WEBVTT 00:00:00.000 --> 00:00:03.140 How do you take a simple scheduled automation, 00:00:03.520 --> 00:00:05.820 you know, a set of rules that just runs on a 00:00:05.820 --> 00:00:07.700 timer, and turn it into something that can actually 00:00:07.700 --> 00:00:11.140 think? Yeah. Something that can adapt and execute 00:00:11.140 --> 00:00:15.240 really complex actions. Today, we're moving past 00:00:15.240 --> 00:00:16.839 that basic script. We're going to look at the 00:00:16.839 --> 00:00:19.519 specific tools that give your workflow a brain 00:00:19.519 --> 00:00:23.700 and hands. Welcome back to the Deep Dive. So 00:00:23.700 --> 00:00:26.100 if you were with us for part one, you've already 00:00:26.100 --> 00:00:28.480 got the foundation down, the scheduling, the 00:00:28.480 --> 00:00:30.839 triggers, all that core logic. That was the body, 00:00:30.879 --> 00:00:32.659 right? We built the robot. Exactly. You built 00:00:32.659 --> 00:00:34.640 the body. It's standing up. It's ready to go. 00:00:34.780 --> 00:00:36.799 And now we give it a mind. Now we give it a mind. 00:00:36.939 --> 00:00:40.479 Our mission today is all about mastering these 00:00:40.479 --> 00:00:43.240 final seven advanced nodes. We're talking nodes 00:00:43.240 --> 00:00:46.179 11 through 17. Okay. And we've broken it down 00:00:46.179 --> 00:00:49.920 for you into three. Really critical areas, advanced 00:00:49.920 --> 00:00:52.299 data handling, connecting to the outside world, 00:00:52.399 --> 00:00:55.380 and the big one, the AI engine itself. These 00:00:55.380 --> 00:00:57.759 are the tools you need to build a real AI agent. 00:00:57.820 --> 00:01:00.159 These are the superpowers. Let's get into it. 00:01:00.320 --> 00:01:02.240 So when we built that foundation, we talked a 00:01:02.240 --> 00:01:04.680 lot about splitting data. Right. Taking a big 00:01:04.680 --> 00:01:06.700 stream of information and breaking it into smaller 00:01:06.700 --> 00:01:09.180 chunks to work on. Yeah. But once you've done 00:01:09.180 --> 00:01:11.459 all that work. Yeah. You kind of need the opposite. 00:01:11.579 --> 00:01:13.359 You need to put it all back together. That is 00:01:13.359 --> 00:01:16.799 exactly where Node 11, the aggregate node, comes 00:01:16.799 --> 00:01:19.340 in. We call it the cleanup crew. The cleanup 00:01:19.340 --> 00:01:21.239 crew. I like that. It's all about presentation. 00:01:22.060 --> 00:01:26.099 So say you split a huge spreadsheet into, I don't 00:01:26.099 --> 00:01:29.579 know, 50 rows. You run an AI summary on each 00:01:29.579 --> 00:01:31.900 one. Right. You don't want your workflow spitting 00:01:31.900 --> 00:01:34.840 out 50 separate emails or notifications. No, 00:01:34.859 --> 00:01:38.420 that would be complete chaos. Total chaos. So 00:01:38.420 --> 00:01:40.620 you use aggregate to combine all those little 00:01:40.620 --> 00:01:44.340 pieces back into one single list or one item. 00:01:44.459 --> 00:01:46.280 Like putting the marbles back in the bag. Perfect 00:01:46.280 --> 00:01:48.819 analogy. It makes sure that final report for 00:01:48.819 --> 00:01:51.859 your boss lands as one polished. neat document, 00:01:51.939 --> 00:01:54.299 not 50 little fragments. It just looks professional. 00:01:54.459 --> 00:01:57.040 Okay. That makes total sense. So that brings 00:01:57.040 --> 00:01:59.980 us to node 12, the code node, the calculator. 00:02:00.280 --> 00:02:02.939 Ah, the scary one. This is the one. I think most 00:02:02.939 --> 00:02:05.000 people see it. They see that cursor blinking, 00:02:05.239 --> 00:02:07.140 waiting for JavaScript or Python, and they just, 00:02:07.159 --> 00:02:09.740 they run away. They do. And I get it. It lets 00:02:09.740 --> 00:02:12.879 you inject these small bits of deterministic 00:02:12.879 --> 00:02:15.419 code. But the thing we really want you to get 00:02:15.419 --> 00:02:18.039 today is that you do not need to be a coder to 00:02:18.039 --> 00:02:20.710 use this. And I can. I can admit something here. 00:02:20.830 --> 00:02:23.810 Go for it. I still wrestle with, you know, really 00:02:23.810 --> 00:02:26.430 complex tech stuff. So I'll just open up an AI 00:02:26.430 --> 00:02:30.050 tool and literally type, write me N8N code to 00:02:30.050 --> 00:02:32.430 fix this date format so it's year, month, day. 00:02:33.719 --> 00:02:35.900 It just works. And that's the modern workflow. 00:02:36.080 --> 00:02:38.400 You're just outsourcing the scripting part. You 00:02:38.400 --> 00:02:41.240 use this node when the simple built -in functions 00:02:41.240 --> 00:02:43.639 just aren't enough. Like really weird text formatting. 00:02:43.879 --> 00:02:47.560 Exactly. Or, you know, some kind of complex math 00:02:47.560 --> 00:02:49.900 that you need to run. I saw a great example of 00:02:49.900 --> 00:02:53.340 this recently with messy API outputs. Oh, yeah. 00:02:53.500 --> 00:02:55.879 Like when you scrape a YouTube transcript, what 00:02:55.879 --> 00:02:58.259 you get back is often just a wall of text. Right. 00:02:58.300 --> 00:03:00.180 It's all mixed together. Time stamps, the words. 00:03:00.379 --> 00:03:03.080 All of it. And you need to split the actual. 00:03:03.080 --> 00:03:05.819 text from the timestamp into two clean fields 00:03:05.819 --> 00:03:08.919 so you can analyze it the code note is perfect 00:03:08.919 --> 00:03:11.240 for that fiddly little job you might go back 00:03:11.240 --> 00:03:12.819 and forth with the AI a little bit right give 00:03:12.819 --> 00:03:15.479 it the input show it the wrong output it gave 00:03:15.479 --> 00:03:17.759 you and explain why it's wrong a little refinement 00:03:17.759 --> 00:03:20.400 yeah but once that script is in the code note 00:03:20.400 --> 00:03:23.319 it's purely deterministic reliable predictable 00:03:24.030 --> 00:03:26.550 every single time. And that's the real power. 00:03:26.750 --> 00:03:29.270 It handles that stuff way faster and cheaper 00:03:29.270 --> 00:03:32.389 than asking a big expensive language model to 00:03:32.389 --> 00:03:35.229 try and fix the text on the fly. That just adds 00:03:35.229 --> 00:03:37.949 cost and slowness. So that brings up a question 00:03:37.949 --> 00:03:41.110 for me. If the AI is good enough to write the 00:03:41.110 --> 00:03:44.550 code for the node, why not just ask the AI agent 00:03:44.550 --> 00:03:47.330 itself to fix the formatting directly? Cut out 00:03:47.330 --> 00:03:49.250 the middleman. Great question. Because the code 00:03:49.250 --> 00:03:51.550 node is faster, cheaper, and perfectly consistent 00:03:51.550 --> 00:03:54.129 every time. okay so we've cleaned up our data 00:03:54.129 --> 00:03:56.189 now it's time to connect our workflow to the 00:03:56.189 --> 00:03:59.270 entire outside world and we start with note 13 00:03:59.270 --> 00:04:01.889 the webhook the easiest way to think of this 00:04:01.889 --> 00:04:05.129 is like a permanent digital mailbox and this 00:04:05.129 --> 00:04:07.289 is what lets your workflow act almost like its 00:04:07.289 --> 00:04:10.389 own service precisely it gives you a unique url 00:04:10.389 --> 00:04:13.729 that is always always listening And when any 00:04:13.729 --> 00:04:16.670 outside app sends data to that address, could 00:04:16.670 --> 00:04:19.509 be anything, a new order, a form submission, 00:04:19.870 --> 00:04:22.670 your workflow just wakes up instantly. So this 00:04:22.670 --> 00:04:24.709 is the magic bridge for connecting to apps that 00:04:24.709 --> 00:04:26.670 don't have a native integration. You got it. 00:04:26.769 --> 00:04:29.329 If there's an API, you can usually tell it to 00:04:29.329 --> 00:04:32.050 send a ping to a webhook. Right. So maybe you're 00:04:32.050 --> 00:04:34.529 scraping leads with some custom tool. Instead 00:04:34.529 --> 00:04:37.110 of, you know, exporting a CSV file and dragging 00:04:37.110 --> 00:04:39.110 it over, you just tell the tool to send the data 00:04:39.110 --> 00:04:41.490 straight to the webhook URL. And it goes right 00:04:41.490 --> 00:04:44.399 into your system. Like Google Sheets or a CRM. 00:04:44.459 --> 00:04:47.459 Instantly. It's all real time. So if the webhook 00:04:47.459 --> 00:04:51.500 is the ear that's listening, then node 14, respond 00:04:51.500 --> 00:04:53.920 to webhook, must be the mouth that answers back. 00:04:54.100 --> 00:04:55.920 It's the other half of the conversation. It's 00:04:55.920 --> 00:04:59.279 a crucial two -way street. This node sends data 00:04:59.279 --> 00:05:01.699 back to whatever app triggered the webhook in 00:05:01.699 --> 00:05:04.060 the first place. It completes the loop. And this 00:05:04.060 --> 00:05:06.360 is absolutely critical for anything that needs 00:05:06.360 --> 00:05:08.620 to happen in real time, like a website chatbot. 00:05:08.939 --> 00:05:12.100 Oh, without a doubt. Imagine a user types a message. 00:05:12.379 --> 00:05:15.259 The website sends it to your webhook. Your AI 00:05:15.259 --> 00:05:18.259 thinks for a second. If you don't use the respond 00:05:18.259 --> 00:05:20.879 node to send the answer back. The user's just 00:05:20.879 --> 00:05:23.680 stuck. Their browser just hangs, waiting. It 00:05:23.680 --> 00:05:25.939 just hangs forever. It'll time out. And there's 00:05:25.939 --> 00:05:28.060 a really important technical step here that people 00:05:28.060 --> 00:05:30.790 miss all the time. Okay, what is it? In the webhook 00:05:30.790 --> 00:05:33.709 trigger node, the respond setting usually defaults 00:05:33.709 --> 00:05:36.629 to immediately. Which means it responds before 00:05:36.629 --> 00:05:39.189 it does any work. Exactly. You have to change 00:05:39.189 --> 00:05:42.509 that setting to using respond to webhook node. 00:05:42.689 --> 00:05:45.110 That one little change tells the other system, 00:05:45.209 --> 00:05:47.250 hey, I got your message. Hang on. I'm working 00:05:47.250 --> 00:05:49.629 on the answer. That feels like a really important 00:05:49.629 --> 00:05:52.790 detail. It is. We run simple tests with tools 00:05:52.790 --> 00:05:54.930 like Postman all the time, just sending a message 00:05:54.930 --> 00:05:57.410 and using the respond node to send it right back 00:05:57.410 --> 00:05:59.370 to make sure that handshake is working perfectly. 00:05:59.399 --> 00:06:01.579 So what's the bottom line for the user if we 00:06:01.579 --> 00:06:04.420 forget to use that respond node? The system they're 00:06:04.420 --> 00:06:06.920 using will stall indefinitely. They'll get a 00:06:06.920 --> 00:06:09.879 timeout error or just think it's broken. Mid 00:06:09.879 --> 00:06:12.319 -roll sponsor, read placeholder for insertion. 00:06:12.399 --> 00:06:14.319 All right, now for the part that really changes 00:06:14.319 --> 00:06:17.139 the whole game. The three nodes that give your 00:06:17.139 --> 00:06:20.259 workflow a real intelligence. The AI engine. 00:06:20.480 --> 00:06:23.689 We start with node 15, the AI agent. This is 00:06:23.689 --> 00:06:26.009 the brain. So let's just define that. The AI 00:06:26.009 --> 00:06:28.269 agent node is what connects your workflow to 00:06:28.269 --> 00:06:31.069 a large language model, an LLM. Right, a program 00:06:31.069 --> 00:06:34.550 like ChatGPT or Claude. And it needs two main 00:06:34.550 --> 00:06:37.230 things from you. First is the system message. 00:06:37.449 --> 00:06:39.170 That's the instruction manual, the personality. 00:06:39.389 --> 00:06:41.790 You are a helpful assistant who only speaks in 00:06:41.790 --> 00:06:44.670 haikus. Exactly. And the second part is the input. 00:06:44.769 --> 00:06:47.490 So the user's actual question or the data you're 00:06:47.490 --> 00:06:49.170 feeding it. And this is where automation just 00:06:49.170 --> 00:06:51.529 fundamentally shifts from if this, then that. 00:06:51.980 --> 00:06:54.680 to an intelligent worker that can actually understand 00:06:54.680 --> 00:06:57.220 context. A good system message is everything, 00:06:57.339 --> 00:06:59.519 isn't it? It's what shapes the reality for the 00:06:59.519 --> 00:07:02.120 AI. If you just say, be helpful, you'll get random 00:07:02.120 --> 00:07:04.720 results. But if you specify the tone, the format, 00:07:04.839 --> 00:07:06.980 the constraints, that's where you get reliability. 00:07:07.459 --> 00:07:09.740 Okay, so the agent is the brain. That means node 00:07:09.740 --> 00:07:13.720 16, AI tools, must be the hands. That's it, because 00:07:13.720 --> 00:07:16.699 without this node, the AI is just a brain in 00:07:16.699 --> 00:07:19.750 a jar. It can think. But it can't do anything. 00:07:19.910 --> 00:07:22.230 It can't act on the world. The AI tools node 00:07:22.230 --> 00:07:25.089 is what connects that brain to other nodes in 00:07:25.089 --> 00:07:27.230 your workflow, giving it actual capabilities. 00:07:27.730 --> 00:07:29.410 And this is really the definition of an agent 00:07:29.410 --> 00:07:32.129 in the jargon. It's an LLM that has access to 00:07:32.129 --> 00:07:34.230 tools. Exactly right. And this is where you get 00:07:34.230 --> 00:07:37.240 that. that moment of wonder. Yeah. Imagine you 00:07:37.240 --> 00:07:40.160 give the AI a tool called Google Calendar and 00:07:40.160 --> 00:07:43.579 you ask it, hey, am I free at 2 p .m.? Instead 00:07:43.579 --> 00:07:46.620 of guessing, the AI recognizes it needs to use 00:07:46.620 --> 00:07:49.339 the tool. It actually checks your real calendar 00:07:49.339 --> 00:07:51.980 and gives you an accurate answer based on reality. 00:07:52.480 --> 00:07:57.000 Whoa. Wait. So imagine scaling that across an 00:07:57.000 --> 00:07:59.240 entire business. Yeah. You're giving an AI hands 00:07:59.240 --> 00:08:02.680 to check inventory or send a refund. That's a 00:08:02.680 --> 00:08:05.120 different game entirely. It moves the AI from 00:08:05.120 --> 00:08:08.100 being a passive writer to an active part of your 00:08:08.100 --> 00:08:10.879 operations. But that kind of power needs some 00:08:10.879 --> 00:08:12.959 guardrails. Right. Which brings us to our last 00:08:12.959 --> 00:08:15.920 node, number 17, the structured output parser. 00:08:15.920 --> 00:08:18.040 We call it the translator. And the problem it 00:08:18.040 --> 00:08:20.139 solves is that AI models love to talk. They're 00:08:20.139 --> 00:08:22.199 trained in human language, so they ramble. They 00:08:22.199 --> 00:08:25.420 ramble, they use emojis, but databases and computers, 00:08:25.620 --> 00:08:28.639 they need strict structure data, not poetry. 00:08:28.759 --> 00:08:31.279 They need data points. Exactly. So the parser 00:08:31.279 --> 00:08:34.860 forces the AI agent to reply in a specific machine 00:08:34.860 --> 00:08:38.120 -readable format, usually JSON. You define the 00:08:38.120 --> 00:08:40.320 exact structure it has to follow. Now, I hear 00:08:40.320 --> 00:08:42.990 this a lot. People say, well, in my system message, 00:08:43.110 --> 00:08:45.269 I just tell the AI to always use JSON, so why 00:08:45.269 --> 00:08:47.750 do I need this extra node? And that is the most 00:08:47.750 --> 00:08:49.970 important technical point here. A system message 00:08:49.970 --> 00:08:53.590 requests JSON. It does not guarantee it. LLMs 00:08:53.590 --> 00:08:56.230 can have prompt drift or they can just mess up 00:08:56.230 --> 00:08:58.450 the syntax. Just one missing bracket. And the 00:08:58.450 --> 00:09:00.730 whole thing breaks. The next node fails. A whole 00:09:00.730 --> 00:09:03.149 thing fails. The parser is a hard validator. 00:09:03.250 --> 00:09:06.129 It forces the output to be clean. So instead 00:09:06.129 --> 00:09:08.149 of the AI saying, well, I looked at the email 00:09:08.149 --> 00:09:10.789 and John Smith seems kind of angry. You get clean, 00:09:10.789 --> 00:09:13.169 actionable data. You get a JSON object that says 00:09:13.169 --> 00:09:17.169 sender. John Smith sentiment. Angry. It guarantees 00:09:17.169 --> 00:09:20.070 reliability. If you are saving to a database. 00:09:20.639 --> 00:09:23.259 It has to be perfect every single time. The parser 00:09:23.259 --> 00:09:25.480 makes that happen. So even if we tell the AI 00:09:25.480 --> 00:09:28.539 and the system message to use JSON, we still 00:09:28.539 --> 00:09:30.960 need the parser. The parser provides guaranteed 00:09:30.960 --> 00:09:34.480 reliability. It forces clean data for consistent 00:09:34.480 --> 00:09:37.600 database input. We have covered so much ground. 00:09:37.779 --> 00:09:39.940 If you take the foundation nodes we did in part 00:09:39.940 --> 00:09:43.139 one and you add these seven intelligent superpowers 00:09:43.139 --> 00:09:45.419 we just talked about. The webhooks, the code 00:09:45.419 --> 00:09:48.220 node, the whole AI suite. You now have these 00:09:48.220 --> 00:09:51.200 17 essential nodes. And this is really the big 00:09:51.200 --> 00:09:53.600 idea we want you to leave with. These 17 nodes 00:09:53.600 --> 00:09:55.960 are the Lego bricks. You don't need to learn 00:09:55.960 --> 00:09:58.159 all 500 plus nodes that are out there. I can 00:09:58.159 --> 00:10:01.120 feel overwhelming. Totally. But with just these 00:10:01.120 --> 00:10:05.159 17, you can build probably 90 % of any complex 00:10:05.159 --> 00:10:07.820 automation you can dream up. Content generators, 00:10:08.159 --> 00:10:10.440 support bots, business dashboards, you name it. 00:10:10.480 --> 00:10:12.720 So the path from beginner to builder isn't about 00:10:12.720 --> 00:10:15.120 knowing everything. It's about mastering these 00:10:15.120 --> 00:10:17.559 powerful few. It's about connecting the logical 00:10:17.559 --> 00:10:19.980 body of your workflow to the intelligent mind 00:10:19.980 --> 00:10:21.820 of the agent. I love how the source material 00:10:21.820 --> 00:10:24.820 puts it. Focus determines reality. So your next 00:10:24.820 --> 00:10:27.100 step isn't to read another guide. It's to pick 00:10:27.100 --> 00:10:29.200 one of these nodes, maybe the respond to webhook, 00:10:29.279 --> 00:10:32.500 maybe the AI agent, and just try to build something 00:10:32.500 --> 00:10:35.059 with it. Start small, but start now. Thank you 00:10:35.059 --> 00:10:38.039 for joining us for this deep dive. Now go build 00:10:38.039 --> 00:10:38.639 something amazing.