WEBVTT 00:00:00.000 --> 00:00:03.020 You know that feeling? It's visceral. You've 00:00:03.020 --> 00:00:05.519 just finished a sales call and it wasn't just 00:00:05.519 --> 00:00:08.339 good, it was a killer call. The chemistry was 00:00:08.339 --> 00:00:11.439 electric. You could actually see the prospect's 00:00:11.439 --> 00:00:14.640 pupils dilate when you mentioned the ROI. You 00:00:14.640 --> 00:00:18.019 hang up and you are buzzing with adrenaline and 00:00:18.019 --> 00:00:22.379 then... Silence. The crash. The crash. You open 00:00:22.379 --> 00:00:24.519 your laptop, you click on PowerPoint, and there 00:00:24.519 --> 00:00:27.179 it is. That white rectangle. The blinking cursor. 00:00:27.300 --> 00:00:29.719 Yeah. The blank page. And suddenly that adrenaline 00:00:29.719 --> 00:00:32.359 turns into absolute dread because you realize 00:00:32.359 --> 00:00:34.140 you have to spend the next three hours remembering 00:00:34.140 --> 00:00:37.539 exactly what you said, organizing it, and making 00:00:37.539 --> 00:00:39.560 it look pretty. It's the worst feeling in sales. 00:00:39.679 --> 00:00:42.399 Honestly, it's a precise moment where the job 00:00:42.399 --> 00:00:46.740 shifts from solving problems to moving pixels 00:00:46.740 --> 00:00:48.950 around. Right. And that's where most deals die. 00:00:49.090 --> 00:00:50.909 They just lose momentum because the follow -up 00:00:50.909 --> 00:00:53.829 takes way too long. Well, today we are declaring 00:00:53.829 --> 00:00:55.950 the death of the blank page. We're doing a deep 00:00:55.950 --> 00:00:59.229 dive into a specific architecture detailed by 00:00:59.229 --> 00:01:02.950 Max Ann that claims to build a 2026 AI system. 00:01:03.149 --> 00:01:06.750 Yeah. And the promise here is bold. A system 00:01:06.750 --> 00:01:09.790 that listens to your meeting and delivers a 90 00:01:09.790 --> 00:01:12.769 % finished slide deck to your inbox three minutes 00:01:12.769 --> 00:01:15.209 after you hang up. And just to be clear for everyone 00:01:15.209 --> 00:01:17.030 listening, this isn't some vague, you know, use 00:01:17.030 --> 00:01:20.049 chat GPT to summarize notes advice. We are getting 00:01:20.049 --> 00:01:22.689 under the hood of a fully automated architecture 00:01:22.689 --> 00:01:28.450 using NAN, Fireflies, and Gamma. It's about building 00:01:28.450 --> 00:01:30.430 a machine that does the grunt work so you can 00:01:30.430 --> 00:01:32.829 stay in that deal closing mindset. So let's map 00:01:32.829 --> 00:01:34.450 this out. We're going to break this architecture. 00:01:35.019 --> 00:01:37.620 down into the technical specifics. We'll look 00:01:37.620 --> 00:01:39.980 at the ears of the operation, how to capture 00:01:39.980 --> 00:01:42.480 data without losing nuance. Right. We'll talk 00:01:42.480 --> 00:01:45.780 about the brain, specifically a very clever AI 00:01:45.780 --> 00:01:48.219 persona designed to write persuasive copy, not 00:01:48.219 --> 00:01:50.200 just summaries. The most important part. And 00:01:50.200 --> 00:01:52.319 finally, the hands that actually build the slides. 00:01:52.439 --> 00:01:56.099 But first, the philosophy. The source material 00:01:56.099 --> 00:01:58.140 makes a really big point about splitting this 00:01:58.140 --> 00:02:01.640 into two workflows. Why? Why not just one straight 00:02:01.640 --> 00:02:04.980 line from audio to slides? This is the classic 00:02:04.980 --> 00:02:07.599 newbie automation mistake. You try to build one 00:02:07.599 --> 00:02:09.900 giant Rube Goldberg machine where one trigger 00:02:09.900 --> 00:02:12.979 does everything. Right. The source argues for 00:02:12.979 --> 00:02:15.800 a modular approach. You have Workflow 1, which 00:02:15.800 --> 00:02:19.199 is the logger. Its only job is to listen, clean 00:02:19.199 --> 00:02:22.080 the data, and store it. Then you have Workflow 00:02:22.080 --> 00:02:25.250 2, the generator. which creates the outline. 00:02:25.490 --> 00:02:28.969 It uses that analogy of Lego blocks versus carving 00:02:28.969 --> 00:02:31.830 a sculpture. Exactly. If you carve a statue out 00:02:31.830 --> 00:02:34.009 of marble and you make a mistake or you decide 00:02:34.009 --> 00:02:36.189 later you want to change the arm, you have to 00:02:36.189 --> 00:02:38.129 start over. You've ruined the whole stone. Yeah. 00:02:38.310 --> 00:02:40.550 If you build with Legos, you just swap a block. 00:02:41.270 --> 00:02:43.490 By separating the logging from the generating, 00:02:43.750 --> 00:02:47.210 you create scalability. Today, you might want 00:02:47.210 --> 00:02:51.060 a sales proposal, but tomorrow... Maybe you need 00:02:51.060 --> 00:02:53.479 meeting minutes for an internal sync. Or a project 00:02:53.479 --> 00:02:55.680 plan or something. Or a project plan for a kickoff. 00:02:55.840 --> 00:02:58.740 If you have that foundational logger workflow 00:02:58.740 --> 00:03:01.300 running separately, you can plug in different 00:03:01.300 --> 00:03:03.139 outcomes without breaking the whole system. So 00:03:03.139 --> 00:03:05.240 you're basically building a data reservoir first 00:03:05.240 --> 00:03:07.000 before you even decide where the pipes are going 00:03:07.000 --> 00:03:09.159 to go. Exactly. You're decoupling the input from 00:03:09.159 --> 00:03:11.960 the output. It's a safety net for growing businesses. 00:03:12.300 --> 00:03:15.539 So why is this separation specifically so critical 00:03:15.539 --> 00:03:17.719 for a business that's scaling up? Because it 00:03:17.719 --> 00:03:21.259 lets you add new deliverables like project plans 00:03:21.259 --> 00:03:24.120 or summaries without risking the stability of 00:03:24.120 --> 00:03:26.599 the core recording process. Let's get into the 00:03:26.599 --> 00:03:28.479 weeds because that's where this gets really interesting. 00:03:28.620 --> 00:03:32.169 Workflow number one, the ears. This starts with 00:03:32.169 --> 00:03:35.729 Fireflies .ai. Now, most people know Fireflies 00:03:35.729 --> 00:03:37.770 is that little bot that joins your Zoom calls, 00:03:37.990 --> 00:03:40.909 but connecting it to an automation system like 00:03:40.909 --> 00:03:44.550 NAN isn't exactly plug and play, is it? No, not 00:03:44.550 --> 00:03:46.389 at all. And this is where people get stuck. You 00:03:46.389 --> 00:03:49.430 set up a webhook in NAN, which is basically a 00:03:49.430 --> 00:03:52.110 digital catcher's mitt, and you tell Fireflies, 00:03:52.250 --> 00:03:53.930 hey, throw the ball here when the meeting is 00:03:53.930 --> 00:03:56.129 done. You check the box for transcription complete. 00:03:56.289 --> 00:03:58.930 Seems simple. It sounds simple. Yeah. But there's 00:03:58.930 --> 00:04:01.550 a glitch. A significant one. It's the webhook 00:04:01.550 --> 00:04:04.590 response problem. Fireflies is incredibly eager. 00:04:05.009 --> 00:04:07.710 It pings your automation the second the audio 00:04:07.710 --> 00:04:10.349 file is processed to say, hey, I'm done. But 00:04:10.349 --> 00:04:13.349 at that exact second, it sends almost no data. 00:04:13.509 --> 00:04:16.170 Just an ID. It's like a waiter running to your 00:04:16.170 --> 00:04:18.189 table to tell you your food is ready, but his 00:04:18.189 --> 00:04:20.129 hands are empty. So you have the notification, 00:04:20.370 --> 00:04:23.990 but not the meal. Precisely. And the problem 00:04:23.990 --> 00:04:27.459 is... The AI summaries, the action items, the 00:04:27.459 --> 00:04:30.959 gist, the sentiment analysis, those take another 00:04:30.959 --> 00:04:34.620 30 to 60 seconds to bake after the audio is processed. 00:04:34.879 --> 00:04:37.540 If your automation grabs the data the moment 00:04:37.540 --> 00:04:40.500 that webhook fires, you get blank fields. You 00:04:40.500 --> 00:04:42.480 get broken automation. So how do we solve this? 00:04:42.519 --> 00:04:44.360 The source describes something called a polling 00:04:44.360 --> 00:04:47.180 loop. It's a very clever bit of engineering that 00:04:47.180 --> 00:04:50.579 mimics human patience. First, you insert a wait 00:04:50.579 --> 00:04:53.639 node. You force the system to just pause. Just 00:04:53.639 --> 00:04:56.519 breathe for 30 seconds. Okay. Then you make the 00:04:56.519 --> 00:05:00.160 request to Fireflies. But, and this is key, you 00:05:00.160 --> 00:05:02.759 don't just assume it's ready even then. You use 00:05:02.759 --> 00:05:05.240 an IF node. You ask the system, does the summary 00:05:05.240 --> 00:05:07.620 actually exist yet? And if it doesn't? The system 00:05:07.620 --> 00:05:09.740 loops back. It waits another 30 seconds and it 00:05:09.740 --> 00:05:12.079 asks again. It keeps checking until the data 00:05:12.079 --> 00:05:14.319 is actually there. So it's essentially just checking 00:05:14.319 --> 00:05:16.100 the oven to see if the cake is ready. That's 00:05:16.100 --> 00:05:18.980 it. It prevents the system from serving you raw, 00:05:19.120 --> 00:05:22.180 unfinished data. It ensures reliability. Why 00:05:22.180 --> 00:05:24.220 do we need to build a loop, though, instead of 00:05:24.220 --> 00:05:27.170 just waiting, say, five minutes to be safe? Because 00:05:27.170 --> 00:05:29.709 speed matters. Yeah. You want the data as soon 00:05:29.709 --> 00:05:31.769 as it's ready, not five minutes later. But you 00:05:31.769 --> 00:05:34.189 can't grab it before it exists. I love that. 00:05:34.310 --> 00:05:37.069 It's a patience mechanism built right into code. 00:05:37.209 --> 00:05:39.829 Now, once we have the data, the source mentions 00:05:39.829 --> 00:05:42.310 data hygiene. We're not just dumping raw text 00:05:42.310 --> 00:05:45.209 into a file. Right. But the source suggests using 00:05:45.209 --> 00:05:47.949 Claude the AI model to actually write the code. 00:05:48.129 --> 00:05:50.730 Yeah. This is a massive unlock for non -technical 00:05:50.730 --> 00:05:53.970 people. The data coming out of Fireflies is in 00:05:53.970 --> 00:05:56.730 JSON format. Okay. So. For those who aren't developers, 00:05:56.949 --> 00:06:00.170 what exactly is JSON? Think of JSON as a very 00:06:00.170 --> 00:06:03.629 specific, very rigid way of organizing data. 00:06:03.810 --> 00:06:06.149 It's like a filing cabinet where every folder 00:06:06.149 --> 00:06:08.730 has to be labeled exactly right. It can be messy 00:06:08.730 --> 00:06:11.329 and really hard to read for humans. You might 00:06:11.329 --> 00:06:13.750 need to extract just the list of attendees or 00:06:13.750 --> 00:06:15.670 format the date so it doesn't look like computer 00:06:15.670 --> 00:06:17.949 gibberish. And usually you'd need to know JavaScript 00:06:17.949 --> 00:06:21.720 to clean all that up. Correct. But here... The 00:06:21.720 --> 00:06:24.879 hack is that you take that raw, messy JSON, you 00:06:24.879 --> 00:06:27.939 paste it into Claude or ChatGPT, and you say, 00:06:28.079 --> 00:06:32.160 I am using NA10. Write me a code node in JavaScript 00:06:32.160 --> 00:06:34.939 to extract the speaker names from this mess. 00:06:35.199 --> 00:06:37.139 And it just does it. It writes the code for you. 00:06:37.180 --> 00:06:40.019 You copy it, you paste it into NA10, and it works. 00:06:40.240 --> 00:06:42.019 You don't need to learn to code. You just need 00:06:42.019 --> 00:06:44.120 to know how to ask for the code. That just lowers 00:06:44.120 --> 00:06:47.939 the barrier to entry so significantly. So we 00:06:47.939 --> 00:06:50.870 clean the data, and then we log it. The destination 00:06:50.870 --> 00:06:54.089 here is a Google Sheet. Why log it to Sheets 00:06:54.089 --> 00:06:56.170 if we're just making a slide deck? Why not go 00:06:56.170 --> 00:06:57.990 straight to the slide generator? Because you 00:06:57.990 --> 00:07:00.550 want a master brain. By logging every single 00:07:00.550 --> 00:07:03.889 meeting into a structured spreadsheet date, title, 00:07:04.089 --> 00:07:07.529 attendees, gist, meeting ID, you're creating 00:07:07.529 --> 00:07:09.689 a historical database. It creates a paper trail. 00:07:09.810 --> 00:07:12.410 I see. But more importantly, that new row in 00:07:12.410 --> 00:07:14.670 the spreadsheet acts as the trigger for the next 00:07:14.670 --> 00:07:17.209 step. it ensures you never lose a meeting's history 00:07:17.209 --> 00:07:19.850 even if the slide generation fails for some reason 00:07:19.850 --> 00:07:22.629 so what's the primary strategic reason for logging 00:07:22.629 --> 00:07:25.209 everything to a spreadsheet first it creates 00:07:25.209 --> 00:07:28.050 a master database and acts as a reliable trigger 00:07:28.050 --> 00:07:30.709 for any future automation so nothing is lost 00:07:30.709 --> 00:07:33.230 okay so we have our logger running every meeting 00:07:33.230 --> 00:07:35.670 is being recorded cleaned and neatly filed into 00:07:35.670 --> 00:07:38.470 a spreadsheet row now we move to workflow number 00:07:38.470 --> 00:07:42.009 two the generator this is where the magic happens 00:07:42.009 --> 00:07:45.579 the trigger is that new row in the spreadsheet. 00:07:45.920 --> 00:07:49.000 Correct. But here is a nuance. We don't want 00:07:49.000 --> 00:07:51.060 to just blast out proposals for every single 00:07:51.060 --> 00:07:53.019 meeting. Right. If I have a 15 -minute coffee 00:07:53.019 --> 00:07:55.500 chat with a colleague, I don't need a 10 -slide 00:07:55.500 --> 00:07:57.519 commercial deck sent to my email. That would 00:07:57.519 --> 00:08:00.540 be a little aggressive and a waste of computing 00:08:00.540 --> 00:08:04.040 credits. Very. So we introduce a Humean loop. 00:08:04.300 --> 00:08:06.759 Yeah. The automation sends the message to Slack. 00:08:06.899 --> 00:08:09.920 It says, meeting X just ended. Do you want to 00:08:09.920 --> 00:08:12.420 generate a proposal? And it gives you two buttons, 00:08:12.620 --> 00:08:14.839 yes or no. I find this so interesting because 00:08:14.839 --> 00:08:17.540 usually automation is all about removing friction. 00:08:17.860 --> 00:08:20.600 But here we are deliberately inserting friction. 00:08:20.899 --> 00:08:23.459 A button you have to click. It's strategic friction. 00:08:24.339 --> 00:08:27.019 You want the creation to be automated, but the 00:08:27.019 --> 00:08:30.519 intent to be human. If you click no, the system 00:08:30.519 --> 00:08:32.779 updates the spreadsheet to say generation declined, 00:08:33.179 --> 00:08:37.480 and it stops. But if you click yes, that's when 00:08:37.480 --> 00:08:39.559 we wake up the agent. The proposal generator 00:08:39.559 --> 00:08:41.919 agent. Now, this isn't just a generic prompt 00:08:41.919 --> 00:08:44.200 that says summarize this meeting. The source 00:08:44.200 --> 00:08:47.000 material is very specific about the persona we're 00:08:47.000 --> 00:08:49.720 assigning to the AI. This is critical. I mean, 00:08:49.779 --> 00:08:52.059 prompt engineering is often the difference between 00:08:52.059 --> 00:08:54.679 a mediocre output and a really great one. The 00:08:54.679 --> 00:08:56.919 prompt here explicitly defines the AI's role. 00:08:57.440 --> 00:09:00.179 Expert Senior AI Solutions Consultant for UpIt 00:09:00.179 --> 00:09:02.899 AI. UpIt AI being the fictional company in this 00:09:02.899 --> 00:09:04.559 example. Right. You'd swap that for your own 00:09:04.559 --> 00:09:07.860 company. But the key is the title. Senior Consultant. 00:09:08.320 --> 00:09:10.799 It tells the AI to be confident, authoritative, 00:09:10.860 --> 00:09:13.700 and strategic. It changes the tone from a junior 00:09:13.700 --> 00:09:16.340 note -taker to a senior partner. The constraints 00:09:16.340 --> 00:09:18.259 listed in the prompt are interesting, too. It 00:09:18.259 --> 00:09:21.159 says do not include internal notes, do not mention 00:09:21.159 --> 00:09:23.840 this was AI -generated, and my favorite, make 00:09:23.840 --> 00:09:26.659 confident assumptions. Ah, yeah. That confident 00:09:26.659 --> 00:09:29.360 assumptions part is scary but necessary. Is it? 00:09:29.440 --> 00:09:31.879 I mean, that feels risky. If the AI guesses wrong 00:09:31.879 --> 00:09:34.259 on a number, couldn't that blow the deal? It's 00:09:34.259 --> 00:09:37.159 a calculated risk. Look, if the client didn't 00:09:37.159 --> 00:09:39.620 explicitly say we lose $50 ,000 a year on this 00:09:39.620 --> 00:09:42.279 problem, but they hinted at it, you want the 00:09:42.279 --> 00:09:44.919 AI to estimate it. Why? Because a proposal with 00:09:44.919 --> 00:09:47.919 placeholders like insert number here looks unfinished. 00:09:48.200 --> 00:09:50.519 It looks like a template. A proposal that says 00:09:50.519 --> 00:09:54.159 estimated annual loss $50 ,000 looks proactive. 00:09:54.500 --> 00:09:56.820 It gives the client something to react to. So 00:09:56.820 --> 00:09:59.460 it's easier to correct a wrong number than to 00:09:59.460 --> 00:10:01.899 fill in a blank one. Exactly. It shifts the conversation. 00:10:02.159 --> 00:10:04.490 If the number is wrong. The client corrects you, 00:10:04.590 --> 00:10:06.370 and now you have the real number. If you leave 00:10:06.370 --> 00:10:09.149 it blank, the client just ignores it. The structure 00:10:09.149 --> 00:10:12.009 demanded by the prompt is also very rigid. Nine 00:10:12.009 --> 00:10:14.509 specific sections. Executive summary, problem 00:10:14.509 --> 00:10:18.789 solution, ROI, roadmap. It forces the AI to speak 00:10:18.789 --> 00:10:21.289 the language of business value. It's not just 00:10:21.289 --> 00:10:24.029 transcribing, it's translating. It takes, we 00:10:24.029 --> 00:10:26.690 hate doing this manual data entry, and turns 00:10:26.690 --> 00:10:29.470 it into operational bottleneck, resulting in 00:10:29.470 --> 00:10:32.360 15 wasted hours per week. So what is the main 00:10:32.360 --> 00:10:34.779 psychological shift we're trying to force the 00:10:34.779 --> 00:10:37.600 AI to make with this specific persona? We are 00:10:37.600 --> 00:10:40.120 forcing it to be persuasive and client -facing, 00:10:40.299 --> 00:10:42.659 moving away from passive description to active 00:10:42.659 --> 00:10:45.419 sales. So we have this persuasive text generated. 00:10:45.480 --> 00:10:47.559 Now we need to get it into a slide deck. And 00:10:47.559 --> 00:10:49.960 we're using a tool called Gamma for this. Gamma 00:10:49.960 --> 00:10:52.080 is fantastic because it's not just a text -to 00:10:52.080 --> 00:10:55.080 -slide tool. It understands design. We use an 00:10:55.080 --> 00:10:58.840 HTTP request node, an NAN, to talk to Gamma's 00:10:58.840 --> 00:11:01.379 API. This sounds technical. Walk us through what 00:11:01.379 --> 00:11:03.179 that actually means. It's a little bit technical, 00:11:03.200 --> 00:11:05.559 but totally manageable. You're basically sending 00:11:05.559 --> 00:11:08.559 a digital package to Gamma. Inside that package, 00:11:08.659 --> 00:11:10.820 you tell it the theme, so it uses your company's 00:11:10.820 --> 00:11:13.279 colors and fonts, not some random template. Okay. 00:11:13.379 --> 00:11:15.700 You tell it the source for images. We're using 00:11:15.700 --> 00:11:18.480 DAO E3 here for unique visuals. And you paste 00:11:18.480 --> 00:11:21.350 in that structured text from the AI agent. There's 00:11:21.350 --> 00:11:23.250 a specific replace function mentioned in the 00:11:23.250 --> 00:11:25.169 source, a bit of code cleanup before sending 00:11:25.169 --> 00:11:27.730 to Gamma, something about new lines. Yeah, this 00:11:27.730 --> 00:11:30.730 is a classic gotcha in automation. AI models 00:11:30.730 --> 00:11:34.769 love to output text with line breaks, tabs, quotation 00:11:34.769 --> 00:11:38.490 marks. If you feed that raw text into a JSON 00:11:38.490 --> 00:11:42.620 payload for an API, It breaks the code. It confuses 00:11:42.620 --> 00:11:44.320 the system because it thinks the code has ended 00:11:44.320 --> 00:11:46.720 when it hasn't. So the Cisco just crashes. It 00:11:46.720 --> 00:11:49.320 crashes hard. So you use a simple dot replace 00:11:49.320 --> 00:11:51.799 function in JavaScript to smooth out those characters. 00:11:52.179 --> 00:11:54.399 It's like sanding down the rough edges of a puzzle 00:11:54.399 --> 00:11:57.580 piece so it sits into the slot. It replaces actual 00:11:57.580 --> 00:11:59.980 line breaks with the symbols for line breaks 00:11:59.980 --> 00:12:02.220 so the machine can read it. And the result? The 00:12:02.220 --> 00:12:04.419 result is a link. It arrives in your inbox. You 00:12:04.419 --> 00:12:06.259 click it and there's a fully built presentation. 00:12:06.899 --> 00:12:09.340 Title page with the client's name. An executive 00:12:09.340 --> 00:12:13.190 summary. A slide calculating the ROI. A roadmap 00:12:13.190 --> 00:12:16.009 slide. Is the output actually ready to send right 00:12:16.009 --> 00:12:18.370 that second? It's 90 % there. And that's the 00:12:18.370 --> 00:12:20.629 honest truth. You still need to review it. Maybe 00:12:20.629 --> 00:12:22.769 the AI estimated the savings a bit too high. 00:12:22.909 --> 00:12:25.149 Maybe the tone is slightly too formal. Right. 00:12:25.230 --> 00:12:27.929 But you are editing, not creating. Yeah. And 00:12:27.929 --> 00:12:30.009 that is a fundamentally different cognitive load. 00:12:30.190 --> 00:12:33.309 So ultimately, does this output actually remove 00:12:33.309 --> 00:12:35.929 the human from the process? No, it just changes 00:12:35.929 --> 00:12:38.850 the human's role from writer to editor, which 00:12:38.850 --> 00:12:41.000 is much, much faster. We're going to take a pause 00:12:41.000 --> 00:12:44.000 right here. We've built the machine. But when 00:12:44.000 --> 00:12:45.600 we come back, we're going to talk about a feature 00:12:45.600 --> 00:12:48.700 that I think is the hidden gem of this entire 00:12:48.700 --> 00:12:51.340 system, the on -demand trigger. Because what 00:12:51.340 --> 00:12:53.799 happens when you accidentally click no, we are 00:12:53.799 --> 00:12:55.820 back on the deep dive. We've built the logger. 00:12:55.840 --> 00:12:57.460 We've built the generator. We have our Slack 00:12:57.460 --> 00:13:00.320 button. But the source material brings up a very 00:13:00.320 --> 00:13:04.960 real human problem. What if I click no in Slack? 00:13:05.740 --> 00:13:08.019 But then a week later, I changed my mind. Right. 00:13:08.179 --> 00:13:10.059 Actually, that guy was interested. I really should 00:13:10.059 --> 00:13:12.200 send a proposal. If your system only works in 00:13:12.200 --> 00:13:14.139 real time, you're out of luck. The moment has 00:13:14.139 --> 00:13:17.120 passed. So we need an on -demand feature, a way 00:13:17.120 --> 00:13:19.159 to go back in time. The solution is a manual 00:13:19.159 --> 00:13:22.360 trigger, a form submission node in N8n. You go 00:13:22.360 --> 00:13:24.379 to your Google Sheet, remember our master database, 00:13:24.539 --> 00:13:26.980 dash copy the meeting ID, paste it into a simple 00:13:26.980 --> 00:13:29.320 form, and hit go. But here's the engineering 00:13:29.320 --> 00:13:32.159 challenge. You now have two different ways to 00:13:32.159 --> 00:13:35.769 start the same process. Path A is the automatic 00:13:35.769 --> 00:13:39.269 path from the spreadsheet. Path B is the manual 00:13:39.269 --> 00:13:42.929 path from the form. How does the AI agent know 00:13:42.929 --> 00:13:45.090 which one to listen to? This is where we get 00:13:45.090 --> 00:13:47.669 into what the author calls technical magic. It's 00:13:47.669 --> 00:13:50.769 all about standardizing inputs. Okay. Imagine 00:13:50.769 --> 00:13:54.649 two different roads merging onto a highway. The 00:13:54.649 --> 00:13:57.169 AI agent is driving on the highway. It doesn't 00:13:57.169 --> 00:13:59.049 care which road the car came from. It just needs 00:13:59.049 --> 00:14:01.309 the car. So we need a traffic controller. Exactly. 00:14:01.450 --> 00:14:04.149 We use a set node as a middle layer. It sits 00:14:04.149 --> 00:14:06.429 between the triggers and the AI. It looks at 00:14:06.429 --> 00:14:08.269 path A and path B and says, okay, whoever sent 00:14:08.269 --> 00:14:10.590 data, I'm going to map it to this one variable 00:14:10.590 --> 00:14:13.610 called JSON .transcript. So whether the transcript 00:14:13.610 --> 00:14:15.950 came from the live automation or the manual form, 00:14:16.149 --> 00:14:18.830 the AI agent just looks for JSON .transcript. 00:14:18.870 --> 00:14:20.789 Right. It decouples the source from the processor. 00:14:21.110 --> 00:14:23.179 It's elegant. Because it means you don't have 00:14:23.179 --> 00:14:25.500 to duplicate the entire AI workflow for every 00:14:25.500 --> 00:14:27.740 different trigger. You just map the inputs to 00:14:27.740 --> 00:14:30.080 the same variable names. So why is this specific 00:14:30.080 --> 00:14:32.539 set node technique considered technical magic? 00:14:32.799 --> 00:14:35.899 It allows you to use the same AI brain for multiple 00:14:35.899 --> 00:14:38.100 different triggers without duplicating your work. 00:14:38.259 --> 00:14:41.580 Let's zoom out. We've built this system. It works. 00:14:42.039 --> 00:14:45.240 What's the actual impact? The source does some 00:14:45.240 --> 00:14:47.220 ROI math here. Yeah, the math is compelling. 00:14:47.360 --> 00:14:49.700 It assumes you write about two proposals a week. 00:14:51.050 --> 00:14:53.429 Manually, that's maybe three hours each. That's 00:14:53.429 --> 00:14:56.370 six hours a week. Which is nearly a full workday. 00:14:56.509 --> 00:14:59.309 Right. Over a year, if you're doing volume, say, 00:14:59.450 --> 00:15:03.029 100 proposals, that's 300 hours. That is seven 00:15:03.029 --> 00:15:05.690 and a half work weeks. Wow. Almost two months 00:15:05.690 --> 00:15:08.090 of working time. Just spent formatting slides. 00:15:08.210 --> 00:15:10.830 Just spent staring at a screen, copy pasting, 00:15:10.850 --> 00:15:13.990 adjusting font sizes, aligning boxes. And the 00:15:13.990 --> 00:15:17.419 competitor who has this system. they're using 00:15:17.419 --> 00:15:20.240 those 300 hours to meet more clients, to refine 00:15:20.240 --> 00:15:22.620 their pitch, or just to rest so they're sharper 00:15:22.620 --> 00:15:24.799 on the next call. And the finished product isn't 00:15:24.799 --> 00:15:27.740 just text. It includes that ROI slide, the roadmap. 00:15:28.000 --> 00:15:29.840 It looks professional. It looks like you spent 00:15:29.840 --> 00:15:32.419 four hours on it. That perception of effort, 00:15:32.539 --> 00:15:35.340 it matters to clients. If you turn around a high 00:15:35.340 --> 00:15:38.379 -quality deck in 20 minutes, it signals competence. 00:15:38.740 --> 00:15:41.059 It signals that you have your act together. It 00:15:41.059 --> 00:15:43.519 strikes me that the real value here isn't just 00:15:43.519 --> 00:15:46.860 time. It's cognitive bandwidth. Oh, absolutely. 00:15:47.039 --> 00:15:49.679 The blade page syndrome is draining. It causes 00:15:49.679 --> 00:15:52.419 procrastination. You put off writing the proposal 00:15:52.419 --> 00:15:54.779 because you know it's going to be a slog. With 00:15:54.779 --> 00:15:57.919 this system, the friction is almost zero. You 00:15:57.919 --> 00:16:01.799 click yes and slack. That's it. It changes your 00:16:01.799 --> 00:16:04.230 entire relationship with the work. You know, 00:16:04.250 --> 00:16:07.289 we talk a lot about AI replacing jobs, but this 00:16:07.289 --> 00:16:09.250 feels different. This feels like AI enabling 00:16:09.250 --> 00:16:12.590 relationships. That's the big idea here. AI agents 00:16:12.590 --> 00:16:14.950 don't replace the handshake. They don't replace 00:16:14.950 --> 00:16:17.110 the strategy. They replace the busy work. They 00:16:17.110 --> 00:16:19.809 eliminate the barrier between having a good meeting 00:16:19.809 --> 00:16:22.710 and asking for the sale. If you are still formatting 00:16:22.710 --> 00:16:25.110 slides for two hours, you are technically doing 00:16:25.110 --> 00:16:27.509 the job, but you are falling behind. You're fighting 00:16:27.509 --> 00:16:29.509 a war with a musket while the other side has 00:16:29.509 --> 00:16:32.799 laser guided missiles. A vivid image. So here 00:16:32.799 --> 00:16:34.500 is a final thought to mull over. We've talked 00:16:34.500 --> 00:16:36.899 about using this for sales proposals, but what 00:16:36.899 --> 00:16:39.220 happens when you turn this lens inward? What 00:16:39.220 --> 00:16:42.299 if you use the same logger and generator architecture 00:16:42.299 --> 00:16:45.360 for your own internal decision making? How do 00:16:45.360 --> 00:16:47.879 you mean? Imagine an AI that listens to your 00:16:47.879 --> 00:16:50.659 team's internal debates and automatically generates 00:16:50.659 --> 00:16:53.159 a pros and cons document. or a decision memo, 00:16:53.360 --> 00:16:55.919 completely free of the bias of whoever usually 00:16:55.919 --> 00:16:58.659 takes the notes. That is fascinating. An objective 00:16:58.659 --> 00:17:01.139 historian for the company. That could change 00:17:01.139 --> 00:17:03.360 how meetings are run entirely. Something to explore. 00:17:03.840 --> 00:17:06.220 Thank you for listening to this deep dive. Don't 00:17:06.220 --> 00:17:07.960 just read about the Lego blocks. Go build them. 00:17:08.460 --> 00:17:09.680 See you in the next deep dive.