Google's New Model One Shots Traditional Software Development (Gen UI is coming)

Lights, camera, action. Okay, so I've been working a lot on systems because how are you going to scale all of your efforts if you don't have systems in place? I'm going to talk about those systems and also talk about a little bit of the news in this video. Everyone has news fatigue. I'm a builder just so happen to yap. So, let's get into it. Very excited about this video, actually. buckle up. Maybe put your seatelt on, grab yourself a little water, give yourself some grace because there's a lot of stuff to cover. So, first of all, we're in Canva because Canvas is great. And first, I want to talk about Microsoft. Microsoft, Microsoft, Microsoft. They came to play. I have been yearning for the moment that I don't have to cobble together all of these different software tools to get a nice productive workflow. Two weeks ago, I started tool hopping. I did this on behalf of our lovely people over at Vibe with AI and we got a bunch of stuff going on there. A lot of smart people. Don't know why they're listening to me, but no, I'm just the shepherd. But I put together a bunch of stuff around all the different tools and the best thing and all that. And then I'm just not shipping any work because I'm just tool hopping and talking about the news. So that's going to change. That's why the videos have slowed a little bit. But yeah, so this is what matters though is that Microsoft should simplify all of these things. One, they're open sourcing their product. Two, they've taken all the patterns that are successful and made them better for developers. Let's talk about the language model tool API. What does this do? A language model tool is a function that can be invoked as part of a language model request. For example, you might have a function that retrieves information from a database, performs some calculation, or performs some online API. When you contribute a tool in a VS Code extension, agent mode can invoke the tool based on the context of the conversation. So, this has probably been around. I just have not even noticed. And I've been using VS Code, and I'm I hate to say it. I don't want to call it, but I'm really enjoying it because the customizability and the native feel of not using a fork. I've noticed cursor has slowed down significantly since 050. I've noticed they're under a lot of pressure. You either crack or you become a diamond. I think they're cracking, but not going to hate. Love cursor. But that one piece is such a game changer because I can open up my extensions and I'll pop this open and let's say I do a forward slash. I have all these nice little commands. Or if I do an at, let's do a new one. If I do an at and I had extensions with the support for the language API, then I can just have the extension be talking to directly the thing that I want. So an example of this, let me just type in Ruby LSP. This is funny enough like the one I figured it out with, but this has an extension with support for your chat and it's really nice because it will just bake in tools. The second example is the ability to have this instructions built in. So, we know cursor rules, but this it's just right here and it's a pattern. It's a new primitive. It's not MDC that has this weird flavor of markdown that kind of crashes. You have to learn this meta thing. No, it just says a glob and that's it. And that's really nice. And the way that you do that is GitHub instructions and then you can paste these in. You can say when you want to apply them, when you do not. It's really nice. They also have a change log. So in the docs with the right tuning, you can have any request go and update the change log. That's great. So it just cleans things up. And then other things that Microsoft is doing that I really like. I know there's tons of news. I just want to talk about the stuff that I like. So yes, the co-pilot extensibility. So you have agent mode. You have MCP management. Now I have a rare rare condition where I'm allergic to MCPs, but I like the way that they do it where if I want to do an MCP. First of all, when I open this up, I can see the instructions. So, I can see all the different instructions that I have in here. And I can create a new one and just type that in. But if I went and I did MCP servers, then something should happen. How do I do MCP list servers? There you go. So, now this is pulling from everywhere. And then if I wanted to, I could click on one and then see all the available tools. It's just right there. It's really nice. So, I was using 1 MCP today and it was awesome. And this flow is something that I was working on before where I was like, cool, I'm going to have to rely on cursor agent. So, I built this CLI that was going to try to communicate with the cursor agent. It's just easier just to literally just tag the thing and have you go through the AI software development life cycle from idea to PRD to PRD plus. PRD plus just pokes holes in whatever it is you're using. I actually have a read me in here that explains what model to use and when. Use Google Gemini and or 03 for this. Use 04 mini for Python tax task execution. When you get to PRD plus I recommend using the sequential thinking MCP. Let me sequential. Yeah, posted this today. This is what the prompt looks like. You won't have the tools, but it is interesting to reverse and basically just see how it works. And so when I fired off, the way that I got this was I fired the MCP to run and it was not authorized by VS Code. So it's, hey, it's trying to do this and here's the outputs of it. So that's a cool thing. You can see exactly what's going on. So very excited about this. I'm hoping that I can just only use Copilot. I really am. I think that I've been just ripping on it in the past, but it's good now. And also, they just own the ecosystem. I think it's a safe bet. Like I'm in a place where I want to focus on shipping and making money. All the people around me and the community and just I just want to ship more basically. And so this seems like it's at a level now where I can ship more and not have to tool up. You can also customize the chat responses. That's what this instruction file was. And if you want depth on any of these, that's going to be part of the content strategy which I'll talk about in a minute. Basically, I'm going to be doing videos on how VS Code showed up, how notebook LM is cracked, how the Google or sorry, github.com copilot is bananas, open sourcing the basically all my projects and then people that join the community get access to those repos can commit to them. Trying to build build that thing up. I'm also building Discord bots with brains using AI to design. These are all the things that I've been doing the last couple days. So, that's why the videos have slowed a little bit. talk about how Gemini cooked, how Chrome cooked, how to keep it simple, how GitHub copilot is good now, how the Android XR and Warby and Warby Parker partnership is going to just smoke Apple. So, these are all the videos and wow, I thought you make your thumbnails with AI. Yeah, but they look like trash. I did a couple of them. So what you do is you make these templates and you see how they perform and then I'll use these in pillow right for automation because I mean I made all these manually before. Great. So excited about that. I'm going to go into depth on all of those. The coding agent pretty dope. So now on to Google. You don't need to watch this video to get every Google news because there's a bunch of videos for that. This is just stuff I like. First of all, the new AI mode. I think it's a game changer. I think that everyone in SEO is going to be freaking out, but they shouldn't because at the end of the day, if you post consistent content that is consistently good rather than occasionally great, like you just have to show up. It's rule one of content is show up. And you should be making content. I sound like Gary Vee, but it is true. Like, I've made it for two months and a lot of doors have opened like way more than had I not. So the way that you get around like the fear of AI mode ending your world is make stuff that people people care about and they care about it because it's either solving a problem through a lack of education or a problem that is search intent. Oh, how do I fix my door? Or entertainment value. I don't do entertainment value, but yeah. And then visual search is exploding. They didn't launch any new models, but diffusion's coming. So, on to the things that I care about again. Chrome updates. These are good. I'm back on Chrome, as you can see here. And it just has little things like view transitions when they they did some I'm assuming performance optimizations since the last time that I was using it. But if I haven't opened a tab in a while, then it's going to like maybe get cash. I don't know. Someone out nnered me here. But basically, like I come back and it has this nice little fade transition in and it's cool. And it just feels a lot snappier. But beyond that, I was doing research today for market research and I popped this bad boy open and I'm like, "Cool, I want all the styles for all the text. Give me the tailwind so that I can paste it in to my codebase and my Tailwind config." So, you could just literally rip that thing. I know there's other ways to do it. You can get a Chrome extension, but that's nice. And then they have debuggers. Like, basically, you have a little AI following you around everywhere in here. There's one of them. This isn't all new. Like these have existed before, but now they're better. They're tied into performance. They're tied into everything. And then there is a way to dashboard these. So if you have the React profile or profiler, I haven't set that up, but that's very cool. Other things is they added I don't know like better carouseling. Yeah, there's just like a lot of stuff that they added. Here's Copilot. It showed up, right? It's literally everywhere and it's not going to change. So get on the train. Let's get back to Canva. Notebook. LM updates are they go so hard. Like I think a lot of us forget the R&D phrase has the word or the letter R in it. So we skip over research. But I just did a video on the main channel. This is a daily channel and I cover different levels of how I'm using Notebook LM from the updates. So you can open up the mind map. All right, you can see this. And then if I click on one of these nodes that does not have an arrow, then it will go and research that. So that's pretty cool. And then that led me to how do I like make a pipeline out of this? I found a Reddit article. Someone had done it. I'm like, I don't even know what that is. Like it's reverse engineering, but am I using something in like the cyber security world? Is it like wire shark or what? How do I do that? So I went here, asked how to do it. It gave me all the answers to the quiz. I knew it was going to be playright, but I didn't know how to do it. So, it gave me all the things aka the keywords that I needed to then go and use Notebook LM. And one of the newer things that they added is the ability to do discover. So, I just went over to discover and then I pasted in like reverse engineer a private API using XYZ. And then I got a bunch of awesome resources. And another one is for fast API because I'm horrible at fast API. I'm brand new. Basically did a master class or a speed run on it over the last month. I started this new project a month ago and it's it's live but not very good live for me because I use fast API. I wanted to learn more about it and I wanted a complex project. So this guy did real time analytics and it uses a bunch of cool stuff. So I was like I don't have 4 hours and 44 minutes to watch this. Let me just ram it in here and then tune the output of the podcast because I don't want to dive in the deep end of the pool. I just want to like passively learn some stuff. And so I made a really technical one. I prompted around it so it would be longer. Again, can click into here, see what are the technologies that are discussed in here. You get this nice mind map. Yeah, go watch the video after you're done watching this one because I think yeah, notebook LM now will be something that I frequent like really frequent. Haven't gotten to play with the new flash. Excited for diffusion. So, generative UI was originally proposed in a next comp like two years ago, maybe a year ago, and it was cool. It was pretty useless at the same time. Like, who's going to ask an LLM for the weather when I don't know. I can just look here. It just doesn't make sense. Why would I send a token for that? But it's it's a use case. Where I think this is going is because diffusion is five times faster than Flash. And I don't know, Flash is pretty damn fast. then you do open the door to doing generative UI. Now, why should you listen to me? I literally ran into this issue when I was building map, which is a multigenic platform that helps you accomplish your goals. studio.youtube.com. Basically, I got to a point where I was trying to do so much that was multi-gentic and had like generative UI that I found it to be the most useless pattern because it didn't do more than This demo is going to be three parts. In the first one, we're going to cover. So, in this case, I would come in and I'd explain my goals and it would load in. It would do an ETL off of all the different wearables. So, an Aura Ring, a Whoop, and an Apple Watch. And then it would standardize them because heart rate's tracked in a different way than over here. Still the same heart rate, but when you know Apple calls it HK heart rate, Whoop has a way or as a way, whatever. Get all that in there and then conditional to whether or not you connected things. It would show you different things in the platform. That's not aic, but you would basically get this plan, right? And the plan was based on different questions that were asked. Okay, so here multi-aggentic gen UI stuff. Here's a decent example I guess is perplexity. There's an open source version of this and this is me explaining all the different steps of it. How cool. You have the quer from the user. You have the task manager. Task manager basically uses Zod validation to figure out like where do we go with this thing? But then when we get to here, so this would be a decent example where it's okay based on how you want to run this marathon in the next 90 days. because we know that you at that point run 20 m a week and you have these health stats like this is a realistic plan and I had trained a pipeline in Azure actually crazy and I was using Kafka for no reason Apache Spark all these things so overkill but it would then decide what it's going to show you when you select one of these so yeah now I ended up running into issues and exox and I could just question why the heck I would use this but now a day with diffusion. I think it really opens the door to something far more interesting. So in my head we have different protocols. We have MCP which is now the USBC to AI. Think of it that way. But what I'm thinking is in the future if I am a developer, I'm the dev and I make the features and I have full API support for let's call it 15 methods on the YouTube API and it's going to be oh is this video scheduled? Is it published now? Is it unlisted? Is it private? All these things around it. Do we need subtitles? Is it made for kids? Is it monetized? Where are we? Is it mid rolls? There's all these things that you see when you go to an API, but as a developer, you're going to prioritize the ones that you think are going to have the most impact, but you might in some cases write the API support for it, but then not surface that in the front end because of a resource constraint or because whatever the case is. But let's assume that out of the you did 15 of them, but there's actually 50 methods. What I propose is going to happen is because you technically have the ability to do this and again this is not a fully baked idea but I think it's interesting there would need to be guardrails for this. The buyer of said software could see their 15 features right and they'd have access to those 15. But then there's this magical thing that I propose will happen which is that you could have the other 35 in some sort of list where it's look this I don't have subtitles as a feature that's baked but we have support for it. So do you want to add that? If you have the API contract in place, you have examples in your codebase, there's probably going to be some sort of set of rules that have already defined the system patterns and how to drive GenUI, then theoretically as a user, you could say, "Yeah, I actually do want that." And then it would know based on the design library, the choices that you've made, the screen availability, the real estate, a lot of things to figure out there. But you have truly custom software and it's not feature flags. Those are different. We've had those. We have used feature flags for showing different pricing, different buttons, all these things based on geo and whatever it was that we're tracking in the event taxonomy. But this is fundamentally different and I'm excited about this idea. Someone this will be like the MCP in 2029, I think, is is where this this all goes. So that's my my bit on generative UI. Now we get into some of my strategy. Holy smokes, just minutes. Do I get into this? No, I don't. I saved this for another video. This is the sauce, though. So, sorry. But I've been busy at work building so much stuff that I'm excited about. Look at my hair. Bananas. But yeah, I'm gonna save this for tomorrow and that way I can give my take on on generative UI packages and we will tomorrow be covering some of the things that I found with prompting around marketing, prompting around research. So offer creation, content engine, that's what I keep teasing about like the content is going to be changing a lot. ICP, Reddit scraping, naming exercise, pricing tiers, Discord automations, updates on Echo, updates on buy with AI, updates on Earthang. So, I'm going to just do that. If you enjoyed this video, of course, you should like and subscribe and join the community because things are about to change and they're going to be a gajillion times better. Thank you for watching and I hope you have a lovely day. See you.