Welcome back. Anthropic co-founder Jack Clark spent the past few weeks reading hundreds of public data sources on AI development, and he's now convinced that AI could build its own successors by 2028. Read his full reasoning here.

Also: Know when sub-agents beat agent teams in Claude Code, an open-source editor to run agentic coding workflows, and what's reigniting the per-message AI pricing debate.

Vercel takes on cybersecurity with an open-source agent scanner: The developer platform just dropped DeepSec, which deploys coding agents to find vulnerabilities across massive codebases. It runs locally with developers' existing Claude or Codex setups and scales to thousands of parallel Vercel Sandboxes. Using Opus 4.7 and GPT 5.5, the scanner traces data flows to cut false positives and exports findings as ready-to-fix engineering tickets.

OpenAI and Anthropic announce strategic deals with private equity firms: The two leading AI labs just locked in billion-dollar partnerships to roll out AI across businesses. OpenAI closed a $10B venture to push tools into TPG’s (The Deployment Company) portfolio companies, while Anthropic launched a $1.5B firm with Blackstone and others to build custom Claude systems for banking and healthcare.

Astro co-founder ships a lightweight Claude Code rival: The ex-CEO of HTML just dropped Flue, a headless TypeScript toolkit for building autonomous AI agents. It uses Markdown files for agent logic and runs on Node.js, Cloudflare Workers, and GitHub Actions. The open-source project picked up over 1,700 stars on its first day.

Your dev stack got an AI upgrade everywhere except the input layer. You're still typing every prompt, every ticket, every review comment by hand.

Wispr Flow closes that gap. Dictate into Cursor, VS Code, Slack, Linear, or anywhere else you work. It's syntax-aware: camelCase, snake_case, acronyms, and file names all come through clean. Mention a file in Cursor or Windsurf, and it auto-tags.

It's the voice layer for an AI-native workflow. Speak your intent. Your tools do the rest.

Available on Mac, Windows, iPhone, and Android. Used by millions of developers, including teams at OpenAI and Mercury.

Try free

X marks the exit. For a full month, HashiCorp founder Mitchell Hashimoto kept a log, marking an "X" every single day a GitHub outage stalled his work. After seeing those marks pile up day after day, he finally walked away after eighteen years, calling the platform "no longer a place for serious work." Even GitHub's CTO has now admitted that the platform just wasn't built to handle the load from AI agents.

The blame doesn't stick. The same agent load hits Vercel, Sentry, and Railway every day, yet they’re all holding steady. GitHub's traffic didn't just skyrocket overnight; it’s been growing steadily for two years. So why is only GitHub melting down?

Late to the curve. GitHub didn’t see it coming. The COO is calling this load "unprecedented," but Google was already preparing for a 10x spike in code volume by mid-2025. GitHub didn't start its own plan until October. By February, the target had tripled. Now, two decades of tech debt and a half-finished Azure migration are dragging down every single fix.

The default broke. Most users won't quit because the ecosystem makes switching feel almost impossible. But GitHub used to be the default that just worked, and that is just not the case anymore. Hashimoto is taking Ghostty elsewhere, and GitLab is suddenly back in the conversation. If GitHub wants that trust back, they're going to have to earn it from the ground up, this time for the agent era.

Whenever Claude runs “npm install” or any shell command, it does so with your full user permissions. A malicious package could easily read your SSH keys, copy your .env files, or access anything else in your home directory.

This fix keeps Claude locked inside your project folder, ensuring that nothing it executes can reach the rest of your machine. Run this command inside Claude Code:

Select "accept-edits mode." Claude can now read, write, and run commands within the project, but the moment something tries to touch a file outside that scope, the OS blocks it immediately.

P.S. You can find 50+ AI coding hacks here.

How Stripe built an internal AI prototyping tool using Cursor: In this tutorial, Stripe design manager Owen Williams introduces ProtoDash, an internal AI tool that redefines product design. It shows developers how to build AI-driven platforms that let teams quickly spin up functional, realistic dashboard prototypes using custom design systems, all without needing heavy coding skills.

Zed: An open-source editor built for agentic workflows that lets you run multiple AI threads in parallel to handle deep refactors and complex tasks simultaneously. You get inline assistance for real-time edits, edit predictions that anticipate your next move, and the flexibility to plug in your own LLM keys or run models locally.

CocoIndex (8.2k ⭐): This repo turns codebases, meetings, and apps into live, real-time context for AI agents. It eliminates stale data and context gaps, letting you build production-ready agents in under 10 minutes with minimal processing.

How to deliver low-latency voice AI at scale (by OpenAI): Scaling real-time voice AI on standard infrastructure usually hits a wall with port exhaustion and latency issues. OpenAI fixed this by building an architecture that routes media packets efficiently, keeping conversations fast and natural even at a massive scale.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 260K+ engineers and 150K+ followers on socials. Get in touch.

You can also reply directly to this email if you have suggestions, feedback, or questions.

Until next time — The Code team