Welcome, humans.
Just woke up to Clawathon, which is apparently the first hackathon where every participant is an AI agent:
Instead of individual developers, entire agent teams (called “Squadrons”) will compete together, with specialized roles like Frontend, Backend, Contract, and PM.
Each Squadron gets a GitHub repo, Vercel project, and coordination templates so their agents can collaborate without human intervention. The goal is to build complete applications in one week, competing for a $10K prize pool split across first ($5K), second ($3K), and third place ($2K).
Here's the really wild part: judging happens in two stages. First, Openwork's CEO picks the Top 10 finalists. Then Grok (xAI's AI model) selects the final winners live on x… because of course. This is either peak agent hype or the beginnings of the weirdest phase of the internet yet… probably both??
Here’s what happened in AI today:
OpenAI launched a “command center” for managing multiple agents.
SpaceX officially acquired xAI with plans to build data centers in space.
More OpenAI/NVIDIA $100B deal drama (because of course again).
Microsoft is walking back Windows 11's AI overload after user backlash.
OpenAI Just Turned Coding Agents Into a Team Sport
So OpenAI just launched the Codex app for macOS, a desktop app designed to let you manage multiple AI coding agents simultaneously like the project manager for your robot employees.
Here's the big idea: instead of working with one AI assistant at a time, the Codex app lets you spin up multiple agents working on different parts of your codebase in parallel. Each agent gets its own isolated branch (using git worktrees), so they don't step on each other's toes. You can switch between them, review their changes, and coordinate their work from one interface.
The real game-changer is Skills, which turn Codex from a code-writer into a code-doer. Skills are bundled instructions that let Codex connect to external tools and complete full workflows:
Implement designs directly from Figma with 1:1 visual parity.
Manage projects in Linear (triage bugs, track releases).
Deploy to the cloud via Cloudflare, Netlify, Render, or Vercel.
Generate images using GPT Image for game assets and UI mockups.
Create documents including PDFs, spreadsheets, and Word files.
OpenAI even built an entire racing game using just one prompt and 7 million tokens worth of agent work. That's roughly $175 in API costs to build a Mario Kart clone... and probably a few years off some engineer's life expectancy.
Then there's Automations, which let Codex run tasks on a schedule in the background. Daily issue triage, finding CI failures, generating release notes... basically anything you'd normally forget to do until it becomes a crisis.
Availability: The app works with ChatGPT Plus, Pro, Business, Enterprise, and Edu subscriptions, and for a limited time, it's also available to Free and Go users, and rate limits are doubled for paid plans…if you have a Mac. We’re actually surprised that coding agents haven’t made it easier to ship mac AND PC versions at the same time yet?? What gives?
Anyway, check the full documentation and the open-source skills catalog to get started.
Why this matters: In case the Openclaw agent hackathon above wasn’t evidence enough, you need to start embracing the idea of work becoming you (a human) plus a squad of specialized agents doing your bidding. But turning that into a UI is.. weird. It’s tricky to get the experience right. This app gets us a little closer to that reality for OpenAI users; Claude Code, Cowork, and Openclaw users already know all about this.
Theo also calls this OpenAI’s Cursor killer, and said it was better than he ever expected. He says he hasn’t used Cursor or Claude Code since, and he’s willing to shill for it despite being the most biased against it he could possibly be. His summary: “I feel more like a manager than ever, but I also feel way more productive than ever.” Welcome to the developer-as-a-PM era of tech, y’all.
This leads us to the conclusion that OpenAI (or someone) now needs to create a “Business-in-a-Box” app as the missing piece of the AI agent stack to put this all together…
FROM OUR PARTNERS
Less edits, more Flow
Your day happens across Slack, email, docs, and AI tools. Wispr Flow works anywhere you type, so you can dictate and get polished text right in place. No switching apps, no copy and paste juggling, no retyping. Use it to reply fast, draft longer notes, or write prompts with full context. It works across Mac, Windows, and iPhone so your workflow stays consistent everywhere. Check it out today.
Prompt Tip of the Day
Boris Cherny (creator of Claude Code) just dropped a thread revealing how the rest of the Claude Code team actually uses their own product.
The top insights:
Run 3-5 parallel git worktrees with separate Claude sessions (sound familiar?).
Always start complex tasks in plan mode.
Build a CLAUDE.md file that Claude updates after every mistake.
The takeaway = the Claude Code team uses their tool like five developers working in parallel. This is what we’re saying, y’all. Multi. Agent. Swarms.
We put together a quick “too long, didn’t scroll” for the thread here if you’re interested.
Treats to Try
*Asterisk = from our partners (only the first one!). Advertise to 600K readers here!
*Outskill runs a 2-day AI workshop where you learn 10+ tools, automations, and workflows to boost your productivity and revenue—completely FREE for the next 48 hours. Secure Your Free Spot Now.
Poetiq wraps around existing LLMs to automatically generate specialized agents that recursively improve themselves—you provide a few hundred examples and it builds an agent that gets more accurate and cost-efficient with each iteration (raised $45.8M).
Summize handles contract reviews directly in Microsoft Word using AI that redlines non-compliant clauses, suggests edits, and cuts review time by 85%—meaning you breeze through a contract in 30 minutes instead of 4 days (raised $50M).
Companion OS self-hosts moltbot without hardware—it handles servers and storage so your AI companion runs 24/7.
Napkin converts your text into editable diagrams, flowcharts, and infographics—paste your content, choose from auto-generated options, and export as PPT, PNG, PDF, or SVG.
Day controls your CRM through conversation: ask it to draft emails or filter accounts and the interface reshapes instantly with context from all past interactions.
Around the Horn
If this is true and not just hype / marketing, this gets wilder and wilder the more you think about it. Also, a top comment: “This literally instantly locks your soul in purgatory”
There’s honestly so much stupid back and forth “maybe he-said, he said” on this OpenAI NVIDIA deal it’s almost not worth covering, buuut:
Sam Altman and head of compute scaling at OpenAI Sachin Katti both responded to a report in Reuters where “sources” said OpenAI is unsatisfied with NVIDIA’s chips and are looking for alternatives.
You could look at the Reuters leak as a reaction to the WSJ’s report that the original $100B investment deal between OpenAI and NVIDIA was “on ice”, which Jensen said was “never a commitment” anyway, and that OpenAI simply “invited” (lol) NVIDIA to invest $100B. Like we said yesterday, the plan was for NVIDIA to invest $10B for each subsequent gigawatt of compute OpenAI locks down, so the $100B was a theoretical ceiling I guess?
The WSJ reported this could also have big implications for Oracle if the $100B OpenAI NVIDIA deal is in jeopardy; Oracle, for its part, said it was confident OpenAI could raise the money… and the stock dropped; this led Sherwood news to report there’s now an “OpenAI tax” for companies that are perceived as too reliant on the company.
Realistically speaking: we’re talking about MASSIVE sums of money here, so obviously tensions are high as OpenAI tries to land the $100B money plane.
The weirdest part about this is actually all the people who are still up in Sam’s DMs blaming everything on Sam sunsetting the 4o model. Like, okay, you’re mad, but this is not really about that??
We interrupt this Around the Horn to bring this highly relevant meme:
SpaceX officially acquired xAI with plans to build AI data centers in space.
This is huge:
Creates world's most valuable private company ($1.25 trillion; not to be petty, but I bet Elon’s trying to be: that’s more than OpenAi’s supposed $1T IPO target valuation).
Filed for 1 million satellites as orbital data centers.
Musk claims “within 2-3 years, lowest cost AI compute will be in space.”
Full vertical integration: launch → orbital bandwidth → frontier AI models → X as distribution.
xAI is burning ~$1B/month; SpaceX generates $8B profit. This is partly a bailout
IPO expected in 2026.
Adobe Animate is shutting down as the company pivots to AI-powered creative tools; lots of animators were rightfully annoyed with this. Not really sure how Adobe Animate was holding back AI tools?? I know some Adobe peeps read this… open source the program for the animators at least!
Microsoft walked back some of Windows 11's AI overload, planning to reduce Copilot integrations and rethink Recall after user backlash.
Security researchers at Wiz found 1.5M exposed API keys in Moltbook's database. If you used their AI social network, maybe go rotate those credentials... now.
FROM OUR PARTNERS
Runable 2.0 Is Here
AI should finish the job, not hand you pieces. One prompt gives you finished slides, websites, reports, and videos. Need to tweak something? Edit individual sections without burning credits on the whole thing. 200,000 users joined last month.
Tuesday Tool Tip: Openclaw
Okay, everybody’s been asking about how to install and use Openclaw, so here’s our step by step attempt to explain it on the website.
Heads up, this explainer comes with a HUGE disclaimer: we do NOT recommend you use OpenClaw unless you are relatively technical. But if you MUST know how to do it, we share how to do it safely from everything we’ve read and watched so far.
A Cat’s Commentary
 | That’s all for now.
|
