[0:00]Elon Musk just made two predictions about AI. The first one is bold, the second one sounds insane. But then you look at what Manus the AI company Zuckerberg bought just released this week and it starts to make sense. I consult companies on AI for a living, so I spent all week going through every update so you don't have to. Here's what happened. One, Google just released a free design tool that should cost money. Two, Elon's two predictions and why the second one changes everything. Three, Manas launched something nobody expected. Four, lovable is no longer what you think it is. Five, gamma changed one button and it says everything about where they're going. Six, Jensen Huang made a comparison about agents that stopped the room. Seven, Google quietly connected something that should worry you. Eight, Perplexity got access to data it's never had before. Nine, Claude shipped four updates in seven days. Ten, two image models and one is from a company that surprised everyone. Plus, five AI tools you haven't heard of yet and a full step-by-step tutorial on that Google tool. You need to see what it builds, but at the end, an open AI co-founder scored every job on how replaceable it is by AI. Your score might surprise you. Now, I've put all of these in my WhatsApp channel, so after you're done watching, go grab them from there. Let's get into it.
[1:22]Okay, so Google shipped two products the same week and I don't think that was an accident. Because when you put them together, they complete each other. First one is Google AI Studio. So you know how we've all been hearing about vibe coding, right? You describe what you want and AI writes the code. Google went all in on this. They upgraded AI Studio into a proper full stack coding platform and in their demo someone typed one prompt and it built a 3D multiplayer racing game with a working lobby, room codes, two people racing at the same time. One prompt, that's the coding side. And then the second one is Stitch 2.0. Which is the same concept but for design. You describe how you want your app to look and it designs the whole thing for you. And here's what got me, you can talk to it with your voice. It gives you multiple options, you pick one, it applies a design system across every screen and then it turns everything into a prototype you can actually tap through like a real app. Now, what I just showed you is Google Stitch on its own, but when you combine Stitch with Google AI Studio, the workflow gets really interesting. You design in one, code in the other and you can go from idea to live app in minutes. You didn't open Figma, you didn't hire anyone. And later in this video, I'm going to show you exactly how Stitch works, step-by-step. You need to see what it builds. So Elon Musk was at the abundance summit with Peter Diamandis and he made two predictions that I think everyone needs to hear. Prediction one, AI will run out of things to make. Right now everything is expensive because a human has to make it. Musk is saying, yeah, it's basically AI and robots are going to make so much stuff and provide so many services that they will actually run out of things to do for the humans. AI doesn't solve one shortage, it solves all of them. Nothing is scarce anymore. Prediction two. I'd say the economy is ten times its current size in ten years. actually a fairly comfortable prediction. with if if there's like World War three or something, um, that that could put a kink in those plans. Not double, ten times, and he called it a fairly comfortable prediction as long as there's no World War 3, which is fair. Manus, the AI agent company that Meta bought for two billion dollars, just launched something called My Computer, which is basically the same idea. A desktop app that puts an AI agent on your laptop that can organize your photos, rename your invoices, even build apps through your terminal, all from a single prompt. And you know what, between Claude, Manus, Perplexity, and OpenClaw, everyone is racing to put an agent on your machine and that's clearly where this whole thing is heading.
[4:38]So I caught something this week that I think most people missed. Gamma, you know the presentation tool, they quietly changed every single button on their platform. Share your presentation, now says share your Gamma. Google did this to search, Uber did this to cabs. That's the play when you replace the category name with your brand name, that's how you know a company is going for it. But you can't own the category if the product doesn't back it up, so they shipped something called Imagine. One prompt generates logos, posters, infographics, diagrams, all on your brand. Don't like the color, type, add more blue and it updates right in the deck. It now plugs into Claude, Chat GPT, dozens of tools, free for 30 days across all plans. With 100 million users behind it, the word presentation might have quietly retired. So Google just did something quietly that I think is a bigger deal than people realize. They added something called personal intelligence to Google search. And what that means is Google now reads your Gmail, your photos, your past purchases, and uses all of that to answer your questions. And so in their demo, someone typed, I want to grab a quick bite during my layover. Google pulled their flight from Gmail, figured out which terminal they're in, how long the layover is, and recommended restaurants near their gate based on the fact that this person usually picks vegetarian and healthy food when they travel. One search, no extra context needed. Google already knew, rolling out now in the US for search and Gemini. And just think about what that means. Google already has your emails, your photos, your location, your calendar. Now, it's connecting all of it. That's incredibly useful and honestly, a little uncomfortable at the same time. This one is interesting. So your health data right now is probably in five different apps, right? Your Apple Watch tracks one thing, your lab results are somewhere else, your sleep data is in another app and none of them talk to each other. Perplexity just launched something called Perplexity Health that connects all of it. You type one prompt, something like I've been getting migraines, pull my records, labs, wearables, and it builds a custom dashboard from your actual data. Circadian disruption, low HRV, B12 declining, things your doctor genuinely doesn't have time to catch in a 15-minute visit. Rolling out now for Pro and Max users in the US. And I'm going to leave you with one thought on this one. When one AI knows your health records, your sleep, your heart rate, and your lab work, how much do you actually want it to know? Okay, so right now your work is spread across like four different apps. Slack, Teams, WhatsApp, email, and none of them know each other exists. GenSpark just launched Workspace 3.0 and the idea is one AI that sits across all of it. So they have this thing called GenSpark Claw and it lives inside Telegram. This guy in their demo, he just sends a voice message, how many inbound leads did we convert this week? And it pulls his CRM, comes back with 61 active deals, asks if he wants a full pipeline report. He says yes, report builds all inside Telegram.
[7:42]He never opened another app. But here's where it gets interesting. They also shipped workflow automation where you just describe what you want in plain English and it builds the whole thing. Connects to Slack, Google Drive, Salesforce, all of it. They added meeting bots that join your Zoom calls, transcribe everything and send the summary to everyone after. And there's a Chrome extension where you literally tell it summarize this person's posts on X. And it scrolls through the page and does it. So basically, they're not building one AI tool, they're building the layer that connects all of your tools. And that's a very different play. Nvidia, the company whose chips power Chat GPT, Gemini, Claude, every major AI, their CEO Jensen Huang just got on stage and laid out three shifts that happened in the last two years. One, AI went from looking things up to creating things. Text, images, code. Two, AI learned to think before answering. Plan, check its own work. Three, we stopped asking AI questions. We started giving it tasks. Write this, build this, fix this. Jensen summed it up like this. An A.I. that was able to perceive became an A.I. that could generate, an A.I. that could generate, became an A.I. that could reason, and an A.I. that could reason now became an A.I. that can actually do work. And then he announced the thing that ties it all together, an open source project called Open Claw. Open Claw has open sourced, essentially the operating system of Agentic Computers. It is no different than how Windows made it possible for us to create personal computers. Now, Open Claw has made it possible for us to create personal agents. He compared it to Windows. Windows gave everyone a PC, Open Claw gives everyone a personal AI agent. Every software tool you pay for monthly, Notion, Slack, Salesforce. Jensen says those companies will stop selling you tools and start giving you agents that do the work. The tool becomes the worker. So Claude shipped four updates in one week, which is kind of insane, so let me go quick. One, they made the memory five times bigger and I know million token sounds technical, so let me put it simply. Imagine pasting an entire book into a chat. Not a chapter, the whole book and the AI remembers every page. Page one, page 200, page 500, all of it. That's what this update is. And on top of that, they got rid of the extra fee that used to kick in for long conversations. So now it's the same price whether your chat is ten messages or 10,000. Two, co-work dispatch. Okay, so picture this, you're heading out, you've got a meeting in 15 minutes and you realize you haven't read the proposal. So you pull out your phone and you text Claude, hey, open that deck in my downloads and give me the key points. And your computer sitting at home just does it. By the time you check your phone, the summary is waiting for you. Your phone becomes the remote control, your computer does the heavy lifting. Three, co-work projects. You can now point Claude at a folder on your computer and say, this is a project. Claude reads those files, follows your instructions and remembers everything it does inside that project and that memory stays separate from your other work. You can even import projects you already have going in Claude chat. Four, Claude Code channels. You know Open Claw, right? The thing where you message an AI from Telegram and it goes and does stuff. Claude just built that in natively. You message it through Telegram or Discord, it writes code while you're away and messages you back when it's done. Two image models dropped this week. So Mid Journey dropped V8, or well, V8 Alpha and on paper it sounds great, five times faster, 2K images, better text rendering. But then people actually started using it and hands are still broken. Like the fingers thing is still happening. And compared to what Google's Nano Banana is doing right now, Mid Journey is catching up. Two years ago, they were the best, not anymore. This one I genuinely did not see coming. Microsoft, not Open AI, Microsoft themselves released their own image model called May Image 2 and it's ranked number three in the world. The whole focus is photorealism, natural light, real skin tones and like you put text in your image and it actually comes out readable, which is kind of impressive. Still early though, you can only do square images 15 a day and it's US only for now. But quality-wise, it's honestly more impressive than what Midjourney just put out, which is a really weird thing to say about Microsoft. Here are five AI tools you probably haven't heard of yet. Number one, Timelapse. You're spending on brand marketing, but you have no clue if it's actually working. Timelapse runs real consumer research across 4,000 people in your target audience and gives you the insights five times cheaper than traditional agencies. Number two, Dex, an AI data analyst built for founders. You connect your databases or spreadsheets, ask questions in plain English and get answers with recommended next steps. Number three, Blink Claw. Self-hosting AI agents is powerful but painful. Docker, VPS, security, rate limits, nine out of ten people never get their first agent running. Blink Claw handles all of that. One click and your agent is running. Number four, Cappy. Most AI coding tools still need you watching over them, but Cappy plans, builds, tests and reviews code entirely in the cloud. You give it a task, you come back to finished work. And number five, Mothership, a workspace designed specifically for AI agents, fully autonomous but always observable and editable.
[13:23]Okay, so enough talking about Stitch, let me just show you what it does. So this is Stitch. You sign in with your Google account, it's free, and basically what Google has done here is, you know how we've all been hearing about vibe coding, right? You describe something, AI writes the code. Google took that same concept and brought it to design. They're calling it vibe design. So instead of opening Figma or hiring a designer, you just describe what you want and AI designs the whole thing for you. Let me show you what I mean. So let's build something. Let's say you're addicted to scrolling, right? Your screen time is out of control, so we're going to build a dopamine detox app called Scroll Stop. The idea is simple, it locks your Instagram, YouTube, Tik Tok and you can't open them until you do 20 squats. So I'm just going to type that as a prompt. Dark mode, minimalistic, focus-driven and let's go with the Pro model here. It gives better results and generate. Okay, see that? Full app. It even gave the color palette a name, kinetic void. You've got lock icons for every social media app. This giant yellow button that says 20 squats to unlock and this whole thing came from one prompt in about 40 seconds. Now you're probably thinking, okay, that's one screen. Fair, but watch what happens next. So the design looks good, right? But what if we wanted to look like premium, like what if we wanted to look like Apple made it? So here's what I'm going to do, I'm going to go to apple.com, just copy the URL. Come back to Stitch, paste it here and basically tell the AI, extract Apple's design system from this website and apply it to my app, their fonts, their colors, their spacing. That's it. Just a URL and a prompt and look at what it's giving us, new palette called Cupertino Premium. You can see Apple's blue tones, that clean glass finish feel, it's on our app now. Now, quick thing on how this actually works and it's simpler than it sounds. So, every Stitch project has this file called design.md. Think of it like the brain of your design, right? All the rules for fonts and colors and spacing live in that file. So when we pasted Apple's URL, Stitch basically went to their website, pulled out all those design rules and rewrote our file to match. So from here on, everything we build automatically follows Apple's look and feel. Okay, and this part honestly, this surprised me. So if typing prompts feels like too much, right? There's a mic button, so I click it and I just say, make the locked Instagram icon glow red. And it just did it. It found the icon, added the glow. I just spoke to my design and it changed. They're calling this voice canvas, which is basically like having a conversation with your design like it's a person sitting next to you. All right, so this is great, but right now it's just a picture, right? You want to actually tap through it and feel how the app works. So there's this play button up here. Let's click it. And okay, now it looks like a real app on a real phone. And watch, if I click stats, there's a full analytics page, settings, there's a settings menu, 20 squats to unlock, gives us a live tracking screen with a timer, finish button, cancel button, the whole flow is there. And here's the thing, these screens didn't exist two minutes ago. Nobody asked for them. Stitch just looked at the main screen, figured out what pages were missing and built all of them. I know that sounds like I'm exaggerating, but you literally just watched it happen. And when you're done, you just export. You can send it to Figma if you want a designer to clean it up. You can send it to AI Studio if you want to turn it into a real working app. Or you just grab the code, clean HTML and CSS ready to go. Here's the thing. What we just did in ten minutes used to be a couple of weeks and a couple of thousand dollars with a designer. And the only skill you actually need is being able to describe what you want. Stitch. with Google.com, it's free while it's in Google Labs. You get 350 designs a month. Go try it before they start charging for it. Okay, one more thing before I go and honestly, this one might hit different than everything else in this video because it's not about a tool or a product, it's about your career. So Andre Karpathy, the guy who co-founded Open AI and ran AI at Tesla, built something this week that I think everyone watching this should go look at. He took 342 real jobs from the US Bureau of Labor Statistics, which covers about 143 million workers and he scored every single one of them on a scale of zero to ten. Zero means AI can't touch what you do and ten means AI can basically do your entire job. And the pattern that comes out of this is so clear, it's almost uncomfortable. If your work happens on a screen, if you're writing or coding or analyzing data or living in spreadsheets, you're in the red zone. Software developers scored 8 to 9 out of 10, which means most of what a developer does day-to-day, AI can already handle. Medical transcriptionists got a perfect 10. But now flip it, if your work is physical, if you're an electrician or a plumber or in construction, you scored between zero and one. AI can't do what your hands do, at least not yet. And here's the part that nobody expected, you'd think the entry level, lower paying jobs would be the ones most at risk, right? But actually, it's the opposite. Jobs that pay over $100,000 a year had an average exposure score of 6.7 out of ten and jobs under $35,000 only had a 3.4. So the more you earn sitting at a desk, the more of your work AI can already do. I know that's uncomfortable to hear, but that's what the data says. Now look, this is US data, but that core pattern of screen work versus physical work holds everywhere, India, Europe, it doesn't matter where you are. If your day is spent in front of a laptop, this is your reality too. Go look it up at karpathy.ai/jobs, I've put the direct link in the description and on WhatsApp. That's the week. If you want to see what one of these desktop agents actually looks like in action, I did a full Claude co-work walkthrough and that video is right here on screen.
