OpenAI's Codex SUPERAPP LEAKED, Sam Altman's House Attacked & Claude Takes Over Microsoft Office!

Open AI is turning Codex into a super app. Sam Altman had a Molotov cocktail

app. Sam Altman had a Molotov cocktail thrown at his house and wrote a blog post about it. And Anthropic just completed their full Microsoft Office takeover. Let's get into it.

takeover. Let's get into it.

Open AI is quietly turning their Codex app into something much bigger. And

based on what's been spotted [music] in the code, it could land as soon as next week. Right now, Codex is known as a

week. Right now, Codex is known as a coding tool. That's it. But what's being

coding tool. That's it. But what's being built underneath it tells a completely different story. According to the leaks,

different story. According to the leaks, Open AI is building chat, Codex, and Open Claw all into a single Mac app, one platform, and they're using Codex as the foundation to build everything else on

top of it. So, what does that actually look like? There are four things being

look like? There are four things being added. First, a rendering layer for

added. First, a rendering layer for images and video. So, this isn't just a text environment anymore. Second, a

heartbeat system. So, if you use Open Claw, you know what this is. It's a

background automation that runs on a schedule, checks in on your work, surfaces what needs attention, and one of the screenshot shows an automation called the next move scout running every

30 minutes, finding loose ends and flagging the next useful action. That is

not a coding assistant. That has become a coding agent that works alongside you.

Third, model and thinking mode selection per task. So, you'll be able to pick

per task. So, you'll be able to pick which model handles which job, almost like an agent manager inside the app.

And fourth, UI changes designed to make Codex feel less like a tool for developers and more like something anyone can use. The last one is the tell. Open AI isn't building a better

tell. Open AI isn't building a better coding assistant, they're building a super app, and Codex is just the shell they're using to get there. On top of the super app story, Codex itself also

shipped version 0.12 yesterday. Agents

can now stream live progress while they're still running, so you're not sitting there waiting with no feedback.

MCP tools now support output schema for type results, and the TUI separates live and completed hooks, which makes it easier to track what's done versus what's still in flight. So, the CLI

version of the Codex keeps getting more capable while the Mac app is being rebuilt into something that goes way beyond coding. Tibo, who actually works

beyond coding. Tibo, who actually works at Codex, kind of hinted at what next week is going to look like. He said

it'll be more than cooking. Then

somebody made some predictions regarding the Codex super app combining the chat, coding, personal agents, and browsing with a killer feature being how they're all work together. He also predicted

GPT-5.5, codenamed Spud, drops the same week. I don't know about that one, but

week. I don't know about that one, but Greg Brockman described it as the culmination of 2 years of research built with new pre-training methods. Spud is

supposed to be the first model that actually feels like it thinks the way a human does, which I think is a bold claim, and we'll cover it properly when it actually drops. But the Codex super

app is the biggest structural story.

Open AI has ChatGPT on mobile, but they have the API and they have the canvas.

The Mac desktop has been fragmented. If

they pull this off, they have one native app that handles conversation, coding, and long-running agents all using the right model for the task. That's the

real product. And if the timing is accurate, we're about a couple of days away from seeing whether the execution matches what they are saying.

This video is sponsored by Hostinger. If

you're running AI tools and automations, you need somewhere reliable to host them, and that's where a VPS comes in. A

virtual private server gives you your own dedicated slice of a cloud server, full root access, dedicated resources, and is always on. No need to keep your laptop running, no interruptions, no

worrying about your machine going to sleep in the middle of a workflow.

Hostinger is the platform to host your VPS. The dashboard is clean and

VPS. The dashboard is clean and beginner-friendly. There's a built-in

beginner-friendly. There's a built-in browser terminal, so you don't need any extra tools to get started, and their VPS plans are some of the most affordable ones out there for what you actually get. Whether you're

actually get. Whether you're self-hosting automation tools, running AI agents, or just want a reliable cloud environment to work out of, Hostinger makes it simple. And here's the cool

part. You can deploy cloud code directly

part. You can deploy cloud code directly on a Hostinger VPS. Head into your dashboard, open the browser terminal, and grab the single curl command from the cloud code docs. Paste it in, hit

enter, and cloud code is installed in about 10 seconds. Then just type cloud in the terminal, sign in with your pro or max cloud subscription, and you're live. Cloud code running in the cloud

live. Cloud code running in the cloud 24/7 from anywhere. Once it's running, you've essentially got a genius AI developer living on your server around the clock. You can use it to manage your

the clock. You can use it to manage your VPS, clone GitHub repos, spin up Docker containers, build apps, whatever you need. Check the link in the description

need. Check the link in the description and use code Universe of AI at checkout to grab a discount on your first Hostinger VPS plan. All right, now back to the video. Sam Altman posted

something early yesterday morning that he almost didn't publish. The reason he was up writing at 3:45 a.m. was because

someone threw a Molotov cocktail at his house. It bounced off, luckily nobody

house. It bounced off, luckily nobody got hurt, but he was awake and pissed, and decided to say some things he'd been sitting on.

The post covers three things: what he believes, personal reflections, and his take on the industry. And the industry section is the part people are talking about. He wrote that his takeaway from

about. He wrote that his takeaway from the last several years of drama between AI companies comes down to one line.

Once you see AGI, you can't unsee it.

And he describes it as having a ring of power dynamic. Not that AGI is the ring

power dynamic. Not that AGI is the ring itself, but the philosophy of being the one to control it. That mindset, he's saying, is what makes otherwise rational

people do irrational things. It explains

a lot of the feuds, the lawsuits, the public blow-ups you've seen in this space. His proposed solution is

space. His proposed solution is straightforward. Don't let anyone have

straightforward. Don't let anyone have the ring. Spread the technology broadly,

the ring. Spread the technology broadly, keep democratic institutions more powerful than the companies building it.

He's explicit that he doesn't think it's right for a handful of AI labs to make the most consequential decisions about humanity's future.

He also gets personal. He says he's proud of holding the line against Elon Musk during the early days of Open AI, when Elon reportedly pushed for unilateral control over the company.

He's not proud of being conflict-averse in other situations, which he admits caused serious damage to the company and to people around him. He calls himself a flawed person in an exceptionally

complex situation, trying to get a little better each year. There's also a brief section where he addresses the Molotov cocktail more directly. He

posted a photo of his family, says they're normally private, and says he's sharing it hoping it might dissuade the next person. He says there was an

next person. He says there was an article about him a few days before the attack, and someone warned him it was making things more dangerous. He brushed

it off, now he's not brushing it off anymore. That's the state of things

anymore. That's the state of things right now. The CEO of the most prominent

right now. The CEO of the most prominent AI company in the world is writing blog posts at 4:00 a.m. after an arson attempt asking people to turn down the temperature. Whatever you think of Sam

temperature. Whatever you think of Sam Altman or Open AI, that's a genuinely strange moment to be living through right now.

Claude is now inside Microsoft Word.

They had Excel since October of last year, PowerPoint launched in February, and now Word is live in beta, which means Claude is natively embedded across the entire classic Office suite. The way

it works is a persistent sidebar inside Word. You draft, edit, and revise

Word. You draft, edit, and revise documents directly from there, and all of Claude's edits appear as native Word track changes. So, you can review and

track changes. So, you can review and accept or reject them the same way you would with any human editor. You had

Excels, numbering schemes, formatting rules, all of it stays intact. But the

more interesting thing here is what happens across all three apps together.

Claude maintains shared context across Word, Excel, and PowerPoint in a single conversation thread. So, you can start

conversation thread. So, you can start by pulling data in Excel, build an exact deck in PowerPoint using the same analysis, then jump to Word to write the draft accompanying the report. And

Claude carries the full context the entire time. So, you don't have to

entire time. So, you don't have to copy-paste, and you don't have to re-explain the same data three times.

Claude for Word also lets you pull numbers directly from an open Excel model into a Word memo, or summarize a document into presentation slides without manually moving anything between

the apps. Early users have noticed

the apps. Early users have noticed smoother document handling and more coherent multi-app context flow compared to Microsoft's own 365 Copilot, which

already has deep Word integration.

That's a real comparison to make because Copilot is deeply embedded into Microsoft ecosystem by default.

Anthropic is coming in from the outside and still making that argument credible.

Access right now is gated to team and enterprise plan subscribers. Anthropic

says broader rollout is coming. So, if

you're on one of those plans, you can install it now directly from Microsoft app store. But the main message over

app store. But the main message over here is that Anthropic has quietly built a native productivity layer across the tools that a huge portion of office workers use every day. Cloud code for

developers, Claude for office for everyone else. That's their real

everyone else. That's their real enterprise strategy.

But that's it for today's video. Make

sure you guys subscribe to the channel.

Follow our new newsletter as well at universeofai.beehive.com,

universeofai.beehive.com, as well subscribe to the main channel World of AI, and support us on X by following the Universe of AIZ as well.

Until then, I'll see you guys in the next video.