WEBVTT 00:00:00.000 --> 00:00:04.059 What if you could ask an AI just a simple question, 00:00:04.139 --> 00:00:07.519 like one sentence, and still get consistently 00:00:07.519 --> 00:00:09.839 better results than someone else writing this 00:00:09.839 --> 00:00:12.320 huge, complicated essay prompt? Yeah, it sounds 00:00:12.320 --> 00:00:14.699 kind of wild, but it's actually where things 00:00:14.699 --> 00:00:17.359 are heading. The secret isn't just the instruction 00:00:17.359 --> 00:00:20.059 anymore. It's not just what you type. It's really 00:00:20.059 --> 00:00:22.980 about what the AI already knows about you, about 00:00:22.980 --> 00:00:24.800 your business. And that's really the core of 00:00:24.800 --> 00:00:27.160 this deep dive, isn't it? We're seeing this shift. 00:00:27.760 --> 00:00:30.960 away from, let's call it traditional prompt engineering. 00:00:31.260 --> 00:00:33.859 The old way, yeah. Towards something new, context 00:00:33.859 --> 00:00:37.280 engineering. Exactly. And the mission here for 00:00:37.280 --> 00:00:39.979 you listening is how to take that AI that, you 00:00:39.979 --> 00:00:43.020 know, kind of generic forgetful assistant and 00:00:43.020 --> 00:00:45.520 really transform it, make it into this personalized, 00:00:45.600 --> 00:00:47.979 always on team member. Turning a general tool 00:00:47.979 --> 00:00:50.780 into a specialist colleague. I like that. So 00:00:50.780 --> 00:00:53.799 we've broken it down into five levels, a framework. 00:00:54.000 --> 00:00:55.240 Okay. And we're going to walk you through them. 00:00:55.579 --> 00:00:58.060 Starting from the AIs. like basic knowledge, 00:00:58.100 --> 00:01:00.340 all the way up to giving it this permanent business 00:01:00.340 --> 00:01:03.240 memory. But maybe we should start with why this 00:01:03.240 --> 00:01:05.459 matters so much. Because it's not just about 00:01:05.459 --> 00:01:07.459 getting slightly better answers, is it? There's 00:01:07.459 --> 00:01:10.180 some pretty serious research coming out. Yeah, 00:01:10.219 --> 00:01:12.819 absolutely. Places like MIT, they're finding 00:01:12.819 --> 00:01:15.140 some disturbing things about passive AI use. 00:01:15.319 --> 00:01:18.340 I saw that. It's actually, well, shocking. Using 00:01:18.340 --> 00:01:21.939 AI like a crutch, the studies show it can make 00:01:21.939 --> 00:01:25.870 people measurably... Measurably dumber. It's 00:01:25.870 --> 00:01:27.930 a real effect. They call it cognitive outsourcing. 00:01:28.230 --> 00:01:30.409 Basically, if you let the machine do all the 00:01:30.409 --> 00:01:32.469 heavy lifting, all the synthesis, the analysis, 00:01:32.829 --> 00:01:36.189 your own brains, you know, mental muscles, those 00:01:36.189 --> 00:01:38.950 neural pathways, they start to weaken. Atrophy, 00:01:39.090 --> 00:01:41.849 like a muscle you don't use. Precisely. You lose 00:01:41.849 --> 00:01:44.209 that habit of deep thinking, of connecting the 00:01:44.209 --> 00:01:47.209 dots yourself. So if I'm just asking the AI for 00:01:47.209 --> 00:01:49.390 strategic summaries all the time, and I'm not 00:01:49.390 --> 00:01:51.230 really wrestling with the information myself. 00:01:51.980 --> 00:01:54.239 challenging it, I'm actually hurting my own abilities. 00:01:54.439 --> 00:01:56.719 That's the risk. It degrades the very skills 00:01:56.719 --> 00:01:59.420 you need for like critical thinking and making 00:01:59.420 --> 00:02:02.140 smart decisions. But the solution isn't to just 00:02:02.140 --> 00:02:04.640 stop using AI. Not at all. It's about changing 00:02:04.640 --> 00:02:07.640 how you use it, making it an active partner, 00:02:07.739 --> 00:02:09.539 something that sharpens you, doesn't replace 00:02:09.539 --> 00:02:12.340 you. Okay. That really sets the stage. Active 00:02:12.340 --> 00:02:15.159 engagement through context. So let's dive into 00:02:15.159 --> 00:02:19.099 level one, the foundation, the AI's basic knowledge. 00:02:19.139 --> 00:02:22.800 Right. Level one, the training data. This is 00:02:22.800 --> 00:02:25.780 that massive ocean of information the AI learned 00:02:25.780 --> 00:02:29.139 from. Trillions of documents, websites, books, 00:02:29.419 --> 00:02:32.020 everything. So it knows a lot of general stuff. 00:02:32.240 --> 00:02:35.780 A ton. But critically, zero personal context. 00:02:35.840 --> 00:02:38.439 It doesn't know you, your company, your goals. 00:02:38.620 --> 00:02:40.280 It's like that brilliant stranger at a party 00:02:40.280 --> 00:02:43.039 idea. Yeah. You know? Super smart, can talk about 00:02:43.039 --> 00:02:46.259 anything. Yeah. But their advice, it's often 00:02:46.259 --> 00:02:48.379 totally off because they don't know your specific 00:02:48.379 --> 00:02:50.699 situation, your budget, your team, your history. 00:02:50.840 --> 00:02:52.580 Exactly. We see this all the time with like a 00:02:52.580 --> 00:02:54.939 generic marketing plan prompt. You ask for one. 00:02:55.060 --> 00:02:56.419 And you get something that looks impressive. 00:02:56.479 --> 00:02:58.319 Lots of pages, bullet points. Looks comprehensive. 00:02:58.379 --> 00:03:02.080 Yeah. But dig in. It's often flawed. The audience 00:03:02.080 --> 00:03:04.060 might be wrong. The tactics totally irrelevant 00:03:04.060 --> 00:03:06.300 for you. Maybe it even makes up details about 00:03:06.300 --> 00:03:08.479 your product. And this kind of leads into that 00:03:08.479 --> 00:03:10.620 whole model size myth, right? People get so obsessed. 00:03:10.759 --> 00:03:13.699 My AI has trillions of parameters. No, mine has 00:03:13.699 --> 00:03:16.599 more trillions. It's mostly noise at this point. 00:03:16.620 --> 00:03:20.099 Honestly, a distraction. Context beats complexity 00:03:20.099 --> 00:03:23.139 every single time. What matters isn't the raw 00:03:23.139 --> 00:03:25.979 size or who made the model. Hold on. Are you 00:03:25.979 --> 00:03:29.639 actually saying. That maybe a smaller model, 00:03:29.759 --> 00:03:31.719 maybe even one that's a couple of years old. 00:03:31.780 --> 00:03:35.139 Yeah. If I feed it all my specific customer data, 00:03:35.219 --> 00:03:38.919 my internal documents, that could be better than 00:03:38.919 --> 00:03:42.120 the absolute latest, biggest flagship model out 00:03:42.120 --> 00:03:44.710 there right now. Yes. That's exactly what I'm 00:03:44.710 --> 00:03:46.810 saying. It sounds counterintuitive, I know, because 00:03:46.810 --> 00:03:48.150 that's not what the vendors are shouting about. 00:03:48.330 --> 00:03:50.229 Right. But it's the reality we see in the results. 00:03:50.430 --> 00:03:52.810 A smaller model, if it's super rich with your 00:03:52.810 --> 00:03:55.370 specific context, will consistently give you 00:03:55.370 --> 00:03:57.870 more relevant, more useful output than a giant 00:03:57.870 --> 00:03:59.669 model that's just guessing about who you are. 00:03:59.770 --> 00:04:01.969 It's about relevance, not raw host power. Okay. 00:04:02.150 --> 00:04:05.210 Okay. That's a big shift in thinking. So if model 00:04:05.210 --> 00:04:07.590 size isn't the be all and end all, and we haven't 00:04:07.590 --> 00:04:10.210 added our own data yet, what's the very first 00:04:10.210 --> 00:04:12.430 thing we can do? Like right now? To get better 00:04:12.430 --> 00:04:14.409 results without adding anything personal. We 00:04:14.409 --> 00:04:16.009 need to get smarter about what's already there. 00:04:16.350 --> 00:04:20.009 Move past the generic training data and start 00:04:20.009 --> 00:04:24.009 exploiting the AI's own internal rulebook. Ah, 00:04:24.170 --> 00:04:27.689 okay. That takes us to level two. The hidden 00:04:27.689 --> 00:04:30.209 system prompts. You're saying every AI kind of 00:04:30.209 --> 00:04:32.889 has this secret instruction manual. Pretty much. 00:04:33.480 --> 00:04:35.639 Think of it like a hidden document, maybe 100, 00:04:35.839 --> 00:04:39.420 120 pages long, built by the developers. It defines 00:04:39.420 --> 00:04:41.939 its personality, its safety rules, how it should 00:04:41.939 --> 00:04:43.879 behave by default. And we can't just go in and 00:04:43.879 --> 00:04:46.620 edit that. No, no. That's locked down for safety, 00:04:46.680 --> 00:04:49.120 security reasons. But you can definitely exploit 00:04:49.120 --> 00:04:52.060 it by learning its triggers. Specific words or 00:04:52.060 --> 00:04:54.579 phrases that kind of poke the AI and make it 00:04:54.579 --> 00:04:58.089 access deeper. more analytical parts of its programming. 00:04:58.209 --> 00:05:00.069 Okay, interesting. So for something like Claude, 00:05:00.069 --> 00:05:01.410 you've mentioned there are specific keywords. 00:05:01.649 --> 00:05:03.110 Yeah, there are several. You can kind of group 00:05:03.110 --> 00:05:05.750 them. Some are about depth, words like in -depth 00:05:05.750 --> 00:05:08.430 or comprehensive. Makes sense. Others are more 00:05:08.430 --> 00:05:11.250 about actions, like analyze, evaluate, assess. 00:05:11.829 --> 00:05:14.149 These tell it not just to spit information back, 00:05:14.230 --> 00:05:16.790 but to actually think about it. structurally. 00:05:16.970 --> 00:05:19.029 And then there are triggers for the output itself. 00:05:19.370 --> 00:05:21.889 Right. Like asking it to research something or 00:05:21.889 --> 00:05:24.329 specifically make a report, just using one or 00:05:24.329 --> 00:05:27.350 two of these. It can seriously boost the quality, 00:05:27.449 --> 00:05:30.470 a huge jump for minimal effort. So back to that 00:05:30.470 --> 00:05:32.509 generic marketing plan example. Yeah. Instead 00:05:32.509 --> 00:05:34.629 of just asking for a plan, you say, give me an 00:05:34.629 --> 00:05:36.949 in -depth, comprehensive marketing strategy. 00:05:37.230 --> 00:05:40.029 And the difference is what? It's night and day. 00:05:40.189 --> 00:05:43.709 You go from maybe a simple list of ideas to a 00:05:43.709 --> 00:05:46.699 much more thoughtful analysis. It might start 00:05:46.699 --> 00:05:49.439 talking about market size, competitors, real 00:05:49.439 --> 00:05:52.879 business factors, actual strategy. Wow. Okay. 00:05:53.040 --> 00:05:55.540 Tiny change, big impact. So that's level two, 00:05:55.660 --> 00:05:58.399 exploiting the hidden rules. How do we then take 00:05:58.399 --> 00:06:00.959 like permanent control, shape how it talks to 00:06:00.959 --> 00:06:02.860 us all the time? That brings us to level three, 00:06:02.939 --> 00:06:04.959 user preferences. This is your control panel, 00:06:05.100 --> 00:06:07.160 your set it and forget it layer. Like custom 00:06:07.160 --> 00:06:09.879 instructions in ChatGPT or personal preferences 00:06:09.879 --> 00:06:13.120 in Cloud. Exactly those. You define how you want 00:06:13.120 --> 00:06:15.759 it to communicate once and it sticks. It applies 00:06:15.759 --> 00:06:18.220 globally to all your future chats. And you focus 00:06:18.220 --> 00:06:21.240 on three main changes here that are transformative. 00:06:21.620 --> 00:06:24.240 Yeah. Three core things. First, format preference. 00:06:24.660 --> 00:06:28.300 Be demanding. Tell it you always want answers 00:06:28.300 --> 00:06:30.980 in, say, bullet points with nested sub bullets. 00:06:31.519 --> 00:06:33.819 For clarity. Okay. Consistent structure. Yeah. 00:06:33.879 --> 00:06:37.180 Second, enforce conservative analysis. You literally 00:06:37.180 --> 00:06:40.220 instruct it. Give me realistic, grounded advice. 00:06:40.579 --> 00:06:43.300 No pie in the sky stuff. It keeps the output 00:06:43.300 --> 00:06:46.120 anchored in reality. I like that. Practical advice. 00:06:46.379 --> 00:06:48.980 And third, this one's critical. Certainty levels. 00:06:49.139 --> 00:06:51.240 Certainty levels. Yeah. Tell the AI that for 00:06:51.240 --> 00:06:53.220 every recommendation it makes, it must state 00:06:53.220 --> 00:06:55.920 its confidence level. Like, I'm 70 % certain 00:06:55.920 --> 00:06:58.459 this approach will work based on the data. So 00:06:58.459 --> 00:07:01.790 if it says it's only like... 65 % sure about 00:07:01.790 --> 00:07:04.009 a pricing strategy. Exactly. That's your cue. 00:07:04.129 --> 00:07:06.930 It forces you, the user, to dig deeper, to ask 00:07:06.930 --> 00:07:09.269 clarifying questions right where the AI is weakest 00:07:09.269 --> 00:07:11.709 or the data is maybe less clear. It encourages 00:07:11.709 --> 00:07:13.790 that active engagement we talked about. Okay, 00:07:13.850 --> 00:07:15.750 so level three locks in the communication style, 00:07:15.949 --> 00:07:20.100 the format, the tone, globally. Since this changes 00:07:20.100 --> 00:07:22.680 everything permanently, how does setting that 00:07:22.680 --> 00:07:26.100 foundation get the AI ready for the really specific 00:07:26.100 --> 00:07:28.779 stuff like our own business intelligence? Well, 00:07:28.920 --> 00:07:31.060 by getting level three right, you guarantee a 00:07:31.060 --> 00:07:33.740 certain baseline quality and consistency. The 00:07:33.740 --> 00:07:35.779 output is reliable. It's structured. It's clear. 00:07:35.920 --> 00:07:39.019 That makes the AI much better prepared to ingest 00:07:39.019 --> 00:07:42.240 and then intelligently use large amounts of complex, 00:07:42.339 --> 00:07:45.139 specialized, high stakes information without, 00:07:45.199 --> 00:07:46.899 you know, getting confused or giving you messy 00:07:46.899 --> 00:07:49.910 results. Midroll sponsor. read all right we're 00:07:49.910 --> 00:07:53.230 back and we've reached level four project knowledge 00:07:53.230 --> 00:07:55.410 you said this is maybe the most transformative 00:07:55.410 --> 00:07:57.529 level absolutely this is where we really give 00:07:57.529 --> 00:08:00.269 the ai that memory upgrade a working reliable 00:08:00.269 --> 00:08:02.370 memory so it sucks being just a general tool 00:08:02.370 --> 00:08:04.970 right and it starts acting like a like a fully 00:08:04.970 --> 00:08:06.829 onboarded team member someone who's been with 00:08:06.829 --> 00:08:09.029 you a while knows the ropes knows our products 00:08:09.029 --> 00:08:11.470 our pricing key metrics who our competitors are 00:08:12.110 --> 00:08:14.490 That level of detail. Exactly. It assimilates 00:08:14.490 --> 00:08:16.329 your specific business context for a particular 00:08:16.329 --> 00:08:19.209 area of work. And how do we actually do that? 00:08:19.250 --> 00:08:21.569 How do we implement it? It usually involves creating 00:08:21.569 --> 00:08:23.949 project specific workspaces, you know, like those 00:08:23.949 --> 00:08:26.629 separate chat sidebars you see in Claude or dedicated 00:08:26.629 --> 00:08:29.730 project folders and other tools. And then into 00:08:29.730 --> 00:08:32.450 that specific space, you upload your documents, 00:08:32.789 --> 00:08:35.210 your proprietary stuff. Like what kind of documents? 00:08:35.470 --> 00:08:38.149 Could be anything relevant. Past client proposals, 00:08:38.629 --> 00:08:41.950 market research reports, detailed lists. of customer 00:08:41.950 --> 00:08:44.710 pain points, testimonials, your strategic planning 00:08:44.710 --> 00:08:47.909 docs, competitor analysis files, Google Docs, 00:08:47.909 --> 00:08:51.649 PDFs, whatever holds that key project knowledge. 00:08:51.929 --> 00:08:55.190 You know, I have to admit, I still wrestle with 00:08:55.190 --> 00:08:58.059 prompt drift myself sometimes. You get deep into 00:08:58.059 --> 00:09:00.779 a conversation and you realize the AI has kind 00:09:00.779 --> 00:09:02.480 of forgotten what you were talking about 10 turns 00:09:02.480 --> 00:09:04.000 ago. Oh, totally. It happens to everyone. But 00:09:04.000 --> 00:09:06.480 level four, building these dedicated project 00:09:06.480 --> 00:09:09.379 spaces, it forces an organizational structure. 00:09:09.460 --> 00:09:12.259 It gives the AI and you an anchor, making the 00:09:12.259 --> 00:09:14.840 whole process way more reliable. It combats that 00:09:14.840 --> 00:09:16.700 drift. So how does it work technically? Does 00:09:16.700 --> 00:09:19.519 it just read the documents once? It's often using 00:09:19.519 --> 00:09:21.860 something called Arrage. That's retrieval augmented 00:09:21.860 --> 00:09:25.299 generation. Basically, when you upload documents 00:09:25.299 --> 00:09:29.220 to a specific project, the AI builds a mini searchable 00:09:29.220 --> 00:09:32.299 knowledge base just for that project. And then 00:09:32.299 --> 00:09:35.159 for every single question or task you give it 00:09:35.159 --> 00:09:38.539 within that project space, it automatically searches 00:09:38.539 --> 00:09:41.360 and pulls relevant information from those documents 00:09:41.360 --> 00:09:43.620 you uploaded. It's like an always -on dedicated 00:09:43.620 --> 00:09:46.899 memory for that specific task or client. Whoa. 00:09:47.000 --> 00:09:50.220 Okay, hang on. Imagine you ask it for a quarterly 00:09:50.220 --> 00:09:53.470 strategy review. And it automatically pulls in 00:09:53.470 --> 00:09:55.429 the latest competitive analysis you uploaded 00:09:55.429 --> 00:09:58.809 last week and references specific customer quotes 00:09:58.809 --> 00:10:01.929 from testimonial docs without you having to copy 00:10:01.929 --> 00:10:04.250 paste or remind it about any of that background. 00:10:04.450 --> 00:10:06.690 That's it. Exactly. That's genuine scaling. It's 00:10:06.690 --> 00:10:08.129 not just an assistant. That's like a personalized 00:10:08.129 --> 00:10:10.950 copilot. Precisely. Upload once, leverage it 00:10:10.950 --> 00:10:13.029 forever within that context. It becomes deeply 00:10:13.029 --> 00:10:15.549 specialized. Okay. Mind kind of blown there. 00:10:15.610 --> 00:10:17.409 So what's left? What's level five? Level five 00:10:17.409 --> 00:10:20.029 is the peak. The apex, really. AI -generated 00:10:20.029 --> 00:10:21.889 prompts, or you could call it meta -prompting. 00:10:21.909 --> 00:10:24.129 Meta -prompting. Yeah. Now we're actually using 00:10:24.129 --> 00:10:26.830 the AI itself to help us engineer the perfect 00:10:26.830 --> 00:10:29.870 prompt for a really complex task, using the machine 00:10:29.870 --> 00:10:32.210 to build the instructions for the machine. Okay, 00:10:32.289 --> 00:10:33.809 walk me through that. What does that initial 00:10:33.809 --> 00:10:36.620 request look like? It sounds... Complicated. 00:10:36.799 --> 00:10:39.299 It is a bit. The initial request is high level. 00:10:39.440 --> 00:10:42.960 You might ask the AI, OK, generate a new comprehensive 00:10:42.960 --> 00:10:45.940 prompt template for creating a marketing strategy 00:10:45.940 --> 00:10:49.240 for this specific new customer segment. And you'd 00:10:49.240 --> 00:10:51.039 add constraints. Oh, yeah. Yeah. You tell it. 00:10:51.059 --> 00:10:53.379 This prompt needs to ensure the final strategy 00:10:53.379 --> 00:10:56.159 includes specific team OKR's objectives and key 00:10:56.159 --> 00:10:59.120 results. It must analyze insights from our past 00:10:59.120 --> 00:11:02.039 customer complaints database and factor in recent 00:11:02.039 --> 00:11:05.440 competitor wins. You get the idea. You're defining 00:11:05.440 --> 00:11:07.860 the structure of the ideal prompt. Got it. So 00:11:07.860 --> 00:11:11.480 the AI takes that complex request and it uses 00:11:11.480 --> 00:11:13.559 everything it knows now. The base training data, 00:11:13.700 --> 00:11:16.179 the system triggers, my preferences, and all 00:11:16.179 --> 00:11:18.700 the specific project documents. All four layers, 00:11:18.779 --> 00:11:21.759 exactly. And it uses them to generate this incredibly 00:11:21.759 --> 00:11:25.000 detailed, tailored prompt template specifically 00:11:25.000 --> 00:11:27.220 designed for that task. And then you take that 00:11:27.220 --> 00:11:29.480 AI generated template. You feed it back into 00:11:29.480 --> 00:11:32.000 the AI in a new session, maybe tweak it slightly. 00:11:32.509 --> 00:11:35.450 And boom, the output isn't just a strategy doc. 00:11:35.549 --> 00:11:38.129 It might be a whole package. Sales enablement 00:11:38.129 --> 00:11:40.610 guides, scripts for handling objections, even 00:11:40.610 --> 00:11:44.009 frameworks for calculating ROI, all hyper -specific 00:11:44.009 --> 00:11:46.730 to that context. Feels like a massive leap in 00:11:46.730 --> 00:11:49.830 efficiency. It is. Think about it. That level 00:11:49.830 --> 00:11:52.870 of thoughtfulness, of cross -referencing documents, 00:11:53.169 --> 00:11:55.710 aligning with goals, that used to take a team 00:11:55.710 --> 00:11:58.049 maybe two, three weeks. Easily. Lots of meetings 00:11:58.049 --> 00:12:00.529 back and forth. Now you can get that quality 00:12:00.529 --> 00:12:03.350 of output, that integrated thinking in minutes, 00:12:03.549 --> 00:12:06.590 that time saving. That's where the real competitive 00:12:06.590 --> 00:12:08.730 edge comes from. So just to be clear, this isn't 00:12:08.730 --> 00:12:11.169 about context replacing prompt engineering. No, 00:12:11.169 --> 00:12:13.700 not at all. They're partners. Think of it like 00:12:13.700 --> 00:12:17.320 this. Context, levels one to four, is the perfectly 00:12:17.320 --> 00:12:19.960 tuned, fully loaded race car. It's got the engine, 00:12:20.039 --> 00:12:22.860 the fuel, the specs. Prompting, level five in 00:12:22.860 --> 00:12:24.919 your ongoing instructions, is the skilled driver 00:12:24.919 --> 00:12:26.960 who knows the track and how to get the absolute 00:12:26.960 --> 00:12:29.220 best performance out of that car. You need both. 00:12:29.399 --> 00:12:31.820 That's a great analogy. Okay, so we have this 00:12:31.820 --> 00:12:33.899 powerful framework, levels one through five, 00:12:33.980 --> 00:12:36.679 potentially saving weeks of work. Given all that 00:12:36.679 --> 00:12:39.309 power... What's the single biggest mistake people 00:12:39.309 --> 00:12:41.090 make when they start trying to implement this? 00:12:41.269 --> 00:12:44.399 What did we absolutely avoid? Stop trying to 00:12:44.399 --> 00:12:46.980 write war and peace in your prompts. Seriously. 00:12:47.480 --> 00:12:49.879 Once you've put in the work to establish the 00:12:49.879 --> 00:12:52.600 context, especially levels three and four, your 00:12:52.600 --> 00:12:55.100 instructions should actually get simpler, shorter, 00:12:55.299 --> 00:12:58.519 more direct. If the AI already knows your entire 00:12:58.519 --> 00:13:00.639 business strategy level four, you don't need 00:13:00.639 --> 00:13:02.919 to re -explain it every time. Just give it the 00:13:02.919 --> 00:13:05.700 core task. Let the context do the heavy lifting. 00:13:05.720 --> 00:13:07.759 Simplify the instructions because the context 00:13:07.759 --> 00:13:10.200 is doing the work. Got it. Okay, let's recap 00:13:10.200 --> 00:13:12.120 the big picture here. The fundamental shift. 00:13:12.620 --> 00:13:15.340 The secret sauce. It's setting up all this context 00:13:15.340 --> 00:13:17.100 before you even ask the main question. That's 00:13:17.100 --> 00:13:18.960 the core idea. We went through the five levels, 00:13:19.120 --> 00:13:21.779 starting with the AI's generic knowledge, L1, 00:13:21.919 --> 00:13:24.379 exploiting its hidden rules, L2, setting our 00:13:24.379 --> 00:13:26.620 preferences, L3, giving it project -specific 00:13:26.620 --> 00:13:29.940 memory, L4, and finally, getting the AI to help 00:13:29.940 --> 00:13:32.720 write its own instructions, L5. And the good 00:13:32.720 --> 00:13:35.559 news is you listening can start doing this right 00:13:35.559 --> 00:13:38.080 away. We've got a simple three -step roadmap 00:13:38.080 --> 00:13:40.940 for you, your context mastery roadmap. Okay, 00:13:41.000 --> 00:13:43.399 step one, what's the immediate action? Today, 00:13:43.600 --> 00:13:47.100 take maybe 10, 15 minutes. Go online and research 00:13:47.100 --> 00:13:49.700 the system prompt for the AI tool you use most 00:13:49.700 --> 00:13:53.019 often. Just search your AI tool name plus system 00:13:53.019 --> 00:13:55.639 prompt. Find those trigger words we talked about. 00:13:55.840 --> 00:13:59.299 Like in -depth, analyze, evaluate. Exactly. And 00:13:59.299 --> 00:14:01.809 just finding them isn't enough. The key is to 00:14:01.809 --> 00:14:04.509 start using them. Try adding just one or two 00:14:04.509 --> 00:14:06.370 to prompts you already use. See the difference 00:14:06.370 --> 00:14:08.549 yourself. It's usually immediate. Okay, quick 00:14:08.549 --> 00:14:11.970 win, step two. Tomorrow morning, go into your 00:14:11.970 --> 00:14:14.909 AI settings, find those user preferences or custom 00:14:14.909 --> 00:14:17.649 instructions, and lock in those three core settings 00:14:17.649 --> 00:14:20.429 we discussed. Demand bullet points or your preferred 00:14:20.429 --> 00:14:23.429 format, insist on conservative, grounded advice, 00:14:23.649 --> 00:14:25.769 and require it to state its certainty level for 00:14:25.769 --> 00:14:28.110 recommendations. Set it and forget it for consistent 00:14:28.110 --> 00:14:30.769 quality. And step three. By the end of this week. 00:14:31.120 --> 00:14:33.460 Pick one project, one area of your work. Create 00:14:33.460 --> 00:14:35.759 a dedicated workspace for it in your AI tool. 00:14:35.919 --> 00:14:38.360 And just upload three to five relevant documents. 00:14:38.539 --> 00:14:40.820 Just start small. Start small. Give it something 00:14:40.820 --> 00:14:43.259 specific. Maybe your company's mission statement, 00:14:43.500 --> 00:14:46.000 a key product description, a list of customer 00:14:46.000 --> 00:14:49.259 FAQs. Something rich so it stops giving you generic 00:14:49.259 --> 00:14:51.620 fluff and starts understanding your world. Right. 00:14:51.679 --> 00:14:54.659 Build that initial memory. So research triggers 00:14:54.659 --> 00:14:57.419 today, set preferences tomorrow, upload a few 00:14:57.419 --> 00:14:59.940 files by week's end. That's the start. It feels 00:14:59.940 --> 00:15:02.059 like organizations that really lean into this, 00:15:02.120 --> 00:15:05.440 that master this context engineering, they're 00:15:05.440 --> 00:15:07.500 going to gain a pretty significant edge, won't 00:15:07.500 --> 00:15:10.159 they? In just speed of decision making, strategy. 00:15:10.539 --> 00:15:12.580 Huge advantage. Almost unfair, like you said. 00:15:12.679 --> 00:15:15.059 When your AI understands your business deeply, 00:15:15.220 --> 00:15:17.620 the speed and quality of strategic work just 00:15:17.620 --> 00:15:20.000 accelerates dramatically. So the final thought 00:15:20.000 --> 00:15:22.980 here is... Do this work. Put these context layers 00:15:22.980 --> 00:15:26.019 in place thoughtfully, and your AI will finally, 00:15:26.019 --> 00:15:29.100 truly start working with you as a partner, not 00:15:29.100 --> 00:15:31.580 just churning out generic text for you. So maybe 00:15:31.580 --> 00:15:34.039 think about it. What's the first area of your 00:15:34.039 --> 00:15:35.940 work you're going to transform by giving your 00:15:35.940 --> 00:15:37.539 AI a real memory and context?