WEBVTT 00:00:00.000 --> 00:00:04.019 So if you're using AI just to write slightly 00:00:04.019 --> 00:00:06.259 better emails, or maybe summarize some text, 00:00:06.639 --> 00:00:09.419 you might be missing out on, well, a lot. That 00:00:09.419 --> 00:00:12.240 kind of basic stuff is fine, sure, but it really... 00:00:12.009 --> 00:00:14.070 barely scratches the surface of what's possible 00:00:14.070 --> 00:00:17.390 when you get into more complex work. That's really 00:00:17.390 --> 00:00:20.250 true. You know, when you look at any real project 00:00:20.250 --> 00:00:22.870 like building software, planning a marketing 00:00:22.870 --> 00:00:25.690 campaign, even something physical like a carpenter 00:00:25.690 --> 00:00:28.109 building a table, it all follows this fundamental 00:00:28.109 --> 00:00:31.489 pattern. Input, then processing, then output 00:00:31.489 --> 00:00:34.969 IPO. And our goal really is just to maximize 00:00:34.969 --> 00:00:37.549 the value of that final output. Systematically 00:00:37.549 --> 00:00:40.909 using AI helps us do that. Welcome to the Deep 00:00:40.909 --> 00:00:43.320 Dive. Today our mission is to help you turn AI 00:00:43.320 --> 00:00:46.200 from maybe a random chat partner into a real 00:00:46.200 --> 00:00:48.909 systematic partner for design for strategy. We're 00:00:48.909 --> 00:00:50.429 pulling the best ideas from the source material 00:00:50.429 --> 00:00:52.729 we looked at on how to build these powerful workflows. 00:00:53.189 --> 00:00:55.130 Yeah, we've got a pretty clear path laid out. 00:00:55.229 --> 00:00:58.049 First, we're going to properly define that IPO 00:00:58.049 --> 00:01:00.689 structure, input, processing, output, what it 00:01:00.689 --> 00:01:02.929 means. Second, we'll talk about the essential 00:01:02.929 --> 00:01:05.430 tools you kind of need and some setup tips you 00:01:05.430 --> 00:01:08.250 can't skip, especially around AI memory. That's 00:01:08.250 --> 00:01:10.870 key. And finally, we'll walk through two really 00:01:10.870 --> 00:01:13.730 specific examples using AI and product design. 00:01:13.920 --> 00:01:16.019 and then for creating social media content at 00:01:16.019 --> 00:01:18.840 scale. Okay, let's unpack this IPO model first 00:01:18.840 --> 00:01:23.180 then. The simple brilliance here is that AI can 00:01:23.180 --> 00:01:26.560 actually help at all three stages. It's not just 00:01:26.560 --> 00:01:28.560 something you plug in at the end to maybe clean 00:01:28.560 --> 00:01:30.980 up your writing. It can speed things up everywhere. 00:01:31.260 --> 00:01:34.260 Exactly. So start with input. This is where AI 00:01:34.260 --> 00:01:36.599 tackles, well, the messiness of real life. Tools 00:01:36.599 --> 00:01:38.879 like VoicePal or Grain were mentioned. They're 00:01:38.879 --> 00:01:40.799 important because they take messy, stuff -spoken 00:01:40.799 --> 00:01:42.739 ideas, rambling voice notes, maybe an hour -long 00:01:42.739 --> 00:01:45.540 meeting, and turn it into organized text the 00:01:45.540 --> 00:01:47.620 AI can actually understand and work with. They 00:01:47.620 --> 00:01:49.519 handle that transcription and maybe some initial 00:01:49.519 --> 00:01:52.219 source. Right. So that cleaned up data then moves 00:01:52.219 --> 00:01:54.359 into processing. This is like the central hub, 00:01:55.000 --> 00:01:57.340 the brain of the operation. You need a strong 00:01:57.340 --> 00:02:00.239 foundation here. That's why something like Claude 00:02:00.239 --> 00:02:03.120 was recommended. What's important about Claude 00:02:03.120 --> 00:02:06.500 or similar advanced models is that big context 00:02:06.500 --> 00:02:08.860 window. Yeah. And that context window is absolutely 00:02:08.860 --> 00:02:11.259 essential. It holds the memory for a complex 00:02:11.259 --> 00:02:14.509 project. It does the heavy lifting. organizing, 00:02:14.650 --> 00:02:16.990 checking data fast, planning things out, even 00:02:16.990 --> 00:02:20.289 drafting, all while keeping the project's history 00:02:20.289 --> 00:02:22.409 sort of in mind. And then get the output. This 00:02:22.409 --> 00:02:24.330 is where the finished thing comes out. Polished. 00:02:24.370 --> 00:02:27.789 We saw some neat specialized examples. Firecut 00:02:27.789 --> 00:02:30.550 for video edits or Gamma turning text drafts 00:02:30.550 --> 00:02:32.810 into presentation slides pretty quickly. So the 00:02:32.810 --> 00:02:34.639 toolkit kind of mirrors the structure. Right. 00:02:34.939 --> 00:02:36.939 Input tools for cleaning up, central processing 00:02:36.939 --> 00:02:39.300 brain, and then output tools for getting it done 00:02:39.300 --> 00:02:41.680 fast. OK. So thinking about that central processing 00:02:41.680 --> 00:02:43.340 hub, the one holding all the project memory. 00:02:43.360 --> 00:02:46.180 Yeah. What would you say is the absolute most 00:02:46.180 --> 00:02:48.560 crucial first step before you even start a complex 00:02:48.560 --> 00:02:52.340 project? Ah. You have to teach the AI who it 00:02:52.340 --> 00:02:55.080 is and what the mission is. Precisely, yeah. 00:02:55.479 --> 00:02:57.580 If you want the AI to act like a professional 00:02:57.580 --> 00:02:59.659 partner, you absolutely have to personalize it. 00:02:59.979 --> 00:03:02.379 The most critical setup step is making sure that 00:03:02.379 --> 00:03:05.460 memory is solid. This really means avoiding having 00:03:05.460 --> 00:03:08.340 to explain your job or the company voice or the 00:03:08.340 --> 00:03:10.439 whole project background every single time you 00:03:10.439 --> 00:03:13.460 log in. That step is so vital because otherwise 00:03:13.460 --> 00:03:17.120 the AI just defaults to generic textbook answers. 00:03:17.379 --> 00:03:19.639 Right, right. And speaking of context, another 00:03:19.639 --> 00:03:22.759 key setup piece is file uploads. You need to 00:03:22.759 --> 00:03:26.300 feed it your own data, PDFs, spreadsheets, style 00:03:26.300 --> 00:03:29.139 guides, so it can check things against your actual 00:03:29.139 --> 00:03:32.310 real -world situation. your constraints. Mm -hmm. 00:03:32.569 --> 00:03:34.569 You're basically setting up a clear data structure, 00:03:34.710 --> 00:03:37.310 not just randomly dumping files in. You're telling 00:03:37.310 --> 00:03:40.810 the AI, this is the hierarchy of knowledge for 00:03:40.810 --> 00:03:43.770 this specific task. You know, I still... Well, 00:03:43.969 --> 00:03:45.770 actually, I still wrestle with prompt drift myself 00:03:45.770 --> 00:03:47.930 sometimes. I remember trying to map out a new 00:03:47.930 --> 00:03:50.590 hiring plan in a chat where I'd just been drafting 00:03:50.590 --> 00:03:53.150 like... casual messages to friends. And suddenly 00:03:53.150 --> 00:03:55.150 the AI started talking about aligning core values, 00:03:55.189 --> 00:03:57.409 but using like really aggressive slang. Total 00:03:57.409 --> 00:03:59.069 meat of the tongue. Got completely confused. 00:03:59.389 --> 00:04:01.509 Oh, yeah, that's a classic problem. And that's 00:04:01.509 --> 00:04:03.949 exactly why the source material really emphasized 00:04:03.949 --> 00:04:06.310 the project's feature tip. You've got to create 00:04:06.310 --> 00:04:09.430 separate, dedicated spaces, like one called Q4 00:04:09.430 --> 00:04:12.050 marketing strategy, another personal blog ideas. 00:04:12.550 --> 00:04:14.849 It keeps the information quarantined, stops the 00:04:14.849 --> 00:04:16.509 AI from getting rixed up, keeps things focused. 00:04:16.610 --> 00:04:18.790 So when we use these dedicated projects, these 00:04:18.790 --> 00:04:22.000 separate spaces. Yeah. What's the key confusion 00:04:22.000 --> 00:04:24.519 we're stopping the AI from making? It stops the 00:04:24.519 --> 00:04:26.720 AI from blending your different job contexts 00:04:26.720 --> 00:04:29.259 together, keeps work separate. Okay, this is 00:04:29.259 --> 00:04:30.800 where it starts to get really practical. Let's 00:04:30.800 --> 00:04:34.259 test this theory. Workflow one, using AI as a 00:04:34.259 --> 00:04:36.540 product manager. Let's imagine we're designing 00:04:36.540 --> 00:04:38.759 a simple budget app for students. We'll call 00:04:38.759 --> 00:04:41.839 it student money. Right. So step one needs really 00:04:41.839 --> 00:04:44.180 clear instructions in the prompt. Define the 00:04:44.180 --> 00:04:47.259 user, the persona. You tell the AI to act like 00:04:47.259 --> 00:04:50.149 an expert user researcher. And you ask for specifics, 00:04:50.509 --> 00:04:53.470 deep specifics, target audience. Vietnamese university 00:04:53.470 --> 00:04:56.430 students may be 18 to 22 years old. And crucially, 00:04:56.750 --> 00:04:59.589 you define their main pain point. Maybe they 00:04:59.589 --> 00:05:02.009 feel lazy and just avoid writing down every single 00:05:02.009 --> 00:05:04.810 expense. That emotional detail matters. Yeah, 00:05:04.810 --> 00:05:07.290 if you don't give it that level of emotional 00:05:07.290 --> 00:05:09.730 context about the user, the AI will probably 00:05:09.730 --> 00:05:11.790 just give you a generic feature list, not a real 00:05:11.790 --> 00:05:14.769 solution. Exactly. Then step two is designing 00:05:14.769 --> 00:05:18.209 the user flow. Based on that lazy persona, let's 00:05:18.209 --> 00:05:21.610 call him Anne, you prompt the AI to design the 00:05:21.610 --> 00:05:24.670 absolute easiest flow possible. Goal, add an 00:05:24.670 --> 00:05:28.069 expense in under five seconds. So maybe the AI 00:05:28.069 --> 00:05:30.750 guesses the category. If Anne types fuk long, 00:05:31.290 --> 00:05:33.769 does it just assume coffee? OK, but hang on. 00:05:33.790 --> 00:05:36.250 If we make it that simple, just relying on guesswork, 00:05:37.170 --> 00:05:39.689 don't we lose accuracy? How do we make sure Anne 00:05:39.689 --> 00:05:41.649 isn't just tagging everything as coffee and maybe 00:05:41.649 --> 00:05:45.389 missing big things, like rent or a tuition payment? 00:05:45.550 --> 00:05:47.870 That's the exact question you posed to the AI. 00:05:48.110 --> 00:05:50.129 You ask it to analyze that specific trade -off. 00:05:50.449 --> 00:05:52.269 You set the hard constraint under five seconds, 00:05:52.329 --> 00:05:54.850 and the AI has to suggest the compromise, how 00:05:54.850 --> 00:05:56.670 to balance speed and accuracy. See, it shifts 00:05:56.670 --> 00:05:58.709 your job from designing every little screen yourself 00:05:58.709 --> 00:06:01.829 to basically challenging the AI's design assumptions. 00:06:02.269 --> 00:06:04.709 Ah, OK. That makes a lot of sense. So we've defined 00:06:04.709 --> 00:06:06.790 the user, the flow, debated the trade -offs. 00:06:07.370 --> 00:06:10.189 Mid -roll sponsor, read, placeholder. Welcome 00:06:10.189 --> 00:06:12.089 back to the Deep Dive. We've got our user Anne 00:06:12.089 --> 00:06:14.470 and the quick expense adding flow defined for 00:06:14.470 --> 00:06:18.269 our student money app. What's step three? Ah, 00:06:18.550 --> 00:06:22.000 right. Ask for a simple model. A prototype. So 00:06:22.000 --> 00:06:24.459 we instruct the AI to create a clickable prototype, 00:06:24.680 --> 00:06:26.779 but we specify exactly what screens we need, 00:06:27.079 --> 00:06:29.459 and the key elements, like home screen needs 00:06:29.459 --> 00:06:32.939 a huge O plus button, input screen, only show 00:06:32.939 --> 00:06:34.899 the number keyboard, keep it minimal, and then 00:06:34.899 --> 00:06:37.319 maybe an automatic done screen after input. Yeah, 00:06:37.379 --> 00:06:39.500 this is where the AI does that translation, takes 00:06:39.500 --> 00:06:41.540 those abstract rules you set and turns them into 00:06:41.540 --> 00:06:43.920 something tangible, something visual you can 00:06:43.920 --> 00:06:46.699 react to. You're not painstakingly drawing boxes 00:06:46.699 --> 00:06:48.500 and arrows, you're giving the system the rule 00:06:48.500 --> 00:06:50.790 to design for you. And then we hit step four, 00:06:51.110 --> 00:06:53.810 test and fix. iteration, we challenge the AI's 00:06:53.810 --> 00:06:56.290 prototype, throw an edge case at it, like, how 00:06:56.290 --> 00:06:58.629 does this super simple app handle a loan Anne 00:06:58.629 --> 00:07:01.509 gave to a friend? We make the AI analyze the 00:07:01.509 --> 00:07:03.649 trade -off again. Do we add complexity, like 00:07:03.649 --> 00:07:05.730 a loan section, or does that break the core simplicity 00:07:05.730 --> 00:07:08.069 needed for Anne? And the big lesson here, really, 00:07:08.069 --> 00:07:10.050 is you don't need to be a coder. You don't even 00:07:10.050 --> 00:07:12.350 need to be great at wireframing tools. You just 00:07:12.350 --> 00:07:14.389 need to be incredibly clear about describing 00:07:14.389 --> 00:07:16.329 the requirement, the problem you're trying to 00:07:16.329 --> 00:07:19.649 solve for the user. The AI handles the translation 00:07:19.649 --> 00:07:22.660 into a testable model. So what core function 00:07:22.660 --> 00:07:25.180 is the AI really performing in this whole design 00:07:25.180 --> 00:07:28.160 loop then? It quickly translates abstract ideas 00:07:28.160 --> 00:07:31.259 into a tangible model for review. Okay, let's 00:07:31.259 --> 00:07:35.139 switch gears. Workflow 2. AI as a content strategist. 00:07:35.579 --> 00:07:38.180 Thinking about fast platforms like TikTok or 00:07:38.180 --> 00:07:41.540 Reels, often the hardest bit isn't making the 00:07:41.540 --> 00:07:43.959 video. It's coming up with ideas that actually 00:07:43.959 --> 00:07:46.800 connect, that resonate. Step one here is about 00:07:46.800 --> 00:07:48.740 getting emotionally effective hook ideas. We 00:07:48.740 --> 00:07:51.100 don't just want generic topics. The prompt needs 00:07:51.100 --> 00:07:53.660 to tell the AI acting now as a social media marketing 00:07:53.660 --> 00:07:56.779 strategist to generate maybe 20 ideas and then 00:07:56.779 --> 00:07:58.600 crucially sort them by psychological triggers 00:07:58.600 --> 00:08:01.019 that would work for our user and things like 00:08:01.019 --> 00:08:04.259 pain point, curiosity, quick win. I really like 00:08:04.259 --> 00:08:07.259 that focus. It forces the ideas to be relevant 00:08:07.259 --> 00:08:10.259 right from the start. Something like why you're 00:08:10.259 --> 00:08:12.860 always broke by the 20th and how to fix it in 00:08:12.860 --> 00:08:15.699 30 seconds. hits a pain point, offers a quick 00:08:15.699 --> 00:08:18.720 win. Exactly. Then comes step two, which is critical. 00:08:19.220 --> 00:08:21.660 The human filter. You have to apply your feeling. 00:08:22.060 --> 00:08:24.439 The AI gives you 20 psychologically sorted ideas, 00:08:24.720 --> 00:08:26.459 yeah. But you only pick the best, maybe three 00:08:26.459 --> 00:08:28.579 to five that genuinely resonate with you, that 00:08:28.579 --> 00:08:30.399 feel right for your voice. You're the curator 00:08:30.399 --> 00:08:33.779 here. Wow. Just imagine scaling that. Testing 00:08:33.779 --> 00:08:36.460 thousands of those kinds of psychologically angled 00:08:36.460 --> 00:08:39.740 concepts in like minutes, that speed for strategic 00:08:39.740 --> 00:08:41.480 planning, it's kind of mind blowing compared 00:08:41.480 --> 00:08:43.320 to traditional brainstorming. It really is a 00:08:43.320 --> 00:08:45.580 total acceleration of that early creative phase. 00:08:46.379 --> 00:08:48.779 Okay, so now step three, creating a detailed 00:08:48.779 --> 00:08:51.279 script. Right. You take one of your chosen hooks, 00:08:51.639 --> 00:08:53.360 let's use this while you're always broke by the 00:08:53.360 --> 00:08:56.039 20th, and you prompt the AI for a full, say, 00:08:56.159 --> 00:08:59.019 30 second script. The key here is using a structured 00:08:59.019 --> 00:09:01.659 format, like a table, time column, words column, 00:09:01.879 --> 00:09:04.700 visual column, and you need specific visual directions 00:09:04.700 --> 00:09:07.379 in there, like zero three seconds, close up shot 00:09:07.379 --> 00:09:10.639 of your face looking serious, or 10, 15 seconds, 00:09:11.019 --> 00:09:13.580 fast cuts showing three small, impulsive buys 00:09:13.580 --> 00:09:16.480 coffee snacks. Which leads perfectly into step 00:09:16.480 --> 00:09:18.600 four, and this is maybe the most important point, 00:09:18.960 --> 00:09:21.690 adaptation. Do not just copy and paste that script. 00:09:21.809 --> 00:09:23.990 Please don't. Read it. Understand the structure, 00:09:24.110 --> 00:09:26.049 the flow, but then you absolutely have to adapt 00:09:26.049 --> 00:09:28.590 it. Use your own voice, your tone, your humor. 00:09:29.009 --> 00:09:31.490 The final video, the final content, it must feel 00:09:31.490 --> 00:09:33.970 like it came from you, a real person. Beyond 00:09:33.970 --> 00:09:36.309 just making it relatable, why is keeping your 00:09:36.309 --> 00:09:39.070 own specific voice so crucial in that final step? 00:09:39.210 --> 00:09:41.649 What risk are we avoiding? Authenticity ensures 00:09:41.649 --> 00:09:45.110 the content feels human. Avoid sounding generic 00:09:45.110 --> 00:09:48.190 or predictable like an algorithm wrote it. So 00:09:48.190 --> 00:09:51.070 just to quickly recap the big ideas for you listening, 00:09:51.750 --> 00:09:54.309 AI really shines when you integrate it properly 00:09:54.309 --> 00:09:57.350 into that input processing output system. That 00:09:57.350 --> 00:10:00.090 random one -off use, it needs to evolve. Yeah, 00:10:00.129 --> 00:10:03.169 and critically, AI shifts from just being a tool 00:10:03.169 --> 00:10:06.110 to being a real partner, like a product manager 00:10:06.110 --> 00:10:08.470 or a content strategist, but only when you give 00:10:08.470 --> 00:10:11.149 it super detailed role -specific instructions. 00:10:11.210 --> 00:10:12.929 You got to treat it like a highly skilled con... 00:10:12.750 --> 00:10:15.350 and Sultan you've hired, be specific. And these 00:10:15.350 --> 00:10:16.789 workflows we talked about, they're really just 00:10:16.789 --> 00:10:18.529 the beginning. The source material hinted at 00:10:18.529 --> 00:10:21.090 more advanced stuff. Things like using AI to 00:10:21.090 --> 00:10:22.970 understand feedback from thousands of customers 00:10:22.970 --> 00:10:25.570 automatically, or even building out a whole online 00:10:25.570 --> 00:10:28.090 course structure from just a basic outline. Pretty 00:10:28.090 --> 00:10:30.409 powerful extensions. So here's a final thought 00:10:30.409 --> 00:10:33.169 to chew on. If you get really good at defining 00:10:33.169 --> 00:10:36.110 the problem perfectly, and defining the user 00:10:36.110 --> 00:10:38.850 persona with deep clarity, and setting those 00:10:38.850 --> 00:10:41.090 strict rules for the AI to execute the plan, 00:10:41.850 --> 00:10:45.230 Are you still the creator in the old sense? Or 00:10:45.230 --> 00:10:47.370 have you become something else, maybe a highly 00:10:47.370 --> 00:10:50.830 effective curator of intelligence? Definitely 00:10:50.830 --> 00:10:52.669 something to think about. We really encourage 00:10:52.669 --> 00:10:55.509 you to stop using AI just randomly. Start thinking 00:10:55.509 --> 00:10:58.149 about and maybe building your own IPO systems 00:10:58.149 --> 00:11:00.429 today. Thanks so much for joining this deep dive. 00:11:00.730 --> 00:11:01.289 Until next time.