Google AI Death Star is Aimed at Every Startup (what it means for you)

Let's talk about Google and how they're about to delete maybe a trillion dollars worth of startup capital, startup enterprise value at their event next week. Now, you're probably thinking, "Oh, that's exaggerated. Let's just talk about it." So, this is a daily upload and I'm Parker Rex and I go through AI news, I go through some Q&A, and I go through AI strategy on my way to 100K in profit a month. So, first of all, yeah, Google, the reason that you clicked this video. So, they have their IO event next week and it's slated to release a lot of stuff. We're going to go through the events themselves and then we're going to talk about some of the stuff that's been leaked on Twitter. So, first of all, you should check this out. You make a little developer account and then you can start adding things to your own calendar. But the keynote, it's going to be the the classic. Get the CEO up there. Let him yap for a little bit. See how many times he says AI. But things that I'm excited about are the Chrome releases. So it is rumored that Gemini will be put into Chrome. This is probably a good move. A lot of people are hesitant about it. You already have AI in Chrome with the dev tools, but it can only get better. They have so many captive users that don't use Gemini. So they have a distribution problem and they just need to put a nice interface in front of everybody. Currently I use Brave and when I hit commandB this thing opens up and it is useless. Excited to see how that comes along. And also I look at Arc as a browser and I'm like, "Oh man, like that sucks because it was so good for a small period of time and they released a new product called DIA and it doesn't look that great because they're really going after, oh, you can ask questions about things in here. You can only do it if you are at the browser level, but then Chrome's going to release this." So they're going to talk about Google Cloud, which is stuff that I'm very excited about as I build out my product called Echo. And Echko solves a problem for me where making content at scale that is personalized and high quality just requires a whole lot of work and it has been holding back my ability to serve my own needs and SAS because there's just so much time in a day. So I was like okay let me build SAS that will help with that that will build all of the things around content that's annoying beyond just recording this. So that is both for when you have the idea and you want to get it from there to the internet with all the metadata and then having it followed on with all the different mediums. So taking long form clipping to short form taking the script turning it into blogs to every other platform. There's 18 phases and you can check out my strategy on a previous video for that. But I'm excited about the cloud stuff because there's just so much going on in there that is the one-stop shop. Yes, it's really me. It's the one-stop shop for anything that you could possibly want. That sounds hyperbolic, but it really is. Like you have storage. And then here's all the tabs that I keep open. Let me move this a little bit so you can see. But these are all the ones that I frequent. So you obviously have billing, you have cloud storage, you have cloud run, you have the workspace, you have vertex. Vert.ex is already seeing updates pretty much every other day it seems where they have the ability now to use Imagin 4 as like a preview and it's amazing. You have all the different tools you could possibly want to do with AI in here. You have the agent garden that I'm very excited about because it has great examples in it. And then they're going to be building the UI for their new agent ADK. And that's why I've been going all in on Python. The last couple of weeks have been a struggle for me as a TypeScript dev. But then realizing I don't want to hit the same wall that I've run into before where TypeScript cannot deliver on the same features and basically capabilities as Python. And it's not a language problem. It's a community problem. When you have Google and OpenAI releasing new stuff that can only be accessed by Python, you need to learn Python. But they'll be adding in a web UI for the actual Agentic stuff. I think this is going to be kind of like how NADN had a big moment and is having a moment. I think that this will be our moment as developers. So very excited about that. And then all the other things in here, we'll probably see updates. So excited about all that stuff. The AI stack for developers, what's new in web, what's new in Firebase. I'm sure they'll be talking about the Firebase Studio. It from my understanding was the flop. Automating work with Google Workspaces. They've built almost a Zapier competitor. Pretty cool. And yeah, there's just so much that's going to be covered. What's new in Go? Yeah, very excited. If you didn't see, they also have a new model that basically combines Flash and Gemini for algorithmic optimization. Big buzzword set there, but basically they're crushing. Even their Vzero competitor now has a bunch of cool canvasing features. If I go to Jeff Dean's page and see what he's talking about, you should definitely follow him. I have him on a list, but you can see him talking about the canvas and how amazing it is. But you can build a whole lot of stuff in here. It's just Google's basically cooking. As soon as Sergey went in and said, "Look, we need to cook," they started cooking. And if you don't know about Jeff Dean, go research the guy. He is a legend. So excited about all that stuff with Google. Again, I'm very bullish on them because they have the science, they have the team, they have the TPUs, so they don't have a bottleneck there. You don't see them flying over to the Saudi summit to both network but also raise money. They don't have to because they're printing money. They have, I think it was either 70 or 90 billion capital allocation to AI infrastructure just for this coming year. And that's on top of them making more money in their earnings than typical. So that whole thing around, oh, they're going to be bleeding from the five blue links. No, not at all. They're making big strides there. So that's on the Google side. Other things I wanted to talk about, Grock Web is getting tasks. So we can take a look at this. Grock Web will soon get tasks. So chat GPT tasks basically a cron job for a query that you want to have. There's now MCPS within terminal using warp. So pretty cool. If you're a warp user now you can do this. I'm not very bullish on MCPS. I feel like I'm missing something but maybe it's a skill issue. I have people in our community who are like obsessed with them and also people who are on my side where they're like, I just don't get it. So, we have a whole channel in here for Vibe with AI where basically people are going off about, oh no, it's the best thing since sliced spread or oh, this thing just doesn't work for me. But regardless of my opinion, it's a fact that they're getting a lot of adoption and companies like Warp who make a great terminal are putting time into it. So there must be something there. There's something in the water. Up next onto some cursor stuff. So cursor has background agents now. And I don't have access to it, but I made a video about that yesterday, but I'm really excited about it. They're also cooking up dashboards just to see. This kind of reminds me of Waca Time. If you're not familiar with Waca Time. If I just click into my Waca Time profile and I log in, it'll show you your basically your activity across the week. And so this will be specific to your cursor usage, which is cool. And you can see I've had a big shift from using no neomm to all cursor. And then I've had a shift from using only TypeScript to Python. So it'll be cool to get some analytics around that. They also what else? They put out a recent blog article about how they choose their models. So that's probably worth it the video on its own. Let's see. We get a lot of questions on which model to use in cursor. Yep. So they have a selection tree. So what's most important to you right now? I want the model to figure out or I want the control over what the model does. What kind of task are you working on? Is it a small scope to change? Then use cloud 35. If it's a larger task with clear instructions, then 3.5 or 4.1. If it's complex and very ambiguous then you use 03 very expensive model you use that with Klein or is it a routine or general use then 37 sonnet 25 or 41 so that's how they treat their selection tree and it maps to the things that I do which is nice to see I'm doing things correctly other things co-pilot actions are coming so it's basically computer use on chat GBT wild. But I'm interested to see how people use this. I know that they had launched computer use for the $200 a month subscription and I didn't see much hype around it. People that I've spoken to did not find it that helpful just because it's like really slow and I don't know, I wouldn't trust to make a dinner reservation which says something. Another interesting thing that I found was this product that basically goes and it scrapes all the negative reviews around an app for consumer apps on the app store and then comes up with potential ideas for that. So cool. Again, this is like a customer research, customer discovery tool, and you know, it's nice. It pulls in all the different reviews and then you find the gaps in there. This reminds me of I want to say it's like gummy search which is basically like Reddit audience profiling. So that could be interesting if you're looking for ideas but I don't need new ideas. I need execution. Mistrol is rolling out connections. This basically is what it sounds like where you can basically toggle things on. So you could have mistrol connect to your Gmail or your Google calendar. Now I had done this manually before and I'm just curious like if they don't have every tool defined then it's going to make a mess of these high trust high importance things like your communications. And I also just wonder like why Gmail because Gmail is just going to launch Gemini for all their workspace users. Oh, wait. They already did. So, hesitant about that. I also think just with the Google calendar, like Google's going to do that very quickly. So, now let's jump into AI strategy things that I'm thinking about. Oh, sorry. Q&A. Almost forgot. So, I had a question come in or comment that was about it was from a guy named Dan. Great guy. He works at Google actually. And he just mentioned on a mono repo post that I had made to look at Google 3 and how bananas it is. So when you go and you research monor repos, you'll come across a product that's called Basil, which ironically enough is my lady's name, but Basil is a monor repo build tool that Google uses and they needed it because they apparently have more than half of their code. So two billion lines of the Google codebase is on one monor repo. Crazy. Yeah, it says Android, Golang, Chromium, and a few others were separate, but there are two billion lines of code in one repository. So, that is crazy. Other comments were just supportive. Thank you for the support. If you guys have questions, please ask them. They'll be answered in order of time. So, on to AI strategy. This is something I've just been thinking about a lot as I had taken a week away from making the daily videos and there's just this thing that's happening where there the word vibe comes out and everyone's hyped about it. They everyone forgets that the guy who coined it invented a lot of the AI classes at Stanford and has been coding since 2005. So his prompting is going to be very different than yours. And when I got started forever ago with coding, like building my first thing, like you would work really hard and struggle to get this on like a blank screen. That was the whole app. And now the same amount of effort is like a fancy looking website with the header and a call to action. You might even get like a game of tic-tac-toe or Minecraft. You could even you can make like actual legitimate things off the rip, which is tricky because then people think that they're coders, but they're not. And they still have to learn fundamentals of how the systems work. How do you connect a database? How do you start a database? How do you structure a database? How do you connect a database to a front end? How do you write a basic backend? What is a backend? What is a front end? What is a framework? What is a meta framework? What is a library? What is an package manager? All these things have different big concepts that if you don't know, you're going to go and you're going to vibe code your way into a hole. And it it's almost like you have to go backwards because you thought that because you had gotten here that now you can go to the next level. And I think a lot of people are struggling with this where I meet people where they think that the AI is going to be so good soon that it doesn't make any sense to learn the principles of programming. But that doesn't make sense because you need to know the principles in order to get more leverage. And you're going to pay for it either way. you're going to pay for it in tokens by asking a million questions, which is what I would do if I was trying to learn. And you're still like flattening that learning curve, but you're either going to do that route or you're going to do the route which is just trying to prompt your way around it without trying to learn. So if I look at it and it's like time to learn or time to competency and this is let's say one month, two month, three month, four month and then you have let's call it just like skill. We're going to use orange for the old way. We're going to use yellow for the new way, which is using AI to learn. And then we're going to use purple for just puritin vibe coder. And so I think that it's tough because you as a vibe coder, if you just refuse to actually try to learn these things and just ask questions, you're only in agent mode and you're only just like drilling it and being like, where's the next prompt? You're going to just like your skill level will just stay down here for a very long time. and you might just quit or it's going to stay down here for a very long time. And then by nature of you prompting your way around it, you will finally learn because you're basically just pattern matching. You're reading a lot of code. You're trying to pattern match. So it might take you about four months. And these numbers are just guesstimates. I actually did a talk on this where the time to competency is basically onethird of what it used to be. But now in the old way, your time to competency is your skill level. Let's say it's just like at the same level, but it would take you about I'm assuming like basically the same amount of time like just to get to some level of understanding. But my whole theory here is that if you were to just ask questions hundreds of times and just live in the ask mode on cursor, wind surf, ader, you name it, that your time to competency will be half because your skill level will probably seem like it's lower because the outputs that you're getting. You're not getting the fancy stuff fast. But I really think that it's going to just save you time and then your skill level goes up and then that pays dividends because you have better prompts because at the end of the day the things that will not change in five years time is your basically like your competence will always matter. competence. This is going to be a premium. Like you need to be competent. I can't say it enough. And that has a couple different areas. So that comes to product strategy. You get better at product by being more technical. That is a fact. You're listening to someone who's done product for a very long time. Had wins, had losses. But I got better at product the more technical I got. your architecture skills, you're gonna want those. Your system thinking so. So, I don't care if this thing is a genius that can oneshot things. It will oneshot it better. You will get a better output if you have better product strategy, better architecture, and better systems thinking. Because system thinking bleeds into everything. It's not just coding. It's not just making new features. It's also debugging. It's also the way you think about business. You're also as you learn these things, your prompt engineering gets better because you can call things out that others would not. And there's evidence of this now where if I look at a paper about multiple steering, like basically when you have multiple turns on an LLM, the quality of it degrades very fast. So let's just find this. Yeah. So LLMs get lost in multi-turn conversation. And this is a new paper where basically they cover how quickly things degrade. and those things degrade because you are not aware because you have skill issues. So the paper investigates how LLM perform in realistic multi-turn conversational settings where user instructions are often underspecified and clarified over several turns. Keep telling devs to spend time preparing those initial instructions. Prompt engineering is important. So you can see the unreliability shifts, the aptitude goes down. The authors conduct large scale simulations across 15 LMS 41, 25 Pro, Sonnet 37 over six generation tasks, code, math, SQL, API calls, data to text, and document summarization. So you can get more into this paper. I'm not going to dig into every little detail of it, but the gist of it is users are better off consolidating all requirements into a single prompt rather than clarifying over multiple turns. If conversation goes off track, start a new session with a consolidated summary that'll lead to better outcomes. System builders and model developers are urged to prioritize reliability in multi-turn context, not just raw capability. especially true if building complex agentic systems where the impact of these issues are more prevalent. You can check out the paper on RVX. I give it a look and I just can't say this enough, but like I feel this every day when I'm working and I'm just like, "Okay, everybody, like we want to have the most specific one shot, the most specific thing possible." And this is an example of what you can get when you're really specific. And this comes out of asking a lot of questions, creating a lot of new chat windows, coming up with a plan, figuring out the file tree, running ESA, which allows you to get that file tree so you can see what the proposed changes would be, having that brought into a new chat window, just using basic markdown files. I've tweaked the way that the memory bank kind of pattern works. I like using less tools when I'm learning something. But then this pattern that I've landed on working great includes the ability to pipe out the actual code snippets that are going to be used so that it can just go. It doesn't have to worry. You're not steering. And then I'll do one task and then move on to the next chat window. So this one ran a little bit long. There was a lot to cover. If you found this video helpful, definitely subscribe because I make videos like this all the time. If you learned something new, make sure you like the video. If you want to join the AI builders, the people that are trying to figure out how to be at the leading edge, get the most leverage on both coding and content, then you can check out the link in the description. And I'll see you in the next one.