WEBVTT 00:00:00.000 --> 00:00:02.100 We've all been there, haven't we? You're deep 00:00:02.100 --> 00:00:03.879 in your code, you've got your AI helper running, 00:00:04.740 --> 00:00:07.280 and you ask it something. Maybe refactor a component, 00:00:07.679 --> 00:00:11.019 debug a little function, and it just completely 00:00:11.019 --> 00:00:12.759 misses the point. Oh yeah. Is your old code, 00:00:12.939 --> 00:00:16.260 or it's like, it's got digital amnesia. Forgets 00:00:16.260 --> 00:00:17.760 everything you just told it about the project 00:00:17.760 --> 00:00:20.859 structure, like five minutes ago. Digital amnesia, 00:00:20.859 --> 00:00:24.120 I like that. It's spot on, really. You're in 00:00:24.120 --> 00:00:26.940 the zone, things are clicking, and then the AI 00:00:26.940 --> 00:00:30.359 just... loses the plot entirely, super frustrating. 00:00:30.519 --> 00:00:34.460 Exactly, and that frustration, that constant 00:00:34.460 --> 00:00:36.740 need to re -explain things, that's really what 00:00:36.740 --> 00:00:39.119 we're diving into today. Our goal is basically 00:00:39.119 --> 00:00:41.880 mastering AI context so we can all be more productive, 00:00:42.359 --> 00:00:46.060 because the core issue is that lack of... continuous 00:00:46.060 --> 00:00:48.679 context. These tools often feel like a blank 00:00:48.679 --> 00:00:50.600 slate every time you prompt them. Which waste 00:00:50.600 --> 00:00:53.060 so much time. It really does. And it's a known 00:00:53.060 --> 00:00:54.780 problem, right? People are thinking about this. 00:00:54.840 --> 00:00:58.039 There's even this kind of futuristic idea floating 00:00:58.039 --> 00:01:00.899 around called MCP servers, Model Context Protocol 00:01:00.899 --> 00:01:04.260 servers. The idea is these servers would let 00:01:04.260 --> 00:01:07.579 AIs automatically tap into a project's full context. 00:01:07.939 --> 00:01:11.200 Like, really understand. Think of it like a permanent 00:01:11.200 --> 00:01:13.560 memory for the AI. Deep understanding. Yeah, 00:01:13.780 --> 00:01:17.079 and it's important to say, while these dedicated 00:01:17.079 --> 00:01:19.420 MCP servers themselves might still be mostly 00:01:19.420 --> 00:01:21.659 concepts, maybe just around the corner, the ideas 00:01:21.659 --> 00:01:24.340 driving them, those are incredibly valuable right 00:01:24.340 --> 00:01:27.140 now. They give us a sort of roadmap for how we 00:01:27.140 --> 00:01:30.859 can... today, teach our current AIs, chat, GPT, 00:01:30.959 --> 00:01:33.040 Gemini, Claude, whatever you use, teach them 00:01:33.040 --> 00:01:36.099 to grasp our projects on a much deeper level. 00:01:36.299 --> 00:01:38.359 Exactly. So that's our mission for this deep 00:01:38.359 --> 00:01:41.079 dive. We're going to unpack 10 practical techniques, 00:01:41.200 --> 00:01:43.280 all based on those MCP ideas, things you can 00:01:43.280 --> 00:01:45.840 actually use now, to give your AI a better memory, 00:01:45.980 --> 00:01:48.099 basically, a clearer picture of your project, 00:01:48.299 --> 00:01:50.379 which should really boost your productivity, 00:01:50.540 --> 00:01:52.579 get the AI working with you, not against you. 00:01:52.840 --> 00:01:56.000 So think of this as just a calm, curious look 00:01:56.000 --> 00:01:59.549 at how we can make AI truly work for us, moving 00:01:59.549 --> 00:02:02.269 beyond those simple one -off questions towards 00:02:02.269 --> 00:02:04.409 something more like an informed partnership. 00:02:05.150 --> 00:02:08.210 Let's dig in. OK, so first up, something really 00:02:08.210 --> 00:02:12.069 fundamental, the project structure itself. This 00:02:12.069 --> 00:02:14.669 first idea comes from the concept of a file system 00:02:14.669 --> 00:02:17.710 MCP. Imagine an AI server that could just read 00:02:17.710 --> 00:02:19.509 and understand your whole project directory. 00:02:19.870 --> 00:02:22.210 Know where every file is automatically. Exactly. 00:02:22.409 --> 00:02:24.650 So how do we do that now, practically speaking? 00:02:24.770 --> 00:02:26.430 Well, it's actually pretty straightforward. Before 00:02:26.430 --> 00:02:28.629 you ask the AI to do something involving files, 00:02:29.009 --> 00:02:31.169 give it the structure first. You can just run 00:02:31.169 --> 00:02:34.689 a tree if you're on Windows or Ellsvoord R on 00:02:34.689 --> 00:02:37.430 Mac or Linux. Copy that output. Paste it right 00:02:37.430 --> 00:02:39.310 into your prompt. It's like, here are the blueprints. 00:02:39.509 --> 00:02:41.310 Gotcha. So you'd say something like? Yeah, you'd 00:02:41.310 --> 00:02:43.430 say, here's my project's directory structure. 00:02:43.530 --> 00:02:46.090 Then paste the tree output. And then follow up 00:02:46.090 --> 00:02:49.090 with, OK, I just moved button .tsx from Cirque 00:02:49.090 --> 00:02:51.409 Components over to Cirque Design System Adams. 00:02:52.069 --> 00:02:54.310 Based on the structure I gave you, find all the 00:02:54.310 --> 00:02:57.569 files and Cirque pages using this button and 00:02:57.569 --> 00:03:00.490 rewrite their import paths. Ah, OK. You're giving 00:03:00.490 --> 00:03:04.169 it a concrete map. No guesswork. Precisely. So 00:03:04.169 --> 00:03:06.870 why is knowing that file location so crucial 00:03:06.870 --> 00:03:09.710 for the AI? Well, without that map, it's just 00:03:09.710 --> 00:03:12.509 guessing, right? Especially in a big project, 00:03:12.530 --> 00:03:15.129 or one that changes a lot. Those guesses about 00:03:15.129 --> 00:03:17.930 where files are, that leads to wrong suggestions, 00:03:18.090 --> 00:03:21.490 broken code, just misunderstanding what you even 00:03:21.490 --> 00:03:23.909 want. Giving it the structure avoids all that 00:03:23.909 --> 00:03:26.610 hassle. So basically, it prevents AI from making 00:03:26.610 --> 00:03:28.949 assumptions about file paths. That's the core 00:03:28.949 --> 00:03:30.990 of it. OK, so that's where things are. What about 00:03:30.990 --> 00:03:33.460 how they got there? The history. Good point. 00:03:33.819 --> 00:03:36.419 That leads us straight into Git context. This 00:03:36.419 --> 00:03:39.280 comes from the Git MCP idea, an AI that knows 00:03:39.280 --> 00:03:42.080 your commit history, branches, recent changes, 00:03:42.439 --> 00:03:44.800 the whole story of the code. Like it's been pair 00:03:44.800 --> 00:03:46.479 programming with you the whole time. Kinda, yeah. 00:03:46.539 --> 00:03:48.520 So the practical step here is running commands 00:03:48.520 --> 00:03:52.039 like git log or git diff locally, and then just 00:03:52.039 --> 00:03:54.259 copying that output into the prompt. Exactly. 00:03:54.340 --> 00:03:57.599 You give the AI a clear snapshot of the project's 00:03:57.599 --> 00:03:59.699 evolution. OK, so an example might be... You 00:03:59.699 --> 00:04:01.939 can say, here's the commit history between tag 00:04:01.939 --> 00:04:04.659 v1 .2 .0 and the current main branch. Then paste 00:04:04.659 --> 00:04:08.379 the output of, say, git log v1 .2 .0 .head1 line. 00:04:08.520 --> 00:04:11.680 Then ask, based only on these commits, please 00:04:11.680 --> 00:04:14.879 draft some concise release notes for v1 .3 .0. 00:04:15.240 --> 00:04:17.339 Categorize them into new features, bug fixes, 00:04:17.639 --> 00:04:19.740 and improvements. Nice. You're giving it the 00:04:19.740 --> 00:04:22.139 actual changelog, not just bits of code. So how 00:04:22.139 --> 00:04:24.759 does this historical context make the AI a better 00:04:24.759 --> 00:04:27.060 collaborator? Well, it shifts it from just generating 00:04:27.060 --> 00:04:31.500 code to being more like a knowledgeable teammate. 00:04:31.620 --> 00:04:34.360 OK. It sees the why behind changes, the problem 00:04:34.360 --> 00:04:36.160 solved, the direction you're heading. Right. 00:04:36.220 --> 00:04:38.300 So the code it suggests, or the summaries it 00:04:38.300 --> 00:04:40.100 writes, they're just much more relevant because 00:04:40.100 --> 00:04:42.519 it gets the intent. It lets AI understand project 00:04:42.519 --> 00:04:44.860 evolution and recent changes. Yeah, which is 00:04:44.860 --> 00:04:47.040 crucial. Which leads nicely into the next idea, 00:04:47.300 --> 00:04:50.540 creating a kind of memory bank. Ah, yes. The 00:04:50.540 --> 00:04:53.620 Memory Bank, the concept, the Memory Bank MCP, 00:04:53.779 --> 00:04:56.680 was this server that holds all your project decisions, 00:04:56.819 --> 00:04:59.019 your coding conventions, architectural rules, 00:04:59.019 --> 00:05:01.180 and it remembers them across sessions. Dream 00:05:01.180 --> 00:05:03.360 it, right? No more repeating yourself. Definitely 00:05:03.360 --> 00:05:06.459 the dream. So how do we mimic that today? You 00:05:06.459 --> 00:05:09.319 can use features like custom instructions if 00:05:09.319 --> 00:05:12.420 your AI tool supports them. Or, and this works 00:05:12.420 --> 00:05:14.920 anywhere, just start every conversation with 00:05:14.920 --> 00:05:18.240 a really clear system prompt. Okay. Lay down 00:05:18.240 --> 00:05:21.490 the law from the start. Establish the project's 00:05:21.490 --> 00:05:24.230 DNA for the AI. So a system prompt might look 00:05:24.230 --> 00:05:26.350 like. Yeah, like right at the top. System prompt. 00:05:27.110 --> 00:05:29.069 Okay, for this whole chat, stick to these project 00:05:29.069 --> 00:05:32.689 rules. One, language is KypeScript. Two, use 00:05:32.689 --> 00:05:36.209 Axios for all API calls. Three, React components 00:05:36.209 --> 00:05:39.040 are function components with hooks. Clear rule. 00:05:39.360 --> 00:05:41.079 Now, help me create a user profile component. 00:05:41.600 --> 00:05:43.600 I have to admit, I still wrestle with prompt 00:05:43.600 --> 00:05:46.220 drift myself sometimes. Keeping the AI consistently 00:05:46.220 --> 00:05:49.319 on track can be genuinely tough. Well, for sure, 00:05:49.319 --> 00:05:52.180 it happens. But this memory bank approach, setting 00:05:52.180 --> 00:05:54.939 it up, what's the biggest benefit? Does it outweigh 00:05:54.939 --> 00:05:56.899 that initial effort? That's a fair question. 00:05:57.000 --> 00:05:58.560 It does take a little setup. But think about 00:05:58.560 --> 00:06:01.639 how much time you spend correcting the AI later, 00:06:01.779 --> 00:06:04.660 right? Re -explaining conventions, fixing code 00:06:04.660 --> 00:06:06.879 that doesn't fit the pattern. That drift adds 00:06:06.879 --> 00:06:09.980 up. So this memory bank approach, it front loads 00:06:09.980 --> 00:06:13.300 that effort, but it pays off over time with consistency. 00:06:13.519 --> 00:06:15.959 It ensures consistent AI behavior across all 00:06:15.959 --> 00:06:18.439 interactions. Exactly. Consistency is key. And 00:06:18.439 --> 00:06:21.540 speaking of keys, knowing how the pieces connect. 00:06:21.759 --> 00:06:24.220 Absolutely. Which brings us to analyzing code 00:06:24.220 --> 00:06:27.060 relationships. This is inspired by the knowledge 00:06:27.060 --> 00:06:30.339 graph memory idea, an AI server that automatically 00:06:30.339 --> 00:06:32.720 maps out all the dependencies. Which functions 00:06:32.720 --> 00:06:35.800 call which? Which components use others? Like 00:06:35.800 --> 00:06:38.040 a big interconnected web. Wow, that would be 00:06:38.040 --> 00:06:39.920 powerful. How do we approximate that? We can 00:06:39.920 --> 00:06:42.699 use our IDEs, actually, features like find all 00:06:42.699 --> 00:06:45.540 references or go to definition. Ah, okay. Use 00:06:45.540 --> 00:06:48.079 the tools we already have. Right. Get that dependency 00:06:48.079 --> 00:06:50.839 info and then feed it to the AI. So you could 00:06:50.839 --> 00:06:53.800 say... You might prompt it like this. Okay, I'm 00:06:53.800 --> 00:06:56.660 refactoring the Calculaprice function in utilscalculations 00:06:56.660 --> 00:07:00.240 .js. My IDE shows it's used in the components 00:07:00.240 --> 00:07:05.139 card .js, page checkout .js, and order service 00:07:05.139 --> 00:07:08.339 .js. Based on that, what are the potential risks 00:07:08.339 --> 00:07:11.139 here, and what steps should I take to avoid breaking 00:07:11.139 --> 00:07:13.519 things during the refactor? Okay, so you're giving 00:07:13.519 --> 00:07:16.399 it the dependency map. Why is understanding these 00:07:16.399 --> 00:07:18.620 links so important for the AI? Because it lets 00:07:18.620 --> 00:07:21.399 the AI predict the ripple effects. If it knows 00:07:21.399 --> 00:07:23.459 everywhere that calculate price function is used, 00:07:23.759 --> 00:07:25.600 it can warn you before you break something downstream. 00:07:25.920 --> 00:07:28.180 Oh, yeah. It can suggest safer ways to refactor. 00:07:28.319 --> 00:07:31.220 It shifts it from just coding to being more preventative. 00:07:31.720 --> 00:07:34.420 It helps AI foresee impact and avoid breaking 00:07:34.420 --> 00:07:36.379 things. Which is always good, less debugging. 00:07:36.620 --> 00:07:38.920 What about external info, like docs? Yeah, good 00:07:38.920 --> 00:07:41.420 question. Technique five is about retrieving 00:07:41.420 --> 00:07:44.910 web and API documentation. This comes from the 00:07:44.910 --> 00:07:48.310 Fetching Context 7 MCP ideas. The vision was 00:07:48.310 --> 00:07:50.329 an AI that could just pull the latest docs for 00:07:50.329 --> 00:07:53.149 any library, automatically. Instant up -to -date 00:07:53.149 --> 00:07:55.430 info. Okay, so for now, it's on us. For now, 00:07:55.529 --> 00:07:58.350 yeah. Instead of hoping the AI finds the right 00:07:58.350 --> 00:08:01.110 docs, or worse, makes something up. Hallucinates. 00:08:01.230 --> 00:08:03.230 Exactly. You go to the official docs yourself, 00:08:03.769 --> 00:08:05.449 copy the relevant bit, could be text, could be 00:08:05.449 --> 00:08:07.750 a code example, and paste that directly into 00:08:07.750 --> 00:08:09.529 your prompt. You're curating the info. Okay, 00:08:09.550 --> 00:08:12.620 so like... I'm using chart .js version 4 .4. 00:08:12.699 --> 00:08:15.220 Here's the official docs snippet for donut charts. 00:08:16.199 --> 00:08:18.980 Then you paste the snippet. Right. And then based 00:08:18.980 --> 00:08:21.500 on this specific documentation, write the code 00:08:21.500 --> 00:08:25.100 for a donut chart showing this data. Sales 60, 00:08:25.360 --> 00:08:28.680 marketing 25, dev 15. You're giving it the ground 00:08:28.680 --> 00:08:31.360 truth. How does providing that specific curated 00:08:31.360 --> 00:08:34.379 info prevent those AI hallucinations? Well, it 00:08:34.379 --> 00:08:36.399 directly counters them. You're providing concrete 00:08:36.399 --> 00:08:39.100 facts from a trusted source. The AI doesn't have 00:08:39.100 --> 00:08:41.519 to guess or rely on possibly outdated training 00:08:41.519 --> 00:08:43.240 data. Right. It has the authoritative answer 00:08:43.240 --> 00:08:45.600 right there. Dramatically boosts accuracy and 00:08:45.600 --> 00:08:48.759 reliability. It provides concrete factual knowledge 00:08:48.759 --> 00:08:51.100 for the AI. Makes sense. Less time debugging 00:08:51.100 --> 00:08:54.200 the AIs on the app than the creativity. Exactly. 00:08:54.960 --> 00:08:58.659 OK, next up, leveraging intelligent search. This 00:08:58.659 --> 00:09:01.720 stems from the Tavoli MCP concept, a specialized 00:09:01.720 --> 00:09:04.379 search engine for developers. Understood code? 00:09:04.570 --> 00:09:06.789 problems. That sounds amazing. So how do we get 00:09:06.789 --> 00:09:09.009 close to that now? If you're using an AI with 00:09:09.009 --> 00:09:13.509 web browsing built in, like Gemini or the latest 00:09:13.509 --> 00:09:16.730 chat GPT, you can guide it search. Don't just 00:09:16.730 --> 00:09:19.049 ask a generic question. Tell it how to search. 00:09:19.529 --> 00:09:22.190 Instruct it to search, synthesize, and pull answers 00:09:22.190 --> 00:09:24.529 from reliable sources only. Oh, OK. Like giving 00:09:24.529 --> 00:09:26.250 it research instructions. Pretty much. So you 00:09:26.250 --> 00:09:28.580 could say, I'm getting a course. Pre -flight 00:09:28.580 --> 00:09:31.620 request did not succeed error with AWS Lambda 00:09:31.620 --> 00:09:34.779 and API Gateway. Please search the web, specifically 00:09:34.779 --> 00:09:37.820 looking for official AWS documentation and reputable 00:09:37.820 --> 00:09:40.259 tech articles on configuring cores for this setup. 00:09:40.720 --> 00:09:42.580 Then summarize the top three common solutions 00:09:42.580 --> 00:09:45.110 you find. OK, so it's targeted problem solving. 00:09:45.330 --> 00:09:46.830 What's the key difference between doing that 00:09:46.830 --> 00:09:49.370 and just Googling it myself? The synthesis part 00:09:49.370 --> 00:09:52.409 is key. The AI isn't just giving you links. Right. 00:09:52.549 --> 00:09:54.950 It's reading multiple sources, figuring out the 00:09:54.950 --> 00:09:57.250 common patterns, the most frequent fixes, and 00:09:57.250 --> 00:09:59.529 presenting a concise summary. Distilling it down. 00:09:59.590 --> 00:10:02.509 Yeah. It's a focused, solution -driven search, 00:10:02.590 --> 00:10:04.970 not just information retrieval. It's a focused, 00:10:05.190 --> 00:10:07.889 synthesis -driven search, specifically for solutions. 00:10:07.970 --> 00:10:09.710 That could save a lot of time sifting through 00:10:09.710 --> 00:10:12.490 search results. Definitely. OK, number seven. 00:10:13.210 --> 00:10:16.450 Connecting personal notes. Inspired by Obsidian 00:10:16.450 --> 00:10:20.470 MCP, the idea of an AI plugging directly into 00:10:20.470 --> 00:10:23.009 your personal knowledge base. Your notes, your 00:10:23.009 --> 00:10:25.529 requirements. Your brain, basically. Sort of, 00:10:25.610 --> 00:10:27.509 yeah. How do we do that, practically? You just 00:10:27.509 --> 00:10:30.019 copy and paste. Open your notes app, Obsidian, 00:10:30.200 --> 00:10:32.759 Notion, Plain Text, whatever. Find the notes 00:10:32.759 --> 00:10:34.620 relevant to what you're working on, copy them, 00:10:34.720 --> 00:10:37.120 paste them into the prompt. Give the AI a peek 00:10:37.120 --> 00:10:39.279 into your specific thoughts. Exactly. So you 00:10:39.279 --> 00:10:41.340 could write, here are my notes on the login page 00:10:41.340 --> 00:10:44.080 requirements, and paste your bullet points, email 00:10:44.080 --> 00:10:47.779 password fields, standard email validation, password 00:10:47.779 --> 00:10:51.360 eight chars, Google sign in button, then ask. 00:10:52.340 --> 00:10:55.000 Based only on these requirements, generate the 00:10:55.000 --> 00:10:58.299 basic HTML and CSS for this form. How does giving 00:10:58.299 --> 00:11:00.080 it these personal notes, which aren't really 00:11:00.080 --> 00:11:02.159 code, how does that personalize the AIs help? 00:11:02.580 --> 00:11:05.919 It makes the output truly specific to your project, 00:11:06.100 --> 00:11:09.279 your thinking. Often, the little details, the 00:11:09.279 --> 00:11:11.700 nuances, the specific constraints, they only 00:11:11.700 --> 00:11:15.059 live in those notes. Giving the AI that context 00:11:15.059 --> 00:11:17.200 helps to generate something that actually matches 00:11:17.200 --> 00:11:20.399 your vision, not just a generic template. AI 00:11:20.399 --> 00:11:23.460 gains insight into your unique project specifics 00:11:23.460 --> 00:11:25.580 and thoughts. That's really powerful for getting 00:11:25.580 --> 00:11:28.659 tailored results. Now, what about really complex 00:11:28.659 --> 00:11:31.340 tasks? Good transition. Let's talk about applying 00:11:31.340 --> 00:11:33.789 sequential thinking. This comes from the sequential 00:11:33.789 --> 00:11:37.809 thinking MCP idea, an AI that automatically breaks 00:11:37.809 --> 00:11:41.350 down big problems into logical steps, like how 00:11:41.350 --> 00:11:43.669 an experienced developer thinks. OK. How do we 00:11:43.669 --> 00:11:46.009 encourage that? This one's actually super easy 00:11:46.009 --> 00:11:48.429 to apply and really effective. You just explicitly 00:11:48.429 --> 00:11:51.710 ask the AI to think step by step or break down 00:11:51.710 --> 00:11:53.350 the problem first. Just tell it to slow down 00:11:53.350 --> 00:11:55.509 and show its work. Pretty much. It prompts it 00:11:55.509 --> 00:11:57.649 to use a more structured approach and actually 00:11:57.649 --> 00:11:59.929 explain its reasoning. So an example. You could 00:11:59.929 --> 00:12:02.190 say, I need to build a shopping cart feature. 00:12:02.350 --> 00:12:05.389 Please think step -by -step and outline a detailed 00:12:05.389 --> 00:12:08.970 plan. Include, one, the data structure for the 00:12:08.970 --> 00:12:13.889 cart. Two, key API endpoints needed. Add, remove, 00:12:14.070 --> 00:12:18.009 update. Three, the main client -side logic. Ah, 00:12:18.169 --> 00:12:21.549 okay. Forces it to be methodical. Does asking 00:12:21.549 --> 00:12:23.870 it to think step -by -step actually make the 00:12:23.870 --> 00:12:26.730 AI, like, smarter? or is something else going 00:12:26.730 --> 00:12:29.169 on? It's less about making it intrinsically smarter 00:12:29.169 --> 00:12:31.330 and more about forcing it to show its process 00:12:31.330 --> 00:12:33.990 and follow a logical path. Okay. It reduces the 00:12:33.990 --> 00:12:35.850 chance of it jumping to conclusions or skipping 00:12:35.850 --> 00:12:38.190 crucial steps. Right. It guides the AI towards 00:12:38.190 --> 00:12:40.470 outputs that are more structured, more reliable, 00:12:40.870 --> 00:12:43.070 easier for us to understand and verify. Less 00:12:43.070 --> 00:12:45.570 rework. It guides the AI toward more structured, 00:12:45.570 --> 00:12:47.909 reliable outputs. That are more trustworthy code. 00:12:48.090 --> 00:12:49.990 Makes sense. Okay. Technique 9 deals with something 00:12:49.990 --> 00:12:52.450 critical. Handling environment variables securely. 00:12:52.759 --> 00:12:55.659 based on the InVault MCP concept, an AI server 00:12:55.659 --> 00:12:57.639 that helps access secrets and environment variables 00:12:57.639 --> 00:12:59.620 without ever exposing them. Super important. 00:12:59.779 --> 00:13:01.980 And the rule today is absolutely critical, right? 00:13:02.240 --> 00:13:06.799 Paramount. You never, ever paste API keys, passwords, 00:13:07.100 --> 00:13:10.500 any secrets directly into an AI prompt. Full 00:13:10.500 --> 00:13:13.399 stop. Never, ever. So what do we do? You instruct 00:13:13.399 --> 00:13:16.309 the AI. to write code that retrieves secrets 00:13:16.309 --> 00:13:18.750 from environment variables using standard methods, 00:13:19.250 --> 00:13:23.830 like process .env in Node or os .getinv in Python. 00:13:24.129 --> 00:13:27.090 OK. Tell it how to get the secret securely. Exactly. 00:13:27.269 --> 00:13:29.450 So a prompt might be, write a Python function 00:13:29.450 --> 00:13:32.850 to call the OpenAI API. Do not put the API key 00:13:32.850 --> 00:13:35.110 in the code. Read it from an environment variable 00:13:35.110 --> 00:13:37.909 called OpenIAPI key. If that variable isn't set, 00:13:37.990 --> 00:13:40.090 the function should raise an error. Perfect. 00:13:40.289 --> 00:13:42.850 Keeps the secret safe, out of the code, out of 00:13:42.850 --> 00:13:46.009 the prompt. Ooh. Imagine securely managing secrets 00:13:46.009 --> 00:13:48.470 for like a billion AI queries automatically, 00:13:48.769 --> 00:13:51.190 never risking exposure. That scale is mind -boggling. 00:13:51.269 --> 00:13:53.529 It really is. But even on our scale, why is the 00:13:53.529 --> 00:13:56.269 security aspect just so vital right now? It's 00:13:56.269 --> 00:13:58.769 just fundamental secure coding practice. Hard 00:13:58.769 --> 00:14:00.769 coding secrets or accidentally pasting them into 00:14:00.769 --> 00:14:03.370 a prompt, that's how major breaches happen. Compromised 00:14:03.370 --> 00:14:06.730 accounts, stolen data. So teaching the AI and 00:14:06.730 --> 00:14:09.720 ourselves to use environment variables. properly 00:14:09.720 --> 00:14:12.200 protects that sensitive info. It prevents those 00:14:12.200 --> 00:14:14.600 potentially catastrophic, really costly mistakes. 00:14:15.039 --> 00:14:17.159 It's about building secure habits. It protects 00:14:17.159 --> 00:14:19.860 sensitive information, preventing costly data 00:14:19.860 --> 00:14:21.919 breaches. Absolutely priceless in the long run. 00:14:22.039 --> 00:14:24.509 OK, one more. Last one, number 10. Integrating 00:14:24.509 --> 00:14:28.250 API contracts. Inspired by APSpec MCP, an AI 00:14:28.250 --> 00:14:30.750 that could read, say, an OpenAPI or Swagger file 00:14:30.750 --> 00:14:33.590 to know exactly how an API endpoint works. The 00:14:33.590 --> 00:14:36.210 precise structure, requests, responses. Exactly. 00:14:36.230 --> 00:14:38.250 So practically, we copy the relevant part of 00:14:38.250 --> 00:14:40.950 our spec file. That's the idea. Find the specific 00:14:40.950 --> 00:14:43.929 endpoint definition in your OpenAPI .json or 00:14:43.929 --> 00:14:47.389 Swagger .yaml. Copy that whole definition, the 00:14:47.389 --> 00:14:49.990 YAML or JSON block. Paste it directly into the 00:14:49.990 --> 00:14:52.610 prompt. Give the AI the undeniable contract for 00:14:52.610 --> 00:14:54.980 that endpoint. OK, so, like, I need a JavaScript 00:14:54.980 --> 00:14:57.559 function. Here's the OpenATI 3 .0 spec for the 00:14:57.559 --> 00:14:59.659 user's user endpoint. Then paste the spec lock. 00:14:59.759 --> 00:15:02.960 Right. Then, based strictly on this spec, write 00:15:02.960 --> 00:15:05.179 a fetch function called getUserBide that takes 00:15:05.179 --> 00:15:07.960 userIde. Make sure it handles the 200 success 00:15:07.960 --> 00:15:11.019 response and the 404 error case correctly, as 00:15:11.019 --> 00:15:14.720 defined. Got it. How does giving it that explicit 00:15:14.720 --> 00:15:18.799 API contract make the interaction more precise. 00:15:18.960 --> 00:15:21.039 It just removes all the ambiguity. The AI doesn't 00:15:21.039 --> 00:15:23.639 have to guess about parameters or what the response 00:15:23.639 --> 00:15:26.259 looks like or which status codes mean what. It 00:15:26.259 --> 00:15:28.580 has the blueprint. Exactly. The code it generates 00:15:28.580 --> 00:15:30.980 for talking to your API will be perfectly aligned 00:15:30.980 --> 00:15:34.100 with your backend. Fewer integration bugs, smoother 00:15:34.100 --> 00:15:37.919 development. It gives AI an exact blueprint for 00:15:37.919 --> 00:15:41.570 API communication. So that's 10 techniques, 10 00:15:41.570 --> 00:15:44.250 ways to give our AI better context, all inspired 00:15:44.250 --> 00:15:48.389 by those future -facing MCP server ideas. And 00:15:48.389 --> 00:15:51.490 the big idea, the main takeaway here, is pretty 00:15:51.490 --> 00:15:53.490 simple, really. You don't need to wait for those 00:15:53.490 --> 00:15:55.549 futuristic servers. You can effectively be the 00:15:55.549 --> 00:15:58.019 MCP server for your AI. right now. That's it, 00:15:58.059 --> 00:16:00.539 exactly. The power isn't in waiting for new tools, 00:16:00.639 --> 00:16:03.019 it's in how we use the tools we have. Adopting 00:16:03.019 --> 00:16:05.120 that context -first mindset every time you write 00:16:05.120 --> 00:16:06.919 a prompt, it changes everything. And when you 00:16:06.919 --> 00:16:09.240 start doing this, providing file structure, get 00:16:09.240 --> 00:16:11.879 history, that memory bank system prompt, the 00:16:11.879 --> 00:16:14.399 API specs, you will see a difference, a really 00:16:14.399 --> 00:16:16.539 noticeable difference. The quality of the AI's 00:16:16.539 --> 00:16:20.460 output goes way up, and your productivity. It 00:16:20.460 --> 00:16:22.419 really does get a significant boost. It's about 00:16:22.419 --> 00:16:25.159 having a richer, more informed, and just more 00:16:25.159 --> 00:16:27.960 reliable conversation with these tools. And maybe 00:16:27.960 --> 00:16:30.200 a final thought to leave you with. What if, by 00:16:30.200 --> 00:16:32.539 giving AI this deeper context, we're not just 00:16:32.539 --> 00:16:35.740 making the AI more useful, but we're also subtly 00:16:35.740 --> 00:16:38.240 changing how we approach problem solving, how 00:16:38.240 --> 00:16:39.720 we structure our own thoughts about the code? 00:16:40.080 --> 00:16:43.279 That's deep. But yeah, maybe. I'd say just try 00:16:43.279 --> 00:16:45.000 one or two of these techniques. Next time you're 00:16:45.000 --> 00:16:46.960 working with an AI, pick one, give it a shot, 00:16:47.159 --> 00:16:48.740 see how it feels, see what happens. Thank you 00:16:48.740 --> 00:16:50.940 for joining us for this deep dive on mastering 00:16:50.940 --> 00:16:53.379 AI context. We really hope it helps you build 00:16:53.379 --> 00:16:55.940 smarter, build faster, and maybe with a little 00:16:55.940 --> 00:16:56.559 less frustration.