OpenAI’s bug-hunting AI earns a ‘Critical’ risk rating

Welcome back. OpenAI just launched a new security model, and it’s already helping fix some major vulnerabilities across the internet. Meanwhile, a new wearable promises to rethink the way you collab with your AI agents. 

Also: How to use Notion without a subscription, a personal AI agent that claims to be unlike anything else, and why Microsoft is building a new data center every four days.

OpenAI’s new security model earns a ‘Critical’ risk rating: The ChatGPT maker just rolled out GPT-6 Cyber, which can hunt down vulnerabilities and suggest fixes. It has built working exploits for 948 Linux kernel flaws and uncovered 81 high-severity Firefox bugs that Mozilla is fixing. Only vetted Daybreak Red customers get access. But critics say OpenAI could be doing more. According to this report, its defender program only reaches thousands, and its Linux bug hunt cost $1.2 million versus the roughly $6 million it spent on a fluid dynamics problem.

This startup wants you to wear your agents: If you’re tired of unlocking your phone every time an agent needs your approval, former Ultrahuman exec Apoorv Shankar has a fix. His hardware startup Aina built a $399 ring with a screen, joystick, mic, and speaker. It puts Zoom controls on your finger and lets you dictate or talk to agents. You can also write custom triggers in plain language and route unsupported apps to agents like Muse via virtual browsers. Ships January 2027. Preorder here.

Google’s new video model lets you edit through chat: The search giant just rolled out Gemini Omni 1.2 Flash to all developers and businesses, making AI video creation cheaper and easier. The model generates 15-second clips at $0.07 per second for 720p, down 30% from before. You can now tweak videos by chatting, changing backgrounds or lighting, without generating a new clip from scratch. It’s handy if your app needs dozens of variations. You can try it in Google AI Studio or start building with the Gemini API here.

Zep Agent Skills turns your agents' traces into step-by-step skills they read at runtime, over MCP or the Zep SDKs.

Start building with Zep Agent Skills →

50x faster. So what? Imagine making Rust builds 50 times faster and deciding it’s not worth a blog post. Well, that’s exactly what senior engineer Dan Luu did. While devs complained about slow Rust compiles, he had an AI agent build an interpreter that made cold builds roughly 50x faster, by his informal estimate. Five years ago, that would have been a big deal. Today, Luu didn’t think it deserved a writeup. That says a lot about how devs approach AI.

The old math is broken. In a new essay, Luu argues that devs should be building bigger, more ambitious software now that agents are around. But most still size projects by how much work they used to take. Luu also built a regex engine and a ripgrep fork with no experience in either. Work he’d once have set aside a year for took a few hours.

Chase the slow bits. Luu’s tips for picking what to build:

Slop needs a safety net. Luu admits most AI rewrites are slop, so testing is everything. He relies on fuzzing and property-based tests rather than traditional unit tests. And your own coding skills still matter. With every new model, Luu tackles harder problems where AI still struggles. He argues that programming skills are more valuable than ever. You can read the full essay here.

How this dev uses Grok Bot to run his coding workflow: This tutorial shows how to use Grok Bot as the command center for your coding agents. You’ll learn how to orchestrate projects, delegate work across T3 Code, Codex, and Claude Code, and manage the whole workflow from one interface instead of bouncing between agents.

Untangle Codebase: A new agent skill for cleaning up messy codebases before they become a nightmare to maintain. Point Claude Code, Cursor, or Codex at a repo, and it helps remove copy-paste, leaky abstractions, temporary workarounds, and pointless wrappers so the code is easier for both you and your agents to navigate and change.

How to make your CI 20x faster: Boardy's head of engineering breaks down how his team slashed PR wait times from 40 minutes down to just 2. You will learn where your CI pipeline actually loses time, what is worth caching or running in parallel, and how to run smarter tests so your coding agents never get stuck waiting on builds. 

A dev shared a fix for quota that resets half-unused: point Codex’s /goal command at the cruft months of vibe coding left behind.

The full thread has four more targets: duplicated logic, unused dependencies, stale sessions, and zombie scheduled tasks.

These 15 hands-on projects teach you to secure AI agents from cyber attacks. A must read for all developers.

If you enjoyed this newsletter, you should definitely check out our IT and cybersecurity newsletters. Subscribe using the links below:

We put together 20+ hands-on guides covering coding agents, agentic engineering, and leadership frameworks to help you level up your career. You can get them here.

Advertise with us: Join companies like Google, IBM, and Datadog. Showcase your product to our 350K+ 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