WEBVTT 00:00:00.000 --> 00:00:02.859 Okay, welcome back to the Deep Dive. You sent 00:00:02.859 --> 00:00:05.040 us a really interesting stack of sources, and 00:00:05.040 --> 00:00:08.119 there's one in here that it really grabbed me. 00:00:08.160 --> 00:00:10.519 It talks about something that sounds pretty major 00:00:10.519 --> 00:00:12.160 happening right now in software development. 00:00:12.400 --> 00:00:16.379 Yeah, it describes it as this, you know, shocking 00:00:16.379 --> 00:00:20.820 yet surprisingly quiet revolution. It feels big, 00:00:21.000 --> 00:00:24.019 but maybe not everyone's talking about it loudly 00:00:24.019 --> 00:00:26.949 just yet. Right. Because we hear about AI doing 00:00:26.949 --> 00:00:29.809 a lot of stuff like writing code or helping debug. 00:00:30.089 --> 00:00:34.130 But the core idea here is, well, it's a bit wild. 00:00:34.490 --> 00:00:36.630 It suggests you might not actually need to build 00:00:36.630 --> 00:00:39.070 that traditional visual front end anymore. Because 00:00:39.070 --> 00:00:41.869 the AI becomes your front end. Yes. The AI becomes 00:00:41.869 --> 00:00:43.929 your front end. That's the headline. So our mission 00:00:43.929 --> 00:00:46.109 here is to dive into this. What exactly does 00:00:46.109 --> 00:00:47.950 that mean? What are the technologies driving 00:00:47.950 --> 00:00:50.210 this? We need to look at what's actually happening 00:00:50.210 --> 00:00:52.329 today. And like, what does this potentially mean 00:00:52.329 --> 00:00:54.170 for you? How you build applications, how you 00:00:54.170 --> 00:00:56.130 interact with them? Exactly. It's a different 00:00:56.130 --> 00:00:57.909 way to frame the whole thing. So traditionally, 00:00:58.109 --> 00:00:59.950 let's think about the front end developer role 00:00:59.950 --> 00:01:02.869 for a second. The sources mentioned this is often, 00:01:02.969 --> 00:01:04.629 you know, the most time consuming, sometimes 00:01:04.629 --> 00:01:06.890 frustrating part of the job. What's involved 00:01:06.890 --> 00:01:08.489 there? Oh, you know, it's building the whole 00:01:08.489 --> 00:01:11.469 visual layer, the buttons, the forms, managing 00:01:11.469 --> 00:01:13.909 application state, wrestling with CSS to get 00:01:13.909 --> 00:01:15.950 everything just right, pixel perfect, dealing 00:01:15.950 --> 00:01:19.890 with like browser compatibility across. tons 00:01:19.890 --> 00:01:22.489 of different devices. It's incredibly detailed 00:01:22.489 --> 00:01:24.530 work, right? It's like building this bespoke 00:01:24.530 --> 00:01:26.730 little house just for the user to live and click 00:01:26.730 --> 00:01:29.269 in. Pretty much. But the central argument here 00:01:29.269 --> 00:01:32.129 is that you have this convergence now of powerful 00:01:32.129 --> 00:01:35.950 AI models and importantly, this new interoperability 00:01:35.950 --> 00:01:39.230 protocol called MCP that suggests a really fundamental 00:01:39.230 --> 00:01:41.750 shift away from that traditional model. Okay, 00:01:41.810 --> 00:01:44.049 MCP, we definitely need to unpack that. But the 00:01:44.049 --> 00:01:46.049 shift itself, you're saying it's moving from 00:01:46.049 --> 00:01:48.569 building those visual interfaces for humans to 00:01:48.569 --> 00:01:50.769 building structured capabilities, basically API 00:01:50.769 --> 00:01:53.609 functions, for AI agents to control directly. 00:01:53.890 --> 00:01:56.189 That's the heart of it. Instead of building that 00:01:56.189 --> 00:01:58.569 layer that translates human clicks into backend 00:01:58.569 --> 00:02:01.290 actions, you're focusing on making the backend 00:02:01.290 --> 00:02:04.530 actions themselves directly consumable and understandable 00:02:04.530 --> 00:02:07.549 by an AI. I love the analogy the article uses 00:02:07.549 --> 00:02:11.289 here. It's like moving from being an instrumentalist, 00:02:11.289 --> 00:02:14.310 playing one specific piece of the orchestra perfectly, 00:02:14.490 --> 00:02:17.710 to being the conductor. Right. You're orchestrating 00:02:17.710 --> 00:02:20.509 this whole symphony of AI and APIs. It's definitely 00:02:20.509 --> 00:02:22.909 operating at a higher level of abstraction. Yeah. 00:02:23.229 --> 00:02:25.169 Okay, let's talk about APIs for a second because 00:02:25.169 --> 00:02:28.750 the material is clear. APIs are really the foundation 00:02:28.750 --> 00:02:32.550 of pretty much all modern apps, right? Web, mobile, 00:02:32.750 --> 00:02:35.590 desktop. They're all using APIs to talk. They 00:02:35.590 --> 00:02:37.729 are. They're the digital glue, the messengers 00:02:37.729 --> 00:02:40.750 behind the scenes. You have internal APIs, which 00:02:40.750 --> 00:02:43.150 are your app's core logic, saving a document, 00:02:43.330 --> 00:02:45.710 creating a user, calculating something. Right. 00:02:45.810 --> 00:02:47.870 And then external APIs, which connect you to 00:02:47.870 --> 00:02:51.069 outside services. Google Maps, Stripe, open weather 00:02:51.069 --> 00:02:53.189 map, that kind of thing. And the traditional 00:02:53.189 --> 00:02:56.349 front end, that visual layer, has been that key 00:02:56.349 --> 00:02:58.909 visual bridge, right? Taking human action, like 00:02:58.909 --> 00:03:01.930 clicking submit and translating it into the correctly 00:03:01.930 --> 00:03:04.330 formatted API call for the back end. Right. It's 00:03:04.330 --> 00:03:06.430 been that translator for user actions. But as 00:03:06.430 --> 00:03:09.210 the sources point out, MCP, model control protocol, 00:03:09.409 --> 00:03:12.330 is the technology that really changes this equation. 00:03:12.650 --> 00:03:15.110 So what is MCP? Like, what does it do that's 00:03:15.110 --> 00:03:17.240 different? The article calls it a standardized 00:03:17.240 --> 00:03:21.479 way to make your APIs agent controllable. Think 00:03:21.479 --> 00:03:23.560 of it kind of like a universal translator mixed 00:03:23.560 --> 00:03:25.379 with an instruction manual, but specifically 00:03:25.379 --> 00:03:28.930 for an AI. A standard ATI speaks a very technical 00:03:28.930 --> 00:03:32.930 language, right? HTTP, JSON. An AI, which processes 00:03:32.930 --> 00:03:35.490 things in natural language, doesn't intuitively 00:03:35.490 --> 00:03:38.250 grasp how to use that rigid structure to take 00:03:38.250 --> 00:03:41.469 action. So MCP bridges that gap. Exactly. It 00:03:41.469 --> 00:03:43.629 presents your API's capabilities, your functions 00:03:43.629 --> 00:03:46.289 to the AI agent in a way that the AI can understand 00:03:46.289 --> 00:03:48.669 and leverage. It essentially tells the AI, hey, 00:03:48.729 --> 00:03:50.669 I have these tools you can use. There's one called 00:03:50.669 --> 00:03:52.710 GetWeather4City. If you want to use it, you need 00:03:52.710 --> 00:03:54.909 to give me a city name. Do that and I'll give 00:03:54.909 --> 00:03:57.000 you back the weather data. Oh, OK. So it's not 00:03:57.000 --> 00:03:59.099 just giving the AI data to read, which is more 00:03:59.099 --> 00:04:02.580 like ARID's retrieval augmented generation. Yeah, 00:04:02.680 --> 00:04:04.439 that's a really key distinction the material 00:04:04.439 --> 00:04:07.400 makes. ARID is about giving the AI access to 00:04:07.400 --> 00:04:11.060 data. MCP is about giving the AI access to capability. 00:04:11.460 --> 00:04:15.099 The control part is crucial. It moves the AI 00:04:15.099 --> 00:04:17.660 from just being able to read and synthesize information 00:04:17.660 --> 00:04:21.060 to being able to actively do things by calling 00:04:21.060 --> 00:04:24.189 APIs. That makes total sense. Because if an AI 00:04:24.189 --> 00:04:26.350 agent can understand what you want in natural 00:04:26.350 --> 00:04:29.509 language, what's the weather in Paris? Or find 00:04:29.509 --> 00:04:31.129 me the best price for this thing and buy it. 00:04:31.170 --> 00:04:33.350 And then it can intelligently figure out which 00:04:33.350 --> 00:04:37.170 API tool it needs and use MCB to call it. You're 00:04:37.170 --> 00:04:39.529 potentially just bypassing the need for a visual 00:04:39.529 --> 00:04:41.649 interface entirely for that task. It kind of 00:04:41.649 --> 00:04:44.750 melts away the need to navigate a GUI by clicking 00:04:44.750 --> 00:04:47.350 around. Exactly. Which brings us back to the 00:04:47.350 --> 00:04:49.449 developer role, the whole instrumentalist versus 00:04:49.449 --> 00:04:52.009 conductor idea. What does the conductor actually 00:04:52.009 --> 00:04:55.310 do in this new AI -centric world? Well, the traditional 00:04:55.310 --> 00:04:58.269 instrumentalist was focused on perfecting their 00:04:58.269 --> 00:05:00.509 specific instrument, the pixel -perfect UI, the 00:05:00.509 --> 00:05:03.129 state management code. The conductor's focus 00:05:03.129 --> 00:05:05.050 is much broader. They're not playing every note. 00:05:05.089 --> 00:05:06.649 They're leading the whole performance. They're 00:05:06.649 --> 00:05:08.680 not building every button themselves. No. the 00:05:08.680 --> 00:05:11.660 sources break their role down into like four 00:05:11.660 --> 00:05:15.139 key parts first they write the score which is 00:05:15.139 --> 00:05:17.939 really about api design designing clean logical 00:05:17.939 --> 00:05:21.160 apis that represent the core actions or capabilities 00:05:21.160 --> 00:05:23.920 of your application that's the fundamental music 00:05:24.379 --> 00:05:26.019 Got it. And then they recruit the musicians, 00:05:26.139 --> 00:05:29.160 which is selecting tools. Right. Choosing which 00:05:29.160 --> 00:05:31.220 internal and external APIs are going to be part 00:05:31.220 --> 00:05:33.240 of your orchestra. Maybe Stripe is your percussion 00:05:33.240 --> 00:05:35.540 section for payments. And Google Maps handles 00:05:35.540 --> 00:05:38.079 the strings with location data, you know. I like 00:05:38.079 --> 00:05:40.790 that analogy a lot. And then. provide the sheet 00:05:40.790 --> 00:05:44.089 music, which is MCP generation. Precisely. You 00:05:44.089 --> 00:05:47.209 wrap those APIs in MCP servers. You're basically 00:05:47.209 --> 00:05:49.970 creating the clear, standardized manual that 00:05:49.970 --> 00:05:52.689 tells the AI conductor exactly how to use each 00:05:52.689 --> 00:05:54.209 one of those instruments, each of those tools. 00:05:54.550 --> 00:05:57.399 And finally... Lead the performance, which is 00:05:57.399 --> 00:05:59.660 orchestration and prompting. This is where the 00:05:59.660 --> 00:06:02.180 AI agent is doing the work, but the developer 00:06:02.180 --> 00:06:04.680 is still directing it. Yes. You're guiding the 00:06:04.680 --> 00:06:07.180 AI with higher level goals, strategies, constraints. 00:06:07.680 --> 00:06:10.540 You're enabling it to intelligently chain those 00:06:10.540 --> 00:06:14.240 API musicians together to perform complex tasks 00:06:14.240 --> 00:06:17.600 based on user requests. So the core process described 00:06:17.600 --> 00:06:22.279 is you wrap your APIs as an MCP server. You make 00:06:22.279 --> 00:06:24.810 it accessible to the AI agent. like running it 00:06:24.810 --> 00:06:27.170 locally, you attach that server to the agent, 00:06:27.269 --> 00:06:29.009 and then you configure the agent so it knows 00:06:29.009 --> 00:06:31.170 about these new tools. That's the workflow, and 00:06:31.170 --> 00:06:33.089 it lets you build apps in a totally different 00:06:33.089 --> 00:06:35.689 way. Let's make this really concrete. The sources 00:06:35.689 --> 00:06:38.209 provide a step -by -step walkthrough for building 00:06:38.209 --> 00:06:41.250 a simple weather app using this method, but crucially, 00:06:41.250 --> 00:06:43.029 without building a traditional visual interface. 00:06:43.310 --> 00:06:44.930 This part's pretty neat. You want a weather app 00:06:44.930 --> 00:06:46.629 where you can just ask for the weather. Yeah. 00:06:46.689 --> 00:06:49.810 So step one, find and select your API. Okay, 00:06:49.850 --> 00:06:52.529 so the article suggests using a tool like Postman, 00:06:52.649 --> 00:06:55.750 which is popular for working with APIs. They 00:06:55.750 --> 00:06:58.730 apparently have an API network and MCC generator 00:06:58.730 --> 00:07:01.430 built in. This is where you select your instruments. 00:07:01.670 --> 00:07:04.470 Okay. You go into their network, search for, 00:07:04.509 --> 00:07:06.810 say, open weather map. And you'd find the relevant 00:07:06.810 --> 00:07:09.209 API collection and pick which specific functions, 00:07:09.329 --> 00:07:11.689 which endpoints you want the AI to be able to 00:07:11.689 --> 00:07:13.470 use, like getting the current weather or the 00:07:13.470 --> 00:07:15.990 forecast. You're choosing the capabilities. Exactly. 00:07:15.990 --> 00:07:17.970 You import those into your workspace. That's 00:07:17.970 --> 00:07:21.009 step one. Done. Okay, step two, generate the 00:07:21.009 --> 00:07:23.350 MCP server. This is where you get the sheet music. 00:07:23.730 --> 00:07:27.329 Using the MCP generator feature in Postman, you 00:07:27.329 --> 00:07:29.410 point it to that open weather map collection 00:07:29.410 --> 00:07:32.610 you just imported. You just hit generate. And 00:07:32.610 --> 00:07:34.550 it automatically writes server code for you. 00:07:34.610 --> 00:07:37.149 That's kind of wild. It is. It produces the necessary 00:07:37.149 --> 00:07:40.370 code for a Node .js server, including a file 00:07:40.370 --> 00:07:43.230 to manage dependencies and a template for configuration 00:07:43.230 --> 00:07:46.129 secrets. You download that package. Okay, so 00:07:46.129 --> 00:07:48.050 you have the code for the server. Step three. 00:07:48.399 --> 00:07:50.930 Set up the server on your machine. Right. You'd 00:07:50.930 --> 00:07:53.310 take that generated code, extract it, put it 00:07:53.310 --> 00:07:55.709 somewhere on your computer. You'd open a terminal 00:07:55.709 --> 00:07:59.089 and run a standard command like npm install to 00:07:59.089 --> 00:08:01.870 install any necessary supporting software based 00:08:01.870 --> 00:08:04.910 on the package Postman gave you. Got it. And 00:08:04.910 --> 00:08:07.209 there's that really crucial security step they 00:08:07.209 --> 00:08:09.970 emphasize here. Yes. Super important. You have 00:08:09.970 --> 00:08:12.949 a template file for configuration, often called 00:08:12.949 --> 00:08:16.350 .env .example or something. A critical step is 00:08:16.350 --> 00:08:19.970 renaming that to just .env and putting your sensitive 00:08:19.970 --> 00:08:22.850 information like your actual api key from open 00:08:22.850 --> 00:08:25.959 weather map in that file securely So you don't 00:08:25.959 --> 00:08:28.540 accidentally check your secret key into version 00:08:28.540 --> 00:08:31.480 control or share it? Absolutely critical. Without 00:08:31.480 --> 00:08:34.100 that key, the server can't actually talk to the 00:08:34.100 --> 00:08:36.139 open web and map service. Once that's configured, 00:08:36.299 --> 00:08:37.679 you just start the server running, typically 00:08:37.679 --> 00:08:39.580 from that same terminal window with something 00:08:39.580 --> 00:08:42.519 like NodeMCP server .js. And it would just run 00:08:42.519 --> 00:08:45.000 locally, like listening on a port, say, port 00:08:45.000 --> 00:08:47.179 3000. Exactly. You need to keep that running 00:08:47.179 --> 00:08:49.500 because your AI agent is going to talk to this 00:08:49.500 --> 00:08:52.000 local server. Right. Server's running. Step four, 00:08:52.240 --> 00:08:55.750 connect it to your AI agent. The source uses 00:08:55.750 --> 00:08:58.929 an AI native editor called Cursor as an example, 00:08:59.090 --> 00:09:02.009 but others might work too. Yeah. In an editor 00:09:02.009 --> 00:09:04.210 like that, you'd go into its settings and find 00:09:04.210 --> 00:09:06.570 where it manages external tools or servers. You'd 00:09:06.570 --> 00:09:08.570 add a new connection. You'd give it a name, say, 00:09:08.750 --> 00:09:11.990 My Weather Agent. Tell it where to find the server 00:09:11.990 --> 00:09:14.610 code you just set up. Maybe the command to start 00:09:14.610 --> 00:09:16.970 it if needed. So you're pointing the AI agent 00:09:16.970 --> 00:09:19.009 to the location of this new manual, this new 00:09:19.009 --> 00:09:21.610 set of tools. Precisely. You enable it. And the 00:09:21.610 --> 00:09:23.750 editor shows you that the A .I. now knows about 00:09:23.750 --> 00:09:26.730 specific tools like get weather from city name 00:09:26.730 --> 00:09:29.309 tool. The A .I. has been given its sheet music 00:09:29.309 --> 00:09:31.909 basically. And step five is the fun part. You 00:09:31.909 --> 00:09:34.559 test it. Using just natural language. The magic 00:09:34.559 --> 00:09:37.360 happens. You go to the AI chat interface within 00:09:37.360 --> 00:09:39.419 the editor or wherever you interact with the 00:09:39.419 --> 00:09:42.000 agent and just type your request. Like example, 00:09:42.259 --> 00:09:44.080 what's the current weather in London, UK? And 00:09:44.080 --> 00:09:46.419 give it to me in Celsius. Okay, let's quickly 00:09:46.419 --> 00:09:48.899 walk through the AI's thought process here, the 00:09:48.899 --> 00:09:51.940 orchestration bit. Sure. So the AI agent gets 00:09:51.940 --> 00:09:54.139 your natural language query. It figures out your 00:09:54.139 --> 00:09:56.440 intent. You want the weather. It looks through 00:09:56.440 --> 00:09:58.899 its available tools and sees the My Weather Agent 00:09:58.899 --> 00:10:01.179 connection offers a Get Weather from City Name 00:10:01.179 --> 00:10:04.419 tool. Okay. It then extracts the necessary details 00:10:04.419 --> 00:10:07.639 from your sentence, London, UK, for the city, 00:10:07.740 --> 00:10:11.200 and notes the request for Celsius. It understands 00:10:11.200 --> 00:10:13.080 the parameters just from your sentence. That's 00:10:13.080 --> 00:10:15.870 pretty cool. Yeah. That's the AI understanding 00:10:15.870 --> 00:10:18.509 the intent and mapping it to the tool's requirements. 00:10:19.029 --> 00:10:22.090 It then constructs a call to your local MCP server, 00:10:22.350 --> 00:10:24.909 invoking that specific tool with those parameters. 00:10:25.330 --> 00:10:28.710 Your MCP server receives the request, uses the 00:10:28.710 --> 00:10:31.350 secure OpenWeatherMap API key you configured, 00:10:31.490 --> 00:10:33.850 and makes the actual call to the OpenWeatherMap 00:10:33.850 --> 00:10:36.169 API out on the internet. OpenWeatherMap sends 00:10:36.169 --> 00:10:38.029 the data back to your local server. Which sends 00:10:38.029 --> 00:10:40.490 it back to the AI agent. The agent gets the raw 00:10:40.490 --> 00:10:43.049 data, parses it, and then synthesizes a human 00:10:43.049 --> 00:10:44.950 -friendly answer for you based on your original 00:10:44.950 --> 00:10:47.169 request. Something like, the current weather 00:10:47.169 --> 00:10:50.669 in London, UK is. No buttons, no input forms, 00:10:50.809 --> 00:10:53.870 no complex UI state management. Just pure logic, 00:10:54.110 --> 00:10:56.909 intent, and AI -driven functional orchestration. 00:10:57.460 --> 00:10:59.740 That quote from the source really hammers at 00:10:59.740 --> 00:11:01.960 home. You built a functional app interaction 00:11:01.960 --> 00:11:04.840 using the AI as the interface. It's a completely 00:11:04.840 --> 00:11:07.179 different user experience built on the back end 00:11:07.179 --> 00:11:09.519 capabilities, really. And the source talks about 00:11:09.519 --> 00:11:12.360 a major benefit of this, eliminating what they 00:11:12.360 --> 00:11:15.159 call UX debt or front end debt. What are we talking 00:11:15.159 --> 00:11:17.100 about there? It's all that time and effort we 00:11:17.100 --> 00:11:19.639 mentioned earlier. The ongoing cost of getting 00:11:19.639 --> 00:11:22.509 pixels just right. Dealing with the endless churn 00:11:22.509 --> 00:11:24.990 of JavaScript frameworks, browser compatibility 00:11:24.990 --> 00:11:27.970 issues, managing incredibly complex application 00:11:27.970 --> 00:11:31.049 state, all that detailed, often frustrating work 00:11:31.049 --> 00:11:33.450 on the visual layer. And in this agent -controlled 00:11:33.450 --> 00:11:37.600 world, a lot of that debt just goes away. A significant 00:11:37.600 --> 00:11:40.059 amount of it, yes. When the primary interface 00:11:40.059 --> 00:11:43.019 isn't a visual GUI you build by hand, but natural 00:11:43.019 --> 00:11:45.639 language talking to APIs orchestrated by AI, 00:11:45.860 --> 00:11:48.139 your development focus shifts dramatically. You're 00:11:48.139 --> 00:11:49.720 not managing pixels, you're doing functional 00:11:49.720 --> 00:11:52.460 orchestration. So developer time is spent designing 00:11:52.460 --> 00:11:56.059 really clean, powerful APIs, figuring out the 00:11:56.059 --> 00:11:58.559 best strategy for how the AI selects and chains 00:11:58.559 --> 00:12:01.440 those tools. Right. How does the AI intelligently 00:12:01.440 --> 00:12:04.159 combine different capabilities to fulfill complex 00:12:04.159 --> 00:12:07.440 requests? That becomes the main challenge. And 00:12:07.440 --> 00:12:09.899 the user experience isn't about how pretty the 00:12:09.899 --> 00:12:11.879 buttons are. It's about the intelligence, the 00:12:11.879 --> 00:12:15.759 reliability, the effectiveness of the AI's orchestration. 00:12:15.899 --> 00:12:18.320 Does the agent understand what you want? Does 00:12:18.320 --> 00:12:20.620 it pick the right tools? Does it actually complete 00:12:20.620 --> 00:12:23.580 the task successfully? That's the new UX. So 00:12:23.580 --> 00:12:25.820 the conductors judged on the quality of the symphony, 00:12:25.899 --> 00:12:28.559 not how shiny the trumpets are. Exactly. It's 00:12:28.559 --> 00:12:30.799 a deeper... kind of UX measurement, I think. 00:12:30.940 --> 00:12:32.820 And the sources emphasize this isn't just theoretical. 00:12:32.980 --> 00:12:34.940 You can build this today. They give examples 00:12:34.940 --> 00:12:37.600 like the ultimate AI travel assistant. That's 00:12:37.600 --> 00:12:39.879 a great one. Imagine asking for a super complex 00:12:39.879 --> 00:12:42.980 trip plan, specific dates, budget, interests, 00:12:43.320 --> 00:12:45.120 flights from a certain airport, certain type 00:12:45.120 --> 00:12:48.360 of lodging. The AI orchestrates calls to different 00:12:48.360 --> 00:12:52.080 APIs, flight search, hotel APIs, Airbnb, restaurant 00:12:52.080 --> 00:12:54.620 reservations, maybe even museum or tour booking 00:12:54.620 --> 00:12:57.899 APIs. And it synthesizes this entire itinerary 00:12:57.899 --> 00:12:59.700 for you. Or the proactive personal financial 00:12:59.700 --> 00:13:02.299 agent. What were my biggest spending categories 00:13:02.299 --> 00:13:05.840 last month versus the month before? Also. Can 00:13:05.840 --> 00:13:08.559 you pay off my credit card bill in full? Calls 00:13:08.559 --> 00:13:10.639 your financial API via something like Plaid to 00:13:10.639 --> 00:13:13.019 get transaction data, analyzes it, then calls 00:13:13.019 --> 00:13:15.340 your bank's payment API to execute the transfer, 00:13:15.519 --> 00:13:17.779 gives you a summary and confirmation, all from 00:13:17.779 --> 00:13:19.940 one natural language prompt. It's powerful stuff. 00:13:20.259 --> 00:13:23.399 The on -demand AI design assistant. I need a 00:13:23.399 --> 00:13:25.559 logo concept. Make it minimalist with black, 00:13:25.580 --> 00:13:27.620 white, and bronze. Once I pick one, create a 00:13:27.620 --> 00:13:29.960 social media banner using it. It uses an image 00:13:29.960 --> 00:13:33.600 generation API like DALI 3 for concepts. Once 00:13:33.600 --> 00:13:35.460 you choose, it calls something like the Figma 00:13:35.460 --> 00:13:38.279 API to create a design file, place the logo, 00:13:38.559 --> 00:13:41.480 add text to a template, returns a link to the 00:13:41.480 --> 00:13:43.860 finished design maybe. And the automated e -commerce 00:13:43.860 --> 00:13:46.200 shopping assistant. Find me the best price for 00:13:46.200 --> 00:13:48.659 this laptop with these specs across Amazon Best 00:13:48.659 --> 00:13:51.580 Buy and B &H Photo. calls their product apis 00:13:51.580 --> 00:13:53.600 maybe does some intelligent scraping compares 00:13:53.600 --> 00:13:57.580 everything price shipping availability summarizes 00:13:57.580 --> 00:14:00.019 the top options maybe even adds it to your cart 00:14:00.019 --> 00:14:02.820 if you've authorized it the pattern is consistent 00:14:02.820 --> 00:14:06.419 define the functional capabilities with apis 00:14:06.419 --> 00:14:09.700 and ai agents chain them together to solve complex 00:14:09.700 --> 00:14:12.139 problems now the sources are also careful to 00:14:12.139 --> 00:14:14.820 call the death of the front end a provocation 00:14:15.120 --> 00:14:17.399 They aren't saying traditional GUIs are going 00:14:17.399 --> 00:14:19.460 away entirely. Oh, absolutely not. That would 00:14:19.460 --> 00:14:21.720 be an overstatement. This isn't a one -size -fits 00:14:21.720 --> 00:14:24.299 -all solution. GUIs are still superior for a 00:14:24.299 --> 00:14:27.179 lot of tasks. Like highly visual creative work. 00:14:27.259 --> 00:14:30.679 Yeah, think Photoshop, Figma, 3D modeling software 00:14:30.679 --> 00:14:33.940 like Blender. You need direct tactile manipulation 00:14:33.940 --> 00:14:36.600 that natural language can't replicate. At least 00:14:36.600 --> 00:14:38.659 not yet. That kind of interaction is still best 00:14:38.659 --> 00:14:40.879 visually. Or browsing. Like scrolling through 00:14:40.879 --> 00:14:42.919 Netflix or shopping on Amazon. You need that 00:14:42.919 --> 00:14:45.120 visual layout for discovery, for serendipity. 00:14:45.120 --> 00:14:46.799 You don't always know exactly what you want. 00:14:47.120 --> 00:14:50.360 Exactly. Complex data visualization to dashboards, 00:14:50.500 --> 00:14:53.840 things like Tableau or Power BI. Interacting 00:14:53.840 --> 00:14:56.480 directly by clicking and filtering is often much 00:14:56.480 --> 00:14:59.460 more intuitive for exploration than typing queries, 00:14:59.659 --> 00:15:02.159 you know. And high stakes actions like a bank 00:15:02.159 --> 00:15:04.860 transfer or confirming a medical procedure. You 00:15:04.860 --> 00:15:07.340 definitely want a clear visual screen showing 00:15:07.340 --> 00:15:09.860 you exactly what's about to happen before you 00:15:09.860 --> 00:15:12.720 hit confirm. Yeah. For safety, transparency and 00:15:12.720 --> 00:15:15.360 clarity, a GUI is still often the best approach 00:15:15.360 --> 00:15:17.990 there. You need that. explicit confirmation step 00:15:17.990 --> 00:15:20.409 so the most likely future isn't one or the other 00:15:20.409 --> 00:15:22.730 but probably a hybrid model that's what the material 00:15:22.730 --> 00:15:25.029 points towards and it makes a lot of sense you 00:15:25.029 --> 00:15:28.169 combine a visual GUI for tasks where direct manipulation 00:15:28.169 --> 00:15:31.350 discovery or clear confirmation is needed but 00:15:31.350 --> 00:15:33.909 integrate a powerful AI agent may be accessible 00:15:33.909 --> 00:15:36.529 via a chat interface within the GUI itself for 00:15:36.529 --> 00:15:39.669 complex multi -step processes deep searches or 00:15:39.669 --> 00:15:42.350 tricky configurations like using a project management 00:15:42.350 --> 00:15:44.990 app You use the visual board to drag and drop 00:15:44.990 --> 00:15:47.970 tasks, maybe see the overall status. But you 00:15:47.970 --> 00:15:50.690 tell the AI agent, find all my overdue tasks 00:15:50.690 --> 00:15:53.129 in this project, change their priority to high, 00:15:53.289 --> 00:15:55.529 notify the team on Slack that I'm working on 00:15:55.529 --> 00:15:57.690 them, and schedule a follow -up meeting for tomorrow 00:15:57.690 --> 00:16:00.750 morning. Exactly. The GOI gives you the visual 00:16:00.750 --> 00:16:03.309 context you need, and the AI agent handles the 00:16:03.309 --> 00:16:05.889 complex cross -system execution that would take 00:16:05.889 --> 00:16:09.129 you many manual clips and steps otherwise. It 00:16:09.129 --> 00:16:11.600 leverages the strengths of both. So wrapping 00:16:11.600 --> 00:16:13.559 up, the really big takeaway from these sources, 00:16:13.659 --> 00:16:16.480 the transformative idea, it's this fundamental 00:16:16.480 --> 00:16:19.480 shift from building visual front ends for human 00:16:19.480 --> 00:16:23.200 users to architecting structured APIs for AI 00:16:23.200 --> 00:16:25.960 users. That's the paradigm shift. Your design 00:16:25.960 --> 00:16:27.820 language isn't just about aesthetics anymore. 00:16:27.980 --> 00:16:29.860 It's how well -structured and understandable 00:16:29.860 --> 00:16:32.779 your API is. Your user experience is becoming 00:16:32.779 --> 00:16:35.360 less about visual polish and more about the intelligence 00:16:35.360 --> 00:16:37.600 and reliability of your functional orchestration, 00:16:37.740 --> 00:16:40.259 how effectively the AI agent uses your tools. 00:16:40.440 --> 00:16:43.820 And your front end is increasingly the AI agent 00:16:43.820 --> 00:16:46.519 itself, with natural language as the interface. 00:16:46.940 --> 00:16:50.700 The visual front end isn't like dead, but it's 00:16:50.700 --> 00:16:54.240 evolved. It's now just one. possible surface 00:16:54.240 --> 00:16:57.200 you can interact with voice, text, chat, or others 00:16:57.200 --> 00:16:59.700 while the underlying functional interface is 00:16:59.700 --> 00:17:02.179 being built for AI agents. It really makes you 00:17:02.179 --> 00:17:05.220 rethink how you design and build software from 00:17:05.220 --> 00:17:07.779 the ground up. It's a different mindset. It totally 00:17:07.779 --> 00:17:10.180 does. And if you're involved in building software 00:17:10.180 --> 00:17:12.400 at all, this is something you really have to 00:17:12.400 --> 00:17:14.680 grapple with. The sources leave us with this 00:17:14.680 --> 00:17:17.019 really provocative idea, something to think about. 00:17:17.339 --> 00:17:19.619 If you're still building every button by hand, 00:17:19.779 --> 00:17:22.779 maybe try wrapping just one API in MCP and letting 00:17:22.779 --> 00:17:25.640 an AI control it. See how it feels to be the 00:17:25.640 --> 00:17:27.819 conductor, not just the coder. Just try it out. 00:17:27.940 --> 00:17:29.880 Yeah. It definitely shifts your perspective on 00:17:29.880 --> 00:17:32.279 the tools you use, the applications you build, 00:17:32.339 --> 00:17:34.299 and how we might interact with software a lot 00:17:34.299 --> 00:17:36.559 sooner than we think. Something to mull on for 00:17:36.559 --> 00:17:40.019 sure. It's exciting and I guess a little daunting 00:17:40.019 --> 00:17:42.900 at the same time. A big change. Absolutely. Thanks 00:17:42.900 --> 00:17:44.319 for taking this deep dive with us today.