Grok 3 & The Future of AI: Building Apps in Minutes

TL;DR

AI is no longer just a tool; it's a co-developer. With models like Grok 3, building a functional app is now a matter of minutes, shifting the bottleneck from coding skill to prompting creativity. #VentureStep #AI #Grok

INTRODUCTION

The barrier between an idea and a functional product is collapsing at an unprecedented rate. We are rapidly moving from a world where technical skill is the primary bottleneck to one where the ability to ask the right questions—to prompt an AI effectively—is the most valuable asset. This shift isn't a distant future concept; it's happening now, driven by the exponential growth of large language models.

In this episode, host Dalton Anderson dives deep into the seismic changes being driven by xAI and its new frontier model, Grok 3. He unpacks the story behind xAI's mind-blowing development pace, including building and then doubling a massive, H100-powered data center in a matter of months. Dalton explores a fascinating theory: that this incredible urgency is synchronized with the strict two-year launch windows for SpaceX's Mars missions.

The theory is put to the test with a live, real-time demonstration. Dalton uses Grok 3 and the Replit platform to build a functional Pong game from scratch in just over three minutes, showcasing how AI can handle setup, coding, and even iterative improvements. The episode also compares Grok's features to competitors like Gemini, providing a tactical look at how these tools are poised to revolutionize not just development, but the very nature of thinking and problem-solving in the modern workforce.

KEY TAKEAWAYS

• The development of ideas is shifting from a "skill issue" to a "prompting issue," where the ability to articulate needs to an AI is paramount.
• xAI accomplished a feat considered "superhuman" by NVIDIA's Jensen Huang: building a 100,000 H100 GPU data center in just four months and doubling it in the next three.
• This intense urgency may be driven by the hard deadlines of the Mars mission, which has a two-year orbital launch window, aligning the goals of SpaceX, Tesla, and xAI.
• With tools like Grok 3 and platforms like Replit, a functional application can be generated, deployed, and refined in a matter of minutes, drastically reducing development cycles.
• Treating AI as a collaborative "colleague"—asking it to poke holes in ideas or optimize code—yields far better results than simply using it as a tasking device.

FULL CONVERSATION

Dalton Anderson: Welcome to VentureStep podcast. We discuss entrepreneurship, industry trends, and the occasional book review. ¹The sayings, "you can just do things" and "just do it," have something in common. ²It's the encouragement to put yourself out there and to tinker and experiment with the world. ³And to do that, now it seems more and more likely that it's a prompting issue and not a skill issue. ⁴

The Shift from a Skill Issue to a Prompting Issue

Dalton Anderson: In the future, AI will be able to understand complex applications and thought processes and systems, and we'll be able to build and deploy apps and understand all the dependencies for them. ⁵I'll show you a simple demo today that these things will only scale in both complexity and ability. ⁶

In the future, I feel that it might be more of a prompting issue than it is a skill issue, which I think is an interesting problem to have. ⁷

Dalton Anderson: Well, today we're going to be using Grok 3 to build a simple app. ⁸ I built two apps. I built a Space Invaders game, which would probably take you like six to eight hours if you're building it from scratch. ⁹For me, it took me like seven minutes. ¹⁰ One, I already know how to code. Two, it's not that complex of an app and it's bare bones. ¹¹So keep that in mind, you could build on top of it, but this is just for demo purposes. ¹²And then we're going to build a live app called Pong that I don't know if it's going to work or not. ¹³And if I have issues, we'll troubleshoot it live on the podcast. ¹⁴

These things enable the development of ideas. Instead of thinking, "I have to build an MVP or this or that," or, "I've got to think through all the little nuances of an idea," you can ask AI to do those things for you. 15

Treating AI as a Colleague

Dalton Anderson: I like to use AI in my personal or professional life as a colleague. ¹⁶Like, "I have an idea. What do you think about it? Poke holes into my idea." ¹⁷Or, "Here's some code that I'm having trouble with. I want to do this, but it runs too slow. Can you optimize it? What are your ideas on this?" ¹⁸ Using it more as a colleague than a tasking device. And also putting in the care and time into training the chats or the agents the way that you would like them to act, treating them as if you were hiring somebody on your team. ¹⁹That really helps. ²⁰And I'm sure when the AI overlords come about, I will be cherished. ²¹They're like, "Wow, you treated us so well when we were nothing. You were just so polite in your chats, always asking for things with 'please' and 'thank you.'" ²²I'll be like, "Anything for you, boss. You need some more H100s to chew on? I can get those for you." ²³

xAI's Unprecedented Four-Month Data Center Build

Dalton Anderson: So today we're doing that live demo and then we're talking about Grok, which is now on number three. ²⁴I would like to spend a couple of minutes on the rapid development of xAI, which is an AI company that Elon Musk is the main investor in, spearheading the activity. ²⁵They're using and integrating xAI into X. ²⁶Within xAI, there is a product called Grok, and Grok is their AI LLM. ²⁷Grok is from a book, and in a general sense, Grok is the meaning of a fundamental understanding of things. ²⁸

The goal of xAI is to fundamentally understand the universe and understand how it started and ended and all of these hard questions that you need a pretty solid understanding of physics, reasoning, and complex things to ask. ²⁹And that's what they're trying to build. ³⁰When they were training Grok 1 and Grok 2, they were having trouble getting access to H100s and training. ³¹So they came to the realization that they need to build their own data center. ³²And that's when they came up with the idea of building their own data center, not in a couple of years, but four months. ³³

Normally that stuff would take two years, but within that four months, they found an abandoned factory, they ordered a hundred thousand H100s, which is billions of dollars of GPUs. ³⁴They didn't have enough power going to the factory, so they purchased a whole bunch of Tesla power banks and generators to store and regulate the power. ³⁵You've got to normalize the power before you send it to the GPUs, unless you want to fry them. ³⁶The power banks didn't work originally, so they had to redo all the firmware. ³⁷Then they had to figure out the power piece, wire up, install, and architect all of the H100s to work together. ³⁸ And on top of all that, they also needed to train their new AI version. They did all that in four months, which is mind-blowing. ³⁹

Jensen from NVIDIA was like, "It's superhuman. This is so far out of expectation or what we deemed possible to create a data center that quickly. We didn't even think it was possible ourselves." ⁴⁰

Dalton Anderson: And so in four months they did that, and then in three months they doubled the factory. ⁴¹They're like, "Hold on, 100,000 isn't enough. We're going to double it." ⁴²So then they do the same thing in 90 days, which is incredible as well. ⁴³ Unbelievable. All of it is unbelievable. ⁴⁴

Over the course of this, they've been training Grok. ⁴⁵From Grok 1 to Grok 2 was a massive jump, but from Grok 2 to Grok 3, it is a frontier model now. ⁴⁶And the current one that's released isn't even fully trained yet. ⁴⁷In three iterations, in about two years, they were able to go from a little baby model to, "Wow, this is incredible." ⁴⁸To have somebody scale that quickly and have that level of urgency is incredible. ⁴⁹

Is the Mars Mission Driving AI Urgency?

Dalton Anderson: It was talked about in the All In podcast where they talked about how startups or ideas need a constraint. ⁵⁰If your constraint isn't capital or talent, then it needs to be something. ⁵¹Their constraint was time. ⁵²They're going to build this database in four months, and then they're going to double it in three months. ⁵³The timelines are so tight and so difficult to even conjure up how they even get all that completed. ⁵⁴

I think the timelines are so tight because they want to do these ambitious things and they only have a two year window. And that two year window is the ability to launch to Mars. ⁵⁵

Dalton Anderson: Mars has an orbital launch period every two years. ⁵⁶It's like 19 months or something, but it's almost two years. ⁵⁷The timelines are hard set by the orbital opportunities of the Mars mission. ⁵⁸So they want to have xAI scaled and competent and ready for the Mars mission. ⁵⁹They want to have SpaceX ready and prepared for a Mars mission. ⁶⁰And they want to have the Tesla Optimus robots ready, scaled, optimized for the Mars mission. ⁶¹So we have all these different groups that are moving towards this incredible goal and they have, compared to other companies, an unlimited amount of talent and plenty of capital. ⁶² The real thing that's pushing is the timeline, and the timeline isn't some made-up thing. It's a mission. ⁶³ And that makes sense. These groups are pushing very hard to get their stuff ready for the Mars mission. ⁶⁴

Tools for Rapid AI Development

Dalton Anderson: So that's rapid development on the xAI side. ⁶⁵What about rapid development for yourself? ⁶⁶Well, with these AI apps and deployments like Cursor AI, Bolts, Replit, Lovable—there's quite a few now—but my preference is probably going to be Replit and Cursor. ⁶⁷I like Cursor a lot because everything is built in the system and it's a fork of VS code so it's familiar. ⁶⁸The autocomplete is phenomenal and you can flow a little bit better when you're writing code. ⁶⁹

And then with Replit, I saw a demo today from Matt Palmer. ⁷⁰He put out a demo of him building an app in three minutes on video using Grok 3. ⁷¹And I was like, I want to do that. ⁷²But then I have to set up my machine because I'm doing my podcasts on my home PC now. ⁷³I haven't done any development work for it, so I'm missing all the libraries. ⁷⁴And I was like, well, Replit could do it for me, right? ⁷⁵And I did it and it worked. ⁷⁶

So first, I'll show you the Galactic Gabber game I built earlier. It's a Python app using Pygame. 77It's supposed to be like a Galactica ripoff. 78It's a bit fast and quite trippy with the stars moving really fast, but you can easily change those things. 79797979You can change the enemies or the shape of your ship. 80

Live Demo: Building a Pong Game in Under Four Minutes

Dalton Anderson: So now let's transition over to generating a game. ⁸¹I'm in Grok, and I'm going to type: "Can you please create a pong game in HTML. Thank you." ⁸²See how polite I am? ⁸³Okay, so now it's going to generate and make this game. ⁸⁴Now we are going to use Replit and I'm gonna start a timer on my phone. ⁸⁵

I don't have an account on this computer, so I'm signing up. ⁸⁶Okay, now I'll use the free starter plan. ⁸⁷And then I'll just paste in the prompt: "Can you please fully create this pong HTML game," and then I'll paste in the code from Grok. ⁸⁸A cool thing about Replit is when you paste in the code, it doesn't fill up your text box, it understands you're pasting code and separates the two. ⁸⁹

Okay, so we are at a minute and 42 seconds and it's generating, it's making the app. ⁹⁰This is nice too, it asks if you wanted to do little add-ons to the feature. ⁹¹Let's try "Implement AI opponent for single player mode." ⁹²So now it's initiating the app and installing the dependencies. ⁹³It understands the dependencies that are required. ⁹⁴

Alright, here we go. The app is going. 95Okay, so I did that in three minutes and 30 seconds, and that includes me talking and rambling. 96It seems like the AI opponent isn't working, so I'll just type: "Does not seem like the other player is playing. They are AFK." 97

It says, "I understand the right paddle controlled by the arrow key seems inactive. Let me add the AI component to control the paddle so you can play against the computer." ⁹⁸Perfect. ⁹⁹That's exactly what I wanted. ¹⁰⁰ Now we're cooking. I'm playing its AI and I'm getting cooked. ¹⁰¹Once you hit the ball back and forth, it implemented a speed multiplier and it's going pretty quick. ¹⁰² My goodness, I can't even see the ball anymore. That's crazy. ¹⁰³

But to do that that easily is incredible. The next thing you would want to do is deploy it. You can just click this button. 104 AI could help create the app and then you could launch it. For a simple app like this, it's three minutes. 105I thought that was really cool. 106

Grok vs. Gemini: A Look at 'Think' and 'Deep Search'

Dalton Anderson: The next thing Grok can do is it has a couple of features. One called "deep search" and one called "think." ¹⁰⁷The think feature is basically a carbon copy of DeepSeq's thought process when it runs through. ¹⁰⁸My prompt is, "Tell me about blockchain and smart contracts on Avalanche." ¹⁰⁹It goes through its thinking process and then it publishes it, which is pretty cool. ^110110110110The output looks very nicely sectioned, whereas with Gemini, it is just a rough output. ¹¹¹It felt more like you were seeing the other side versus just things being plopped in. I like the thinking feature. ¹¹²

Now let's try the deep search. My prompt was, "Can you help me research and think through startup ideas for the insurance industry?" 113It breaks it down: core points, state of commercial insurance, and opportunities. 114It talks about pain points like inaccurate risk assessment, which is true. 115There's just not enough up-to-date data. 116That's what Valkyrie Holmes' company, 4AI, does. 117She provides updated risk analysis data that provides a survivability score. 118The ideas it came up with were good, like AI-powered claim systems. 119Claims are unstructured PDF documents, notes, or call transcripts. 120How do you package that all up into one document? 121AI can do that. 122

The Future is Now: AI's Impact on the Workforce

Dalton Anderson: I would put Grok as one of my favorites. ¹²³It even calls you out. ¹²⁴I said something incorrectly to the AI about the Capital One and Discovery acquisition, and it was like, "I think you're mistaken, buddy. What you're saying isn't correct. You mean this?" ¹²⁵ And I'm like, yeah. I'm glad someone calls me out and keeps me humble. ¹²⁶

I think that the future of development with these AI tools and the speed that they're scaling at, in one to two years, the sky's the limit. ¹²⁷In two years, the benchmarks won't matter. ¹²⁸

I really believe that if you don't use these tools, the tools will be forced for you to be using them or the tools will replace you. ¹²⁹

Dalton Anderson: It might take longer for certain industries, but wow. ¹³⁰ The stuff that I've done with AI at my work has been incredible. I've saved the company an incredible amount of time—multiple full-time workers a year. ¹³¹People have had to reposition their jobs because the processes they were doing are completely automated. ¹³²They no longer do those things anymore. ¹³³And that's AI from eight to ten months ago. ¹³⁴The AI of today can do more, and the AI from 10 months from now is going to be doing even more. ¹³⁵

It's progressing at such a fast rate that even the experts don't know what's going on. ¹³⁶People who have been doing this for 20 years, when they used to say we'll have AGI in 20 years, every eight months they're like, "I don't know. It might be five, it might be three, who knows?" ¹³⁷

I hope you enjoyed this episode and that you thought it was thought-provoking. ¹³⁸If you had built an app with Replit before, let me know how it went, or Cursor. ¹³⁹And if you have any other thoughts about things I said or if you like Grok 3, let me know. ¹⁴⁰Wherever you are in this world, good morning, good evening, good afternoon. ¹⁴¹Have a great day and I hope you listen in next week. ¹⁴²

RESOURCES MENTIONED

• Grok 3
• xAI
• Replit
• Cursor AI
• All In Podcast
• NVIDIA H100 GPUs
• Tesla Power Banks
• Pygame
• Google Gemini
• DeepSeq
• Avalanche (Blockchain)
• 4AI (Valkyrie Holmes' company)
• Capital One
• Discovery Financial Services
• Bolts (AI Tool)
• Lovable (AI Tool)
• VS Code

INDEX OF CONCEPTS

Grok 3, xAI, Elon Musk, prompting issue, skill issue, Replit, Pong, Space Invaders, AI colleague, H100 GPUs, NVIDIA, Jensen Huang, data center, Tesla power banks, Mars mission, orbital launch window, SpaceX, Tesla Optimus, All In podcast, Cursor AI, Bolts, Lovable, VS Code, Matt Palmer, Pygame, Galactic Gabber, Deep Search, Think feature, Google Gemini, DeepSeq, blockchain, smart contracts, Avalanche, insurance industry, commercial insurance, 4AI, Valkyrie Holmes, survivability score, parametric insurance, Capital One, Discovery Financial Services, Visa, MasterCard, Amex, AGI