WEBVTT 00:00:00.000 --> 00:00:01.520 You know, the average person, I think, spends 00:00:01.520 --> 00:00:04.740 hours, literally hours, every single week just 00:00:04.740 --> 00:00:08.560 sorting, filing, reading emails, so much stuff 00:00:08.560 --> 00:00:11.539 that probably didn't need their immediate human 00:00:11.539 --> 00:00:14.080 attention. It's this constant drain. It really 00:00:14.080 --> 00:00:16.579 is. It's like the silent killer of real productive 00:00:16.579 --> 00:00:19.780 work. But imagine flipping that. Imagine having, 00:00:19.859 --> 00:00:22.600 like, a personal AI assistant doing that heavy 00:00:22.600 --> 00:00:24.399 lifting for you, making those little decisions, 00:00:24.519 --> 00:00:27.660 processing new messages in, what, under 15 seconds? 00:00:27.739 --> 00:00:30.179 Yeah. And the best part, you build it. No code 00:00:30.179 --> 00:00:33.159 required. None. Welcome back to the Deep Dive. 00:00:33.179 --> 00:00:35.700 Today, we are really getting into the foundations. 00:00:35.700 --> 00:00:39.060 We're extracting the core ideas from this really 00:00:39.060 --> 00:00:41.719 detailed no code guide. Our mission is pretty 00:00:41.719 --> 00:00:44.920 clear. Build the core intelligence, the AI brain, 00:00:45.060 --> 00:00:47.579 if you will, for an automated email agent that 00:00:47.579 --> 00:00:50.759 reads, classifies, and basically organizes your 00:00:50.759 --> 00:00:53.369 messages. even while you sleep and this isn't 00:00:53.369 --> 00:00:55.189 just theory right we're diving deep into the 00:00:55.189 --> 00:00:57.350 actual blueprint yeah the tools you need we're 00:00:57.350 --> 00:01:00.409 talking nan gmail open router yeah and the four 00:01:00.409 --> 00:01:02.530 big steps to get this intelligent core up and 00:01:02.530 --> 00:01:05.150 running okay let's unpack that so this time we're 00:01:05.150 --> 00:01:07.109 focusing just on that decision making engine 00:01:07.109 --> 00:01:10.950 the intelligence and the filing system call it 00:01:10.950 --> 00:01:13.719 part one Building the brain. Exactly. Part two, 00:01:13.840 --> 00:01:16.000 that's later. That's going to be all about the 00:01:16.000 --> 00:01:18.159 automated actions, the doing part. Right. So 00:01:18.159 --> 00:01:20.659 the core problem we're tackling today, it's pretty 00:01:20.659 --> 00:01:23.319 universal, isn't it? Information overload, just 00:01:23.319 --> 00:01:27.120 drowning in emails, constantly pulled away from 00:01:27.120 --> 00:01:29.579 the work that actually matters. Totally. We all 00:01:29.579 --> 00:01:31.400 feel it. And the whole system, once it's all 00:01:31.400 --> 00:01:34.340 built out, it follows this really neat kind of 00:01:34.340 --> 00:01:37.280 simple four -stage loop. Stage one, the trigger. 00:01:37.459 --> 00:01:40.349 Okay. That's just a new email landing in your 00:01:40.349 --> 00:01:42.069 inbox. Think of it like the doorbell ringing. 00:01:42.230 --> 00:01:44.609 Okay, the doorbell. Then stage two, that's the 00:01:44.609 --> 00:01:46.209 intelligence part we're building today, right? 00:01:46.269 --> 00:01:49.049 The AI collectification. Yep. The system reads 00:01:49.049 --> 00:01:51.530 the subject, reads the body, the whole body, 00:01:51.629 --> 00:01:54.810 and decides instantly, which bucket does this 00:01:54.810 --> 00:01:56.909 go in? We're starting with four main ones, customer 00:01:56.909 --> 00:01:59.670 support, finance billing, high priority, and 00:01:59.670 --> 00:02:03.129 promotion. And once it knows the bucket. Stage 00:02:03.129 --> 00:02:06.849 three, labeling. Simple as that. The agent slaps 00:02:06.849 --> 00:02:09.500 the right. Gmail label on it automatically. That's 00:02:09.500 --> 00:02:10.960 your filing system right there in the inbox. 00:02:11.060 --> 00:02:12.860 Okay. And stage four, that's for next time. Stage 00:02:12.860 --> 00:02:15.159 four is the automated actions. That's where it 00:02:15.159 --> 00:02:17.020 gets really cool, like sending an AI generated 00:02:17.020 --> 00:02:20.400 reply or maybe pinging your team on Slack, stuff 00:02:20.400 --> 00:02:23.580 like that. The shift here feels significant. 00:02:23.780 --> 00:02:26.300 You're taking this repetitive cognitive sorting 00:02:26.300 --> 00:02:29.979 task and just handing it off to a system that 00:02:29.979 --> 00:02:32.620 doesn't get tired, doesn't get distracted. And 00:02:32.620 --> 00:02:35.449 the source mentioned this test result. processing 00:02:35.449 --> 00:02:38.169 and even responding in about 15 seconds total. 00:02:38.310 --> 00:02:41.389 15 seconds. Whoa. Think about that. Seriously. 00:02:41.650 --> 00:02:44.610 Imagine scaling that. If you get, say, 50 emails 00:02:44.610 --> 00:02:47.009 a day that need sorting, that's real time you 00:02:47.009 --> 00:02:48.909 get back. It just handles the grunt work immediately. 00:02:49.289 --> 00:02:52.090 So if something is genuinely urgent. It's flagged, 00:02:52.090 --> 00:02:54.310 labeled, sitting there waiting for your actual 00:02:54.310 --> 00:02:56.729 human expertise, not buried under newsletters. 00:02:57.129 --> 00:02:59.590 So thinking about that, how does automating the 00:02:59.590 --> 00:03:02.289 sorting fundamentally change how you even define 00:03:02.289 --> 00:03:04.610 what's urgent work? Well, it just cuts through 00:03:04.610 --> 00:03:06.610 the noise, doesn't it? Yeah. It ensures that 00:03:06.610 --> 00:03:08.930 only the stuff that really needs your eyes and 00:03:08.930 --> 00:03:11.849 brain gets through that filter. Everything else 00:03:11.849 --> 00:03:15.069 is neatly filed. Okay. Makes sense. So let's 00:03:15.069 --> 00:03:18.469 talk tools. The blueprint. Right. So to actually 00:03:18.469 --> 00:03:20.990 connect all this and build the brain, you only 00:03:20.990 --> 00:03:23.419 need three main tools. That's the beauty of this 00:03:23.419 --> 00:03:25.180 whole no -code thing. The infrastructure is mostly 00:03:25.180 --> 00:03:28.300 there. First up is N -A -A -N. If you haven't 00:03:28.300 --> 00:03:30.879 used it, think of it like a visual workflow builder. 00:03:31.099 --> 00:03:33.800 You drag, you drop, you connect apps. No traditional 00:03:33.800 --> 00:03:36.080 coding needed. It's where we map out the logic. 00:03:36.300 --> 00:03:38.620 Then, obviously, Gmail. That's our inbox for 00:03:38.620 --> 00:03:40.639 this example, our trigger point. Yeah. But the 00:03:40.639 --> 00:03:42.340 logic works for Outlook, too, pretty much anything 00:03:42.340 --> 00:03:44.539 similar. Okay. And number three, OpenRouter. 00:03:44.639 --> 00:03:46.960 This thing is great. It's basically a gateway 00:03:46.960 --> 00:03:50.110 to lots of different AI models. Instead of juggling 00:03:50.110 --> 00:03:53.909 API keys for GPT, Claude, whatever, OpenRouter 00:03:53.909 --> 00:03:56.430 gives you one key. Simple access to powerful 00:03:56.430 --> 00:04:00.050 models like, say, GPT 4 .1 Mini, or even the 00:04:00.050 --> 00:04:01.889 really cutting -edge ones if you need them. And 00:04:01.889 --> 00:04:04.650 it's usually cheaper. Now, the guide really hammers 00:04:04.650 --> 00:04:06.949 this point home. Wireframing. Mapping it out 00:04:06.949 --> 00:04:08.849 visually before you even open any. You know, 00:04:08.889 --> 00:04:12.370 measure twice, cut once. Yeah. But honestly, 00:04:12.629 --> 00:04:14.870 for a workflow that seems like just four main 00:04:14.870 --> 00:04:18.560 steps, is wireframing really necessary? Doesn't 00:04:18.560 --> 00:04:21.519 it feel a bit like overkill? That's a fair question. 00:04:21.899 --> 00:04:24.920 But yes, absolutely. It matters even for small 00:04:24.920 --> 00:04:29.839 stuff. Yeah. Because automation complexity, it 00:04:29.839 --> 00:04:33.060 doesn't scale linearly. It gets messy fast. Think 00:04:33.060 --> 00:04:34.720 of that blueprint trigger classifier, the four 00:04:34.720 --> 00:04:37.000 branches, the actions. That's your map. It's 00:04:37.000 --> 00:04:39.470 like. Disaster prevention. When something breaks 00:04:39.470 --> 00:04:41.509 later, and it might, you look at the map, you're 00:04:41.509 --> 00:04:43.750 not just clicking around blindly in NANA trying 00:04:43.750 --> 00:04:45.870 to figure out where the logic went wrong. It 00:04:45.870 --> 00:04:47.850 forces you to think through the flow first. So, 00:04:47.870 --> 00:04:50.910 bottom line, why is mapping it out first so critical 00:04:50.910 --> 00:04:53.470 even before touching the tools? It just prevents 00:04:53.470 --> 00:04:55.370 that initial overwhelm. Yeah. You define the 00:04:55.370 --> 00:04:57.269 logic clearly before you get tangled up in the 00:04:57.269 --> 00:04:59.209 connections and settings. Saves a lot of headaches. 00:04:59.370 --> 00:05:01.870 Got it. Okay. Blueprint's ready. Step one, the 00:05:01.870 --> 00:05:04.389 trigger. Setting up that doorbell. Right. So, 00:05:04.509 --> 00:05:08.350 new N8n workflow. You grab the Gmail node and 00:05:08.350 --> 00:05:11.610 you pick the on message received trigger. Simple. 00:05:11.709 --> 00:05:13.329 Then you just connect your Gmail account, right? 00:05:13.370 --> 00:05:15.370 Give a credential. Yep. And Annie Dan will start 00:05:15.370 --> 00:05:17.970 checking usually about every minute for new mail. 00:05:18.050 --> 00:05:20.569 Pretty quick. And for testing, you just send 00:05:20.569 --> 00:05:22.589 yourself an email, something basic? Exactly. 00:05:22.589 --> 00:05:25.730 Like subject, help, and body. My account isn't 00:05:25.730 --> 00:05:28.149 working. Fire that off. Then in Annie Dan, you 00:05:28.149 --> 00:05:31.009 hit fetch test event. That pulls in the data 00:05:31.009 --> 00:05:33.029 from your test email. Okay. And here comes the 00:05:33.029 --> 00:05:34.750 key part you mentioned. Ah, yeah. This is the 00:05:34.750 --> 00:05:37.689 crucial bit. The secret sauce, maybe. By default, 00:05:37.829 --> 00:05:41.370 N8N tries to be helpful. It simplifies the data 00:05:41.370 --> 00:05:43.769 it pulls from Gmail. Right. You absolutely must 00:05:43.769 --> 00:05:47.050 turn the simplify toggle OFF on that Gmail trigger 00:05:47.050 --> 00:05:49.129 node. Okay. That's the expert move. I remember 00:05:49.129 --> 00:05:51.550 wasting, I don't know, probably an hour once 00:05:51.550 --> 00:05:53.129 on a different automation because I only had 00:05:53.129 --> 00:05:55.069 partial data coming through. Turning off simplify 00:05:55.069 --> 00:05:58.029 gets you everything. Everything. The full email 00:05:58.029 --> 00:06:01.430 body, sender info, attachments, and crucially, 00:06:01.490 --> 00:06:04.490 the unique message ID for that specific email. 00:06:04.800 --> 00:06:07.740 And that full context matters for the AI. Vitally 00:06:07.740 --> 00:06:11.639 important. If the AI only sees subject, H -E 00:06:11.639 --> 00:06:14.660 -L -P, I think it's spam. But if it sees the 00:06:14.660 --> 00:06:18.139 body says H -E -L -P, the main server is down, 00:06:18.279 --> 00:06:21.420 that's different. Full text equals accurate classification. 00:06:21.860 --> 00:06:24.439 Right. So fetch that full data, then pin it in 00:06:24.439 --> 00:06:27.000 N8N. That means you can test all the next steps 00:06:27.000 --> 00:06:29.339 using that same data without sending more emails. 00:06:29.759 --> 00:06:32.040 Super handy. Okay, quick recap then. What's the 00:06:32.040 --> 00:06:35.399 absolute essential data the AI needs from that 00:06:35.399 --> 00:06:37.660 trigger for top accuracy? It needs the whole 00:06:37.660 --> 00:06:40.540 picture, the complete unsimplified email body, 00:06:40.639 --> 00:06:43.040 and the subject line for maximum context. Perfect. 00:06:43.259 --> 00:06:45.519 All right. Data stream is flowing. Step two, 00:06:45.740 --> 00:06:48.540 building the actual intelligence, the sorting 00:06:48.540 --> 00:06:50.779 hat. Yep. Add the text classifier node right 00:06:50.779 --> 00:06:52.839 after your Gmail trigger. Now we need to tell 00:06:52.839 --> 00:06:54.600 it what text to analyze. How do you feed it the 00:06:54.600 --> 00:06:56.500 right stuff? You switch the input field to expression 00:06:56.500 --> 00:06:58.910 mode. Yeah. Then you use this specific structure, 00:06:59.170 --> 00:07:03.050 subject .headers .body .text. It looks a bit 00:07:03.050 --> 00:07:05.050 technical, but all it's doing is grabbing the 00:07:05.050 --> 00:07:07.230 subject from the incoming data, grabbing the 00:07:07.230 --> 00:07:09.069 body text, and sticking them together for the 00:07:09.069 --> 00:07:11.589 AI. So it sees both at the same time. Makes sense. 00:07:11.870 --> 00:07:14.529 Exactly. Then comes the really important part, 00:07:14.709 --> 00:07:17.730 defining your categories. This is pure prompt 00:07:17.730 --> 00:07:20.389 engineering. This is the brain, basically. The 00:07:20.389 --> 00:07:23.089 AI only knows what you tell it. Fuzzy instructions 00:07:23.089 --> 00:07:25.839 mean fuzzy results. So those four buckets we 00:07:25.839 --> 00:07:28.379 mentioned, customer support, finance, high priority, 00:07:28.620 --> 00:07:31.220 promotion, we need clear descriptions. Crystal 00:07:31.220 --> 00:07:34.839 clear. Precision is key. For customer support, 00:07:35.019 --> 00:07:38.139 think keywords. Error, issue, help, troubleshoot. 00:07:38.420 --> 00:07:41.279 For finance billing, obviously, invoice, payment, 00:07:41.459 --> 00:07:44.180 refund, subscription. Promotion is easy. Sale, 00:07:44.339 --> 00:07:47.720 offer, discount. And high priority needs words 00:07:47.720 --> 00:07:51.379 like urgent, ASA, critical, outage, emergency. 00:07:51.980 --> 00:07:53.439 And you have to be careful they don't overlap 00:07:53.439 --> 00:07:56.079 too much, right? Like what if an email says urgent 00:07:56.079 --> 00:07:58.779 invoice payment failed? Is that finance or high 00:07:58.779 --> 00:08:01.000 priority? Good point. That's exactly where the 00:08:01.000 --> 00:08:02.720 description part of the prompt becomes critical. 00:08:02.860 --> 00:08:05.300 You need to guide the AI. Maybe you say high 00:08:05.300 --> 00:08:07.560 priority is only for system outages or critical 00:08:07.560 --> 00:08:10.560 deadlines and payment issues go to finance unless 00:08:10.560 --> 00:08:13.079 they use specific emergency language. You have 00:08:13.079 --> 00:08:14.980 to define that nuance. And this is where things 00:08:14.980 --> 00:08:17.360 can go sideways if you're not careful. Oh, yeah. 00:08:17.720 --> 00:08:20.879 Clarity is your debugging tool here. And honestly. 00:08:21.519 --> 00:08:24.100 I still wrestle with prompt drift myself sometimes. 00:08:24.480 --> 00:08:28.100 You know, where the AI kind of slowly misunderstands 00:08:28.100 --> 00:08:30.639 your original intent over time. Yeah, I've seen 00:08:30.639 --> 00:08:32.460 that. Keeping those category descriptions tight 00:08:32.460 --> 00:08:35.840 and precise is your best defense. If classifications 00:08:35.840 --> 00:08:38.159 start going weird, the first place you check 00:08:38.159 --> 00:08:40.460 is that description text. Did I make it ambiguous? 00:08:41.080 --> 00:08:43.740 So why is getting that nuance right, like between 00:08:43.740 --> 00:08:46.500 finance and high priority, so absolutely essential? 00:08:46.960 --> 00:08:49.919 Because explicitly telling the AI how to handle 00:08:49.919 --> 00:08:52.259 those edge cases, those overlapping keywords, 00:08:52.600 --> 00:08:55.320 prevents it from guessing wrong when it sees 00:08:55.320 --> 00:08:58.419 ambiguous emails later on. Clarity prevents errors. 00:08:58.720 --> 00:09:01.419 Okay. Brain structure defined. Now it needs power. 00:09:01.779 --> 00:09:05.460 Step three. Connecting the engine. The text classifier 00:09:05.460 --> 00:09:07.720 node has the instructions, but it needs a large 00:09:07.720 --> 00:09:10.360 language model and LLM to actually do the thinking. 00:09:10.519 --> 00:09:12.279 And that's where OpenRouter comes in. Exactly. 00:09:12.440 --> 00:09:14.080 Set up an account, you'll need to add billing 00:09:14.080 --> 00:09:16.139 info, even just for the free credits they usually 00:09:16.139 --> 00:09:19.309 give you, then generate an API key. Good practice 00:09:19.309 --> 00:09:21.350 to name it something specific like inbox agent 00:09:21.350 --> 00:09:23.990 key. And you just paste that key into N8N. Yep. 00:09:24.049 --> 00:09:25.870 Right into the text classifier node settings 00:09:25.870 --> 00:09:27.809 where it asks for your chat model credentials 00:09:27.809 --> 00:09:30.769 or API key. Connects it right up. What about 00:09:30.769 --> 00:09:33.389 choosing the actual AI model? There are so many 00:09:33.389 --> 00:09:35.789 options. For this kind of classification, the 00:09:35.789 --> 00:09:40.049 guide suggests GPT 4 .1 mini. It's a really good 00:09:40.049 --> 00:09:43.210 balance. It's cheap. It's fast. And it's plenty 00:09:43.210 --> 00:09:45.980 smart enough for sorting emails accurately. But 00:09:45.980 --> 00:09:47.940 the benefit of open routers, you're not stuck 00:09:47.940 --> 00:09:50.240 with that. Precisely. That's the key advantage. 00:09:50.600 --> 00:09:54.080 Agility. If you find GPT 4 .1 Mini isn't quite 00:09:54.080 --> 00:09:56.259 getting the nuance you need, you literally just 00:09:56.259 --> 00:09:59.799 select, say, Claude 4 .5 Sonnet from the dropdown 00:09:59.799 --> 00:10:03.639 in NN. Done. No new accounts, no new keys. Test 00:10:03.639 --> 00:10:06.220 away. That flexibility is huge. It means your 00:10:06.220 --> 00:10:08.679 AI brain isn't locked into one vendor. You can 00:10:08.679 --> 00:10:11.039 experiment, find the best speed, cost, accuracy 00:10:11.039 --> 00:10:13.690 balance for your specific emails. Totally. So 00:10:13.690 --> 00:10:15.750 you set that up, you run the step with your pin 00:10:15.750 --> 00:10:18.090 test data, make sure it correctly outputs customer 00:10:18.090 --> 00:10:20.830 support, and pin that output too for the next 00:10:20.830 --> 00:10:22.710 step. Okay, beyond just maybe saving some money, 00:10:22.789 --> 00:10:25.309 what's the core operational win from using something 00:10:25.309 --> 00:10:27.710 like OpenRouter here? It's really about the speed 00:10:27.710 --> 00:10:30.250 and flexibility. You can easily swap between 00:10:30.250 --> 00:10:32.789 different powerful AI models using just that 00:10:32.789 --> 00:10:35.129 one connection point. Experimentation becomes 00:10:35.129 --> 00:10:38.629 cheap and fast. Gotcha. Intelligence built, powered 00:10:38.629 --> 00:10:42.990 up, tested. Now for step four. The filing cabinet. 00:10:43.559 --> 00:10:45.840 Making it organized. Setting up the labels. Well, 00:10:45.879 --> 00:10:49.279 wait. Crucial pre -step here. Create those labels 00:10:49.279 --> 00:10:52.240 in Gmail first. Go into your Gmail settings and 00:10:52.240 --> 00:10:54.659 actually make the labels. Customer support, finance 00:10:54.659 --> 00:10:56.899 billing, high priority promotion. Right. NEN 00:10:56.899 --> 00:10:58.740 can't create them. It can only apply existing 00:10:58.740 --> 00:11:01.419 ones. Exactly. And a pro tip from the guide. 00:11:01.820 --> 00:11:05.139 Use colors. Make high priority bright red or 00:11:05.139 --> 00:11:07.220 something. Makes it super easy to spot those 00:11:07.220 --> 00:11:09.940 urgent emails visually in your inbox later. Nice 00:11:09.940 --> 00:11:13.730 tip. Okay. Labels exist in Gmail. Back to NAN. 00:11:13.929 --> 00:11:16.470 Add the Gmail. Add label to message action node. 00:11:16.629 --> 00:11:18.830 Put it on the first branch coming off your classifier. 00:11:18.870 --> 00:11:20.470 Let's say the customer support path. And the 00:11:20.470 --> 00:11:23.149 key setting here is telling it which email to 00:11:23.149 --> 00:11:25.990 label. Bingo. This is critical. You need the 00:11:25.990 --> 00:11:28.370 unique identifier for the email that started 00:11:28.370 --> 00:11:30.830 the whole process. Remember that aid in variable 00:11:30.830 --> 00:11:33.299 from the very first Gmail trigger step? The one 00:11:33.299 --> 00:11:35.460 we got by turning simplify off. That's the one. 00:11:35.580 --> 00:11:38.159 You drag that ID variable from the trigger node's 00:11:38.159 --> 00:11:41.179 output data and drop it into the message ID field 00:11:41.179 --> 00:11:44.080 of the add label node. Think of it like the email's 00:11:44.080 --> 00:11:46.279 unique address or, I don't know, social security 00:11:46.279 --> 00:11:48.799 number. It guarantees the label goes on the exact 00:11:48.799 --> 00:11:51.720 right message. No mix -ups. Makes sense. And 00:11:51.720 --> 00:11:53.279 then you just repeat that for the other three 00:11:53.279 --> 00:11:56.080 branches. You could. Or use the don't repeat 00:11:56.080 --> 00:11:58.539 yourself principle. Configure that first add 00:11:58.539 --> 00:12:00.980 label node perfectly. Test it. Then just copy 00:12:00.980 --> 00:12:03.080 and paste it onto the other three branches. All 00:12:03.080 --> 00:12:04.860 you have to change in the copies is which label 00:12:04.860 --> 00:12:07.559 to apply from the dropdown menu. Saves a ton 00:12:07.559 --> 00:12:11.519 of clicking. Ah, nice efficiency move. So, final 00:12:11.519 --> 00:12:15.220 check on this step. Why is grabbing that specific 00:12:15.220 --> 00:12:18.379 message ID so non -negotiable for the ad label 00:12:18.379 --> 00:12:21.279 part? Because that unique ID is the only way 00:12:21.279 --> 00:12:23.360 to guarantee the labeling action connects back 00:12:23.360 --> 00:12:25.740 correctly to the specific email that triggered 00:12:25.740 --> 00:12:28.190 the whole workflow right at the start. Keeps 00:12:28.190 --> 00:12:30.409 everything linked properly. Okay. Wow. We actually 00:12:30.409 --> 00:12:32.490 did it. Mission accomplished for part one. We 00:12:32.490 --> 00:12:34.929 did. We built the brain, the classifier, and 00:12:34.929 --> 00:12:37.429 the filing system, the labeling part. Yeah. Your 00:12:37.429 --> 00:12:40.129 agent can now trigger on a new email, read it 00:12:40.129 --> 00:12:42.529 with full context, make an intelligent decision 00:12:42.529 --> 00:12:45.330 based on your rules, and automatically file it 00:12:45.330 --> 00:12:47.490 with the right label. The system is thinking. 00:12:47.529 --> 00:12:50.049 It's making decisions. It's organizing your inbox 00:12:50.049 --> 00:12:53.789 all automatically. That's a big one. Using just 00:12:53.789 --> 00:12:56.269 no -code tools and some clever AI prompting. 00:12:56.879 --> 00:12:59.179 The heavy lifting, the conceptual part, is basically 00:12:59.179 --> 00:13:01.240 done now. And that sets us up perfectly for part 00:13:01.240 --> 00:13:03.039 two. That's where the agent gets his arms and 00:13:03.039 --> 00:13:05.279 legs, right? Exactly. Next time, we'll build 00:13:05.279 --> 00:13:07.820 out the actions for each of those branches. Things 00:13:07.820 --> 00:13:10.759 like sending automated AI replies, maybe drafting 00:13:10.759 --> 00:13:13.440 responses for you to approve first, pushing those 00:13:13.440 --> 00:13:15.539 high priority alerts to your phone or Slack. 00:13:16.460 --> 00:13:18.539 That's where the real time saving magic happens. 00:13:18.659 --> 00:13:20.320 I'm looking forward to that. But before we wrap 00:13:20.320 --> 00:13:21.700 for today, here's something to think about. A 00:13:21.700 --> 00:13:23.960 final thought for you, the listener. Since we 00:13:23.960 --> 00:13:25.759 saw how critical those category descriptions 00:13:25.759 --> 00:13:28.500 are for accuracy, take a minute later today and 00:13:28.500 --> 00:13:32.720 think. Which emails in your inbox are the trickiest 00:13:32.720 --> 00:13:35.559 for you to categorize quickly? Those weird edge 00:13:35.559 --> 00:13:38.600 cases. How would you explain the difference simply 00:13:38.600 --> 00:13:42.240 to an AI? Ah, interesting exercise. Doing that 00:13:42.240 --> 00:13:44.000 kind of self -reflection, thinking about your 00:13:44.000 --> 00:13:46.740 own sorting logic, can seriously level up your 00:13:46.740 --> 00:13:48.679 prompt engineering game for stuff like this. 00:13:48.879 --> 00:13:51.129 Great point. Well, thank you for sharing the 00:13:51.129 --> 00:13:52.850 source material for this deep dive. It's been 00:13:52.850 --> 00:13:55.409 fascinating to unpack the infrastructure of building 00:13:55.409 --> 00:13:58.470 an AI brain like this. My pleasure. Keep learning. 00:13:58.610 --> 00:14:00.450 Keep building. We'll catch you on the next one 00:14:00.450 --> 00:14:00.889 for part two.