WEBVTT 00:00:00.000 --> 00:00:02.419 Welcome to the Deep Dive. We're here to pull 00:00:02.419 --> 00:00:03.980 out the key bits of knowledge so you can get 00:00:03.980 --> 00:00:06.580 up to speed fast. Today we're looking at something, 00:00:06.740 --> 00:00:09.220 well, really transformative, I think. You know 00:00:09.220 --> 00:00:11.560 that AI agent you used, the text -based one? 00:00:11.679 --> 00:00:13.800 Super smart, super helpful. Well, what if he 00:00:13.800 --> 00:00:15.820 can actually talk back to you? Yeah, not like, 00:00:15.839 --> 00:00:18.460 you know, an old GPS voice. Exactly. Not clunky 00:00:18.460 --> 00:00:21.140 or robotic, but like, with a natural, engaging 00:00:21.140 --> 00:00:24.660 voice. A real personality. Something that sounds, 00:00:24.960 --> 00:00:31.260 well, human. And the thing is, It's something 00:00:31.260 --> 00:00:34.740 you can genuinely build today, like right now, 00:00:34.840 --> 00:00:37.039 and without writing code. That's the amazing 00:00:37.039 --> 00:00:40.000 part, the accessibility. Right. So our mission 00:00:40.000 --> 00:01:00.750 for this deep dive is... So, OK, let's kick things 00:01:00.750 --> 00:01:03.909 off. Method one. We're calling it the AI voicemail 00:01:03.909 --> 00:01:08.010 system expert. This sounds well, it sounds clever, 00:01:08.090 --> 00:01:10.489 like a smart way to handle tasks that aren't 00:01:10.489 --> 00:01:13.730 instant. Right. This asynchronous idea. Exactly. 00:01:13.730 --> 00:01:18.310 Yeah. Think of it like a super intelligent. Voice 00:01:18.310 --> 00:01:21.010 mailbox. You, the user, leave a voice message, 00:01:21.109 --> 00:01:23.650 maybe a complex question or instruction. And 00:01:23.650 --> 00:01:26.950 the AI doesn't rush. It takes its time. It listens. 00:01:27.069 --> 00:01:29.129 It thinks, processes everything, and then responds 00:01:29.129 --> 00:01:31.629 with its own audio message back to you. Oh, OK. 00:01:31.769 --> 00:01:34.109 So this asynchronous thing, it's perfect for 00:01:34.109 --> 00:01:36.890 jobs that need, you know, deeper thought, summarizing 00:01:36.890 --> 00:01:39.329 a really long document maybe or doing some complex 00:01:39.329 --> 00:01:41.530 research. Right. Not that instant back and forth, 00:01:41.689 --> 00:01:43.849 but more considered. Precisely. It's about getting 00:01:43.849 --> 00:01:46.390 a thoughtful, rich response when you don't need 00:01:46.390 --> 00:01:48.319 it that very second. That makes a lot of sense. 00:01:48.319 --> 00:01:50.099 Like having a research assistant, you can brief 00:01:50.099 --> 00:01:52.060 and they come back later with the findings. So, 00:01:52.180 --> 00:01:54.359 OK, what's the actual journey from me hitting 00:01:54.359 --> 00:01:57.000 record to the AI talking back? Walk us through 00:01:57.000 --> 00:01:59.560 this voice message pipeline. Yeah, it's quite 00:01:59.560 --> 00:02:01.620 an elegant flow, actually, like a little assembly 00:02:01.620 --> 00:02:04.319 line. It starts when you send a voice message. 00:02:04.640 --> 00:02:07.060 Let's say you send it to a telegram bot. Then 00:02:07.060 --> 00:02:11.219 N8N, acting as the orchestrator, grabs that audio 00:02:11.219 --> 00:02:15.050 file. downloads it, that raw audio, it goes straight 00:02:15.050 --> 00:02:17.349 to 11 Labs, their models transcribe it, turn 00:02:17.349 --> 00:02:20.289 it into text, and they do it brilliantly. That 00:02:20.289 --> 00:02:23.349 text then gets fed to the AI brain. Could be 00:02:23.349 --> 00:02:25.509 ChatGPT, could be Claude, whatever you're using. 00:02:25.590 --> 00:02:28.389 The AI does its thing, generates the response 00:02:28.389 --> 00:02:31.310 in text, and that text goes back to 11 Labs, 00:02:31.389 --> 00:02:33.909 this time for text -to -speech, turning it into 00:02:33.909 --> 00:02:36.389 that natural -sounding audio. And then finally, 00:02:36.509 --> 00:02:39.189 N8N takes that new audio file and sends it back 00:02:39.189 --> 00:02:41.270 to you in Telegram. It's a full circle. Wow. 00:02:41.370 --> 00:02:43.819 Okay. Okay, that sounds incredibly capable. And 00:02:43.819 --> 00:02:45.560 the bit that really grabs me is the no -code 00:02:45.560 --> 00:02:47.439 part you mentioned. Like, for someone wanting 00:02:47.439 --> 00:02:49.580 to build this, what are the actual pieces, the 00:02:49.580 --> 00:02:52.400 Lego bricks in N8N? How do they, you know, snap 00:02:52.400 --> 00:02:54.780 together? Right, yeah, Lego bricks is a great 00:02:54.780 --> 00:02:56.360 way to put it. They really do feel like they're 00:02:56.360 --> 00:02:59.550 made for each other. You're listening to Ghost. 00:02:59.729 --> 00:03:02.689 In any in -the -ghosted area, like the Telegram 00:03:02.689 --> 00:03:04.990 turn -out, you set it up to catch incoming audio 00:03:04.990 --> 00:03:07.009 files, specifically the voice message format. 00:03:07.270 --> 00:03:10.689 And crucially, it grabs the chatted... The return 00:03:10.689 --> 00:03:13.189 address. Exactly. It's the unique return address, 00:03:13.250 --> 00:03:15.370 so the AI knows where to send its response back. 00:03:15.849 --> 00:03:19.610 Next trick. The universal translator. That's 00:03:19.610 --> 00:03:22.110 an 11 labs node doing the speech -to -text... 00:03:22.169 --> 00:03:25.030 It takes the audio file and just grabs and transcribes 00:03:25.030 --> 00:03:28.770 it. The beauty is NAN handles passing that audio 00:03:28.770 --> 00:03:31.289 data automatically. It's pretty seamless. Okay, 00:03:31.349 --> 00:03:56.710 no complex mapping needed there. Nope. And it 00:03:56.710 --> 00:03:58.930 totally changes the whole feel of the interaction. 00:03:59.310 --> 00:04:02.110 So the system prompt is really the soul of the 00:04:02.110 --> 00:04:04.150 agent. OK, then what? How does it get its voice 00:04:04.150 --> 00:04:07.430 back? Right. So after the AI thinks and writes 00:04:07.430 --> 00:04:10.430 its text response, you use another 11 Labs node. 00:04:10.629 --> 00:04:13.430 This one's for text to speech. You connect the 00:04:13.430 --> 00:04:16.350 AI's text output to this node. And this is where 00:04:16.350 --> 00:04:18.810 you like. Cast your agent. You choose a voice 00:04:18.810 --> 00:04:21.029 from Eleven Labs library. They have tons. Or 00:04:21.029 --> 00:04:23.189 you can use a specific voice ID if you've cloned 00:04:23.189 --> 00:04:25.389 one or have a favorite. Got it. Choosing the 00:04:25.389 --> 00:04:28.449 voice actor. Pretty much. And the final step, 00:04:28.670 --> 00:04:31.410 delivering the message. That's another telegram 00:04:31.410 --> 00:04:33.790 node, a sender this time. You make sure it sends 00:04:33.790 --> 00:04:36.050 the audio file generated by Eleven Labs back 00:04:36.050 --> 00:04:38.189 to the original chat aid so the message goes 00:04:38.189 --> 00:04:39.670 right back to the person who sent the request. 00:04:39.990 --> 00:04:43.579 And then the test flight. The moment of truth, 00:04:43.680 --> 00:04:45.620 as you called it. I bet it's satisfying seeing 00:04:45.620 --> 00:04:47.620 it all light up. Oh, it really is. Watching each 00:04:47.620 --> 00:04:49.879 node activate on the inning and canvas as your 00:04:49.879 --> 00:04:52.420 message flows through. Very cool. But, you know, 00:04:52.480 --> 00:04:54.519 things don't always work first time. What about 00:04:54.519 --> 00:04:59.100 when, Houston, we have a problem? Any quick debugging 00:04:59.100 --> 00:05:01.480 tips? Yeah, absolutely. The bugging and editing 00:05:01.480 --> 00:05:03.620 is usually pretty straightforward. The visual 00:05:03.620 --> 00:05:06.379 flow helps a lot. If a node doesn't light up 00:05:06.379 --> 00:05:08.620 green, you know exactly where it's stopped. First 00:05:08.620 --> 00:05:11.079 thing, check the trigger. Does the Telegram node 00:05:11.079 --> 00:05:13.319 even receive the message? If not, maybe check 00:05:13.319 --> 00:05:16.180 your WhatsApp and Telegram as well. Then follow 00:05:16.180 --> 00:05:18.699 the noodles, those lines connecting the nodes. 00:05:18.800 --> 00:05:21.339 Is everything connected? Did you link an output 00:05:21.339 --> 00:05:24.439 to the right input? Easy mistake to make. But 00:05:24.439 --> 00:05:26.620 honestly, the most common failure point nine 00:05:26.620 --> 00:05:30.370 times out of ten is probably credentials. Ah, 00:05:30.769 --> 00:05:33.550 yes, the classic. Triple check them. Your 11 00:05:33.550 --> 00:05:36.410 labs key, your AI service key, one typo and the 00:05:36.410 --> 00:05:39.449 whole thing falls over. And maybe most importantly, 00:05:39.649 --> 00:05:47.790 read the error message. That's really solid advice. 00:05:47.949 --> 00:05:52.029 OK, so this AI voicemail bot is fantastic for 00:05:52.029 --> 00:05:55.069 those deeper asynchronous things. But what if 00:05:55.069 --> 00:05:58.490 you want that instant back and forth, the conversational 00:05:58.490 --> 00:06:00.689 feel? This is where it gets really interesting, 00:06:00.810 --> 00:06:04.410 I think. We're talking about building a fluid, 00:06:04.730 --> 00:06:08.490 real -time conversational AI, like talking to 00:06:08.490 --> 00:06:10.629 a human assistant on the phone. This sounds like 00:06:10.629 --> 00:06:12.490 the main event, the showstopper. Yeah, this is 00:06:12.490 --> 00:06:15.050 where it feels truly interactive. So how do we 00:06:15.050 --> 00:06:18.430 achieve that, the proper AI assistant experience? 00:06:18.850 --> 00:06:21.290 Okay, so for this real -time system. you need 00:06:21.290 --> 00:06:23.329 a totally different setup, a different kind of 00:06:23.329 --> 00:06:25.269 partnership. I like to think of it using the 00:06:25.269 --> 00:06:28.480 NASA mission control analogy. Ooh, okay. Tell 00:06:28.480 --> 00:06:30.959 me more. Right. In this model, 11 Labs is mission 00:06:30.959 --> 00:06:32.839 control. It's the sophisticated front end. It 00:06:32.839 --> 00:06:34.879 handles the direct chat with the user, the voice, 00:06:34.959 --> 00:06:37.100 the personality, managing the flow of the conversation. 00:06:37.360 --> 00:06:39.480 It's talking to the astronaut, basically. Got 00:06:39.480 --> 00:06:42.279 it. The voice on the comms. Exactly. Now, your 00:06:42.279 --> 00:06:44.420 NA workflow, that's the specialist team back 00:06:44.420 --> 00:06:47.420 in Houston. The engineers, the scientists, they 00:06:47.420 --> 00:06:50.660 don't talk directly to the user. Instead, they 00:06:50.660 --> 00:06:53.660 are a powerful tool that Mission Control 11 Labs 00:06:53.660 --> 00:06:56.740 calls upon when the user asks something complex. 00:06:57.639 --> 00:07:00.379 thing needing research or a specific action performed. 00:07:00.560 --> 00:07:03.740 Ah, I see. So Eleven Labs handles the chat and 00:07:03.740 --> 00:07:05.579 ANN does the heavy lifting in the background 00:07:05.579 --> 00:07:08.339 when needed. Precisely. That's the key takeaway. 00:07:08.639 --> 00:07:11.500 Eleven Labs manages the conversation and ANN 00:07:11.500 --> 00:07:14.839 handles the actions or the deep information retrieval. 00:07:14.939 --> 00:07:18.000 That analogy makes it crystal clear. So if 11 00:07:18.000 --> 00:07:20.819 Labs is mission control, how do we actually build 00:07:20.819 --> 00:07:23.060 it? How do we set up our agent in 11 Labs for 00:07:23.060 --> 00:07:25.160 these live calls? You actually do it right inside 00:07:25.160 --> 00:07:27.019 your 11 Labs account. They have a section called 00:07:27.019 --> 00:07:29.620 Conversational AI. You go there, create a new 00:07:29.620 --> 00:07:31.439 agent, and this is where you play casting director 00:07:31.439 --> 00:07:33.839 again. Give it a name, something fitting. Choose 00:07:33.839 --> 00:07:36.079 a voice from their library that matches the personality 00:07:36.079 --> 00:07:39.529 you want. calm and professional, energetic and 00:07:39.529 --> 00:07:42.110 friendly. And then you craft that initial greeting, 00:07:42.310 --> 00:07:44.370 something simple to start, like, hello, how can 00:07:44.370 --> 00:07:46.029 I help you today? Okay, straightforward enough. 00:07:46.470 --> 00:07:50.129 But how do we connect mission control to the 00:07:50.129 --> 00:07:52.670 specialist team, to NAN? That feels like the 00:07:52.670 --> 00:07:55.250 critical link. It absolutely is. So back in your 00:07:55.250 --> 00:07:58.370 11 Labs agent settings, you scroll down to tools. 00:07:58.959 --> 00:08:01.420 Here, you add a custom webhook tool. This is 00:08:01.420 --> 00:08:03.779 literally the direct phone line to your NAN workflow. 00:08:04.139 --> 00:08:06.439 Okay. You need to give the AI a clear briefing 00:08:06.439 --> 00:08:08.300 about this tool. So you give it a name, like 00:08:08.300 --> 00:08:11.120 NAN Web Researcher, and a description, something 00:08:11.120 --> 00:08:13.839 really clear like, call this tool to search the 00:08:13.839 --> 00:08:16.379 web and find information about any topic. It 00:08:16.379 --> 00:08:19.279 is an expert research assistant. So the AI knows 00:08:19.279 --> 00:08:21.519 what it's for. Exactly. You set the method to 00:08:21.519 --> 00:08:24.420 post and then you paste in the URL from an NAN 00:08:24.420 --> 00:08:27.339 webhook node that connects them. Got it. So the 00:08:27.339 --> 00:08:30.100 agent knows what the specialist team does, but 00:08:30.100 --> 00:08:31.800 it also needs to know when to call them, right? 00:08:31.879 --> 00:08:34.019 Yeah. That brings us to the prime directive, 00:08:34.220 --> 00:08:37.399 the system prompt for the 11Labs agent itself. 00:08:37.639 --> 00:08:40.299 How do we teach it to... Essentially, put the 00:08:40.299 --> 00:08:43.399 user on hold and call the NAN tool at the right 00:08:43.399 --> 00:08:45.980 moment. Yeah, this prompt is, it's like the agent's 00:08:45.980 --> 00:08:48.200 constitution. It dictates everything. 11 Labs 00:08:48.200 --> 00:08:50.799 has a nice generate with AI feature to get you 00:08:50.799 --> 00:08:53.779 started with a base prompt. But you must manually 00:08:53.779 --> 00:08:56.860 add a really crucial instruction. You have to 00:08:56.860 --> 00:09:00.100 explicitly tell the agent. When the user asks 00:09:00.100 --> 00:09:02.139 a question that requires up -to -date information 00:09:02.139 --> 00:09:05.500 or web research, use the N8 Web Researcher tool. 00:09:05.840 --> 00:09:08.360 Tell the user you are searching and then wait 00:09:08.360 --> 00:09:10.860 for the tool's response before continuing. Ah, 00:09:11.059 --> 00:09:14.240 okay, so it's a non -negotiable rule. Absolutely. 00:09:14.240 --> 00:09:16.860 Without that specific instruction, the agent 00:09:16.860 --> 00:09:18.919 might just try to answer from its general knowledge, 00:09:19.039 --> 00:09:20.879 which could be outdated, or it might just get 00:09:20.879 --> 00:09:23.419 confused about how to use the tool. This tells 00:09:23.419 --> 00:09:25.779 it exactly how to handle research requests. That's 00:09:25.779 --> 00:09:28.340 a really important detail. Okay, and what about 00:09:28.340 --> 00:09:31.879 building the specialist team itself, the NANN 00:09:31.879 --> 00:09:34.159 workflow that receives the call from 11 labs? 00:09:34.419 --> 00:09:36.940 What are the key parts there? So the NANN research 00:09:36.940 --> 00:09:39.340 backend is pretty focused. It usually has three 00:09:39.340 --> 00:09:42.710 main parts. First, the researcher. The webhook 00:09:42.710 --> 00:09:45.809 node receives the query from 11 labs. You pass 00:09:45.809 --> 00:09:48.889 that query straight to, say, a perplexity node. 00:09:49.049 --> 00:09:51.029 You'd probably use one of their sonar models. 00:09:51.169 --> 00:09:52.990 They're designed for real -time web search, pulling 00:09:52.990 --> 00:09:55.450 back sourced info. Okay, so it gets the raw data. 00:09:55.610 --> 00:09:59.649 Right. But that raw data can be a lot. Maybe 00:09:59.649 --> 00:10:02.330 too much for a quick conversational answer. So 00:10:02.330 --> 00:10:05.570 next comes the editor. You pass Perplexity's 00:10:05.570 --> 00:10:08.289 output to another AI agent node, its only job, 00:10:08.450 --> 00:10:11.570 to be a ruthless editor. You give it a sign prompt 00:10:11.570 --> 00:10:14.289 like, summarize the following information concisely, 00:10:14.289 --> 00:10:17.289 no more than three sentences. Nice. Keep it brief. 00:10:17.490 --> 00:10:21.250 Exactly. And finally, the report back. This is 00:10:21.250 --> 00:10:24.570 just a respond to webhook node in ANN. It takes 00:10:24.570 --> 00:10:26.629 that short, concise summary from the editor AI 00:10:26.629 --> 00:10:28.690 and sends it straight back to the 11 labs agent 00:10:28.690 --> 00:10:31.529 waiting on the line. Loop closed. This sounds 00:10:31.529 --> 00:10:34.370 seriously powerful when put together. Let's make 00:10:34.370 --> 00:10:36.769 it real. Can you walk us through a quick test 00:10:36.769 --> 00:10:38.809 flight, a hypothetical conversation so we can 00:10:38.809 --> 00:10:41.370 hear how it flows? Yeah, sure. Imagine you call 00:10:41.370 --> 00:10:43.210 your agent. You could be using your browser, 00:10:43.389 --> 00:10:45.809 your phone. You start. Hello, I'd like to do 00:10:45.809 --> 00:10:48.009 some research on the company NVIDIA. The agent, 00:10:48.129 --> 00:10:50.409 11 Labs, responds smoothly using the voice you 00:10:50.409 --> 00:10:52.629 chose. I can certainly help with that. Is there 00:10:52.629 --> 00:10:54.409 anything specific you'd like to know about NVIDIA? 00:10:54.549 --> 00:10:57.720 Okay, nice and natural. Then you say. Yeah, let's 00:10:57.720 --> 00:11:01.759 look at their Q4 2025 forecast. Now, the agent 00:11:01.759 --> 00:11:04.120 recognizes this needs the specialist tool because 00:11:04.120 --> 00:11:06.320 of that prompt rule. So it says, understood. 00:11:07.059 --> 00:11:10.039 I'll search for NVIDIA's Q4 2025 forecast now. 00:11:10.259 --> 00:11:12.940 Please give me just a moment. Ah, putting you 00:11:12.940 --> 00:11:15.759 on hold politely. Exactly. And behind the scenes, 00:11:15.879 --> 00:11:20.000 bam, the N8N workflow fires up. Perplexity searches, 00:11:20.299 --> 00:11:22.799 the AI summarizes. The specialist team is working. 00:11:22.980 --> 00:11:25.519 Right. Then maybe five, 10 seconds later, the 00:11:25.519 --> 00:11:28.759 11 labs agent comes back. Okay. I have found 00:11:28.759 --> 00:11:31.740 some information on NVIDIA's quarter four 2025 00:11:31.740 --> 00:11:36.639 forecast. NVIDIA reported revenue of whatever 00:11:36.639 --> 00:11:38.899 the summary is. Wow. It's a seamless conversation, 00:11:39.059 --> 00:11:41.139 even with that complex lookup happening in the 00:11:41.139 --> 00:11:43.379 background. That's genuinely impressive. It completely 00:11:43.379 --> 00:11:46.559 changes the game for AI interaction. So thinking 00:11:46.559 --> 00:11:49.250 bigger. What does this mean for expanding? You 00:11:49.250 --> 00:11:51.230 mentioned a pro -level upgrade, the multi -specialist 00:11:51.230 --> 00:11:52.789 idea. Yeah, this is where it gets really powerful. 00:11:52.990 --> 00:11:54.809 You're not limited to just one NANA specialist 00:11:54.809 --> 00:11:56.990 tool. You can create multiple NAN workflows, 00:11:57.389 --> 00:11:59.370 each starting with a webhook, each designed for 00:11:59.370 --> 00:12:01.370 a different task. Like what? Well, you could 00:12:01.370 --> 00:12:04.570 have your web researcher, but also maybe an internal 00:12:04.570 --> 00:12:06.970 database checker that looks at customer info. 00:12:07.740 --> 00:12:10.139 or a calendar scheduler, or even one that sends 00:12:10.139 --> 00:12:13.320 emails. Okay. You add each of these as separate 00:12:13.320 --> 00:12:15.759 webhook tools in the 11 Labs agent settings, 00:12:15.899 --> 00:12:18.679 each with a clear name and description. Then 00:12:18.679 --> 00:12:21.259 your 11 Labs agent, Mission Control, becomes 00:12:21.259 --> 00:12:24.000 much smarter. Based on your conversation, it 00:12:24.000 --> 00:12:26.240 will figure out which specialist tool is the 00:12:26.240 --> 00:12:28.679 right one to call for that specific task. So 00:12:28.679 --> 00:12:30.720 it routes the request intelligently. Exactly. 00:12:30.919 --> 00:12:33.159 It transforms your agent from just a researcher 00:12:33.159 --> 00:12:36.090 into a truly versatile assistant. That's incredible, 00:12:36.250 --> 00:12:39.309 giving the AI a whole team. Okay, so for people 00:12:39.309 --> 00:12:41.289 wanting to actually deploy this, move beyond 00:12:41.289 --> 00:13:03.639 just... Also, consider rate limiting on your 00:13:03.639 --> 00:13:05.539 webhook to prevent accidental or intentional 00:13:05.539 --> 00:13:09.139 overload. And keep a close eye on your API usage 00:13:09.139 --> 00:13:13.460 11 labs, your AI model provider perplexity. Those 00:13:13.460 --> 00:13:15.340 costs can add up if you're not monitoring them. 00:13:15.519 --> 00:13:18.419 And technically, remember to switch your NAN 00:13:18.419 --> 00:13:20.799 workflow from the test URL to the production 00:13:20.799 --> 00:13:23.179 URL. Just remove the test part and make sure 00:13:23.179 --> 00:13:25.480 the workflow toggle is set to active. Little 00:13:25.480 --> 00:13:29.019 details, but crucial. Okay, beyond the text setup. 00:13:29.320 --> 00:13:32.159 There's the experience itself, crafting the perfect 00:13:32.159 --> 00:13:34.519 conversational experience. This sounds like an 00:13:34.519 --> 00:13:37.019 art. Any tips on voice selection, keeping the 00:13:37.019 --> 00:13:40.100 flow smooth? It absolutely is an art. For voice 00:13:40.100 --> 00:13:42.320 selection, really think about the agent's role. 00:13:42.500 --> 00:13:45.000 Is it a formal research assistant? Maybe a clear, 00:13:45.019 --> 00:13:47.419 neutral voice? A creative brainstorming partner? 00:13:47.620 --> 00:13:49.820 Perhaps something more expressive? And definitely 00:13:49.820 --> 00:13:52.620 test voices with real content, not just, hello, 00:13:52.820 --> 00:13:55.259 a voice that sounds great for one sentence might 00:13:55.259 --> 00:13:58.230 get grading over a longer explanation. For conversation 00:13:58.230 --> 00:14:00.789 flow, keep the agent's responses conversational 00:14:00.789 --> 00:14:03.750 but also concise. Aim for maybe two or three 00:14:03.750 --> 00:14:06.269 sentences for simple answers. Don't let it ramble. 00:14:06.529 --> 00:14:10.080 And crucially, plan for errors. What happens 00:14:10.080 --> 00:14:14.000 if the NAN workflow times out or fails? The agent 00:14:14.000 --> 00:14:16.419 shouldn't just hang up. It needs a graceful exit, 00:14:16.539 --> 00:14:18.500 like, I'm sorry, I seem to be having trouble 00:14:18.500 --> 00:14:20.379 connecting to my research tool right now. Could 00:14:20.379 --> 00:14:22.340 you try again in a moment? Handling failures 00:14:22.340 --> 00:14:24.679 gracefully, yeah, that's key for a professional 00:14:24.679 --> 00:14:26.720 field. And this brings us back to the prompt, 00:14:26.779 --> 00:14:29.000 really. How do you ensure it behaves consistently? 00:14:29.600 --> 00:14:32.120 That's advanced prompt engineering for voice. 00:14:33.059 --> 00:14:35.639 The system prompt for your 11 Labs agent is its 00:14:35.639 --> 00:14:38.679 constitution. It needs explicit rules. Define 00:14:38.679 --> 00:14:41.120 the persona clearly. Set rules for conversation 00:14:41.120 --> 00:14:43.679 management, like always ask clarifying questions 00:14:43.679 --> 00:14:46.340 if a request is vague, keep information concise, 00:14:46.500 --> 00:14:48.720 use natural language. And most importantly for 00:14:48.720 --> 00:14:52.159 this setup, explicit tool usage guidelines. Define 00:14:52.159 --> 00:14:55.059 exactly when to use each N8N tool, how to introduce 00:14:55.059 --> 00:14:57.320 it, and what to tell the user if a tool fails 00:14:57.320 --> 00:14:59.000 or takes too long. It sounds like the prompt 00:14:59.000 --> 00:15:00.860 is doing a lot of heavy lifting in managing the 00:15:00.860 --> 00:15:03.399 whole interaction. It really is. It's the brain 00:15:03.399 --> 00:15:06.340 governing the conversation flow and tool orchestration. 00:15:06.440 --> 00:15:09.220 So this blueprint, it really unlocks some serious 00:15:09.220 --> 00:15:12.299 superpowers. These advanced voice agent patterns. 00:15:12.539 --> 00:15:14.860 We're talking multi -tool agents doing research, 00:15:15.100 --> 00:15:18.620 scheduling, emailing. Checking databases, analyzing 00:15:18.620 --> 00:15:21.700 sales data. Yeah. And context -aware conversations. 00:15:22.059 --> 00:15:24.179 Yeah. Connecting to memory systems like ZEP or 00:15:24.179 --> 00:15:27.179 Supabase so it remembers past chats. Exactly. 00:15:27.179 --> 00:15:29.519 So you can pick up where you left off or it can 00:15:29.519 --> 00:15:31.919 build knowledge over time. Makes it feel much 00:15:31.919 --> 00:15:34.820 more intelligent. And then my favorite, Star 00:15:34.820 --> 00:15:38.200 Trek mode. Voice -activated workflows. Using 00:15:38.200 --> 00:15:41.360 the voice agent as, like, a master controller 00:15:41.360 --> 00:15:44.379 for other inane automations. Precisely. Imagine 00:15:44.379 --> 00:15:46.240 just saying, computer, run the morning sales 00:15:46.240 --> 00:15:48.659 report automation. Oh, man. For anyone who's 00:15:48.659 --> 00:15:50.639 manually pulled reports every morning, that sounds 00:15:50.639 --> 00:15:52.919 like pure magic. Have you actually seen teams 00:15:52.919 --> 00:15:54.379 implement that kind of thing? Does this save 00:15:54.379 --> 00:15:56.399 a lot of time? Oh, absolutely. E -efficiency 00:15:56.399 --> 00:15:58.620 gains can be huge, especially for repetitive 00:15:58.620 --> 00:16:00.960 tasks that can be triggered by a simple voice 00:16:00.960 --> 00:16:03.379 command. It frees people up for more complex 00:16:03.379 --> 00:16:39.039 work. And the agent talks too much, doesn't know 00:16:39.039 --> 00:16:41.259 when to stop. That's almost always a prompt engineering 00:16:41.259 --> 00:16:43.960 issue. You need to go back to that system prompt 00:16:43.960 --> 00:17:12.339 and add stricter rules. Also, sales and lead 00:17:12.339 --> 00:17:15.509 qualification. An AI voice agent can handle initial 00:17:15.509 --> 00:17:18.710 outreach, answer basic questions, qualify leads, 00:17:19.009 --> 00:17:21.730 and then pass the really warm ones over to a 00:17:21.730 --> 00:17:25.309 human salesperson. Very efficient. And internally, 00:17:25.549 --> 00:17:28.069 think voice -controlled tools for teams, allowing 00:17:28.069 --> 00:17:31.369 people to query databases, run reports, or trigger 00:17:31.369 --> 00:17:33.509 workflows completely hands -free while they're 00:17:33.509 --> 00:17:35.750 doing other things. The possibilities really 00:17:35.750 --> 00:17:38.390 do seem vast. Yeah. Okay, let's try and bring 00:17:38.390 --> 00:17:40.430 this all together. The bottom line seems pretty 00:17:40.430 --> 00:17:42.980 clear. The future is conversational. And think 00:17:42.980 --> 00:17:45.619 about it. A few years back, the idea that a solo 00:17:45.619 --> 00:17:48.119 creator, maybe even just you listening now, could 00:17:48.119 --> 00:17:50.740 build an AI agent that holds a natural real -time 00:17:50.740 --> 00:17:52.900 voice chat, hooks into the Internet, performs 00:17:52.900 --> 00:17:55.680 tasks, that was pure sci -fi. Absolutely. Star 00:17:55.680 --> 00:17:58.109 Trek stuff. Right. And today. It's just another 00:17:58.109 --> 00:18:00.289 project you can build in AAN. You literally have 00:18:00.289 --> 00:18:02.890 the blueprint now to create AI systems that don't 00:18:02.890 --> 00:18:05.609 just crunch data, but actually engage, interact, 00:18:05.809 --> 00:18:08.529 connect on a much more human level. It democratizes 00:18:08.529 --> 00:18:10.789 some seriously advanced capabilities. Totally. 00:18:10.869 --> 00:18:13.069 So here's your mission briefing, your takeaway 00:18:13.069 --> 00:18:15.869 challenge, how you use this blueprint. Are you 00:18:15.869 --> 00:18:18.890 going to cast your character? Maybe a witty, 00:18:18.930 --> 00:18:21.650 sarcastic J -A -R -V -I -S like Iron Man's AI? 00:18:22.670 --> 00:18:26.190 Or perhaps a calm, professional, endlessly patient 00:18:26.190 --> 00:18:29.210 Star Trek computer voice. The personality is 00:18:29.210 --> 00:18:33.809 half the fun. It really is. And think, what specific 00:18:33.809 --> 00:18:38.029 nagging, recurring... problem in your life, your 00:18:38.029 --> 00:18:40.470 work, your business could be totally transformed 00:18:40.470 --> 00:18:43.150 by adding a voice. By batting one of these agents, 00:18:43.230 --> 00:18:44.910 imagine connecting it to your smart home, your 00:18:44.910 --> 00:18:47.369 office, literally saying, hey Jarvis, or whatever 00:18:47.369 --> 00:18:49.309 you call it, turn on the lights in the studio 00:18:49.309 --> 00:18:51.150 and start the coffee machine. The tech is there. 00:18:51.269 --> 00:18:53.309 It's ready. It really is. It's just waiting for 00:18:53.309 --> 00:18:55.650 your creativity. Time to get building. We really 00:18:55.650 --> 00:18:58.069 hope this deep dive has sparked some ideas, maybe 00:18:58.069 --> 00:19:00.470 given you some surprising insights. Keep exploring, 00:19:00.630 --> 00:19:02.609 keep building, and we'll see you on the next 00:19:02.609 --> 00:19:03.029 deep dive.