00:08:17.350 to the Clod app on your phone. Find the Q1 server 00:08:17.350 --> 00:08:19.889 receipt and summarize it. When the AI gets that 00:08:19.889 --> 00:08:22.430 text, it instantly searches the local vector 00:08:22.430 --> 00:08:25.050 index on your desktop back home. It identifies 00:08:25.050 --> 00:08:27.730 the right PDF receipt. It extracts the vendor 00:08:27.730 --> 00:08:30.970 details and the total cost. It prepares a summary 00:08:30.970 --> 00:08:32.990 and sends it right back to your phone. By the 00:08:32.990 --> 00:08:35.549 time you step off the treadmill, it is ready. 00:08:35.840 --> 00:08:37.480 You just forward it to your accountant. It really 00:08:37.480 --> 00:08:39.200 connects the whole system beautifully. Think 00:08:39.200 --> 00:08:42.000 of the evolution here. Skills let the AI work 00:08:42.000 --> 00:08:45.139 on a timer. Dispatch lets you trigger that work 00:08:45.139 --> 00:08:48.740 on demand from anywhere in the world. Exactly. 00:08:48.980 --> 00:08:52.039 Beat. But this raises a glaring logistical issue 00:08:52.039 --> 00:08:55.309 for me. What is that? If my laptop needs to be 00:08:55.309 --> 00:08:58.429 reachable at all times, does this mean my computer 00:08:58.429 --> 00:09:00.450 just stays wide open all day? Well, effectively, 00:09:00.549 --> 00:09:03.070 yes. The system cannot execute local background 00:09:03.070 --> 00:09:05.970 tasks or search your files if the machine's operating 00:09:05.970 --> 00:09:08.909 system goes into deep sleep mode. I see. The 00:09:08.909 --> 00:09:11.549 local daemon needs continuous access to the file 00:09:11.549 --> 00:09:14.289 system to perform the work. So it requires keep 00:09:14.289 --> 00:09:17.409 awake active so background processes run seamlessly. 00:09:17.950 --> 00:09:20.090 It is a trade -off. You use a bit more electricity, 00:09:20.230 --> 00:09:22.549 but you gain a full -time remote researcher. 00:09:22.750 --> 00:09:24.750 This brings us to the actual physical interface. 00:09:25.389 --> 00:09:28.210 Up until now, we have talked about background 00:09:28.210 --> 00:09:31.509 processes. Right. But the April 3rd update expanded 00:09:31.509 --> 00:09:34.649 computer use significantly. This is where things 00:09:34.649 --> 00:09:37.809 get slightly futuristic. Computer use is now 00:09:37.809 --> 00:09:40.629 natively available for both Windows and macOS 00:09:40.629 --> 00:09:43.909 for pro subscribers. The AI actually takes native 00:09:43.909 --> 00:09:46.409 control. It physically moves your mouse cursor. 00:09:46.840 --> 00:09:48.580 Yeah, it types on your keyboard. It can even 00:09:48.580 --> 00:09:51.379 execute complex file drag and drop maneuvers 00:09:51.379 --> 00:09:53.600 between folders. The dragging and dropping was 00:09:53.600 --> 00:09:56.059 actually a huge technical hurdle. Oh, absolutely. 00:09:56.279 --> 00:09:58.639 Pointing and clicking is easy, but maintaining 00:09:58.639 --> 00:10:01.299 a continuous mouse down state while navigating 00:10:01.299 --> 00:10:04.480 requires incredible precision. The big technical 00:10:04.480 --> 00:10:07.940 breakthrough here was reducing the latency. They 00:10:07.940 --> 00:10:10.000 achieved this through something called rolling 00:10:10.000 --> 00:10:13.039 visual context. Let's define that AI jargon. 00:10:13.179 --> 00:10:15.559 Okay, it means caching visual differences to 00:10:15.559 --> 00:10:18.049 reduce the computing time needed. Earlier versions 00:10:18.049 --> 00:10:21.350 of screen control were agonizingly slow. The 00:10:21.350 --> 00:10:23.750 AI would take a full high resolution screenshot, 00:10:24.090 --> 00:10:26.309 send it to the cloud, analyze the whole image, 00:10:26.429 --> 00:10:28.730 and then click. It took several seconds per action. 00:10:28.889 --> 00:10:31.830 Yeah. Now, it is less like the AI taking a static 00:10:31.830 --> 00:10:34.429 photo and more like it is watching a low frame 00:10:34.429 --> 00:10:37.070 rate video stream. It only processes the pixels 00:10:37.070 --> 00:10:39.830 that actually change from the last frame. Exactly. 00:10:39.889 --> 00:10:42.269 That slashes the compute time required to figure 00:10:42.269 --> 00:10:44.590 out where the cursor is, making the interaction 00:10:44.590 --> 00:10:47.450 feel incredibly smooth. I was watching the live 00:10:47.450 --> 00:10:51.169 CapCut video editing demo, and honestly, whoa. 00:10:51.889 --> 00:10:54.789 Imagine the screen glowing orange while it actively 00:10:54.789 --> 00:10:57.129 navigates Finder to splice a video together. 00:10:57.330 --> 00:10:59.629 It is a profound moment the first time you see 00:10:59.629 --> 00:11:01.970 it. The perimeter of your screen actually glows 00:11:01.970 --> 00:11:04.129 orange. That is a built -in security feature, 00:11:04.269 --> 00:11:06.919 right? Right. To clearly indicate. the AI has 00:11:06.919 --> 00:11:09.120 active control. Right, and then the mouse just 00:11:09.120 --> 00:11:11.539 starts flying through menus on its own. You can 00:11:11.539 --> 00:11:14.919 also whitelist or block specific apps. You probably 00:11:14.919 --> 00:11:16.980 don't want Claude wandering into your password 00:11:16.980 --> 00:11:19.620 manager or your banking app. Absolutely not. 00:11:19.679 --> 00:11:22.059 Limiting access keeps the process safe. Yeah. 00:11:22.120 --> 00:11:23.960 But we really need to establish the practical 00:11:23.960 --> 00:11:26.539 rule here. Okay. Just because the AI can visually 00:11:26.539 --> 00:11:28.700 click a mouse doesn't mean it should be your 00:11:28.700 --> 00:11:31.320 default method. There is a strict hierarchy of 00:11:31.320 --> 00:11:33.950 operations. Yes, the hierarchy is crucial. You 00:11:33.950 --> 00:11:35.490 should use connectors first. Those are direct 00:11:35.490 --> 00:11:38.409 API connections. They are invisible, fast, and 00:11:38.409 --> 00:11:41.110 stable. You use browser control second, where 00:11:41.110 --> 00:11:43.429 it navigates web code. And you use computer use 00:11:43.429 --> 00:11:46.830 last. It is strictly for legacy desktop apps 00:11:46.830 --> 00:11:49.490 that do not have APIs. Exactly. So if computer 00:11:49.490 --> 00:11:52.830 use looks so cool, why is it the absolute last 00:11:52.830 --> 00:11:55.210 resort instead of the default? Because visual 00:11:55.210 --> 00:11:58.409 clitting is inherently brittle. If an app updates 00:11:58.409 --> 00:12:00.970 its interface and moves a button two pixels to 00:12:00.970 --> 00:12:04.870 the left, the visual AI might miss it. Ah, that 00:12:04.870 --> 00:12:06.769 makes perfect sense. Direct data connections 00:12:06.769 --> 00:12:09.590 bypass the screen entirely. They move information 00:12:09.590 --> 00:12:12.429 instantly and flawlessly. API connectors are 00:12:12.429 --> 00:12:15.070 simply faster and much more reliable. Which is 00:12:15.070 --> 00:12:17.389 why we need to talk about how you actually set 00:12:17.389 --> 00:12:19.769 up those direct data connections. Right. If computer 00:12:19.769 --> 00:12:23.049 use is the last resort, what is the first? That 00:12:23.049 --> 00:12:26.350 brings us to MCP integrations. MCP stands for 00:12:26.350 --> 00:12:29.269 Model Context Protocol. It is an absolute game 00:12:29.269 --> 00:12:32.029 changer for productivity. Let's define it. It 00:12:32.029 --> 00:12:34.830 is a bridge letting AI talk directly to your 00:12:34.830 --> 00:12:37.470 other software. This protocol completely shifts 00:12:37.470 --> 00:12:39.629 the paradigm. It turns basic text generation 00:12:39.629 --> 00:12:42.950 into actual workflow fulfillment. The AI stops 00:12:42.950 --> 00:12:45.509 just giving you text suggestions. It starts taking 00:12:45.509 --> 00:12:47.990 real action inside your other applications. Yes. 00:12:48.110 --> 00:12:51.190 So if computer use is like a robot sitting in 00:12:51.190 --> 00:12:53.350 your physical desk chair, clicking your physical 00:12:53.350 --> 00:12:57.350 mouse, MCP is like plugging the AI's brain directly 00:12:57.350 --> 00:13:00.090 into the mainframe matrix. It skips the physical 00:13:00.090 --> 00:13:03.269 interface entirely. Let's look at the Apify MCP. 00:13:03.389 --> 00:13:06.870 Apify provides live web scrapers. You can pull 00:13:06.870 --> 00:13:09.350 real -time data from highly dynamic platforms. 00:13:09.610 --> 00:13:11.750 You can scrape YouTube, TikTok, or Instagram. 00:13:12.470 --> 00:13:14.809 But how does it actually get that data without 00:13:14.809 --> 00:13:17.230 getting blocked by login screens? The protocol 00:13:17.230 --> 00:13:19.509 handles the authentication through token exchange. 00:13:19.889 --> 00:13:22.929 You generate a secure API token in Appify, you 00:13:22.929 --> 00:13:25.190 paste it into your co -work settings, and the 00:13:25.190 --> 00:13:27.570 bridge is built. The AI can trigger a scraper 00:13:27.570 --> 00:13:30.190 to run, gather the unstructured data, and feed 00:13:30.190 --> 00:13:32.289 it directly into your local projects. And then 00:13:32.289 --> 00:13:34.629 there's the Xabier MCP. This is the big one. 00:13:34.730 --> 00:13:37.570 This connects the AI to over 9 ,000 different 00:13:37.570 --> 00:13:40.090 apps. We're talking about your CRM, your cloud 00:13:40.090 --> 00:13:42.379 databases, and your email. marketing software. 00:13:42.580 --> 00:13:45.200 The AI can move information between all these 00:13:45.200 --> 00:13:47.279 tools automatically, formatting it perfectly 00:13:47.279 --> 00:13:50.299 along the way. It breaks the AI out of its isolated 00:13:50.299 --> 00:13:53.539 box. It operates directly inside your real business 00:13:53.539 --> 00:13:56.340 workflows. It really does. But honestly, that 00:13:56.340 --> 00:13:59.120 level of access is mind blowing. Yeah. If Zapier 00:13:59.120 --> 00:14:01.639 connects everything, isn't it dangerous to give 00:14:01.639 --> 00:14:04.570 the AI? Full access to it? It would be incredibly 00:14:04.570 --> 00:14:06.870 dangerous if you just opened the floodgates and 00:14:06.870 --> 00:14:09.750 gave it blanket administrative rights. But the 00:14:09.750 --> 00:14:12.269 protocol doesn't work like that. It requires 00:14:12.269 --> 00:14:16.929 explicit, granular permission. The AI never actually 00:14:16.929 --> 00:14:19.929 holds your master passwords. It uses OAuth tokens 00:14:19.929 --> 00:14:22.870 scoped strictly to specific actions. You strictly 00:14:22.870 --> 00:14:25.690 whitelist only specific apps to maintain safe 00:14:25.690 --> 00:14:27.909 boundaries. You choose exactly which apps it 00:14:27.909 --> 00:14:30.230 can touch and exactly what it is allowed to do 00:14:30.230 --> 00:14:33.190 inside those apps. Sponsor read. Welcome back 00:14:33.190 --> 00:14:35.830 to the deep dive. We have covered a massive amount 00:14:35.830 --> 00:14:38.669 of ground today. We really have. From local memory 00:14:38.669 --> 00:14:41.590 to remote execution. And it is very easy to feel 00:14:41.590 --> 00:14:44.090 overwhelmed by all these new capabilities. When 00:14:44.090 --> 00:14:46.769 people first see this, they tend to want to automate 00:14:46.769 --> 00:14:49.370 their entire life by tomorrow morning. Let's 00:14:49.370 --> 00:14:50.610 step back and look at the broader framework. 00:14:51.159 --> 00:14:53.379 We want to avoid creating an over -engineered 00:14:53.379 --> 00:14:56.659 mess. The key to making this work is following 00:14:56.659 --> 00:14:59.720 a strict order of operations. You have to build 00:14:59.720 --> 00:15:01.980 the foundation before you build the robots. Right. 00:15:02.080 --> 00:15:04.139 You always start with context first. That means 00:15:04.139 --> 00:15:06.440 setting up your projects, give the AI memory. 00:15:06.820 --> 00:15:08.960 Let it index your files so it understands your 00:15:08.960 --> 00:15:12.320 world. Then automation comes second. You build 00:15:12.320 --> 00:15:14.960 your skills for repeating tasks. You lock in 00:15:14.960 --> 00:15:17.399 those rigid markdown structures to prevent prompt 00:15:17.399 --> 00:15:20.460 drift. Integrations are third. That is your... 00:15:20.490 --> 00:15:23.490 MCP connections for live data. You plug it into 00:15:23.490 --> 00:15:26.909 Zapier or Appify. Remote access is fourth. You 00:15:26.909 --> 00:15:29.350 set up dispatch for mobile control. You enable 00:15:29.350 --> 00:15:31.250 the background daemon so you can trigger workflows 00:15:31.250 --> 00:15:34.909 from the gym. And screen control is last. You 00:15:34.909 --> 00:15:37.309 only enable computer use when you have a stubborn 00:15:37.309 --> 00:15:40.169 legacy app that absolutely requires visual clicking. 00:15:40.350 --> 00:15:42.950 The best advice here is to start very, very small. 00:15:43.129 --> 00:15:45.950 Do not try to boil the ocean. Just build one 00:15:45.950 --> 00:15:48.809 single project, pointed at one specific folder. 00:15:49.029 --> 00:15:52.210 Exactly. Automate just one repeatable task. Get 00:15:52.210 --> 00:15:54.830 that working perfectly. Watch it run flawlessly 00:15:54.830 --> 00:15:58.110 for a few days before you try to expand. Once 00:15:58.110 --> 00:16:00.470 you trust the practical results, you can scale 00:16:00.470 --> 00:16:03.470 up the complexity safely. So here is your call 00:16:03.470 --> 00:16:07.090 to action. Pick one repeating task today. Instead 00:16:07.090 --> 00:16:08.730 of that morning briefing skill we talked about, 00:16:09.070 --> 00:16:11.210 clean it at your inbox in your calendar. Schedule 00:16:11.210 --> 00:16:14.070 it for 8 a .m. tomorrow morning. Just see the 00:16:14.070 --> 00:16:16.730 system in action for yourself. Let it wake up 00:16:16.730 --> 00:16:19.549 before you do. it will completely change how 00:16:19.549 --> 00:16:22.429 you view your computer. It really will. And it 00:16:22.429 --> 00:16:24.929 leaves me with this final thought. If our operating 00:16:24.929 --> 00:16:27.710 systems can now hold long -term memories, run 00:16:27.710 --> 00:16:30.450 complex workflows while we sleep, and act on 00:16:30.450 --> 00:16:33.570 live web data completely independently, at what 00:16:33.570 --> 00:16:35.590 point does our computer stop being just a tool 00:16:35.590 --> 00:16:37.370 and become our most reliable colleague?