WEBVTT 00:00:00.000 --> 00:00:04.200 Intro music. Welcome to the deep dive. Let's 00:00:04.200 --> 00:00:07.400 jump right in today. Picture this. It is the 00:00:07.400 --> 00:00:12.439 summer of 2026. You are totally overwhelmed at 00:00:12.439 --> 00:00:16.179 work. You ask your brand new agentic AI assistant 00:00:16.179 --> 00:00:19.100 to book a family vacation. You give it the exact 00:00:19.100 --> 00:00:20.660 dates. You give it your credit card. You tell 00:00:20.660 --> 00:00:23.420 it to handle everything. But it makes a critical, 00:00:23.460 --> 00:00:27.260 quiet mistake. It books non -refundable flights 00:00:27.260 --> 00:00:31.260 for November instead of July. $4 ,000 just gone 00:00:31.260 --> 00:00:33.640 in an instant. And fixing that exact catastrophic 00:00:33.640 --> 00:00:36.520 failure is exactly why AI product managers are 00:00:36.520 --> 00:00:38.719 the most sought after hires right now. Companies 00:00:38.719 --> 00:00:40.840 are absolutely desperate for people who can bridge 00:00:40.840 --> 00:00:43.439 this gap. They need people to translate complex 00:00:43.439 --> 00:00:46.640 business needs into raw AI capabilities. Exactly. 00:00:46.799 --> 00:00:48.579 And that is what we are mapping out today. We 00:00:48.579 --> 00:00:51.500 are exploring a very specific, highly detailed 00:00:51.500 --> 00:00:53.679 roadmap. We are looking at how you can get hired 00:00:53.679 --> 00:00:56.399 as an AI product manager. In just four to five 00:00:56.399 --> 00:00:58.479 months. But we are skipping the usual traps. 00:00:58.659 --> 00:01:01.079 No more passively watching tutorial videos for 00:01:01.079 --> 00:01:03.539 hours on end. Right. And no more collecting meaningless 00:01:03.539 --> 00:01:06.989 online certificates. We are focusing purely on 00:01:06.989 --> 00:01:09.010 getting hands -on. You have to actually build 00:01:09.010 --> 00:01:11.609 things to understand this space. Here is how 00:01:11.609 --> 00:01:14.530 we are going to break it down. First, we will 00:01:14.530 --> 00:01:16.849 cover the foundational product manager skills 00:01:16.849 --> 00:01:19.829 you absolutely need. The boring but essential 00:01:19.829 --> 00:01:22.650 stuff. Then we will map out the technical vocabulary 00:01:22.650 --> 00:01:26.109 you must know to survive in 2026. After that, 00:01:26.569 --> 00:01:29.049 we dive into building real product intuition 00:01:29.049 --> 00:01:32.209 and launching actual projects. And finally, we 00:01:32.209 --> 00:01:34.209 will discuss how to make your application stand 00:01:34.209 --> 00:01:37.129 out in a massive pile of 500 resumes. We really 00:01:37.129 --> 00:01:38.870 have to set the stage first though. Yeah, the 00:01:38.870 --> 00:01:41.590 landscape has changed drastically since 2024. 00:01:42.090 --> 00:01:45.799 Think about the early days. AI mostly just summarized 00:01:45.799 --> 00:01:48.379 long text documents. Or it wrote simple, polite 00:01:48.379 --> 00:01:51.680 emails. If it was a novelty. But now we are dealing 00:01:51.680 --> 00:01:54.840 with agentic AI. The paradigm has completely 00:01:54.840 --> 00:01:57.439 shifted. Agentic AI does not just type words 00:01:57.439 --> 00:02:00.560 on a screen. It takes real actions on behalf 00:02:00.560 --> 00:02:03.120 of the user. It navigates messy websites. It 00:02:03.120 --> 00:02:05.680 books those flights. It manages complex calendars 00:02:05.680 --> 00:02:08.750 and spends real money. The stakes are exponentially 00:02:08.750 --> 00:02:11.550 higher now. As a product manager, you have to 00:02:11.550 --> 00:02:13.669 design meticulously for these failure cases. 00:02:13.930 --> 00:02:16.729 Because if an AI makes a massive mistake like 00:02:16.729 --> 00:02:20.310 booking, the wrong flight trust is completely 00:02:20.310 --> 00:02:23.310 broken. And building, or rebuilding, that trust 00:02:23.310 --> 00:02:26.270 is your primary job. Which brings us to a mandatory, 00:02:26.370 --> 00:02:28.909 and maybe surprising, first step on the roadmap. 00:02:29.490 --> 00:02:31.949 What is that? You must spend your first six to 00:02:31.949 --> 00:02:34.969 eight weeks learning traditional, old school 00:02:34.969 --> 00:02:37.129 product management. Because the most advanced 00:02:37.129 --> 00:02:39.590 AI in the world will not save a fundamentally 00:02:39.590 --> 00:02:43.210 bad product. Exactly. It just won't. I have to 00:02:43.210 --> 00:02:45.150 push back here, though. OK, go ahead. It feels 00:02:45.150 --> 00:02:47.650 a bit counterintuitive. In a tech world moving 00:02:47.650 --> 00:02:50.009 at absolute light speed, you're telling people 00:02:50.009 --> 00:02:52.610 to slow down. I am. You want them to sit and 00:02:52.610 --> 00:02:54.810 write a traditional product requirements document. 00:02:54.949 --> 00:02:58.169 I do. Completely. The product requirements document, 00:02:58.409 --> 00:03:01.870 or PRD, is your anchor in the chaos. It forces 00:03:01.870 --> 00:03:04.490 you to explain exactly what you are building. 00:03:04.590 --> 00:03:06.590 And more importantly, why you are building it. 00:03:06.889 --> 00:03:08.969 You have to start with the actual problem. What 00:03:08.969 --> 00:03:11.650 is genuinely hurting the user right now? Right. 00:03:11.689 --> 00:03:15.569 You cannot just say, let's use AI. You have to 00:03:15.569 --> 00:03:18.069 ask what workflow is broken. Then you define 00:03:18.069 --> 00:03:21.490 the goal. How will we quantifiably know we actually 00:03:21.490 --> 00:03:24.090 fixed the problem? You write clear user stories 00:03:24.090 --> 00:03:26.870 to map the experience. Like as a busy manager? 00:03:27.159 --> 00:03:30.139 I want the AI to summarize my weekly meetings. 00:03:30.419 --> 00:03:33.280 So I can save three hours a week. Exactly. And 00:03:33.280 --> 00:03:36.340 then you must define strict success metrics. 00:03:36.539 --> 00:03:39.939 Are we measuring the speed of the task? The accuracy 00:03:39.939 --> 00:03:42.759 of the summary? Or simply overall user happiness? 00:03:43.020 --> 00:03:45.400 You also need a firm grasp on standard product 00:03:45.400 --> 00:03:48.599 metrics. Things like retention, conversion, and 00:03:48.599 --> 00:03:51.259 engagement. Do people actually come back to use 00:03:51.259 --> 00:03:54.099 your AI tool on day two? Do they pull out their 00:03:54.099 --> 00:03:56.280 credit card to actually pay for it? If you are 00:03:56.280 --> 00:03:57.919 starting from scratch, there are a couple of 00:03:57.919 --> 00:04:00.039 gold standard resources. You definitely read 00:04:00.039 --> 00:04:02.159 the book inspired by Marty Kagan. And you should 00:04:02.159 --> 00:04:04.560 follow Lenny's newsletter. Those are the absolute 00:04:04.560 --> 00:04:06.580 best for learning how to think like a classic 00:04:06.580 --> 00:04:09.479 PM. You also need to run what they call discovery 00:04:09.479 --> 00:04:11.180 sprints. This is where you actually draw out 00:04:11.180 --> 00:04:13.620 the product interface and play detective. You 00:04:13.620 --> 00:04:16.180 sit with real users and watch how they try to 00:04:16.180 --> 00:04:18.720 solve the problem today. Without your shiny new 00:04:18.720 --> 00:04:22.709 tool. So how do you actually balance moving fast 00:04:22.709 --> 00:04:25.689 with these traditional PM fundamentals. You move 00:04:25.689 --> 00:04:28.790 fast by knowing exactly what not to build. Those 00:04:28.790 --> 00:04:31.970 fundamentals act as your filter. Master standard 00:04:31.970 --> 00:04:34.610 product management before adding AI to the mix. 00:04:34.829 --> 00:04:37.149 Exactly. They prevent you from wasting months 00:04:37.149 --> 00:04:39.470 of engineering time on flashy features nobody 00:04:39.470 --> 00:04:41.790 wants. Right. And once your traditional foundation 00:04:41.790 --> 00:04:44.750 is incredibly solid, you need to learn a completely 00:04:44.750 --> 00:04:46.910 new vocabulary. You have to be able to talk to 00:04:46.910 --> 00:04:48.769 the engineers who are building the actual engine. 00:04:48.870 --> 00:04:51.639 To be clear, You do not need to know how to code. 00:04:51.899 --> 00:04:55.199 No. But you must thoroughly understand the system 00:04:55.199 --> 00:04:57.759 architecture. Let's define the core jargon you 00:04:57.759 --> 00:05:01.100 need. Keep these simple. First up, what exactly 00:05:01.100 --> 00:05:04.160 is a model? The core reasoning engine that powers 00:05:04.160 --> 00:05:06.579 your product's artificial intelligence. And training. 00:05:07.000 --> 00:05:09.579 Feeding massive datasets to the engine so it 00:05:09.579 --> 00:05:12.420 recognizes did patterns. Inference. The live 00:05:12.420 --> 00:05:14.920 moment when the AI generates an answer for users. 00:05:15.100 --> 00:05:18.139 And fine -tuning. Tweaking a pre -trained model. 00:05:18.350 --> 00:05:21.410 for highly specialized narrow business tasks. 00:05:21.730 --> 00:05:24.750 Two sec silence. Whoa! Think about the concept 00:05:24.750 --> 00:05:26.829 of vector databases for a second. Yeah, they 00:05:26.829 --> 00:05:29.550 are fascinating. Imagine storing complex human 00:05:29.550 --> 00:05:32.769 ideas purely as numbers for lightning fast searches. 00:05:33.269 --> 00:05:36.089 It is wild. That is exactly what a vector database 00:05:36.089 --> 00:05:39.449 does. It maps abstract concepts into a spatial 00:05:39.449 --> 00:05:41.790 coordinate system. So the AI does not just look 00:05:41.790 --> 00:05:45.199 for exact keyword matches? No. It looks for numbers 00:05:45.199 --> 00:05:47.339 that are physically close together in that space 00:05:47.339 --> 00:05:50.379 to find related ideas. Which naturally brings 00:05:50.379 --> 00:05:54.579 us to a massive industry breakthrough. RAG. Retrieval 00:05:54.579 --> 00:05:56.939 Augmented Generation. This seems to be the primary 00:05:56.939 --> 00:05:59.779 way we fix the AI memory problem right now. Yes. 00:06:00.220 --> 00:06:01.920 If you want a simple way to think about RAG, 00:06:02.259 --> 00:06:04.779 it is basically an open book exam for the AI. 00:06:05.050 --> 00:06:07.250 So instead of the AI just guessing an answer 00:06:07.250 --> 00:06:09.889 from whatever it vaguely memorized during training, 00:06:10.310 --> 00:06:12.290 it literally reads a specific book you hand it. 00:06:12.410 --> 00:06:14.589 You connect the AI to your company's private 00:06:14.589 --> 00:06:17.850 database. When a user asks a question, the system 00:06:17.850 --> 00:06:20.089 retrieves the relevant private documents first. 00:06:20.430 --> 00:06:22.610 The AI reads those specific documents in real 00:06:22.610 --> 00:06:25.569 time. And then it generates a highly accurate 00:06:25.569 --> 00:06:28.699 answer based only on that text. This is brilliant 00:06:28.699 --> 00:06:31.639 because it mechanically stops the AI from just 00:06:31.639 --> 00:06:34.339 making things up. It grounds the math in reality 00:06:34.339 --> 00:06:37.259 and drastically cuts down on hallucinations. 00:06:37.500 --> 00:06:40.199 Right, and crucially for a PM, it is so much 00:06:40.199 --> 00:06:42.839 cheaper. You do not have to spend millions of 00:06:42.839 --> 00:06:44.879 dollars training a whole new model from scratch. 00:06:45.120 --> 00:06:47.259 Just to teach you your company's HR policies, 00:06:47.459 --> 00:06:50.800 you just use ARGI. But even with RIGUE, we inevitably 00:06:50.800 --> 00:06:54.000 hit a massive wall in product design. Latency. 00:06:54.300 --> 00:06:57.779 Ah, latency. The waiting time. AI takes significant 00:06:57.779 --> 00:07:00.180 computational time to think and generate its 00:07:00.180 --> 00:07:02.180 answers. And this is where the PM owns their 00:07:02.180 --> 00:07:04.379 salary. You constantly have to choose between 00:07:04.379 --> 00:07:07.019 two distinct paths. Path A is using a massive, 00:07:07.240 --> 00:07:10.300 perfect, super -smart model. But it takes 10 00:07:10.300 --> 00:07:13.139 painfully long seconds to reply. Path B is using 00:07:13.139 --> 00:07:16.100 a much smaller, lighter model. It gives answers 00:07:16.100 --> 00:07:19.069 that are just good enough. but it does it in 00:07:19.069 --> 00:07:20.990 exactly two seconds. I always think of this like 00:07:20.990 --> 00:07:22.930 running a restaurant. That is a great analogy. 00:07:23.189 --> 00:07:26.069 Are we trying to serve a perfect flawless Michelin 00:07:26.069 --> 00:07:28.149 star meal that makes the customer wait two hours? 00:07:28.410 --> 00:07:31.310 Or are we serving a really good solid burger 00:07:31.310 --> 00:07:33.610 in five minutes? In software, the five -minute 00:07:33.610 --> 00:07:36.670 burger almost always wins out. Users desperately 00:07:36.670 --> 00:07:39.569 want digital products to feel snappy and responsive. 00:07:39.750 --> 00:07:42.170 Why does speed almost always win over perfect 00:07:42.170 --> 00:07:46.389 accuracy for users? Because modern human attention 00:07:46.389 --> 00:07:49.069 spans are incredibly short. If they stare at 00:07:49.069 --> 00:07:52.009 a loading spinner for 20 seconds, they just leave. 00:07:52.350 --> 00:07:55.290 Fast and good enough always beats a slow, perfect 00:07:55.290 --> 00:07:58.750 AI answer. Every single time. We will be right 00:07:58.750 --> 00:08:01.670 back after a quick word from our sponsors. Sponsor 00:08:01.670 --> 00:08:04.910 read. All right. We are back. So we understand 00:08:04.910 --> 00:08:07.670 the core technical language now. And we understand 00:08:07.670 --> 00:08:10.470 the severe constraints of latency. Now we must 00:08:10.470 --> 00:08:13.589 translate that into building real product intuition. 00:08:13.829 --> 00:08:16.069 Building that intuition takes deliberate daily 00:08:16.069 --> 00:08:18.389 practice. You cannot just use AI like a normal 00:08:18.389 --> 00:08:20.509 consumer anymore. You have to put on your product 00:08:20.509 --> 00:08:22.810 manager hat every single time you open a tool. 00:08:22.970 --> 00:08:26.019 The roadmap calls this specific approach. Structured 00:08:26.019 --> 00:08:28.920 curiosity. I love that term. It means you do 00:08:28.920 --> 00:08:31.800 not just use the AI to write your emails, you 00:08:31.800 --> 00:08:34.019 actively try to break it completely. You give 00:08:34.019 --> 00:08:37.299 it incredibly short, ambiguous prompts. You feed 00:08:37.299 --> 00:08:39.659 it completely wrong facts on purpose. You push 00:08:39.659 --> 00:08:42.480 it into a corner and watch exactly how it fails. 00:08:42.740 --> 00:08:45.279 Does it apologize? Does it hallucinate confidently? 00:08:45.360 --> 00:08:47.519 You also have to reverse -engineer top products 00:08:47.519 --> 00:08:50.779 to see how their PMs handle these flaws. Notice 00:08:50.779 --> 00:08:54.860 ChatGPT's rapid typing animation. That is not 00:08:54.860 --> 00:08:58.059 just a stylistic choice. It is a brilliant psychological 00:08:58.059 --> 00:09:00.899 trick to mask the latency. You are reading the 00:09:00.899 --> 00:09:03.139 first word while it computes the tenth. Look 00:09:03.139 --> 00:09:05.700 at Claude's notoriously friendly conversational 00:09:05.700 --> 00:09:08.340 tone. It changes the interaction completely. 00:09:08.460 --> 00:09:10.580 It makes users more forgiving when it makes a 00:09:10.580 --> 00:09:13.129 slight error. And look at Perplexity. They realized 00:09:13.129 --> 00:09:15.809 trust is the real product. So they show their 00:09:15.809 --> 00:09:18.190 web sources incredibly clearly right at the top. 00:09:18.330 --> 00:09:21.289 To build immense trust before you even read the 00:09:21.289 --> 00:09:24.049 answer. To truly learn how these mechanics work, 00:09:24.470 --> 00:09:26.990 you must build a real project yourself. You cannot 00:09:26.990 --> 00:09:30.570 skip this. The roadmap outlines two highly specific 00:09:30.570 --> 00:09:33.370 project ideas to get you started. The first project 00:09:33.370 --> 00:09:36.789 is building an AI resume helper. It is brilliant 00:09:36.789 --> 00:09:39.809 because it solves a very common, easily understandable 00:09:39.809 --> 00:09:42.679 problem. You build a system that takes a user's 00:09:42.679 --> 00:09:45.340 uploaded resume and compares it directly against 00:09:45.340 --> 00:09:48.139 the scraped job description. It identifies the 00:09:48.139 --> 00:09:51.340 exact missing skills. Then you prompt the AI 00:09:51.340 --> 00:09:53.820 to rewrite the resume summary to highlight those 00:09:53.820 --> 00:09:56.120 missing skills. But the critical challenge here 00:09:56.120 --> 00:09:58.500 is getting it to do this without inventing a 00:09:58.500 --> 00:10:01.179 fake job history. The second project is a customer 00:10:01.179 --> 00:10:04.360 review analyzer. This is classic business value. 00:10:04.509 --> 00:10:07.809 You take a messy data set of 50 unstructured 00:10:07.809 --> 00:10:10.669 customer reviews. You force the AI to categorize 00:10:10.669 --> 00:10:13.389 them strictly by sentiment, positive, neutral, 00:10:13.509 --> 00:10:16.250 or negative. You have it extract the key quotes. 00:10:16.570 --> 00:10:19.029 Then you prompted to synthesize all of that and 00:10:19.029 --> 00:10:21.190 identify the single biggest product pain point. 00:10:21.419 --> 00:10:24.000 Two -sec silence. I have to admit something here. 00:10:24.179 --> 00:10:26.460 I still wrestle with prompt drift myself when 00:10:26.460 --> 00:10:28.200 building these things. Oh, absolutely. Everyone 00:10:28.200 --> 00:10:30.620 does. It is just so unpredictable. You tweak 00:10:30.620 --> 00:10:32.679 one word in the instructions and suddenly the 00:10:32.679 --> 00:10:34.600 AI acts like a completely different persona. 00:10:34.879 --> 00:10:37.399 That is exactly the messy, frustrating side of 00:10:37.399 --> 00:10:40.740 AI development. The model simply does not always 00:10:40.740 --> 00:10:43.529 do what you explicitly tell it to do. So why 00:10:43.529 --> 00:10:45.710 is hitting a massive wall during these personal 00:10:45.710 --> 00:10:47.870 projects actually a good thing for your career? 00:10:48.149 --> 00:10:51.210 Because the messy, unpredictable reality of AI 00:10:51.210 --> 00:10:54.210 is where the real learning happens. Build real 00:10:54.210 --> 00:10:57.590 projects to learn AI's messy, unpredictable nature. 00:10:57.929 --> 00:11:00.570 When your project breaks, you learn how to fix 00:11:00.570 --> 00:11:03.870 weird edge case bugs and make the user experience 00:11:03.870 --> 00:11:06.629 feel smooth. Now, building a fun little project 00:11:06.629 --> 00:11:09.169 on your laptop is a great start. But keeping 00:11:09.169 --> 00:11:11.879 it alive in the real world is the final hurdle. 00:11:12.019 --> 00:11:14.500 And using it to actually get a job offer. This 00:11:14.500 --> 00:11:16.600 is where we enter the sometimes intimidating 00:11:16.600 --> 00:11:18.659 world of infrastructure operations. You have 00:11:18.659 --> 00:11:21.139 to know how the system breathes and survives 00:11:21.139 --> 00:11:23.779 in reality. The biggest concept here is evaluation. 00:11:24.220 --> 00:11:26.620 In the industry, we just call them evils. Because 00:11:26.620 --> 00:11:28.860 think about it, if you tweak your model to be 00:11:28.860 --> 00:11:31.259 faster, how do you know if your AI is still giving 00:11:31.259 --> 00:11:33.879 good answers? You cannot sit there and read every 00:11:33.879 --> 00:11:36.559 single generated response manually. Right. So 00:11:36.559 --> 00:11:39.159 you build an automated evil system. You basically 00:11:39.159 --> 00:11:42.559 use a smarter, more expensive boss AI to automatically 00:11:42.559 --> 00:11:45.240 grade the daily answers of your cheaper worker 00:11:45.240 --> 00:11:48.679 AI. It is a vital non -negotiable skill for a 00:11:48.679 --> 00:11:51.629 modern PM. Then there is the incredibly difficult 00:11:51.629 --> 00:11:55.090 math of unit economics. Cost versus quality. 00:11:55.490 --> 00:11:58.789 Running AI at scale is staggeringly expensive. 00:11:59.190 --> 00:12:01.210 You have to manage the token budget. Let's look 00:12:01.210 --> 00:12:04.610 at a scenario. Model A costs 10 cents per thousand 00:12:04.610 --> 00:12:07.929 words. It is brilliant, but it takes five seconds 00:12:07.929 --> 00:12:11.370 to run. Meanwhile, Model B costs just one cent 00:12:11.370 --> 00:12:13.730 per thousand words. It is only okay at reasoning, 00:12:13.929 --> 00:12:16.250 but it takes just one second. Your job as a PM 00:12:16.250 --> 00:12:19.149 is to choose. If you are building a high -stakes 00:12:19.149 --> 00:12:22.129 legal document advisor, you gladly pay for model 00:12:22.129 --> 00:12:24.610 A. But if you are just building a quick background 00:12:24.610 --> 00:12:27.590 grammar checker, you use model B to save millions 00:12:27.590 --> 00:12:29.529 of dollars. You also have to obsessively monitor 00:12:29.529 --> 00:12:31.970 the product after it launches. Are users suddenly 00:12:31.970 --> 00:12:34.470 getting an error 500 screen because the API is 00:12:34.470 --> 00:12:36.909 overloaded? Is the AI starting to give weird 00:12:36.909 --> 00:12:39.190 degraded answers over time? We call that model 00:12:39.190 --> 00:12:41.549 drift. You have to stare at those dashboard numbers 00:12:41.549 --> 00:12:44.129 daily. OK, we have the skills. Now let us talk 00:12:44.129 --> 00:12:46.809 about getting noticed. Because right now you're 00:12:46.809 --> 00:12:50.149 sitting in a massive sea of 500 seemingly identical 00:12:50.149 --> 00:12:53.649 resumes. Visibility is honestly just as important 00:12:53.649 --> 00:12:56.830 as your actual technical skill set in 2026. You 00:12:56.830 --> 00:12:58.909 have to prove you can do the work. I want to 00:12:58.909 --> 00:13:01.490 talk about this certification craze. Yeah, it 00:13:01.490 --> 00:13:03.269 seems like collecting certificates is just an 00:13:03.269 --> 00:13:05.389 illusion of progress, right? It feels like busy 00:13:05.389 --> 00:13:08.470 work. Absolutely. It is a huge trap. Most entry 00:13:08.470 --> 00:13:10.929 -level people just post screenshots of completion 00:13:10.929 --> 00:13:13.429 certificates on LinkedIn. It means nothing. It 00:13:13.429 --> 00:13:15.250 completely fails to prove that you can actually 00:13:15.250 --> 00:13:18.029 manage a complex, failing product. So do not 00:13:18.029 --> 00:13:21.250 just post certificates. Post the gritty, behind 00:13:21.250 --> 00:13:23.769 -the -scenes reality of your build process. Share 00:13:23.769 --> 00:13:26.809 the hard parts. Share a public link to your actual 00:13:26.809 --> 00:13:29.659 PRD. Write a post explaining exactly why you 00:13:29.659 --> 00:13:32.320 chose to measure latency instead of accuracy 00:13:32.320 --> 00:13:35.279 for your specific project. Record a quick screen 00:13:35.279 --> 00:13:38.279 capture build video. Show your customized AI 00:13:38.279 --> 00:13:41.720 tool handling a very difficult user request successfully. 00:13:41.879 --> 00:13:44.580 Show the actual mechanics. You should also intentionally 00:13:44.580 --> 00:13:47.179 join the right circles. Find where the actual 00:13:47.179 --> 00:13:49.840 builders are hanging out and sharing their failures. 00:13:50.120 --> 00:13:52.759 The AI Grant Fellowship is an amazing place to 00:13:52.759 --> 00:13:55.299 see exactly what top -tier startups are currently 00:13:55.299 --> 00:13:57.820 struggling with. The GEN Academy is fantastic 00:13:57.820 --> 00:14:00.940 for learning the deep nuances of agentic AI workflows. 00:14:01.139 --> 00:14:03.539 And honestly, just go find local tech meetups 00:14:03.539 --> 00:14:06.139 in your city. There are a few common traps you 00:14:06.139 --> 00:14:08.779 absolutely must avoid during the interview process. 00:14:09.279 --> 00:14:12.399 First, do not fall blindly in love with the raw 00:14:12.399 --> 00:14:16.259 technology. This is so common. Users do not care 00:14:16.259 --> 00:14:19.289 if you use the newest, fanciest model. They only 00:14:19.289 --> 00:14:22.570 care if their specific, painful problem was solved. 00:14:22.870 --> 00:14:25.250 Second, do not trick yourself into thinking you 00:14:25.250 --> 00:14:27.470 need to become a software engineer. Learning 00:14:27.470 --> 00:14:29.929 Python on the weekends is a nice hobby. Right, 00:14:30.149 --> 00:14:32.690 but your job as a PM is to lead the engineering 00:14:32.690 --> 00:14:35.330 team, not to write the production code yourself. 00:14:35.590 --> 00:14:37.870 Do not let learning syntax distract you from 00:14:37.870 --> 00:14:41.159 understanding user behavior. And finally, Never 00:14:41.159 --> 00:14:43.899 ignore the deeply boring stuff. Meticulous data 00:14:43.899 --> 00:14:46.200 cleaning and bulletproof user privacy protocols 00:14:46.200 --> 00:14:48.179 are what actually make an enterprise product 00:14:48.179 --> 00:14:50.580 successful. And test your prototype with real, 00:14:50.679 --> 00:14:53.100 easily confused people as early as possible. 00:14:53.559 --> 00:14:56.100 Watching one real user struggle with your interface 00:14:56.100 --> 00:14:58.759 is worth a hundred hours of solo brainstorming. 00:14:58.960 --> 00:15:01.019 What exactly is a hiring manager looking for 00:15:01.019 --> 00:15:03.539 when they see your LinkedIn build video? They 00:15:03.539 --> 00:15:05.899 are looking for undeniable proof of your product 00:15:05.899 --> 00:15:09.100 mindset. They want to see that you deeply understand 00:15:09.100 --> 00:15:11.879 technical trade -offs. Show companies your messy 00:15:11.879 --> 00:15:15.019 problem -solving process, not just finished certificates. 00:15:15.320 --> 00:15:17.919 Exactly. Let's quickly summarize this intense 00:15:17.919 --> 00:15:20.419 five -month sprint. Let's hear it. Months 1 and 00:15:20.419 --> 00:15:24.120 2 are purely for PM Basics. You master writing 00:15:24.120 --> 00:15:27.559 PRDs and defining rigorous product metrics. Month 00:15:27.559 --> 00:15:30.340 3 is your technical deep dive. You learn LLM 00:15:30.340 --> 00:15:33.700 Basics, RG. and you critically study existing 00:15:33.700 --> 00:15:36.639 AI products. Month four is the crucible. This 00:15:36.639 --> 00:15:38.679 is where you actively build, break, and ship 00:15:38.679 --> 00:15:41.120 your own functional prototype. Month five is 00:15:41.120 --> 00:15:43.840 for setting up evils and monitoring. Then, armed 00:15:43.840 --> 00:15:46.360 with your portfolio, you start applying for jobs. 00:15:46.639 --> 00:15:48.259 I want to leave you with this final thought to 00:15:48.259 --> 00:15:51.000 mull over. Go ahead. If an AI product manager's 00:15:51.000 --> 00:15:53.879 core job is to bridge the gap between computer 00:15:53.879 --> 00:15:57.139 capabilities and human needs, what happens to 00:15:57.139 --> 00:16:00.240 your role in a few years? That is the ultimate 00:16:00.240 --> 00:16:03.129 existential question. for this industry. What 00:16:03.129 --> 00:16:05.909 happens when the AI becomes smart enough to run 00:16:05.909 --> 00:16:09.009 its own discovery sprints, interview users, and 00:16:09.009 --> 00:16:12.070 write its own perfect PRDs? It is wild to think 00:16:12.070 --> 00:16:14.870 about. How do you stay essential when the tool 00:16:14.870 --> 00:16:17.470 eventually learns to manage itself? Well, for 00:16:17.470 --> 00:16:20.649 now, do not freeze up. Pick one small annoying 00:16:20.649 --> 00:16:23.419 problem in your life today. write a simple prompt 00:16:23.419 --> 00:16:26.399 to fix it, and start building. You have exactly 00:16:26.399 --> 00:16:27.919 what it takes to do this, whether you come from 00:16:27.919 --> 00:16:30.240 a background in marketing, teaching, or art. 00:16:30.460 --> 00:16:32.659 Your unique perspective is a massive advantage. 00:16:32.919 --> 00:16:34.840 Thank you for taking this deep dive with us. 00:16:35.019 --> 00:16:35.879 OTRO music.