WEBVTT 00:00:00.000 --> 00:00:02.940 Welcome to the Deep Dive. Imagine for a moment 00:00:02.940 --> 00:00:06.419 having an entire AI content team, you know, right 00:00:06.419 --> 00:00:09.099 there working for you. A researcher, a planner, 00:00:09.300 --> 00:00:13.699 writers, even an editor. Wow. All creating your 00:00:13.699 --> 00:00:16.820 weekly newsletter completely on autopilot. That's 00:00:16.820 --> 00:00:18.859 quite the picture. That's what we're diving into 00:00:18.859 --> 00:00:23.000 today. Okay, so today we're unpacking this really 00:00:23.000 --> 00:00:26.699 comprehensive guide to building a multi -agent 00:00:26.699 --> 00:00:29.539 automation workflow. Yeah, a whole system. Right. 00:00:29.899 --> 00:00:32.240 And our mission really is to understand how you 00:00:32.240 --> 00:00:36.100 transform that sometimes painful process of content 00:00:36.100 --> 00:00:38.619 creation. Oh, yeah. Tedious sometimes. Into something 00:00:38.619 --> 00:00:41.359 seamless, automated. Exactly. So we'll walk through 00:00:41.359 --> 00:00:43.719 the blueprint, the tools you need, and the step 00:00:43.719 --> 00:00:46.020 -by -step build, piece by piece. We'll see how 00:00:46.020 --> 00:00:48.020 these different AI agents actually collaborate 00:00:48.020 --> 00:00:50.979 from like the first idea, the research. All the 00:00:50.979 --> 00:00:52.820 way to the end. All the way to final delivery. 00:00:52.840 --> 00:00:55.460 And then we'll look at the time savings, the 00:00:55.460 --> 00:00:58.159 strategic advantages this brings. So, yeah, let's 00:00:58.159 --> 00:01:02.159 unpack this. This isn't just, you know, one simple 00:01:02.159 --> 00:01:04.920 prompt we're talking about. It's really a full 00:01:04.920 --> 00:01:07.799 digital assembly line. An assembly line. Okay. 00:01:07.859 --> 00:01:11.230 Think of it like a factory. Takes raw info, turns 00:01:11.230 --> 00:01:13.390 it into something valuable, like your newsletter. 00:01:13.629 --> 00:01:15.750 Right. It kicks off with what the guide calls 00:01:15.750 --> 00:01:19.069 an initial research phase. It's like scouting, 00:01:19.090 --> 00:01:21.650 finding the latest news for your topic from, 00:01:21.730 --> 00:01:25.409 say, the past week. So gathering the raw materials 00:01:25.409 --> 00:01:27.829 first. Yeah. Making sure you've got the freshest 00:01:27.829 --> 00:01:29.849 stuff before you even start cooking. Foundational. 00:01:30.189 --> 00:01:33.609 Precisely. Then, from that initial scoop, a specialized 00:01:33.609 --> 00:01:36.629 planning agent steps in. Okay. It analyzes that 00:01:36.629 --> 00:01:39.209 research. comes up with a creative title and 00:01:39.209 --> 00:01:42.269 picks out the maybe three most important topics 00:01:42.269 --> 00:01:44.370 for the newsletter. Got it. After that, there's 00:01:44.370 --> 00:01:47.590 a deep research phase. This goes much deeper 00:01:47.590 --> 00:01:49.450 on each of those topics, really comprehensive 00:01:49.450 --> 00:01:52.049 info gathering. Right. So moving from the kind 00:01:52.049 --> 00:01:54.390 of broad overview into the specifics for each 00:01:54.390 --> 00:01:57.129 point. Exactly. Then individual section writing 00:01:57.129 --> 00:01:59.790 agents get to work. They craft each section. 00:02:00.010 --> 00:02:03.010 And the key part here. The critical part. They're 00:02:03.010 --> 00:02:06.159 told to cite everything. Every fact, every piece 00:02:06.159 --> 00:02:09.520 of data needs a proper citation, even a clickable 00:02:09.520 --> 00:02:13.520 URL. A verification. Nice. Yeah. Then an editorial 00:02:13.520 --> 00:02:16.439 review agent takes over. It grabs those finished 00:02:16.439 --> 00:02:19.639 sections, adds an intro, adds a conclusion, wraps 00:02:19.639 --> 00:02:22.860 it all up in clean HTML. the whole pack the whole 00:02:22.860 --> 00:02:25.599 package and finally automated delivery sends 00:02:25.599 --> 00:02:28.939 a draft straight to your gmail wow and the beautiful 00:02:28.939 --> 00:02:33.400 part it's clean linear totally automated no hands 00:02:33.400 --> 00:02:35.759 needed start to finish that's a lot of coordination 00:02:35.759 --> 00:02:38.159 happening behind the scenes and the guide really 00:02:38.159 --> 00:02:42.879 emphasizes um the secret here is planning You've 00:02:42.879 --> 00:02:44.719 got to plan before you build. Absolutely. Like 00:02:44.719 --> 00:02:47.020 any complex project. Yeah. Think like an architect 00:02:47.020 --> 00:02:49.860 sketching out the factory blueprint. They suggest 00:02:49.860 --> 00:02:52.800 using a whiteboard tool like Excalibur or Miro. 00:02:52.979 --> 00:02:55.400 Yeah. Visual tools help a lot. Map it all out. 00:02:55.439 --> 00:02:57.419 So this blueprint includes what? The weekly trigger. 00:02:57.580 --> 00:02:59.560 Kicks it off. The raw materials intake. That's 00:02:59.560 --> 00:03:01.719 the initial research. Strategic planning. The 00:03:01.719 --> 00:03:04.159 AI planning agent. The architect. The deep research 00:03:04.159 --> 00:03:07.860 division. Digging into topics. Specialized writer's 00:03:07.860 --> 00:03:10.819 room. Final assembly and editing. And then shipping. 00:03:11.719 --> 00:03:14.860 A whole flow. And this planning, it's critical, 00:03:15.000 --> 00:03:18.280 right? It stops that, what's it called? Scope 00:03:18.280 --> 00:03:20.719 creep. Yeah, scope creep, where things just get 00:03:20.719 --> 00:03:24.039 bigger and bigger unplanned. Exactly. Yeah. And 00:03:24.039 --> 00:03:26.439 make sure every single component has a clear 00:03:26.439 --> 00:03:29.159 job. A defined purpose. So you build with intention, 00:03:29.439 --> 00:03:33.000 lay that groundwork, avoid problems later. So 00:03:33.000 --> 00:03:35.879 why is this upfront planning so crucial before 00:03:35.879 --> 00:03:38.379 you even touch any code or any tools? Well, that 00:03:38.379 --> 00:03:41.120 clear blueprint. It just prevents that scope 00:03:41.120 --> 00:03:43.460 creep we mentioned. And it really defines what 00:03:43.460 --> 00:03:45.819 each part is supposed to do precisely. Keeps 00:03:45.819 --> 00:03:47.539 it focused. Got it. Keeps it lean and focused. 00:03:47.759 --> 00:03:49.580 Absolutely. That blueprint's like your insurance 00:03:49.580 --> 00:03:52.240 policy. Make sure every bit of code, every agent 00:03:52.240 --> 00:03:54.780 configuration has a purpose. Okay. Let's talk 00:03:54.780 --> 00:03:57.370 tools then. the builder's toolkit right and the 00:03:57.370 --> 00:04:00.289 cool thing is this whole system uses a pretty 00:04:00.289 --> 00:04:03.069 minimal but really effective tech stack lean 00:04:03.069 --> 00:04:05.770 but powerful exactly yeah at the core you've 00:04:05.770 --> 00:04:09.830 got n8 that's n8n it's this open source platform 00:04:09.830 --> 00:04:12.509 lets you connect different apps build automated 00:04:12.509 --> 00:04:15.469 workflows it's like the central nervous system 00:04:15.469 --> 00:04:18.250 for our factory orchestrating everything yep 00:04:18.250 --> 00:04:21.589 then for research both the initial scan and the 00:04:21.589 --> 00:04:25.870 deep dives we use tavly ai Gmail is just the 00:04:25.870 --> 00:04:28.209 simple, reliable shipping department for the 00:04:28.209 --> 00:04:30.689 drafts. Makes sense. OpenRouter acts as this 00:04:30.689 --> 00:04:33.449 versatile AI switchboard, which is great because 00:04:33.449 --> 00:04:35.750 you get access to lots of different AI models. 00:04:35.970 --> 00:04:37.689 From different providers. Yeah, all through one 00:04:37.689 --> 00:04:40.050 account. Really handy. Oh, interesting. And the 00:04:40.050 --> 00:04:43.110 main brains for writing and analyzing, that's 00:04:43.110 --> 00:04:46.550 GPT -5 and GPT -5 Mini. The big guns and the 00:04:46.550 --> 00:04:49.629 slightly smaller, maybe more efficient gun. Kind 00:04:49.629 --> 00:04:52.089 of, yeah. And this lean stack keeps costs down, 00:04:52.149 --> 00:04:54.730 but you still get, like, enterprise -level automation 00:04:54.730 --> 00:04:57.850 power. Okay, so every automated system, it needs 00:04:57.850 --> 00:05:00.629 that pulse, right? A heartbeat to start things 00:05:00.629 --> 00:05:03.089 off? Absolutely, the starting gun. And for this 00:05:03.089 --> 00:05:05.589 system, that's the scheduled trigger node in 00:05:05.589 --> 00:05:08.230 ANAN. Correct. You set it up for, say, once a 00:05:08.230 --> 00:05:10.709 week, maybe Sunday at midnight. Yeah, something 00:05:10.709 --> 00:05:12.610 like that. Give it plenty of time to run. So 00:05:12.610 --> 00:05:15.410 a draft is waiting Monday morning, ready for 00:05:15.410 --> 00:05:18.589 review. Exactly, like setting your watch. Predictable 00:05:18.589 --> 00:05:22.029 rhythm. But there's a key rule here. The safety 00:05:22.029 --> 00:05:24.509 switch rule. Oh yeah, super important. Always 00:05:24.509 --> 00:05:27.029 keep the workflow set to inactive while you're 00:05:27.029 --> 00:05:29.509 building it. Don't turn it on too early. No way. 00:05:29.589 --> 00:05:31.689 Only flip it to active when you are absolutely 00:05:31.689 --> 00:05:34.290 sure it's built, tested, and ready to go live. 00:05:34.629 --> 00:05:36.769 What's the main reason for that safety switch 00:05:36.769 --> 00:05:38.970 rule during development? Why is that so critical? 00:05:39.310 --> 00:05:41.430 Well, it just stops accidental triggers while 00:05:41.430 --> 00:05:43.439 you're still building. You know, ensures you 00:05:43.439 --> 00:05:46.560 launch a tested, production -ready system, not 00:05:46.560 --> 00:05:49.480 something half -baked. Right. Prevents those 00:05:49.480 --> 00:05:51.699 oops moments. We've all been there, haven't we? 00:05:51.819 --> 00:05:53.699 Yeah. Testing something and suddenly it sends 00:05:53.699 --> 00:05:56.160 a half -done email. Exactly. Build in that safety 00:05:56.160 --> 00:05:58.939 net, even during development. Okay. So the heartbeat's 00:05:58.939 --> 00:06:02.120 set. Now we bring in the scout. That's a Tavoli 00:06:02.120 --> 00:06:05.000 search note. Yep. The first research step. And 00:06:05.000 --> 00:06:07.240 its job isn't to dig super deep at first, right? 00:06:07.300 --> 00:06:11.110 It's just finding the top, say. Three high -level 00:06:11.110 --> 00:06:13.810 news stories from the past week. That's it. Quick 00:06:13.810 --> 00:06:16.370 scan. Find the main relevant headlines. So you'd 00:06:16.370 --> 00:06:19.189 tell it your broad topic, like AI adoption for 00:06:19.189 --> 00:06:22.389 small businesses. Set it to look for news, past 00:06:22.389 --> 00:06:26.209 week, maybe max three results. Exactly. And you 00:06:26.209 --> 00:06:29.089 need your Tavoli API key in an ANAN, of course, 00:06:29.230 --> 00:06:31.990 securely stored. Right, the credentials. So this 00:06:31.990 --> 00:06:34.129 initial search gives you that foundation, points 00:06:34.129 --> 00:06:36.509 you in the right direction. Yeah. And for a pro 00:06:36.509 --> 00:06:38.449 upgrade, you could even make that search query 00:06:38.449 --> 00:06:41.100 dynamic. Have another AI figure out the best 00:06:41.100 --> 00:06:44.060 thing to search for each week. Ooh, okay. Hyper 00:06:44.060 --> 00:06:46.639 -focused newsletters. Interesting. Now, the scout 00:06:46.639 --> 00:06:49.339 brought the raw intel. Time for the architect. 00:06:49.720 --> 00:06:53.339 Another AI agent node. This one creates the actual 00:06:53.339 --> 00:06:56.170 plan for the newsletter. Correct. The master 00:06:56.170 --> 00:06:58.730 blueprint. We use something cost -effective but 00:06:58.730 --> 00:07:02.009 smart, like GPT -5 Mini, maybe via open router 00:07:02.009 --> 00:07:04.730 again. Okay. For the user message, we package 00:07:04.730 --> 00:07:06.970 the research cleanly, use a formatting function. 00:07:07.170 --> 00:07:09.089 It's like a little reusable program. To present 00:07:09.089 --> 00:07:11.730 the info consistently to the AI. Exactly. Much 00:07:11.730 --> 00:07:14.189 more scalable than doing it manually. And the 00:07:14.189 --> 00:07:16.329 system prompt, that's the architect's job description. 00:07:16.569 --> 00:07:19.949 Something like, you're an expert newsletter planner. 00:07:20.230 --> 00:07:23.209 Make a creative title. Find the main topics. 00:07:23.839 --> 00:07:26.639 Keep topics short, three to five words. Clear 00:07:26.639 --> 00:07:28.720 instructions. But here's the really cool part. 00:07:28.939 --> 00:07:32.800 We use Anaheim's structured output with a JSON 00:07:32.800 --> 00:07:36.939 schema. Okay, JSON schema. Sounds technical. 00:07:37.060 --> 00:07:39.519 What does that mean in practice? Think of it 00:07:39.519 --> 00:07:42.420 like giving the AI a very specific form to fill 00:07:42.420 --> 00:07:45.540 out. It forces the AI to give you the title and 00:07:45.540 --> 00:07:49.060 topics in separate, clean data fields. So no 00:07:49.060 --> 00:07:50.959 more trying to pick apart messy paragraphs of 00:07:50.959 --> 00:07:53.899 text from the AI. Exactly. No parsing needed. 00:07:54.079 --> 00:07:56.019 It's structured from the start. This is a non 00:07:56.019 --> 00:07:58.519 -negotiable pro move. Guarantees clean, usable 00:07:58.519 --> 00:08:01.279 data downstream. Saves so many headaches. Why 00:08:01.279 --> 00:08:03.139 is using that structured output for the architect 00:08:03.139 --> 00:08:06.079 considered such a non -negotiable pro move? Because 00:08:06.079 --> 00:08:08.870 it guarantees clean, usable data. like getting 00:08:08.870 --> 00:08:11.129 a perfect blueprint, not just a messy sketch, 00:08:11.329 --> 00:08:13.550 no messy text parsing needed. Takes the guesswork 00:08:13.550 --> 00:08:16.730 out, ensures reliability for the next step. Precisely. 00:08:16.730 --> 00:08:18.670 You know exactly what format the data will be 00:08:18.670 --> 00:08:20.550 in. Okay, architect's done the blueprint. Now 00:08:20.550 --> 00:08:22.550 the investigative journalist's time for the deep 00:08:22.550 --> 00:08:26.050 dive. Yeah. So first, the assignment desk. That's 00:08:26.050 --> 00:08:29.110 a split -out node in NAN. Split -out node. It 00:08:29.110 --> 00:08:31.389 takes the single list of three topics from the 00:08:31.389 --> 00:08:34.009 architect and splits them into three separate 00:08:34.009 --> 00:08:36.649 items. Ah, so each topic can be handled independently. 00:08:36.889 --> 00:08:39.710 Critical step, yeah. Then each topic goes to 00:08:39.710 --> 00:08:42.250 another Tavli search node. This one's configured 00:08:42.250 --> 00:08:45.370 as the investigative journalist. Okay. The search 00:08:45.370 --> 00:08:48.429 query is dynamic now, focusing on just one topic. 00:08:48.769 --> 00:08:52.429 Topic type is general for broader research. Not 00:08:52.429 --> 00:08:55.370 just news this time. Right. And crucially, you 00:08:55.370 --> 00:08:58.649 enable... include raw content. To get the full 00:08:58.649 --> 00:09:00.830 text. Yeah, the full article text. That's the 00:09:00.830 --> 00:09:03.289 raw material the writers need. Got it. So deep 00:09:03.289 --> 00:09:05.389 research done. Now bring in the writers, the 00:09:05.389 --> 00:09:08.330 writers room. Yep. Another AI agent node acts 00:09:08.330 --> 00:09:10.590 as one of these specialist writers. And because 00:09:10.590 --> 00:09:13.549 the topics were split. This runs three times. 00:09:13.629 --> 00:09:15.370 Three separate times, once for each topic. We 00:09:15.370 --> 00:09:17.750 stick with GPT -5 Mini here. For consistency, 00:09:18.090 --> 00:09:20.450 cost efficiency. Makes sense. The user message 00:09:20.450 --> 00:09:22.690 uses another formatting function, gives the writer 00:09:22.690 --> 00:09:24.950 the topic heading and all that raw research text. 00:09:25.269 --> 00:09:27.429 Okay, and the prompt, the instructions for this 00:09:27.429 --> 00:09:29.889 writer. That's its prime directive, non -negotiable 00:09:29.889 --> 00:09:33.460 rules. You're a professional newsletter section 00:09:33.460 --> 00:09:36.039 writer. Always include a clear section heading. 00:09:36.200 --> 00:09:38.500 Right. Do not write an overall title, intro, 00:09:38.639 --> 00:09:42.700 or conclusion. Focus. Total focus. If you reference 00:09:42.700 --> 00:09:46.480 facts, you must cite sources with clickable URLs. 00:09:46.600 --> 00:09:49.299 Do not invent citations. Don't make stuff up. 00:09:49.399 --> 00:09:52.399 Crucial. Yeah, absolutely. And honestly, the 00:09:52.399 --> 00:09:54.960 most important part is often telling the AI what 00:09:54.960 --> 00:09:58.419 not to do. It forces specialization. Yeah, I 00:09:58.419 --> 00:10:00.799 can see that. I still wrestle with prompt drift 00:10:00.799 --> 00:10:03.779 myself sometimes, getting the AI to just stick 00:10:03.779 --> 00:10:06.559 to the task. Me too. So those precise negative 00:10:06.559 --> 00:10:09.600 instructions, vital. Keep it on track. And a 00:10:09.600 --> 00:10:12.259 pro tip. You can even tell it to create a separate 00:10:12.259 --> 00:10:15.379 JSON array of all the facts and sources. Makes 00:10:15.379 --> 00:10:17.440 checking them super easy later. Oh, that's smart. 00:10:17.580 --> 00:10:19.679 So what's the biggest benefit of telling the 00:10:19.679 --> 00:10:21.909 AI writer what not to do in its prompt? It just 00:10:21.909 --> 00:10:24.889 forces the AI to specialize, ensures it really 00:10:24.889 --> 00:10:28.149 excels at its one single specific task instead 00:10:28.149 --> 00:10:32.710 of trying to do too much. Laser focus. Okay, 00:10:32.950 --> 00:10:35.230 writers have done their sections. Time for the 00:10:35.230 --> 00:10:37.470 editor -in -chief. Assembly and final polish 00:10:37.470 --> 00:10:40.370 time. First, an aggregate node. Aggregate. Yeah, 00:10:40.409 --> 00:10:42.190 it just combines those three separate written 00:10:42.190 --> 00:10:45.169 sections back into one single item. Puts them 00:10:45.169 --> 00:10:47.789 together. Got it. Then the final AI agent, the 00:10:47.789 --> 00:10:50.570 editor -in -chief. Right. And for this step, 00:10:50.669 --> 00:10:53.970 we upgrade. Use the full GPT -5 model. Why the 00:10:53.970 --> 00:10:56.870 upgrade here? Better writing quality, more nuance, 00:10:57.169 --> 00:11:00.210 and it's great at advanced HTML formatting. Okay, 00:11:00.250 --> 00:11:02.070 so worth the extra cost for the final polish. 00:11:02.190 --> 00:11:04.169 Definitely. You brief it with the newsletter 00:11:04.169 --> 00:11:07.230 title and those combined sections. Tell it. Add 00:11:07.230 --> 00:11:09.710 an engaging intro, a strong conclusion, format, 00:11:09.909 --> 00:11:12.909 and clean HTML with working links. Make it look 00:11:12.909 --> 00:11:15.250 professional. And structured output again? Yep. 00:11:15.309 --> 00:11:17.870 Two fields. Subject for the email subject line 00:11:17.870 --> 00:11:20.370 and content for the HTML body. Nice and clean. 00:11:20.590 --> 00:11:22.809 And another pro -level upgrade. Give it a brand 00:11:22.809 --> 00:11:24.809 voice document. Tell it to read that and make 00:11:24.809 --> 00:11:26.690 sure the draft sounds like you. Ensure consistency. 00:11:27.129 --> 00:11:29.169 Very cool. So the polished product is ready. 00:11:29.230 --> 00:11:31.029 Now the shipping department. Time to send it 00:11:31.029 --> 00:11:33.669 off. A Gmail note at the end. Yep. End of the 00:11:33.669 --> 00:11:37.600 line. Simple configuration, but one critical 00:11:37.600 --> 00:11:41.340 choice. Which is? The action must be set to create 00:11:41.340 --> 00:11:45.019 draft, not send email. Oh, okay. Create draft. 00:11:45.120 --> 00:11:47.279 You connect the dynamic subject and content from 00:11:47.279 --> 00:11:51.379 the editor, set email type to HTML to keep the 00:11:51.379 --> 00:11:53.899 formatting, send it to your own email address. 00:11:54.340 --> 00:11:57.919 And that create draft setting. Why is that so 00:11:57.919 --> 00:12:00.340 vital? It's that non -negotiable safety feature, 00:12:00.480 --> 00:12:02.740 the human in the loop. It gives you that final 00:12:02.740 --> 00:12:04.820 chance to review everything before it goes out, 00:12:04.860 --> 00:12:06.740 check for errors, maybe add a personal touch, 00:12:06.919 --> 00:12:09.960 keep control. Why is setting the Gmail action 00:12:09.960 --> 00:12:12.580 to create draft so critical for this automated 00:12:12.580 --> 00:12:15.460 system? It's that crucial safety net, that non 00:12:15.460 --> 00:12:17.899 -negotiable human in the loop allowing for that 00:12:17.899 --> 00:12:20.379 vital final review before anything gets published. 00:12:20.620 --> 00:12:22.100 Keeps you in the driver's seat for the final 00:12:22.100 --> 00:12:24.320 check, amplifies effort, doesn't just replace 00:12:24.320 --> 00:12:27.399 it. Exactly. You maintain control. Okay, so we've 00:12:27.399 --> 00:12:29.860 built this thing. A complete content factory 00:12:29.860 --> 00:12:31.799 delivering professional newsletters. Right, your 00:12:31.799 --> 00:12:34.279 inbox. The spec sheet sounds impressive. Professional 00:12:34.279 --> 00:12:37.039 structure, good title, intro, conclusion, three 00:12:37.039 --> 00:12:39.500 focus sections with citations. Yeah, evidence 00:12:39.500 --> 00:12:43.080 -based, clean HTML, ready to go. But the real 00:12:43.080 --> 00:12:46.960 game changer, the ROI is the time saved, right? 00:12:47.039 --> 00:12:49.200 Massive time savings. That's the core benefit. 00:12:49.419 --> 00:12:52.080 It just wipes out, what, 8 to 12 hours of manual 00:12:52.080 --> 00:12:55.220 work every week? Wow, 8 to 12 hours every week. 00:12:55.240 --> 00:12:58.419 That's incredible. Yeah, think about it. automating 00:12:58.419 --> 00:13:00.679 all that tedious research, writing, editing, 00:13:00.860 --> 00:13:03.279 formatting. That doesn't just save time. It completely 00:13:03.279 --> 00:13:05.379 frees you up instead of being chained to the 00:13:05.379 --> 00:13:07.460 keyboard producing content. You could be doing 00:13:07.460 --> 00:13:10.720 higher level stuff, strategy, talking to customers, 00:13:10.899 --> 00:13:13.679 exploring new ideas. It shifts your role from 00:13:13.679 --> 00:13:16.759 producer to... strategist, innovator. Exactly. 00:13:17.039 --> 00:13:19.620 And for performance tuning, making it even better. 00:13:19.820 --> 00:13:21.639 You can add things like a Google Sheet dashboard, 00:13:22.039 --> 00:13:24.360 log everything, track results, use it to tweak 00:13:24.360 --> 00:13:26.679 your prompts. A feedback loop. Yep. Add safety 00:13:26.679 --> 00:13:30.240 features like retry logic for API calls. If Tavoli 00:13:30.240 --> 00:13:32.580 fails, maybe try perplexity automatically as 00:13:32.580 --> 00:13:35.279 a backup. Redundancy. Smart. Upgrade the engine. 00:13:35.460 --> 00:13:38.159 Use multiple research sources. Add custom HTML 00:13:38.159 --> 00:13:40.679 CSS for perfect branding. Take it from good to 00:13:40.679 --> 00:13:43.419 great. Now the professional playbook. Running 00:13:43.419 --> 00:13:46.279 it like a pro. This means adopting that moneyball 00:13:46.279 --> 00:13:49.879 approach to AI models. Moneyball? Like the baseball 00:13:49.879 --> 00:13:52.860 thing? Kinda, yeah. Use the most cost -effective 00:13:52.860 --> 00:13:55.399 model for each specific job. Don't just throw 00:13:55.399 --> 00:13:57.700 the biggest, most expensive model at everything. 00:13:57.919 --> 00:14:01.460 Right. Use the right tool for the job. So, GPT 00:14:01.460 --> 00:14:04.490 -5, mini for planning and section writing. Keep 00:14:04.490 --> 00:14:06.769 cost down there. But then upgrade to the full 00:14:06.769 --> 00:14:09.149 GPT -5 for the final editor where quality and 00:14:09.149 --> 00:14:11.850 formatting really matter. Smart resource allocation. 00:14:12.149 --> 00:14:15.669 And keep experimenting. Try Claude. Try Gemini. 00:14:15.889 --> 00:14:18.470 Find the best balance for your needs. It's not 00:14:18.470 --> 00:14:20.590 one size fits all. It reminds me of those early 00:14:20.590 --> 00:14:23.429 days, just using the biggest model for everything 00:14:23.429 --> 00:14:26.409 and then getting the bill. Like using a sledgehammer 00:14:26.409 --> 00:14:29.370 to crack a nut. Exactly. This money ball idea 00:14:29.370 --> 00:14:32.110 is just smart business. Then the builder's code. 00:14:32.559 --> 00:14:34.940 Best practices. Okay, give us the code. Always 00:14:34.940 --> 00:14:37.960 build and test with a manual trigger first. Don't 00:14:37.960 --> 00:14:39.740 rely on the schedule initially. Makes sense. 00:14:39.980 --> 00:14:42.259 Crucially, pin your data during development. 00:14:42.559 --> 00:14:45.279 Pin data? What's that? It saves the output of 00:14:45.279 --> 00:14:47.139 a node. So if you tweak something later in the 00:14:47.139 --> 00:14:49.059 workflow, you don't have to rerun all the expensive 00:14:49.059 --> 00:14:52.200 AI calls before it. Saves time and money during 00:14:52.200 --> 00:14:55.600 testing. Ah, clever. Like caching results. Kinda. 00:14:55.879 --> 00:14:58.639 Use good descriptive names for your nodes. Don't 00:14:58.639 --> 00:15:01.220 just leave them as AI Agent 3. So you know it's 00:15:01.220 --> 00:15:03.659 what later. Customize your system prompts. Make 00:15:03.659 --> 00:15:05.580 them specific to your industry, your audience. 00:15:05.860 --> 00:15:08.980 Include your brand voice. And iterate. Your first 00:15:08.980 --> 00:15:11.700 prompt is just a draft. Keep refining it. Constant 00:15:11.700 --> 00:15:14.120 improvement. Use workflow templates if you can. 00:15:14.179 --> 00:15:17.039 Don't reinvent the wheel. And use version control, 00:15:17.240 --> 00:15:19.139 like GitHub, for your workflows and prompts. 00:15:19.320 --> 00:15:22.669 Treat it like code. Good practice. Finally! The 00:15:22.669 --> 00:15:26.090 launchpad. Hosting. For this to run reliably 00:15:26.090 --> 00:15:29.129 on schedule, 247, you really need proper hosting, 00:15:29.289 --> 00:15:31.990 like a managed N -A -N -A -B -S, something with 00:15:31.990 --> 00:15:34.950 backups, scalable power. Right. It needs a solid 00:15:34.950 --> 00:15:37.450 foundation to run reliably. So the bottom line, 00:15:37.549 --> 00:15:40.289 the business impact. This isn't just saving time, 00:15:40.330 --> 00:15:42.470 is it? It's a strategic asset. Absolutely. A 00:15:42.470 --> 00:15:44.590 strategic advantage. The immediate benefits, 00:15:44.870 --> 00:15:48.029 the time machine, are clear. Eight 12 -hour saved, 00:15:48.289 --> 00:15:51.009 consistent content, regular engagement. Yep. 00:15:51.230 --> 00:15:54.049 Huge wins right away. But the strategic advantages, 00:15:54.370 --> 00:15:58.850 becoming a thought leader, that's profound. It 00:15:58.850 --> 00:16:01.529 positions you as an expert, creates those valuable 00:16:01.529 --> 00:16:03.750 touch points, and frees you up for high -value 00:16:03.750 --> 00:16:05.990 work, the stuff that actually grows the business. 00:16:06.230 --> 00:16:08.669 It lets you shift from being a writer to being 00:16:08.669 --> 00:16:11.669 a publisher or a strategist, focusing on high 00:16:11.669 --> 00:16:13.970 -level growth. Exactly. And long -term, it's 00:16:13.970 --> 00:16:16.990 a compounding asset. Builds a knowledge base, 00:16:17.110 --> 00:16:19.289 develops brand authority, establishes reliable 00:16:19.289 --> 00:16:22.929 communication. Pays off for years. Beyond just 00:16:22.929 --> 00:16:25.590 saving hours, what's a significant strategic 00:16:25.590 --> 00:16:28.509 shift this system enables for someone? It transforms 00:16:28.509 --> 00:16:31.070 you from just a writer grinding out content. 00:16:31.519 --> 00:16:33.879 into a publisher or a strategist. You're focused 00:16:33.879 --> 00:16:36.480 on high -level growth, not just production. Thinking 00:16:36.480 --> 00:16:38.620 bigger picture. Okay, let's recap the big ideas. 00:16:38.840 --> 00:16:40.960 So we've seen how this multi -agent system, using 00:16:40.960 --> 00:16:44.139 something like NANN, can automate a whole newsletter 00:16:44.139 --> 00:16:46.639 workflow, start to finish. It's basically a digital 00:16:46.639 --> 00:16:48.899 content factory. You've got AI agents researching, 00:16:49.100 --> 00:16:51.299 planning, writing, editing, all specialized. 00:16:51.620 --> 00:16:54.139 Working with you. Yeah. And the real power comes 00:16:54.139 --> 00:16:56.980 from that upfront planning, the smart, strategic 00:16:56.980 --> 00:16:59.860 use of different AI models. The moneyball approach. 00:17:00.059 --> 00:17:03.240 And constantly... tweaking those prompts. Iterative 00:17:03.240 --> 00:17:05.380 improvement. And the result? It frees you from 00:17:05.380 --> 00:17:08.119 the manual grind, lets you step up, become a 00:17:08.119 --> 00:17:11.279 true publisher and strategist for your brand. 00:17:11.619 --> 00:17:14.259 It really is pretty mind -blowing. The idea that 00:17:14.259 --> 00:17:16.259 content gets created while you sleep. It's a 00:17:16.259 --> 00:17:18.420 big shift. But it raises that interesting question, 00:17:18.579 --> 00:17:21.500 doesn't it? What does this mean for human creativity? 00:17:21.900 --> 00:17:28.430 For oversight? Yeah. How do we best leverage 00:17:28.430 --> 00:17:30.950 this power? It's something to think about. Definitely 00:17:30.950 --> 00:17:32.710 something to mull over. Well, thanks for joining 00:17:32.710 --> 00:17:35.430 us for this deep dive into building a multi -agent 00:17:35.430 --> 00:17:38.029 content team. We hope you found this exploration 00:17:38.029 --> 00:17:41.269 as fascinating as we did. Until next time, keep 00:17:41.269 --> 00:17:43.210 exploring, keep questioning. And keep learning.