WEBVTT 00:00:00.000 --> 00:00:02.359 It's the morning after the demo. You've just 00:00:02.359 --> 00:00:04.719 shown the client this incredible AI workflow. 00:00:04.799 --> 00:00:07.059 It's working perfectly. It summarizes emails, 00:00:07.559 --> 00:00:12.919 updates the CRM, drafts responses. It feels like 00:00:12.919 --> 00:00:15.660 magic. The client's loving it. They are smiling. 00:00:16.339 --> 00:00:20.000 But then the smile fades just a little, and they 00:00:20.000 --> 00:00:23.300 ask a simple question. So where does this actually 00:00:23.300 --> 00:00:27.579 live? And who pays if we run this 10 ,000 times? 00:00:27.699 --> 00:00:30.440 Yeah, that's the moment. And right there, the 00:00:30.440 --> 00:00:32.140 project doesn't collapse because the code is 00:00:32.140 --> 00:00:34.920 bad. It collapses because you realize building 00:00:34.920 --> 00:00:37.859 the AI. Yeah. That was the easy part. You built 00:00:37.859 --> 00:00:39.960 a prototype, but you didn't build a system. Right. 00:00:40.039 --> 00:00:42.060 It's that sinking feeling in your stomach. You've 00:00:42.060 --> 00:00:43.920 done the fun part, the prompt engineering, the 00:00:43.920 --> 00:00:46.299 logic puzzles, but you haven't done the delivery 00:00:46.299 --> 00:00:48.380 engineering. And that is a completely different 00:00:48.380 --> 00:00:52.280 beast. Welcome back to the deep dive. Today, 00:00:52.439 --> 00:00:54.399 we are stepping away from the hype cycle. We 00:00:54.399 --> 00:00:55.740 aren't talking about the latest model release 00:00:55.740 --> 00:00:58.799 or some new feature. We are settling into a much 00:00:58.799 --> 00:01:02.539 calmer, maybe more critical analysis of professional 00:01:02.539 --> 00:01:05.219 AI delivery. We're looking at the philosophy 00:01:05.219 --> 00:01:09.379 of systems. How do we move from a cool prototype 00:01:09.379 --> 00:01:12.359 on a laptop to a reliable automation that lives 00:01:12.359 --> 00:01:14.659 inside a business? I love this topic because 00:01:14.659 --> 00:01:16.799 it's where the rubber meets the road. We're seeing 00:01:16.799 --> 00:01:19.280 a massive shift right now. We're moving away 00:01:19.280 --> 00:01:22.359 from prompt engineering, which is just getting 00:01:22.359 --> 00:01:24.599 the model to talk into delivery engineering. 00:01:25.180 --> 00:01:28.540 This is about reliability, security, billing, 00:01:29.079 --> 00:01:31.400 and frankly, keeping your sanity as a builder 00:01:31.400 --> 00:01:34.319 so you don't burn out. We have a really comprehensive 00:01:34.319 --> 00:01:36.819 guide in front of us today, the AI automation 00:01:36.819 --> 00:01:40.129 delivery lifecycle. It argues that most failures 00:01:40.129 --> 00:01:42.750 happen after the demo. So our mission today is 00:01:42.750 --> 00:01:45.209 to really dissect this life cycle. Let's do it. 00:01:45.269 --> 00:01:46.890 We're going to look at the three models of hosting, 00:01:47.049 --> 00:01:49.349 the non -negotiables of security, which, spoiler, 00:01:49.430 --> 00:01:51.590 should be boring, the psychology of billing, 00:01:51.829 --> 00:01:53.629 and then the art of the clean handover. It's 00:01:53.629 --> 00:01:55.670 a roadmap for professionalism. If you follow 00:01:55.670 --> 00:01:57.469 it, you're a systems architect. If you don't, 00:01:57.530 --> 00:01:59.290 you're just a tinkerer with a dangerous toy. 00:01:59.689 --> 00:02:02.530 OK, let's unpack this. The source material starts 00:02:02.530 --> 00:02:06.219 with a very strong claim. Most delivery issues 00:02:06.219 --> 00:02:08.520 come from hosting decisions made way too late. 00:02:08.759 --> 00:02:11.439 Yeah. It seems like a technical detail, but the 00:02:11.439 --> 00:02:14.520 author argues this is the core conflict. It absolutely 00:02:14.520 --> 00:02:17.300 is, because where the automation lives determines 00:02:17.300 --> 00:02:21.240 who owns it, who pays for it, and who gets sued 00:02:21.240 --> 00:02:24.939 if it breaks. The guide outlines three models, 00:02:25.599 --> 00:02:28.060 and mixing them up is where people get into so 00:02:28.060 --> 00:02:29.699 much trouble. Okay, let's walk through them. 00:02:30.140 --> 00:02:33.219 Option one is what the source calls the recommended 00:02:33.219 --> 00:02:36.000 path. This is where the client hosts. Correct. 00:02:36.360 --> 00:02:38.319 Think of it like being a specialized consultant 00:02:38.319 --> 00:02:41.000 or an architect, right? When you hire an architect 00:02:41.000 --> 00:02:43.360 to design a house, they don't build the house 00:02:43.360 --> 00:02:45.560 on their own land and then rent it to you. They 00:02:45.560 --> 00:02:48.400 build it on your land. In automation terms, the 00:02:48.400 --> 00:02:51.000 client buys the software license, the Zapier 00:02:51.000 --> 00:02:54.060 seat, the make account, and you, the builder, 00:02:54.340 --> 00:02:56.120 you just step into their environment to assemble 00:02:56.120 --> 00:02:58.099 the furniture. That makes sense. Keeps the lines 00:02:58.099 --> 00:03:00.680 clean. But then there's option two, internal 00:03:00.680 --> 00:03:02.780 operations. This is where you, the builder, host 00:03:02.780 --> 00:03:05.919 it. Yes, but there's a huge caveat here. You 00:03:05.919 --> 00:03:08.840 can only do this if the automation is for your 00:03:08.840 --> 00:03:11.819 own business operations. Let's say you have an 00:03:11.819 --> 00:03:15.300 agency and you use AI to route your own leads 00:03:15.300 --> 00:03:18.340 or generate reports that you then email to a 00:03:18.340 --> 00:03:20.479 client. That's fine. Because they never log in. 00:03:20.580 --> 00:03:22.740 The client never logs in. They never see the 00:03:22.740 --> 00:03:25.699 gears turning. You're selling the outcome, not 00:03:25.699 --> 00:03:28.169 the machine. OK, but I want to push on this next 00:03:28.169 --> 00:03:31.830 part because option three is the sauce model. 00:03:32.270 --> 00:03:34.189 And honestly, that's the dream for a lot of people 00:03:34.189 --> 00:03:36.349 listening, isn't it? Sure. You build one amazing 00:03:36.349 --> 00:03:39.210 automation, you host it, you charge 50 clients 00:03:39.210 --> 00:03:43.710 a monthly fee, passive income. Why does the guide 00:03:43.710 --> 00:03:46.169 warn against this? Why should we walk away from 00:03:46.169 --> 00:03:47.629 that? I'm not saying walk away from revenue. 00:03:47.750 --> 00:03:49.650 I'm saying know what game you're playing. If 00:03:49.650 --> 00:03:52.050 you host the automation and let clients log in, 00:03:52.330 --> 00:03:54.509 you are no longer a service provider. You are 00:03:54.509 --> 00:03:56.789 a re. You're a software vendor. Which means? 00:03:57.289 --> 00:03:59.810 Which means you are legally responsible for the 00:03:59.810 --> 00:04:02.810 data. If your server gets hacked and 50 clients 00:04:02.810 --> 00:04:05.550 lose their customer lists, you get sued. Not 00:04:05.550 --> 00:04:08.449 OpenAI, not the platform you. You're the data 00:04:08.449 --> 00:04:11.560 controller. And if you look at the terms of service 00:04:11.560 --> 00:04:15.039 for standard API keys from, say, Anthropic or 00:04:15.039 --> 00:04:18.939 OpenAI, they often strictly forbid reselling 00:04:18.939 --> 00:04:20.779 the utility without an enterprise agreement. 00:04:21.240 --> 00:04:24.500 Right. So unless you have a legal team, a VC 00:04:24.500 --> 00:04:27.819 fund for those enterprise licenses, and a cybersecurity 00:04:27.819 --> 00:04:30.920 expert on retainer, don't pretend to be a sauce. 00:04:31.319 --> 00:04:34.060 Be a high -end architect who builds on the client's 00:04:34.060 --> 00:04:37.139 land. So the golden rule here is client work 00:04:37.139 --> 00:04:40.500 means the client hosts. Pretty much, unless you 00:04:40.500 --> 00:04:42.800 want to be a tech CEO with all that liability. 00:04:43.139 --> 00:04:45.560 Why is that distinction between service and product 00:04:45.560 --> 00:04:47.819 so critical for a freelancer? It prevents legal 00:04:47.819 --> 00:04:50.759 risks and expensive licensing fees. Moving on 00:04:50.759 --> 00:04:52.740 to the next layer, we have the foundation, the 00:04:52.740 --> 00:04:55.740 hosting, now we need the walls. Security. The 00:04:55.740 --> 00:04:57.339 source makes a point here that I found really 00:04:57.339 --> 00:04:59.620 interesting. Good security doesn't look impressive. 00:04:59.860 --> 00:05:02.540 It looks boring. I love that line. Security should 00:05:02.540 --> 00:05:04.579 be boring. It should be predictable. If your 00:05:04.579 --> 00:05:06.740 security is exciting, something has gone very 00:05:06.740 --> 00:05:08.420 wrong. Let's talk about credentials. I think 00:05:08.420 --> 00:05:10.019 we've all been guilty of this, especially in 00:05:10.019 --> 00:05:12.600 the early days. Hard coding an API key right 00:05:12.600 --> 00:05:14.980 into a script just to test something. It's easy. 00:05:15.079 --> 00:05:18.379 It's fast. It's the classic Ricky mistake. And 00:05:18.379 --> 00:05:22.629 the rule here is absolute. Never paste API keys 00:05:22.629 --> 00:05:25.730 or passwords directly into workflow steps. Ever. 00:05:26.029 --> 00:05:27.930 Professional automation platforms have what are 00:05:27.930 --> 00:05:30.810 called environment variables or a credentials 00:05:30.810 --> 00:05:34.589 vault. You save the key in the secure vault and 00:05:34.589 --> 00:05:36.829 the workflow just references it by a nickname. 00:05:37.149 --> 00:05:40.689 So the code just says, use my open AI key, but 00:05:40.689 --> 00:05:42.529 it doesn't actually show the string of characters. 00:05:43.209 --> 00:05:44.810 Exactly. It means if you download the workflow 00:05:44.810 --> 00:05:47.350 or share a screenshot or export the file to send 00:05:47.350 --> 00:05:49.810 to a friend, the key isn't there. It's stripped 00:05:49.810 --> 00:05:52.769 out. It's basic hygiene, but it saves you from 00:05:52.769 --> 00:05:54.870 accidentally leaking your credit card to the 00:05:54.870 --> 00:05:56.610 whole internet. And then there's the issue of 00:05:56.610 --> 00:05:59.579 entry points. Webhooks. Now for anyone listening 00:05:59.579 --> 00:06:01.439 who might know the term but not the mechanics, 00:06:02.079 --> 00:06:04.560 a webhook is just a way to trigger an automation 00:06:04.560 --> 00:06:06.660 from the outside world. Right. It's a public 00:06:06.660 --> 00:06:09.699 URL. You send data to that URL and the automation 00:06:09.699 --> 00:06:12.519 wakes up and runs. And that sounds a little dangerous. 00:06:12.699 --> 00:06:14.379 Webhooks are terrifying if you think about them 00:06:14.379 --> 00:06:16.939 too much. A webhook is basically a public door 00:06:16.939 --> 00:06:19.160 to your automation. If I have the URL, I can 00:06:19.160 --> 00:06:21.360 knock on that door. And if it's not locked? I 00:06:21.360 --> 00:06:24.860 can walk right in and run your bot. on my data, 00:06:24.959 --> 00:06:26.879 on your dime. Okay, let's stick with the door 00:06:26.879 --> 00:06:31.319 analogy. If I have the address, the URL, I can 00:06:31.319 --> 00:06:34.560 walk up and knock. How do you actually stop me 00:06:34.560 --> 00:06:37.639 from coming in? Is it just using HTTPS? Well, 00:06:37.660 --> 00:06:40.220 HTTPS is the baseline, but that's just a secure 00:06:40.220 --> 00:06:42.399 tunnel. It stops people from listening in on 00:06:42.399 --> 00:06:44.279 the conversation, but it doesn't stop someone 00:06:44.279 --> 00:06:47.060 from knocking. The mistake most people make is 00:06:47.060 --> 00:06:50.399 thinking the URL itself is a secret. It's not. 00:06:50.519 --> 00:06:53.639 It'll leak. So what's the lock? The lock is something 00:06:53.639 --> 00:06:56.040 called header validation. Translate that for 00:06:56.040 --> 00:06:58.220 me. Think of it like a speakeasy. Knowing where 00:06:58.220 --> 00:07:01.439 the door is isn't enough. When you knock, the 00:07:01.439 --> 00:07:04.579 bouncer asks for a password. In tech terms, the 00:07:04.579 --> 00:07:07.319 sending server -like typeform or stripe includes 00:07:07.319 --> 00:07:09.720 a secret handshake in the data packet headers. 00:07:10.279 --> 00:07:12.300 Your automation shouldn't just open the door. 00:07:12.339 --> 00:07:14.740 It should check for that handshake first. If 00:07:14.740 --> 00:07:16.920 it's missing or it's the wrong password, you 00:07:16.920 --> 00:07:18.600 don't even run the workflow. You just slam the 00:07:18.600 --> 00:07:20.980 door. You know, I have to admit something here. 00:07:21.379 --> 00:07:23.680 Even just discussing this, I feel a bit of that 00:07:23.680 --> 00:07:26.279 anxiety. I still wrestle with the fear of a prompt 00:07:26.279 --> 00:07:30.259 injection or a data leak myself. You hand these 00:07:30.259 --> 00:07:32.540 systems over to a client, but you worry, did 00:07:32.540 --> 00:07:34.860 I close every loop? Is the prompt tight enough? 00:07:35.879 --> 00:07:37.740 What if someone tricks the AI into revealing 00:07:37.740 --> 00:07:40.019 something it shouldn't? That's a very healthy 00:07:40.019 --> 00:07:42.019 fear to have. If you aren't a little scared, 00:07:42.079 --> 00:07:43.899 you aren't paying attention. And honestly, the 00:07:43.899 --> 00:07:46.060 only way to manage that anxiety is through what 00:07:46.060 --> 00:07:49.339 the source calls data minimization. OK. If you're 00:07:49.339 --> 00:07:52.000 processing personal data names, emails only pull 00:07:52.000 --> 00:07:54.500 in exactly what you need for that specific task. 00:07:55.019 --> 00:07:57.379 Don't dump the whole database into the context 00:07:57.379 --> 00:08:00.500 window and limit who can see the execution logs. 00:08:00.879 --> 00:08:03.000 If you limit the surface area, you limit the 00:08:03.000 --> 00:08:05.720 blast radius if something goes wrong. Blast radius? 00:08:05.879 --> 00:08:08.620 That's a vivid image. It's the reality. A small 00:08:08.620 --> 00:08:11.040 pop, not an explosion. That brings us to a different 00:08:11.040 --> 00:08:13.740 kind of risk. One that is maybe less technical, 00:08:14.060 --> 00:08:16.779 but just as dangerous to a project. The money. 00:08:17.560 --> 00:08:20.319 Ah, yes. The awkward conversation. The guide 00:08:20.319 --> 00:08:23.339 calls this the friction point, billing. And the 00:08:23.339 --> 00:08:26.519 rule is... Clients own their API keys, and clients 00:08:26.519 --> 00:08:29.000 pay for their usage. Non -negotiable. I hear 00:08:29.000 --> 00:08:31.439 you on transparency, but let's be real. If I 00:08:31.439 --> 00:08:33.500 hire an expert, I don't want them handing me 00:08:33.500 --> 00:08:35.419 a confusing checklist and saying, go figure out 00:08:35.419 --> 00:08:38.279 OpenAI's dashboard. Isn't that just bad service? 00:08:38.759 --> 00:08:41.519 It feels lazy. Here, you go do the hard part. 00:08:41.639 --> 00:08:43.720 It feels lazy if you just dump it in their lap, 00:08:43.799 --> 00:08:46.740 sure. But think about the alternative. If you 00:08:46.740 --> 00:08:49.779 pay the bill, you're effectively acting as their 00:08:49.779 --> 00:08:52.759 bank. You're floating them credit. But agencies 00:08:52.759 --> 00:08:54.960 pay for ad spend all the time, right? They just 00:08:54.960 --> 00:08:57.340 mark it up. They do. But ad spend is usually 00:08:57.340 --> 00:09:00.519 capped and predictable. With AI, the volatility 00:09:00.519 --> 00:09:04.679 is insane. One bad loop, one recursive error 00:09:04.679 --> 00:09:07.159 where the AI talks to itself, and you could burn 00:09:07.159 --> 00:09:10.860 through $500 in an hour. Oof. Do you want that 00:09:10.860 --> 00:09:13.120 liability on your personal credit card? And more 00:09:13.120 --> 00:09:14.539 importantly, do you want to have the conversation 00:09:14.539 --> 00:09:16.419 where you try to bill the client for that mistake? 00:09:16.840 --> 00:09:19.019 No, I definitely do not want to have that conversation. 00:09:19.200 --> 00:09:21.659 Exactly. When the client uses their own key, 00:09:22.000 --> 00:09:24.399 they see the usage limits. They see the costs 00:09:24.399 --> 00:09:26.840 racking up in their own dashboard. It connects 00:09:26.840 --> 00:09:29.220 the action using the tool to the consequence 00:09:29.220 --> 00:09:31.740 paying for the compute. It stops you from being 00:09:31.740 --> 00:09:34.259 the middleman. So how do we solve the laziness 00:09:34.259 --> 00:09:36.980 problem? How do we make it smooth? Loom videos. 00:09:37.279 --> 00:09:39.679 It's such a simple tactical tip, but it works. 00:09:40.000 --> 00:09:42.480 Record your screen. Show them. Click here. Click 00:09:42.480 --> 00:09:44.700 billing. Click generate key. Copy this string. 00:09:45.080 --> 00:09:47.320 If you hold their hand through the scary dashboard, 00:09:47.539 --> 00:09:49.580 the friction just vanishes. And what if they 00:09:49.580 --> 00:09:51.480 just refuse? What if they say, here's the key 00:09:51.480 --> 00:09:53.659 in an email, just take it? If you absolutely 00:09:53.659 --> 00:09:56.440 must handle it, you treat it like a nuclear launch 00:09:56.440 --> 00:10:00.169 code. Never in plain text use a one -time encrypted 00:10:00.169 --> 00:10:02.610 sharing tool like one password or something similar 00:10:02.610 --> 00:10:05.389 You want that key to be revocable and you want 00:10:05.389 --> 00:10:08.309 zero trace of it in your email or slack history 00:10:08.309 --> 00:10:10.889 So handing over the billing isn't just about 00:10:10.889 --> 00:10:14.090 money. It's about trust Exactly. It removes the 00:10:14.090 --> 00:10:16.629 mystery and puts the client in control. Okay, 00:10:16.809 --> 00:10:18.870 so the client is paying the keys are locked up. 00:10:18.889 --> 00:10:22.070 The headers are validated But we're missing the 00:10:22.070 --> 00:10:24.940 big one The part where the robot actually does 00:10:24.940 --> 00:10:27.519 the work. The logic. The logic. And the guide 00:10:27.519 --> 00:10:29.240 has a term for testing this that I absolutely 00:10:29.240 --> 00:10:31.559 love. It calls it the black box method. This 00:10:31.559 --> 00:10:33.419 is my favorite part of the methodology, because 00:10:33.419 --> 00:10:36.000 it changes how you look at AI. Most people test 00:10:36.000 --> 00:10:38.399 by clicking the run button on a single node and 00:10:38.399 --> 00:10:40.940 seeing if it lights up green. Oh, the API connected 00:10:40.940 --> 00:10:43.379 green light. I'm done. Right. That is not testing. 00:10:43.580 --> 00:10:45.600 That is checking connectivity. Green doesn't 00:10:45.600 --> 00:10:47.620 mean correct. Green just means I didn't crash. 00:10:47.860 --> 00:10:51.299 Exactly. Testing is about what happens when messy, 00:10:51.559 --> 00:10:54.779 chaotic human reality hits your pristine logic. 00:10:55.259 --> 00:10:57.480 The guide says, don't wait until the end. Get 00:10:57.480 --> 00:11:01.529 real data immediately. Real emails. Real CRM 00:11:01.529 --> 00:11:04.629 records. Why is real so important here? Why can't 00:11:04.629 --> 00:11:07.730 I just type test email into the box? Because 00:11:07.730 --> 00:11:11.029 faked data is polite. Real data is rude. Real 00:11:11.029 --> 00:11:13.610 data has missing fields, weird formatting, emojis 00:11:13.610 --> 00:11:15.769 in the subject line, duplicates, spelling errors. 00:11:15.889 --> 00:11:18.129 You need to see how your system handles the rudeness. 00:11:18.330 --> 00:11:20.490 And this leads to the concept of safe failure. 00:11:20.950 --> 00:11:23.759 Safe failure is the mark of a pro. If the system 00:11:23.759 --> 00:11:25.580 breaks, and it will break, does it break loudly 00:11:25.580 --> 00:11:28.960 or quietly? A loud break is. The automation stops, 00:11:29.240 --> 00:11:31.879 the data is lost, and the client calls you screaming. 00:11:32.299 --> 00:11:35.279 A quiet safe break is. The system detects an 00:11:35.279 --> 00:11:38.220 error, logs it, sends an alert to the admin, 00:11:38.580 --> 00:11:41.320 and stops processing that specific item without 00:11:41.320 --> 00:11:43.779 crashing the whole pipeline. The guide suggests 00:11:43.779 --> 00:11:45.940 hiding the complexity during client testing. 00:11:46.460 --> 00:11:48.759 Don't give them the whole note editor. Give them 00:11:48.759 --> 00:11:52.970 a simple form. Yes. A chat box or a simple interface. 00:11:53.129 --> 00:11:56.149 Input goes in. Output comes out. You treat the 00:11:56.149 --> 00:11:58.870 AI as a black box. You don't care how it got 00:11:58.870 --> 00:12:01.149 the answer. You just check if the answer is right. 00:12:01.389 --> 00:12:03.429 And the question is key. The question you ask 00:12:03.429 --> 00:12:05.950 the client is crucial. Don't ask, what do you 00:12:05.950 --> 00:12:09.049 think? It's too vague. Ask, is this factually 00:12:09.049 --> 00:12:12.879 correct? Or is this tone on brand? There is something 00:12:12.879 --> 00:12:15.120 kind of magical about that black box approach 00:12:15.120 --> 00:12:17.279 when it works. It really is. I mean, imagine 00:12:17.279 --> 00:12:19.299 looking at a spreadsheet of a thousand inputs 00:12:19.299 --> 00:12:21.480 customer complaints, let's say, and then watching 00:12:21.480 --> 00:12:23.500 a thousand outputs populate right next to them. 00:12:23.779 --> 00:12:26.340 You're seeing the system thing it at scale. Whoa. 00:12:26.580 --> 00:12:28.460 It's like watching a brain work and fast forward. 00:12:28.539 --> 00:12:30.539 It's reliable. It's not a magic trick anymore. 00:12:30.559 --> 00:12:32.639 It's a machine. That's the goal. When you can 00:12:32.639 --> 00:12:34.860 watch a thousand rows process and trust the output, 00:12:34.960 --> 00:12:37.039 that's when you know you've built a system. So 00:12:37.039 --> 00:12:41.350 we've tested. It works. The data is real. Now 00:12:41.350 --> 00:12:44.850 comes the breakup. The handover. The exit strategy. 00:12:45.029 --> 00:12:47.190 The guide says the handover determines if you 00:12:47.190 --> 00:12:50.309 get repeat work. It's the final impression. But 00:12:50.309 --> 00:12:52.429 there's a darker side to this too. something 00:12:52.429 --> 00:12:55.889 engineers call the bus factor. Oh, the bus factor. 00:12:56.529 --> 00:12:59.590 It's a grim concept, but it's so necessary. If 00:12:59.590 --> 00:13:01.590 you, the builder, walk out of the office and 00:13:01.590 --> 00:13:04.529 get hit by a bus, can the client access their 00:13:04.529 --> 00:13:07.570 own system? If the answer is no, because the 00:13:07.570 --> 00:13:09.730 password is in your head or the workflow is on 00:13:09.730 --> 00:13:12.029 your private laptop, you haven't built a business 00:13:12.029 --> 00:13:14.789 asset. You've built a hostage situation. So how 00:13:14.789 --> 00:13:16.990 do we lower the bus factor? What are the technical 00:13:16.990 --> 00:13:19.929 steps? First, versioning. This is standard in 00:13:19.929 --> 00:13:21.889 software dev, but it's kind of rare in low code. 00:13:22.049 --> 00:13:25.029 You need two versions. Test in production. Never, 00:13:25.049 --> 00:13:27.509 ever experiment in production. Never. Client 00:13:27.509 --> 00:13:29.730 wants a change. You make it in test, you verify 00:13:29.730 --> 00:13:32.330 it, and then you push to prod. And backups. Crucial. 00:13:32.929 --> 00:13:35.029 Don't rely on the tool's cloud storage alone. 00:13:35.690 --> 00:13:38.889 Export the workflows. Save the JSON files to 00:13:38.889 --> 00:13:41.590 Google Drive or GitHub. If the platform goes 00:13:41.590 --> 00:13:43.870 down or the client accidentally deletes the workflow, 00:13:43.950 --> 00:13:46.529 which happens way more than you think, You need 00:13:46.529 --> 00:13:49.309 a hard copy you can restore from. The guide also 00:13:49.309 --> 00:13:52.009 mentions cleanup. Oh, please clean up your code. 00:13:52.409 --> 00:13:55.049 Label the steps. Remove those temporary nodes 00:13:55.049 --> 00:13:57.629 you used for debugging. If another developer 00:13:57.629 --> 00:13:59.909 opens your workflow in six months, they should 00:13:59.909 --> 00:14:02.289 be able to read it like a map. Not a bowl of 00:14:02.289 --> 00:14:04.070 spaghetti. If it looks like a bowl of spaghetti, 00:14:04.269 --> 00:14:07.129 you haven't finished the job. And finally, documentation. 00:14:07.669 --> 00:14:10.029 And I think we both agree here. Nobody reads 00:14:10.029 --> 00:14:13.149 the manual. Nobody reads a 50 -page PDF. But 00:14:13.149 --> 00:14:15.690 they will watch a five -minute video, a quick 00:14:15.690 --> 00:14:17.450 loom walking through the settings and the keys. 00:14:17.789 --> 00:14:20.990 That is gold. It's usable. There's a legal structure 00:14:20.990 --> 00:14:23.850 to this handover as well, defining done. This 00:14:23.850 --> 00:14:25.990 protects everyone. You need a definition of done, 00:14:26.330 --> 00:14:29.139 agreed upon criteria. When X, Y and Z are working, 00:14:29.379 --> 00:14:32.100 the final invoice is sent. And importantly, you 00:14:32.100 --> 00:14:34.159 have to separate the project from the maintenance. 00:14:34.440 --> 00:14:36.320 This seems to be where a lot of freelancers get 00:14:36.320 --> 00:14:38.779 stuck. They build the thing, but then they are 00:14:38.779 --> 00:14:41.600 on the hook forever for every little bug fix. 00:14:41.840 --> 00:14:44.080 Forever. And usually for free. You have to draw 00:14:44.080 --> 00:14:45.779 a line. The project is built. Now, if you want 00:14:45.779 --> 00:14:50.039 me to monitor it, fix bugs, handle updates, that's 00:14:50.039 --> 00:14:52.139 a maintenance retainer, new features, those are 00:14:52.139 --> 00:14:54.480 new projects. If you don't separate these, you 00:14:54.480 --> 00:14:56.879 will absolutely burn out. It sounds like the 00:14:56.879 --> 00:14:59.299 product isn't the AI, but the peace of mind. 00:15:00.080 --> 00:15:03.100 Yes. Clients pay for confidence, not just clever 00:15:03.100 --> 00:15:06.179 prompts. Let's zoom out. We've covered a lot 00:15:06.179 --> 00:15:08.980 of ground. The big idea seems to be this shift 00:15:08.980 --> 00:15:12.899 in mindset. Anyone can connect two tools and 00:15:12.899 --> 00:15:15.539 make AI do something cool for five minutes. That's 00:15:15.539 --> 00:15:18.860 a demo. But professional AI automation is about 00:15:18.860 --> 00:15:21.480 building systems that live inside a business 00:15:21.480 --> 00:15:23.799 without constant babysitting. That is the whole 00:15:23.799 --> 00:15:26.220 thesis. It's the life cycle. Host it correctly 00:15:26.220 --> 00:15:28.779 so ownership is clear. Secure the data so you 00:15:28.779 --> 00:15:30.899 can sleep at night. Make sure the client pays 00:15:30.899 --> 00:15:32.580 the bill so the relationship is transparent. 00:15:32.820 --> 00:15:36.759 Right. test it with real messy data, and hand 00:15:36.759 --> 00:15:39.460 it over clean so it can survive without you. 00:15:39.639 --> 00:15:42.340 It's moving from being a magician to being an 00:15:42.340 --> 00:15:45.279 engineer. Exactly. Magicians are fun for a night. 00:15:45.679 --> 00:15:48.039 Engineers keep the lights on. I want to leave 00:15:48.039 --> 00:15:49.860 our listeners, especially the learners out there, 00:15:49.899 --> 00:15:52.360 with a bit of a challenge. Look at the projects 00:15:52.360 --> 00:15:54.679 you are working on right now. Are you building 00:15:54.679 --> 00:15:57.620 prototypes or are you delivering systems? And 00:15:57.620 --> 00:15:59.539 here's a provocative thought to take with you. 00:15:59.759 --> 00:16:01.960 If you walked away from your automation today, 00:16:02.240 --> 00:16:05.879 just Ghosted. Totally disappeared. Would it survive 00:16:05.879 --> 00:16:08.639 next week without you? If the answer is no, you 00:16:08.639 --> 00:16:10.700 haven't finished the job. A sobering thought 00:16:10.700 --> 00:16:13.220 to end on. Thank you for joining us on this deep 00:16:13.220 --> 00:16:15.399 dive into the mechanics of delivery. We'll catch 00:16:15.399 --> 00:16:16.559 you on the next one. See you then.