WEBVTT 00:00:00.000 --> 00:00:03.279 ever felt just buried under a mountain of digital 00:00:03.279 --> 00:00:05.400 tasks. Look constantly, yeah. Like checking your 00:00:05.400 --> 00:00:08.099 calendar, waiting through emails, trying to find 00:00:08.099 --> 00:00:11.000 that one meeting slot that works. It's endless. 00:00:11.259 --> 00:00:14.359 It really is. I mean, what if you had a super 00:00:14.359 --> 00:00:17.519 smart assistant that could just handle it, not 00:00:17.519 --> 00:00:20.160 just react when you ask, but actually sort of 00:00:20.160 --> 00:00:24.579 anticipate what you need and proactively do things, 00:00:24.820 --> 00:00:26.859 freeing you up for the important stuff? That 00:00:26.859 --> 00:00:29.350 sounds pretty good. Well, stop wishing. This 00:00:29.350 --> 00:00:32.030 isn't science fiction anymore. Well, it's becoming 00:00:32.030 --> 00:00:34.109 a tangible reality, and it's powered by something 00:00:34.109 --> 00:00:37.719 called AI agents. Right. So today, we're diving 00:00:37.719 --> 00:00:40.759 deep into a guide, a really comprehensive one, 00:00:41.020 --> 00:00:43.719 on how you can build your very own AI agent, 00:00:43.719 --> 00:00:46.000 like from scratch. Even without coding experience. 00:00:46.200 --> 00:00:48.119 Exactly. Even if you've never written a line 00:00:48.119 --> 00:00:50.340 of code, the promise from the material we're 00:00:50.340 --> 00:00:53.020 looking at is, frankly, pretty incredible. It 00:00:53.020 --> 00:00:55.280 says by the end of this deep dive, you'll not 00:00:55.280 --> 00:00:59.340 only grasp the core power of AI agents, but you'll 00:00:59.340 --> 00:01:01.179 actually have the know -how to create a functional 00:01:01.179 --> 00:01:03.880 personal assistant, one that manages your schedule 00:01:03.880 --> 00:01:08.099 and emails automatically. Wow. OK. So yeah, let's 00:01:08.099 --> 00:01:10.939 get into it. Yeah, let's do it. What's really 00:01:10.939 --> 00:01:14.819 fascinating here, I think, is how these AI agents 00:01:14.819 --> 00:01:19.120 represent a fundamental shift in how we interact 00:01:19.120 --> 00:01:22.140 with technology. A shift how? Well, they're not 00:01:22.140 --> 00:01:24.819 just about processing information or spitting 00:01:24.819 --> 00:01:27.879 out an answer. They're about taking action. Action, 00:01:27.900 --> 00:01:32.239 OK. Think of an AI agent as an intelligent computer 00:01:32.239 --> 00:01:35.709 program that can perceive its environment. Understand 00:01:35.709 --> 00:01:39.069 what's going on, reason through situations, and 00:01:39.069 --> 00:01:40.969 then act independently to achieve a goal you've 00:01:40.969 --> 00:01:43.150 given it. Ah, so it's moving beyond just simple 00:01:43.150 --> 00:01:45.530 automation. It's more like a digital partner. 00:01:45.849 --> 00:01:47.969 Exactly. Yeah. That's a great way to put it. 00:01:48.469 --> 00:01:50.810 Genuine digital partnership. So it's not just 00:01:50.810 --> 00:01:53.129 smart. It actually does things. That's the game 00:01:53.129 --> 00:01:55.150 changer. That is absolutely the core difference. 00:01:55.310 --> 00:01:57.310 And what really stands out in our source material 00:01:57.310 --> 00:01:59.370 here is that these aren't your typical programs. 00:02:00.030 --> 00:02:02.689 First, they're autonomous. Which means you don't 00:02:02.689 --> 00:02:04.689 need to babysit them, right? You give it a mission 00:02:04.689 --> 00:02:08.189 and it figures out the path and just goes. Precisely. 00:02:08.469 --> 00:02:11.330 And unlike traditional sort of rigid programs, 00:02:11.969 --> 00:02:14.650 AI agents are also incredibly adaptable. Adaptable. 00:02:14.750 --> 00:02:16.969 Meaning? They learn from experience. They're 00:02:16.969 --> 00:02:20.250 constantly figuring out the most optimal way 00:02:20.250 --> 00:02:23.409 to complete a task. Plus, and this is key, They 00:02:23.409 --> 00:02:26.849 have memory. Memory, like remembering past conversations. 00:02:27.069 --> 00:02:29.569 Exactly. They can recall past interactions, which 00:02:29.569 --> 00:02:32.110 lets them make decisions with full context. So 00:02:32.110 --> 00:02:34.229 they become more useful, more personalized over 00:02:34.229 --> 00:02:36.949 time. OK. OK. And here's the bit that feels really 00:02:36.949 --> 00:02:39.969 transformative. They can use tools. Yes. This 00:02:39.969 --> 00:02:42.610 is called out as maybe the most critical differentiator. 00:02:42.849 --> 00:02:44.969 They aren't stuck in their own little digital 00:02:44.969 --> 00:02:47.550 box. Right. They can connect to and actually 00:02:47.550 --> 00:02:49.909 use the applications you use every day, your 00:02:49.909 --> 00:02:54.169 Gmail, Google Calendar, Slack. maybe even Notion, 00:02:54.530 --> 00:02:56.509 hundreds of services. That's literally how they 00:02:56.509 --> 00:02:58.930 get work done for you in the real digital world. 00:02:59.110 --> 00:03:01.650 Yeah, it's a bit like giving a super smart intern 00:03:01.650 --> 00:03:03.770 access to all your office software, isn't it? 00:03:03.909 --> 00:03:08.229 That's a good analogy. This capability, it often 00:03:08.229 --> 00:03:11.090 raises a good question. How do these AI agents 00:03:11.090 --> 00:03:14.469 differ from other AI concepts people might already 00:03:14.469 --> 00:03:17.569 know? Right, like chat bots or automation, easy 00:03:17.569 --> 00:03:19.750 to mix them up. Very easy. Yeah, let's clear 00:03:19.750 --> 00:03:24.409 that up. So chat bots, like say, Chat GPT. They're 00:03:24.409 --> 00:03:26.189 amazing at taking your prompt and giving you 00:03:26.189 --> 00:03:29.090 an answer, right? Text, images, code. Very good 00:03:29.090 --> 00:03:32.030 at generation. You ask, it responds. But the 00:03:32.030 --> 00:03:34.810 limitation is clear. The conversation is kind 00:03:34.810 --> 00:03:37.490 of the endpoint. They can't then go out and like 00:03:37.490 --> 00:03:39.810 take action in other apps. If you ask a chat 00:03:39.810 --> 00:03:42.020 bot, what's the weather like? Huh, it'll tell 00:03:42.020 --> 00:03:43.939 you. But it won't then go book you a meeting 00:03:43.939 --> 00:03:47.319 because it's raining or order you a car. It's 00:03:47.319 --> 00:03:49.560 world stops at the chat window. Right. Its interaction 00:03:49.560 --> 00:03:51.479 is contained. And you have traditional automation 00:03:51.479 --> 00:03:55.400 tools, like Zapier or IFTTT. Ah, yeah, the IFTEN 00:03:55.400 --> 00:03:58.060 stuff. Exactly. These are excellent for executing 00:03:58.060 --> 00:04:00.419 a sequence of actions based on rules you pre 00:04:00.419 --> 00:04:02.740 -configure. Like, if I got a new email from my 00:04:02.740 --> 00:04:05.340 boss, then automatically forward it. Simple, 00:04:05.759 --> 00:04:07.979 effective for specific tasks. But the limitation 00:04:07.979 --> 00:04:10.259 there is they can't really think or adapt, can 00:04:10.259 --> 00:04:13.080 they? Not at all. If something unexpected happens, 00:04:13.379 --> 00:04:16.459 something not in the rule, they just fail. They're 00:04:16.459 --> 00:04:19.000 essentially fixed scripts. They can't improvise. 00:04:19.160 --> 00:04:21.500 Okay. And that's where AI agents really shine. 00:04:21.579 --> 00:04:23.740 They're the perfect combo, it sounds like. Exactly. 00:04:23.920 --> 00:04:26.500 They bring the intelligence, the reasoning of 00:04:26.500 --> 00:04:29.560 a chatbot, together with the action -taking ability 00:04:29.560 --> 00:04:32.160 of automation tools. So they understand complex 00:04:32.160 --> 00:04:35.240 requests, analyze things, make smart decisions, 00:04:35.279 --> 00:04:37.980 and then use those tools to execute actions. 00:04:38.259 --> 00:04:42.160 Yes. Imagine telling an AI agent, OK, check my 00:04:42.160 --> 00:04:44.600 emails from potential clients this week, look 00:04:44.600 --> 00:04:47.800 at my free time, suggest the three best meeting 00:04:47.800 --> 00:04:50.939 slots for next week, and then draft an invitation 00:04:50.939 --> 00:04:54.250 email for each one. Wow. That's a whole other 00:04:54.250 --> 00:04:56.269 level. Traditional automation or chatbots just 00:04:56.269 --> 00:04:58.410 can't do that. Not even close. It's like having 00:04:58.410 --> 00:05:00.629 a smart assistant who can actually do what you 00:05:00.629 --> 00:05:03.329 ask, not just tell you how or give you information. 00:05:03.550 --> 00:05:05.269 Precisely. And if we connect this to the bigger 00:05:05.269 --> 00:05:09.389 picture, an AI agent operates on this powerful 00:05:09.389 --> 00:05:11.910 four -step thinking action loop. That's what 00:05:11.910 --> 00:05:14.290 keeps it moving forward. What are the steps? 00:05:14.649 --> 00:05:17.649 It all begins with you setting a goal. You give 00:05:17.649 --> 00:05:20.149 the agent a specific task or maybe a general 00:05:20.149 --> 00:05:22.790 objective like, help me manage customer support 00:05:22.790 --> 00:05:26.370 requests. Got it. Goal first. Then what? Then 00:05:26.370 --> 00:05:29.269 the agent moves into information gathering. It 00:05:29.269 --> 00:05:31.089 uses its authorized tools, the ones you give 00:05:31.089 --> 00:05:33.790 it access to, to collect all the necessary data. 00:05:34.209 --> 00:05:36.850 So reading emails, scanning the calendar, checking 00:05:36.850 --> 00:05:40.310 files, whatever it needs. Exactly. Accessing 00:05:40.310 --> 00:05:43.310 Google Drive, checking order history in a shop 00:05:43.310 --> 00:05:45.769 system, anything relevant to understand the full 00:05:45.769 --> 00:05:48.980 picture. Okay. Goal. Gather info. Next must be 00:05:48.980 --> 00:05:51.839 thinking. You got it. Decision making. Based 00:05:51.839 --> 00:05:54.139 on the goal and all the info it's gathered, the 00:05:54.139 --> 00:05:57.040 AI model, which is essentially the agent's brain, 00:05:57.160 --> 00:05:59.680 reasons and decides on the best next action. 00:06:00.079 --> 00:06:02.040 Like, should it answer a question directly, ask 00:06:02.040 --> 00:06:04.839 for more details, process a refund, maybe? Precisely. 00:06:04.899 --> 00:06:07.000 It's constantly evaluating the best path forward. 00:06:07.160 --> 00:06:10.000 And finally, the magic happens. Action. Yes. 00:06:10.439 --> 00:06:13.540 The agent doesn't just suggest. It actually uses 00:06:13.540 --> 00:06:16.040 its tools to perform the action it decided on. 00:06:16.319 --> 00:06:18.540 Sending an email, creating a calendar event, 00:06:18.899 --> 00:06:22.259 updating a CRM system. What's a CRM again? Oh, 00:06:22.259 --> 00:06:24.860 sorry. Customer Relationship Management Tool, 00:06:25.540 --> 00:06:29.639 like HubSpot or Salesforce. Businesses use them 00:06:29.639 --> 00:06:32.120 to manage customer interactions. Right, right. 00:06:32.120 --> 00:06:34.459 So it could update that or reply to a customer's 00:06:34.459 --> 00:06:36.639 message. It truly gets things done. OK, that 00:06:36.639 --> 00:06:38.980 loop makes sense. Goal, gather, decide, act. 00:06:39.590 --> 00:06:42.310 So to actually build one of these effective agents, 00:06:42.410 --> 00:06:45.269 what do you need? The guide mentions three essential 00:06:45.269 --> 00:06:47.990 components. It does. Three core parts. First, 00:06:48.110 --> 00:06:50.949 there's the AI model or the brain. This is the 00:06:50.949 --> 00:06:52.949 core intelligence, providing all the reasoning 00:06:52.949 --> 00:06:55.129 and language smarts. And you have choices here, 00:06:55.329 --> 00:06:58.689 like different AI brains. You do. Open AI models 00:06:58.689 --> 00:07:01.870 like GPT -4 .0, Anthropics Cloud 3, Google's 00:07:01.870 --> 00:07:04.470 Gemini, or even Meta's Llama 3, which is interesting 00:07:04.470 --> 00:07:07.029 because it's open source, gives you options for 00:07:07.100 --> 00:07:09.459 Deep customization. Okay, the brain. What's second? 00:07:09.579 --> 00:07:12.399 Second is memory or the experience. This is crucial 00:07:12.399 --> 00:07:14.779 as we touched on. It allows your agent to remember 00:07:14.779 --> 00:07:16.819 previous interactions within the same session. 00:07:16.959 --> 00:07:19.100 So it makes better decisions based on history, 00:07:19.560 --> 00:07:22.519 provides consistency. Exactly. And importantly, 00:07:23.040 --> 00:07:25.899 it learns from the actions it has taken. It gets 00:07:25.899 --> 00:07:28.300 smarter and more useful over time because it 00:07:28.300 --> 00:07:31.240 remembers what happened before. Right. Brain, 00:07:31.560 --> 00:07:35.360 memory, and third must be the tools. You nailed 00:07:35.360 --> 00:07:38.509 it. tools, or the hands of the agent. These are 00:07:38.509 --> 00:07:40.410 what allow it to actually interact with the digital 00:07:40.410 --> 00:07:43.290 world and take action. So connecting to services 00:07:43.290 --> 00:07:45.889 like Google Calendar? To check or create events, 00:07:46.009 --> 00:07:49.129 yeah. Or Gmail to read and send emails. Slack 00:07:49.129 --> 00:07:51.189 for notifications. Google Drive for documents. 00:07:51.329 --> 00:07:53.350 Notion for notes. These are the limbs that let 00:07:53.350 --> 00:07:56.149 it reach out and, you know, do stuff. Perfectly 00:07:56.149 --> 00:07:58.689 put. Thinking about real -world applications. 00:07:59.759 --> 00:08:02.220 The material we're looking at gives some fantastic 00:08:02.220 --> 00:08:04.000 examples of agents you could build right now. 00:08:04.160 --> 00:08:06.259 Yeah, some really practical ones. Like, imagine 00:08:06.259 --> 00:08:09.420 a personal email assistant. One that automatically 00:08:09.420 --> 00:08:12.439 categorizes your emails, maybe filters spam better, 00:08:12.899 --> 00:08:15.379 drafts common replies for you, or even summarizes 00:08:15.379 --> 00:08:17.720 those super long threads. That alone would be 00:08:17.720 --> 00:08:19.560 amazing. No more getting lost in your inbox. 00:08:19.980 --> 00:08:22.959 Right. Or how about a social media content planner? 00:08:23.370 --> 00:08:26.329 Something that researches trends, schedules your 00:08:26.329 --> 00:08:29.250 posts, costs platforms, and maybe even analyzes 00:08:29.250 --> 00:08:32.490 engagement. Or think about business uses. A customer 00:08:32.490 --> 00:08:35.870 service agent that answers FAQs 247, handles 00:08:35.870 --> 00:08:39.169 simple payment issues, or just gathers the initial 00:08:39.169 --> 00:08:42.370 info before escalating to a human, that could 00:08:42.370 --> 00:08:44.629 save so much time. Definitely. You could even 00:08:44.629 --> 00:08:47.629 build a lead management agent. Imagine it tracking 00:08:47.629 --> 00:08:50.149 new leads from your website, sending personalized 00:08:50.149 --> 00:08:52.870 follow -up emails automatically. Scoring those 00:08:52.870 --> 00:08:55.710 leads based on interaction, updating your CRM 00:08:55.710 --> 00:08:59.110 system. Wow. The possibilities really are vast. 00:08:59.289 --> 00:09:01.710 It's about identifying those repetitive or time 00:09:01.710 --> 00:09:03.850 -consuming tasks in your own workflow, isn't 00:09:03.850 --> 00:09:05.490 it? Absolutely, finding those friction points. 00:09:05.730 --> 00:09:07.929 Now let's get to the nitty -gritty, the step 00:09:07.929 --> 00:09:10.470 -by -step guide on building your very first AI 00:09:10.470 --> 00:09:12.629 agent. Okay, the fun part. We're going to aim 00:09:12.629 --> 00:09:14.450 to create that personal assistant we talked about 00:09:14.450 --> 00:09:16.129 earlier, right? The one that manages your calendar 00:09:16.129 --> 00:09:17.970 and emails. The one that checks the calendar, 00:09:18.409 --> 00:09:20.929 reads specific emails, suggests meeting times, 00:09:21.049 --> 00:09:23.700 and sends a report. Exactly. Checks the calendar 00:09:23.700 --> 00:09:26.580 for the next seven days, reads emails labeled 00:09:26.580 --> 00:09:30.120 client, analyzes both to suggest optimal meeting 00:09:30.120 --> 00:09:33.399 times, and then sends you a nice formatted daily 00:09:33.399 --> 00:09:36.159 report email with those suggestions. And it runs 00:09:36.159 --> 00:09:38.240 automatically? Yep. Automatically every day at 00:09:38.240 --> 00:09:40.820 9 a .m. And the best part, the guide says you 00:09:40.820 --> 00:09:44.409 can do this with no code. How? What platform 00:09:44.409 --> 00:09:47.370 are they suggesting? They recommend using a platform 00:09:47.370 --> 00:09:50.809 called N8n. Apparently, it's really visual, like 00:09:50.809 --> 00:09:53.950 a drag -and -drop interface. Ah, N8n. Yeah, I 00:09:53.950 --> 00:09:56.309 know it. It's quite powerful. Visual workflow 00:09:56.309 --> 00:09:59.789 automation. That's the one. No code needed, hundreds 00:09:59.789 --> 00:10:01.950 of app integrations built in, and it's pretty 00:10:01.950 --> 00:10:04.070 affordable too, so perfect for getting started. 00:10:04.389 --> 00:10:06.289 Sounds like a good choice for beginners, definitely. 00:10:06.409 --> 00:10:09.370 Okay, so once you're on N8n, first step, set 00:10:09.370 --> 00:10:11.470 up your trigger. This is what kicks things off. 00:10:11.690 --> 00:10:13.470 Right, the starting gun. For our personal assistant, 00:10:13.529 --> 00:10:16.029 you'd choose a schedule trigger and set it to 00:10:16.029 --> 00:10:19.789 run daily at 9 0 a .m. Simple. Get your brief 00:10:19.789 --> 00:10:21.429 when you start your day. And there are other 00:10:21.429 --> 00:10:23.529 triggers too, presumably. Manual ones for testing. 00:10:23.830 --> 00:10:26.389 Yeah, manual trigger for testing or even a chat 00:10:26.389 --> 00:10:28.129 trigger if you wanted to build something more 00:10:28.129 --> 00:10:30.190 conversational later on. Okay, trigger set. What's 00:10:30.190 --> 00:10:32.769 next? Next, you literally click a plus button 00:10:32.769 --> 00:10:35.549 after the trigger and select the AI agent core 00:10:35.549 --> 00:10:38.570 itself. There's a specific AI agent node. Oh, 00:10:38.570 --> 00:10:40.899 so that's the central hub. The brain box for 00:10:40.899 --> 00:10:43.179 the workflow. Exactly. It acts as the central 00:10:43.179 --> 00:10:45.240 processing unit where all the intelligent decisions 00:10:45.240 --> 00:10:47.919 happen. Got it. After adding that node, you configure 00:10:47.919 --> 00:10:51.139 the AI model, the brain. Inside that AI agent 00:10:51.139 --> 00:10:53.779 node, you pick your AI provider. Let's say open 00:10:53.779 --> 00:10:57.759 AI. OK. And which specific model? The guide recommends 00:10:57.759 --> 00:11:01.899 GPT -4 Mini. Says it's fast, super cost effective, 00:11:02.080 --> 00:11:04.720 and great for beginners and these kinds of daily 00:11:04.720 --> 00:11:06.779 tasks. Makes sense. You probably need an API 00:11:06.779 --> 00:11:09.279 key for that. Yep. You need an OpenAI API key? 00:11:09.519 --> 00:11:12.799 Think of it like a password for NAN to use OpenAI's 00:11:12.799 --> 00:11:15.100 models. You get one from the OpenAI platform 00:11:15.100 --> 00:11:17.980 website and paste it into NADN. And just a heads 00:11:17.980 --> 00:11:21.039 up for listeners, OpenAI API usage isn't free, 00:11:21.399 --> 00:11:23.379 but the mini model is usually very cheap for 00:11:23.379 --> 00:11:26.120 this kind of thing. Good point. Very affordable 00:11:26.120 --> 00:11:28.519 for testing and running daily tasks like this. 00:11:28.860 --> 00:11:31.720 OK, so brain configured. Then you add memory. 00:11:32.120 --> 00:11:34.490 Right, the experience component. Inside that 00:11:34.490 --> 00:11:37.990 same AI agent node, you add simple memory. For 00:11:37.990 --> 00:11:40.409 a daily report like ours, the guide suggests 00:11:40.409 --> 00:11:43.389 a context window length of five is fine. So it 00:11:43.389 --> 00:11:45.789 remembers the last few steps or interactions. 00:11:46.129 --> 00:11:48.110 Exactly. For something more complex, like that 00:11:48.110 --> 00:11:50.110 customer service agent idea, you'd increase it 00:11:50.110 --> 00:11:53.230 maybe 25 or more, so it remembers longer conversations. 00:11:53.429 --> 00:11:56.649 Okay. Memory added. Now for the tools. The hands. 00:11:57.070 --> 00:11:59.750 Precisely. Next up, you connect the Google Calendar 00:11:59.750 --> 00:12:02.230 tool. This lets your agent read your schedule. 00:12:02.509 --> 00:12:05.429 How do you set that up? You configure it to get 00:12:05.429 --> 00:12:08.029 many events from your main calendar. Set the 00:12:08.029 --> 00:12:10.690 start time to now, and the end time to one week 00:12:10.690 --> 00:12:12.830 from now. Makes sense. Covering the upcoming 00:12:12.830 --> 00:12:15.610 week. And crucially, you rename this tool something 00:12:15.610 --> 00:12:18.389 that AI can easily understand, like read appointments. 00:12:18.809 --> 00:12:21.629 Needs to be descriptive. Good tip. Clear naming 00:12:21.629 --> 00:12:24.129 helps the AI know what tool does what. Then you 00:12:24.129 --> 00:12:26.029 connect the Gmail tool for reading emails. This 00:12:26.029 --> 00:12:28.049 is for checking those important client messages. 00:12:28.210 --> 00:12:30.929 OK, similar setup. Pretty much. configure it 00:12:30.929 --> 00:12:33.850 to get many emails, maybe limit it to, say, 50 00:12:33.850 --> 00:12:37.190 so it doesn't pull too much, and filter by a 00:12:37.190 --> 00:12:39.649 specific label, like client. Ah, so you need 00:12:39.649 --> 00:12:41.610 to have that label set up in Gmail first. Yeah. 00:12:41.789 --> 00:12:43.549 And maybe a filter to apply it automatically. 00:12:43.789 --> 00:12:45.629 Exactly. Good catch. You need to create that 00:12:45.629 --> 00:12:48.230 client label in your Gmail settings, and ideally 00:12:48.230 --> 00:12:50.610 set up a filter so relevant emails get tagged 00:12:50.610 --> 00:12:53.629 automatically beforehand. And again, rename the 00:12:53.629 --> 00:12:56.730 tool clearly. Read client emails. OK, calendar 00:12:56.730 --> 00:12:58.909 reading, email reading. What's the last tool? 00:12:59.230 --> 00:13:01.830 Sending the report. You got it. For the final 00:13:01.830 --> 00:13:04.429 hand, you connect another Gmail tool. This one 00:13:04.429 --> 00:13:07.110 for sending the report email. So same tool, different 00:13:07.110 --> 00:13:09.950 function. Right. Set the operation to send. Put 00:13:09.950 --> 00:13:13.009 your own email address as the recipient. And 00:13:13.009 --> 00:13:15.610 here's a key part. Enable the use expression 00:13:15.610 --> 00:13:17.990 toggles for both the subject and the message 00:13:17.990 --> 00:13:20.129 body. Why is that important? Because it lets 00:13:20.129 --> 00:13:22.789 the AI define the subject and the message content 00:13:22.789 --> 00:13:25.799 dynamically each time. It's not just a canned 00:13:25.799 --> 00:13:29.000 message. The AI actually writes the report for 00:13:29.000 --> 00:13:30.720 you based on what it found. Oh, very cool. So 00:13:30.720 --> 00:13:33.000 it crafts the output itself. Yeah, and rename 00:13:33.000 --> 00:13:35.360 this tool too. Maybe send a port email. Calendar, 00:13:35.840 --> 00:13:37.480 reading email, sending emails, got the tools 00:13:37.480 --> 00:13:41.159 connected. What now? The instructions. Now for 00:13:41.159 --> 00:13:43.759 arguably the most critical part. Writing the 00:13:43.759 --> 00:13:46.299 system prompt. This is basically the soul of 00:13:46.299 --> 00:13:48.379 the AI agent. The job description you give the 00:13:48.379 --> 00:13:51.919 AI. Exactly. You tell it clearly. You are a proactive 00:13:51.919 --> 00:13:54.139 scheduling assistant. Your job is to help me 00:13:54.139 --> 00:13:56.940 manage my time and client communications. So 00:13:56.940 --> 00:13:59.440 define its role first. Then you list its available 00:13:59.440 --> 00:14:02.379 tools. read appointments, read client emails, 00:14:02.600 --> 00:14:04.679 send report email, and then you give it a really 00:14:04.679 --> 00:14:07.759 clear step -by -step workflow. Like a mini program 00:14:07.759 --> 00:14:10.279 for the AI to follow. Kind of, yeah. First, read 00:14:10.279 --> 00:14:12.919 appointments. Second, read client emails. Third, 00:14:13.200 --> 00:14:16.299 analyze both. Fourth, suggest optimal meeting 00:14:16.299 --> 00:14:19.250 times based on that analysis. Fifth, draft an 00:14:19.250 --> 00:14:22.129 HTML -formatted report email. Sixth, send it 00:14:22.129 --> 00:14:24.210 using the Send Report Email tool. Wow, really 00:14:24.210 --> 00:14:26.669 specific. And you even specify the exact structure 00:14:26.669 --> 00:14:29.129 you want for that report email. Subject line 00:14:29.129 --> 00:14:31.590 format, sections for schedule summary, email 00:14:31.590 --> 00:14:34.370 summary, meeting suggestions, and maybe recommended 00:14:34.370 --> 00:14:37.549 next actions. That level of detail is absolutely 00:14:37.549 --> 00:14:40.570 paramount. It tells the agent exactly how to 00:14:40.570 --> 00:14:43.409 think and what output you expect. Less ambiguity 00:14:43.409 --> 00:14:46.090 means better results. Totally. OK, prompt written. 00:14:46.330 --> 00:14:49.139 Almost there. Next, you configure the memory 00:14:49.139 --> 00:14:51.460 session ID. Why do you need that? Since our agent 00:14:51.460 --> 00:14:54.460 is triggered by a schedule, not a chat, you need 00:14:54.460 --> 00:14:57.200 a fixed session ID for the memory, like just 00:14:57.200 --> 00:14:59.659 typing in daily report. Ah, so the memory is 00:14:59.659 --> 00:15:02.039 consistent each time it runs automatically. It 00:15:02.039 --> 00:15:04.399 always uses the same memory slot for this daily 00:15:04.399 --> 00:15:06.899 task. Exactly, ensures it has the right context 00:15:06.899 --> 00:15:10.340 each day. Makes sense, and then test it. Absolutely. 00:15:10.910 --> 00:15:13.870 Testing your AI agent is crucial. Before you 00:15:13.870 --> 00:15:16.250 activate the automatic schedule, you swap the 00:15:16.250 --> 00:15:18.730 schedule trigger out for a manual trigger temporarily. 00:15:18.990 --> 00:15:21.350 So you can run it on demand? Yep. Click the test 00:15:21.350 --> 00:15:23.870 workflow button, and you can literally watch 00:15:23.870 --> 00:15:27.230 each node light up as it executes trigger, AI 00:15:27.230 --> 00:15:31.190 agent thinking, calendar fetch, Gmail read, Gmail 00:15:31.190 --> 00:15:33.789 send. Like watching the digital gears turn. It 00:15:33.789 --> 00:15:35.789 really is. And if you get errors, you know where 00:15:35.789 --> 00:15:38.350 to look. Double check your API keys. Maybe Google 00:15:38.350 --> 00:15:40.690 permissions weren't granted correctly or that 00:15:40.690 --> 00:15:43.629 Gmail label wasn't set up right. Standard debugging, 00:15:43.929 --> 00:15:46.350 but in a visual way. Exactly. It's like watching 00:15:46.350 --> 00:15:48.570 a little digital ballet when it works. Nice. 00:15:48.750 --> 00:15:50.889 And once you've got that basic version working, 00:15:51.389 --> 00:15:53.769 the guide provides some excellent ways to expand 00:15:53.769 --> 00:15:56.429 and enhance your agent. Make it even better. 00:15:56.789 --> 00:16:00.029 Oh. Like what? Well, you can improve your system 00:16:00.029 --> 00:16:03.000 prompts, get more specific. Maybe ask for the 00:16:03.000 --> 00:16:05.580 output in a clear format like JSON, which is 00:16:05.580 --> 00:16:08.139 useful if another system needs to read it. Or 00:16:08.139 --> 00:16:10.700 add examples. Yes. Add examples for a few -shot 00:16:10.700 --> 00:16:12.779 prompting. That's where you show the AI a couple 00:16:12.779 --> 00:16:14.759 of good examples of the output you want so it 00:16:14.759 --> 00:16:17.600 learns the pattern. You can also define specific 00:16:17.600 --> 00:16:20.799 tone of voice, formal, friendly, et cetera. That 00:16:20.799 --> 00:16:22.279 makes a big difference. You can also probably 00:16:22.279 --> 00:16:25.879 use expressions. for dynamic data, right? Definitely. 00:16:26.019 --> 00:16:28.059 Allowing your agent to handle dates dynamically 00:16:28.059 --> 00:16:30.799 or pull specific customer names from earlier 00:16:30.799 --> 00:16:33.299 steps in the workflow makes it much more flexible 00:16:33.299 --> 00:16:36.220 than hard coding things. Absolutely. And beyond 00:16:36.220 --> 00:16:38.580 prompts, you can connect more tools, presumably. 00:16:38.679 --> 00:16:41.419 For sure. Expand its capabilities further. Integrate 00:16:41.419 --> 00:16:44.279 with CRM systems like HubSpot or Salesforce, 00:16:44.960 --> 00:16:47.279 project management tools like Trello or Asana. 00:16:47.519 --> 00:16:50.519 Or even databases like Airtable or Google Sheets 00:16:50.519 --> 00:16:53.159 to read or write data. The more tools you give 00:16:53.159 --> 00:16:56.360 it, the more complex tasks it can handle. Exactly. 00:16:57.139 --> 00:16:59.960 And for really critical tasks, maybe things involving 00:16:59.960 --> 00:17:03.259 money or contracts, you can design a human -in 00:17:03.259 --> 00:17:06.119 -the -loop workflow. Ah, so the agent doesn't 00:17:06.119 --> 00:17:08.880 just run wild. Right. The agent proposes an action, 00:17:09.240 --> 00:17:12.079 say, sending a specific contract email, but first, 00:17:12.519 --> 00:17:14.599 it sends an approval request to you, maybe via 00:17:14.599 --> 00:17:17.859 Slack or a simple email. And only after you click 00:17:17.859 --> 00:17:21.650 approve or reply yes, it proceeds. Exactly. It 00:17:21.650 --> 00:17:23.990 waits for your green light before executing the 00:17:23.990 --> 00:17:26.430 sensitive step. It's a smart way to maintain 00:17:26.430 --> 00:17:29.109 control and prevent mistakes on vital tasks. 00:17:29.410 --> 00:17:31.609 That's a really important safety feature, which 00:17:31.609 --> 00:17:33.589 actually brings us to some crucial considerations 00:17:33.589 --> 00:17:35.660 around security. when building these things. 00:17:35.680 --> 00:17:37.359 Yeah, definitely need to talk about that. When 00:17:37.359 --> 00:17:39.619 you grant an AI access to your personal data, 00:17:39.900 --> 00:17:42.200 like email and calendar, privacy has to be top 00:17:42.200 --> 00:17:44.700 of mind. Absolutely. You're allowing a third 00:17:44.700 --> 00:17:47.859 party service like OpenAI to process potentially 00:17:47.859 --> 00:17:50.440 sensitive information. So the principle of least 00:17:50.440 --> 00:17:53.500 privilege is key here, right? Grant only the 00:17:53.500 --> 00:17:56.400 minimum permissions necessary for the task. Don't 00:17:56.400 --> 00:17:58.460 give it access to everything if it only needs 00:17:58.460 --> 00:18:01.279 to read calendar events. Precisely. And treat 00:18:01.279 --> 00:18:05.140 your API keys like gold. Like passwords. Never 00:18:05.140 --> 00:18:07.079 share them publicly. Don't accidentally commit 00:18:07.079 --> 00:18:10.240 them to public code repositories on GitHub or 00:18:10.240 --> 00:18:13.339 places like that. Good point. The material also 00:18:13.339 --> 00:18:15.519 warns about something called prompt injection. 00:18:15.819 --> 00:18:18.299 What's that about? That's where malicious actors 00:18:18.299 --> 00:18:21.259 might try to embed hidden instructions within 00:18:21.259 --> 00:18:24.279 data, the agent processes, like in the body of 00:18:24.279 --> 00:18:26.480 an email it reads. To try and hijack the agent, 00:18:26.579 --> 00:18:28.779 make it do something unintended. Exactly, trying 00:18:28.779 --> 00:18:31.119 to trick it into revealing information or taking 00:18:31.119 --> 00:18:34.019 harmful actions. So you need to be cautious about 00:18:34.019 --> 00:18:37.400 the data sources your agent interacts with, especially 00:18:37.400 --> 00:18:39.680 if they're from unknown or untrusted origins. 00:18:40.039 --> 00:18:42.700 OK, so be careful with permissions, protect keys, 00:18:42.799 --> 00:18:45.519 and be aware of potential injection. Anything 00:18:45.519 --> 00:18:48.400 else just transparency if your agent is going 00:18:48.400 --> 00:18:50.059 to interact with other people like responding 00:18:50.059 --> 00:18:52.000 to customer emails It's really important to make 00:18:52.000 --> 00:18:54.259 it clear. They're communicating with an AI assistant 00:18:54.259 --> 00:18:57.500 builds trust manages expectations exactly. It's 00:18:57.500 --> 00:19:00.259 just good practice Wow OK, we've covered a lot 00:19:00.259 --> 00:19:03.019 of ground today. We went from the foundational 00:19:03.019 --> 00:19:07.039 concepts of AI agents all the way to the practical 00:19:07.039 --> 00:19:09.319 step -by -step process of actually building your 00:19:09.319 --> 00:19:11.579 first one using N8n. Yeah, quite the journey. 00:19:11.839 --> 00:19:14.579 We really dug into the key distinctions, highlighting 00:19:14.579 --> 00:19:17.619 how AI agents are different from chatbots and 00:19:17.619 --> 00:19:20.720 traditional automation, basically bringing intelligence 00:19:20.720 --> 00:19:23.680 and action together. And we explored their core 00:19:23.680 --> 00:19:26.900 principles. that ability to reason, to adapt, 00:19:27.200 --> 00:19:29.579 to remember, and maybe most importantly, to use 00:19:29.579 --> 00:19:31.819 tools to take action on your behalf. And the 00:19:31.819 --> 00:19:33.680 personal assistant example really brought it 00:19:33.680 --> 00:19:36.799 home, I think, showing how achievable this is. 00:19:36.940 --> 00:19:38.799 Yeah, even if you're totally new to this, it 00:19:38.799 --> 00:19:41.420 feels less like magic and more like a practical 00:19:41.420 --> 00:19:43.339 tool you can actually build. And if we connect 00:19:43.339 --> 00:19:45.900 this to the bigger picture, building AI agents 00:19:45.900 --> 00:19:48.900 isn't just about, you know, simple automation 00:19:48.900 --> 00:19:50.900 or digital convenience, is it? Seems like more 00:19:50.900 --> 00:19:53.240 than that. It really is. It's about creating 00:19:53.240 --> 00:19:56.220 genuinely intelligent systems that understand 00:19:56.220 --> 00:19:59.359 and support your work. They can actually transform 00:19:59.359 --> 00:20:02.339 how you approach daily tasks, saving you time, 00:20:02.440 --> 00:20:05.160 reducing errors. So the takeaway for listener 00:20:05.160 --> 00:20:07.880 is? It's that you now have the foundational skills 00:20:07.880 --> 00:20:11.140 to actually be a part of this AI agent revolution, 00:20:11.539 --> 00:20:13.980 not just watching it happen. You're empowered 00:20:13.980 --> 00:20:16.319 to build. That's a powerful thought. And the 00:20:16.319 --> 00:20:18.339 core principles we discussed today. The goal 00:20:18.339 --> 00:20:20.460 setting, the tools, the memory, the prompting, 00:20:20.460 --> 00:20:23.119 they apply to countless other applications. Your 00:20:23.119 --> 00:20:25.880 first AI agent might be simple, managing emails 00:20:25.880 --> 00:20:28.680 and calendars, but it's the foundation for building 00:20:28.680 --> 00:20:31.660 much more sophisticated systems later on. Systems 00:20:31.660 --> 00:20:33.740 that will unlock possibilities you probably haven't 00:20:33.740 --> 00:20:36.039 even imagined yet. A future where our digital 00:20:36.039 --> 00:20:38.940 tools are more like proactive partners. Exactly. 00:20:39.119 --> 00:20:41.839 Truly proactive partners. So the real question 00:20:41.839 --> 00:20:44.920 to leave everyone with is, what AI agent will 00:20:44.920 --> 00:20:48.130 you build next? The only limit, really, is your 00:20:48.130 --> 00:20:48.690 imagination.