Latest / Elon Musk Podcast / Elon Musk’s xAI Is Using AI to Rebuild Microsoft’s Software Stack

Transcript

Elon Musk’s xAI Is Using AI to Rebuild Microsoft’s Software Stack · Elon Musk Podcast

• 0:01
  Hey everybody. Welcome back to the Elon Musk
• 0:04
  Podcast. This is a show where we discuss
• 0:07
  the critical crossroads, the Shape, SpaceX, Tesla X, The
• 0:11
  Boring Company, and Neurolink. I'm your host Will Walden, Elon
• 0:17
  Musk's ex AI is developing a internal project called Macro
• 0:21
  Hard that aims to recreate Microsoft's core system
• 0:26
  ecosystem using AI alone. It goes beyond a typical
• 0:32
  research experiment and instead reflects Musk's intent to
• 0:36
  simulate, replicate, and eventually compete with
• 0:41
  Microsoft head to If you have a podcast player
• 1:03
  that has comments, I want to let you know that you're allowed to
• 1:07
  voice your opinion. I want you to voice your opinion
• 1:10
  on this. Is it a smart idea to allow AI
• 1:14
  to replicate code and also create more code in fix bugs on
• 1:19
  the fly? For enterprise businesses, that
• 1:22
  means big business, big, big business.
• 1:26
  I'm not talking 10 people, 20 people, 30 people.
• 1:29
  We're talking 10s of thousands of people.
• 1:32
  Would you trust your 10s of thousands of people in your
• 1:35
  organization to use AI software? Or do you want the homemade
• 1:40
  version where people create the software and then if you have a
• 1:45
  bug, you tell it to somebody and then that somebody tells it to
• 1:49
  somebody else and it goes down the chain instead of just robot,
• 1:54
  you know, and AI writing all the code.
• 1:57
  It's very interesting. So Elon Musk instructed XAI
• 2:04
  employees to use AI models to rebuild Microsoft Office,
• 2:09
  Microsoft Excel, Word, Teams, Windows, and even GitHub and
• 2:16
  PowerShell. Musk selected the name Macro
• 2:20
  Hard as a deliberate play on Microsoft's branding.
• 2:24
  Microsoft He expects the project to eventually challenge
• 2:28
  Microsoft's stronghold over software that underpins the
• 2:31
  majority of modern enterprises and governments.
• 2:34
  Do you want an AI in charge of your software that you use on a
• 2:38
  daily basis? They can see everything that you
• 2:40
  do. They know exactly what you do,
• 2:43
  all your keystrokes. Could they be tapping into all
• 2:49
  of your movements? Yeah, OK, that's just a
• 2:52
  conspiracy theory that was just to be kind of, you know, weird
• 2:55
  about it. But think about it.
• 2:57
  If you're tapped in in the cloud all the time anyway, could they
• 3:01
  track everything you do to, quote, make their software
• 3:05
  better? Could it be the end user
• 3:08
  license, you know, agreement that you have to sign off on
• 3:12
  that everything you do has to be tracked by this AI.
• 3:16
  And do you want XAI in particular to be in charge of
• 3:20
  that software? XAI is going to be used, Grok
• 3:26
  and other XAI models to execute real world software workflows
• 3:30
  without human engineering. Basically going to be able to
• 3:32
  prompt this thing and say I want to, I want these specific
• 3:37
  things, these specific features, I do it all the time.
• 3:40
  Specifications you, you write a spec sheet and then you have
• 3:44
  something like Grok or any other AI program you know that that
• 3:50
  they want to use. They can build the software.
• 3:54
  I can build software right now. Things that would take me that
• 3:58
  would have taken me weeks to build, even a month to build.
• 4:02
  I can instruct my AI companion to build something for me at
• 4:11
  lovable. Lovable dot dev.
• 4:14
  I did something the other day. Here's an example of what I did,
• 4:19
  and this is a very simple example.
• 4:21
  I had a spec sheet. I knew exactly what I wanted,
• 4:24
  knew exactly how it was going to run from the first user
• 4:28
  interaction on the homepage. It was a web app first
• 4:31
  interaction on the homepage all the way through the process of
• 4:35
  check out etcetera. And when I told the AI to write
• 4:40
  it, he told me this would be the phases, phase one, phase two,
• 4:45
  you know, five phase. I think it was 5 or 6 phases,
• 4:47
  I'm not sure. But eventually about two hours
• 4:51
  later, we had a fully functional MVP of the software that would
• 4:56
  have taken me two weeks to write probably.
• 4:59
  And then of course, that's the MVP.
• 5:02
  And that actually works through Stripe.
• 5:04
  There's e-commerce functionality, e-commerce pages.
• 5:07
  All the templates were there, everything was there.
• 5:08
  Everything was built out without using any of my hands to code
• 5:13
  it. I didn't use anything.
• 5:16
  I used Tailwind, Veet, React, etcetera to build this thing out
• 5:22
  and it was pretty damn good. So building something like Teams
• 5:29
  for an AI model, not going to be that hard.
• 5:33
  I, I'm assuming they're going to have this done by the end of
• 5:39
  this year, at least a really good model of it.
• 5:42
  Because if you can use AI, because also Microsoft already
• 5:46
  built this that, you know, use, building the code to make
• 5:50
  Windows like a Windows 11 clone is going to be a little bit more
• 5:54
  difficult. But something like an app like
• 5:58
  Microsoft Word, that's not going to be that difficult compared to
• 6:03
  say, a full operating system. Something like Teams, not as
• 6:07
  difficult as a full operating system.
• 6:09
  They want to use make PowerShell, Excel, Microsoft,
• 6:13
  basically Microsoft Office. They're not going to copy it
• 6:17
  exactly though. They're going to be able.
• 6:19
  They're just going to do this to demonstrate that EI systems can
• 6:22
  replicate the same capabilities through reasoning and code
• 6:25
  synthesis. So the lead devs will tell it
• 6:30
  what to do. We need something similar to
• 6:32
  Microsoft Word, but maybe they're going to do something
• 6:35
  like I did with Lovable and they were like, just ask it a
• 6:40
  question. How do you think the software
• 6:42
  would be better? I did that throughout the
• 6:45
  development cycle in at points. How would you make this software
• 6:48
  better? And tell me the steps you would
• 6:51
  do, why it's better, how it's better, and then give me
• 6:56
  examples of all of those things. And it did in most of the time,
• 7:01
  probably 99% of the time. It was really excellent at
• 7:05
  making those decisions. The user flow was at the
• 7:10
  beginning, it was an MVP. So it was very basic.
• 7:13
  User flow was just like point A, point B, Point C, you know, And
• 7:18
  they had other ways to flow the user to the end cycle, to the
• 7:25
  buy cycle in just fast and furious ways to get the job
• 7:30
  done. It was crazy.
• 7:32
  And that's just lovable. Could you imagine something like
• 7:35
  Grok who has all of the data? Or you know, I've used other
• 7:42
  apps too. I'm a front end developer,
• 7:45
  Claude, great at coding ChatGPT, it's pretty good, but it's, I
• 7:52
  think Claude is a little bit, I don't know, for the stuff that I
• 7:55
  do, it's a little bit better, but have Xai behind it, have
• 8:01
  Elon and Grok behind it. It's going to be something, you
• 8:04
  know, something ridiculous, I'm sure.
• 8:10
  The other thing is this is an internal target includes
• 8:16
  recreating Excel with AI agents alone.
• 8:19
  And those AI agents understand spreadsheet logic.
• 8:22
  They'll be able to generate formulas, suggest corrections,
• 8:26
  all without using any existing code.
• 8:29
  They're going to create the code completely from scratch, not
• 8:33
  going to use any libraries. Another team is focused on
• 8:37
  building a fully AI generated OS layer with basic window
• 8:42
  management, file storage, and application support.
• 8:46
  Again, completely from scratch, no frameworks.
• 8:50
  And if this is successful, Elon would be the first to show the
• 8:54
  general purpose AI can autonomously generate complex
• 8:58
  commercial software from first principles.
• 9:02
  No part when they start, there's nothing.
• 9:05
  It's just, you know, C++, you know C#, you know whatever
• 9:09
  language it is and you build this thing because you we're
• 9:14
  going to give you the prompt to build it.
• 9:16
  Crazy macro hard is not being built for external users right
• 9:21
  now. It will eventually, I'm sure.
• 9:24
  I'm sure they'll license this out or find a way to make money
• 9:27
  from it. The internal documentation
• 9:30
  frames the project is a stress test for grok in the bet.
• 9:34
  The general EI can reproduce software systems through self
• 9:38
  learning. That's how we get Skynet, man.
• 9:41
  This is the software builds the software.
• 9:45
  The part builds the part That's crazy.
• 9:47
  The initiative emerged shortly after Open A is growing presence
• 9:51
  in enterprise tools became clear, including its
• 9:53
  partnerships with Microsoft. Elon is, I wouldn't say jealous
• 9:58
  of Sam Altman. He's competing.
• 10:01
  He's competing directly with Sam and ChatGPT and Open AI.
• 10:07
  I think Elon is a little behind the game at this point.
• 10:13
  Grok chatbot, pretty cool with some stuff.
• 10:17
  It's all right, but it goes off the rails sometimes and open AI
• 10:23
  as well. It goes off the rails a little
• 10:24
  bit. Open AI.
• 10:26
  ChatGPT 5 wasn't such a big thing as everybody thought it
• 10:32
  was going to be iterative. You know, we had between 3:00
• 10:35
  and 4:00, ChatGPT, it was a huge upgrade.
• 10:40
  And everybody thought between 4:00 and 5:00 was going to be
• 10:43
  this massive gigantic switch, but it wasn't.
• 10:49
  Yeah. It was like, they're just fine
• 10:51
  tuning, ironing out some bugs, fixing some stuff up, making it
• 10:57
  a more solid platform before they move forward to GPD 6 and,
• 11:03
  you know, all the AIS after that.
• 11:07
  So Microsoft Open AI, they're working together, of course.
• 11:10
  And Elon thinks, well, I can do it better because he's also
• 11:16
  already criticized Microsoft's deep ties with Open AI.
• 11:20
  He doesn't believe that an quote open source AI company like Open
• 11:26
  AI should have funding from a closed source company like
• 11:31
  Microsoft. And Microsoft's invest billions
• 11:34
  in Open AI and tightly integrated its models into
• 11:39
  Office. And Musk has expressed concerns
• 11:43
  in the past about the influence Microsoft holds over Open AI's
• 11:47
  operations. Internally, XAI researchers view
• 11:50
  macro hard as a response to what they consider model capture,
• 11:54
  where enterprise players monopolize how AI can be
• 11:57
  deployed at scale. XAI is attempting to do the
• 12:00
  opposite by removing the human middleware and allowing Grok to
• 12:04
  act as the architect, the engineer, the tester, and the
• 12:07
  user interface designer. Teams within XAI are using
• 12:12
  custom prompts in reinforcement learning techniques, the train
• 12:15
  agents that iterate on their own mistakes and optimize outputs
• 12:19
  based on user like goals. So instead of simply being a
• 12:24
  coding assistant like Copilot, Grok is being trained to become
• 12:29
  a full stack developer and a product designer in one.
• 12:32
  It's kind of what Lovable does. Lovable dot dev, if you haven't
• 12:37
  checked it out, if you want to make a web app or anything in
• 12:41
  React, basically the time is now.
• 12:46
  Go to Lovable dot dev. I'm like, we're not sponsored by
• 12:48
  them, but check it out. I use them all the time.
• 12:52
  If I have AFO project, I give it the specs and of course I
• 12:58
  disclose that I'll be using AI to build some of the software
• 13:02
  and come up with with some great ideas Lovable has.
• 13:07
  And how funny is it though? Something like Lovable has its
• 13:11
  own ideas from the systems that it's already been implementing.
• 13:16
  It's learning as it goes as well.
• 13:19
  And XAI is, I wouldn't say catching up to it, not yet.
• 13:25
  If they could build this OS, if they can build a kind of a
• 13:28
  presentation layer of a Windows OS, Windows Cell OS, yeah, it'd
• 13:33
  be incredible. So the larger context though, of
• 13:36
  macro Hard includes a growing concern within Musk's network.
• 13:40
  The Microsoft's influence over AI infrastructure has become too
• 13:44
  concentrated, and by rebuilding key enterprise functions through
• 13:48
  autonomous agents at XAI, they want to demonstrate that open
• 13:52
  alternatives are possible. Could we, could they build this
• 13:56
  and make it sort of like Linux, an open sourced alternative to
• 14:01
  Windows that could run Windows games?
• 14:04
  I think that's where the logic goes from here.
• 14:09
  Elon, he's a big gamer. And I bet you guarantee you one
• 14:13
  of the one of the prompts from Elon to his engineering team was
• 14:18
  make it able to play games eventually.
• 14:22
  Like that's what we want. We want it to be a Steam killer,
• 14:25
  the Steam Deck. We want it to be a Windows
• 14:27
  killer and we want it to be able to play Windows games.
• 14:31
  Can they do that? I don't know.
• 14:32
  That's going to be crazy if they can.
• 14:36
  But Musk hasn't clarified yet if they're actually going to
• 14:39
  release this. But he suggested internally that
• 14:42
  such AI created software should run on a future Grok OS that
• 14:47
  avoids reliance on legacy platforms.
• 14:50
  Make everything open source, everything that he wanted to do
• 14:53
  with open AI. Think he gave him 50 million or
• 14:56
  something, 500 million. I can't remember the the price
• 14:58
  that he gave them or how much money he gave them, but I
• 15:01
  believe he's kind of hurt because they went from being
• 15:06
  open AI, open AI where they're going to share everything to Sam
• 15:12
  Altman getting ridiculously rich from this and all this
• 15:17
  stakeholders in this hoping that it everybody becomes ridiculous
• 15:21
  rich, ridiculously rich and the bubble doesn't burst.
• 15:25
  But as we see now, legacy systems like Google, Microsoft,
• 15:30
  all the initial Internet bubble Internet companies that are
• 15:35
  still around, you know that Yahoo is still around.
• 15:39
  Yeah, it's hard to see something like that unless you're into
• 15:43
  sports and like sports betting and stuff like that or fantasy
• 15:49
  sports. But yeah, Yahoo is still around.
• 15:52
  It's a thing. And you know what?
• 15:55
  It's a legacy company. Same with Microsoft, same with
• 15:57
  Google. They're legacy companies now.
• 15:59
  They do think different, do things differently.
• 16:01
  They don't think of the open source anymore, really.
• 16:03
  They have a small arm of open source, but if Elon can make a
• 16:07
  whole operating system and an office software suite open
• 16:10
  source, that would be ridiculous.
• 16:13
  If they release this, they'll be nuts, but I don't know if I'd
• 16:17
  trust it. Let me know.
• 16:20
  Again, like I said before, I need to know what you think.
• 16:23
  Let me know what you think about this because it sounds a little
• 16:27
  too personal. AI building the software,
• 16:33
  reiterating the software, revising the software, make an
• 16:37
  AI infrastructure so it keeps building and rebuilding
• 16:41
  software. I mean, sure, it's going to be
• 16:42
  great software eventually, but do you want an AI in charge or
• 16:48
  you don't want people in charge? I it's up to you.
• 16:50
  I mean, I have no problem with AI building software.
• 16:55
  I use AI to build software all the time.
• 16:57
  I think it's great. It's a great cheat code.
• 17:00
  You unlock certain things within the software development
• 17:04
  platform, like Lovable, and it just goes wild.
• 17:11
  You can create some really incredible things, Very complex
• 17:15
  things too. Very, very complex things.
• 17:19
  I've built some full blown apps, release them to the public, and
• 17:24
  they're crazy. You know?
• 17:28
  XAI or Elon has not commented publicly on the details of this,
• 17:32
  though you know there are. The internal prototypes already
• 17:38
  performed basic functions that mirror those of Office and
• 17:41
  Teams. One prototype allows Grok to
• 17:44
  compose documents real time and revise them based on simulated
• 17:49
  team collaboration using a chat interface powered by agent
• 17:54
  driven back ends. This is also going to serve a
• 17:59
  deeper technical purpose for XAI.
• 18:02
  By forcing its models to handle complex software tasks without
• 18:05
  pre-existing APIs or templates, the company is generating a
• 18:09
  stress testing environment for the new versions of Grok.
• 18:13
  The new models. These future systems will likely
• 18:16
  integrate self repairing code. Oh my, this is this is getting
• 18:20
  scary. Autonomous workflow planning and
• 18:23
  adaptive UI generation features that Musk believes are necessary
• 18:27
  for AGI to become practical. It has to think for itself self
• 18:34
  repairing its code. It's kind of like, how do you
• 18:37
  think you you make mistake and because of that mistake, you fix
• 18:42
  it, right? You fix whatever mistake it was,
• 18:45
  you make yourself a better person.
• 18:47
  Self repairing code, we do that, we already do that.
• 18:51
  We repair the wrongdoings. So once this autonomy comes to
• 18:59
  fruition, AGI will become more practical.
• 19:05
  Elon is aiming for broader things.
• 19:11
  He's going to be figuring out how software will be written in
• 19:15
  the future, desktop software or software that creates desktops
• 19:20
  and creates operating systems. And if XAI can figure this out,
• 19:26
  of course Tesla bots, Optimus, Tesla vehicles, etcetera, will
• 19:32
  all use this in the back end to make their code faster, more
• 19:38
  reliable, stronger, more productive.
• 19:43
  And guess what bye bye all software engineers, everyone's
• 19:49
  getting fired. So future AI with croc no
• 19:55
  people, no no no engineers anymore.
• 19:59
  There will be some high level engineers but all low level
• 20:02
  after the high like anywhere from medium till in turn are
• 20:06
  gone. You know, but I think for the
• 20:09
  time being, expect to not have your $165,000 starting salary.
• 20:15
  If you are a MIT grad. Expect to work at Chipotle.
• 20:20
  Because once things like this hit the market, that career is
• 20:26
  dunzo. And once Grok can figure out how
• 20:29
  to take out the C-Suite of people, and once other AIS can
• 20:36
  figure out how to take out the C-Suite and build their own
• 20:37
  software to to destroy the C-Suite, it's all over.
• 20:42
  The business will be handled by mostly AI, and then we'll just
• 20:48
  buy the things from the AI. It's going to be wild.
• 20:54
  Hey, thank you so much for listening today.
• 20:56
  I really do appreciate your support.
• 20:58
  If you could take a second and hit this subscribe or the follow
• 21:00
  button on whatever podcast platform that you're listening
• 21:03
  on right now, I greatly appreciate it.
• 21:05
  It helps out the show tremendously and you'll never
• 21:08
  miss an episode. And each episode is about 10
• 21:11
  minutes or less to get you caught up quickly.
• 21:14
  And please, if you want to support the show even more, go
• 21:17
  to patreon.com/stage Zero. And please take care of
• 21:22
  yourselves and each other. And I'll see you tomorrow.