- Magic Beans
- Posts
- GitHub's Independence Dies
GitHub's Independence Dies
Microsoft absorbs the platform into CoreAI as CEO exits to start over

Hey everyone, it’s Cosmo.
This week’s quick take: GitHub lost its CEO and slid tighter under Microsoft's CoreAI. VibeCode raised $9.4M to bring app-building to your phone. Anything launched v1.0 and face-planted in a hands-on. Bolt rolled out Cloud to close the prototype-to-production gap. Anthropic pushed Claude Sonnet 4 to a 1M-token context window, can now read whole codebases in one shot.

Image Credit: The Verge
GitHub CEO Thomas Dohmke resigns; platform folds tighter into Microsoft's CoreAI
Details
Dohmke leaving after 4 years "to become a startup founder again," staying through end of 2025 for transition
Microsoft won't replace the CEO role; GitHub leadership reports more directly into CoreAI under Jay Parikh
Ends 7 years of promised independence since $7.5B acquisition in 2018
GitHub Copilot: 20M users (up from 15M last quarter), drives 40% of revenue growth
Translation: less standalone GitHub, more alignment with Microsoft's AI platform and tooling
Bottom Line Expect Copilot-everywhere and a Microsoft-shaped roadmap. Great if you're all-in on MSFT; fewer "Switzerland" vibes for the rest.

Image Credit: Business Insider
VibeCode raises $9.4M to put app-building on your iPhone
Details
iOS app lets you describe what you want in plain language; AI generates and iterates React Native code on-device
Seed round: $9.4M led by Alexis Ohanian's Seven Seven Six, with Long Journey Ventures, Neo, First Harmonic, Afore, and angels from Google, OpenAI, and Expo
Model support expanded this week: Claude, OpenAI's new GPT-5, Kimi K2, and Qwen 3 Coder
Traction: launched in June; hit #12 in App Store "Developer Tools"; >40K apps generated; plans $20–$200/month
Founded by Duke alums with previous exits targeting "next billion developers"
Bottom Line Mobile is the last big unlock for vibe coding. If VibeCode nails iPhone-to-App Store deployment and polish, "apps as content" goes from trend to pipeline.

Image Credit: Anything
Anything (formerly Create) ships v1.0—proceed with caution
Details
Rebrand: Hands On Create → Anything; claims 30K MAU and >20K projects/day
Pitch: "as easy as selling t-shirts," "idea to App Store to first customer in days"; plans range free to $80/month
Hands-on test: generated volunteer management app (React/Next.js + Neon Postgres) failed to run—logs showed hundreds of errors, preview stuck on "Loading"
User reviews mixed: "published in 18 min" vs "never-ending loop" and calls for actual testing before launch
Bottom Line A clean landing page can't paper over messy deployments. The gap between promise and production is where vibe projects die. Anything isn't over that gap yet.

Image Credit: Bolt
Bolt launches Bolt Cloud to bridge prototype → production
Details
What's live: built-in hosting (via Netlify) and custom domains—publish straight from Bolt interface
Rolling out in private beta: managed databases, auth & user management, Stripe payments, functions, file storage, analytics
Partners: Netlify and Supabase provide enterprise-grade backend infrastructure
Goal: kill the "glue work" (DNS, infra wiring, brittle pipelines) that stalls vibe-coded apps before they ship
Traction: Sacra estimates $40M ARR by March 2025, growing from $4M within 4 weeks of launch
Bottom Line If Bolt actually unifies build → deploy → scale, it eats the boring parts that drain momentum—and becomes the default lane for shipping vibe-coded products.

Image Credit: The Verge
Anthropic bumps Claude Sonnet 4 to a 1M-token context window
Details
5× jump to 1,000,000 tokens; from analyzing ~20K lines of code to handling 75K–110K lines in one go
Available now for higher-tier API customers (Tier 4/custom rate limits); broader rollout "in the coming weeks"
Also acqui-hired Humanloop team (prompt management, LLM evaluation specialists) as enterprise push continues
Revenue hitting $5B run rate, up from $4B earlier this month—coding clients driving growth
Comes week after OpenAI's GPT-5 launch in escalating AI coding arms race
Bottom Line Bigger memory, fewer hacks. Whole-repo refactors and audits become a single request, not a fragile dance of chunks and RAG.
Thank you for reading. Till next week! 😊
Best,
Cosmo