Google's new model hunts and patches security bugs

Welcome back. Every AI lab and startup is racing to win over developers. Yesterday, Factory's CEO accused a VC board advisor of leaking sensitive information to rival Cognition, and developers are grabbing the popcorn. Meanwhile, Google and Grok Bot just rolled out powerful updates for you to try out. 

Also: Anthropic engineer’s game development workflow, Meta engineer shares how to use Sonnet 5.5, and what Apple’s cooking up for your smart home.

Google's new model hunts and patches security bugs: The search giant just launched Gemini 4 Argon in limited access for complex engineering, enterprise, and cybersecurity work. It autonomously finds, verifies, and patches critical security flaws, leading the DeepSWE benchmark at 77.9% while minimizing hallucinations. Google also introduced Skills, letting you turn recurring workflows into reusable templates. Try here.

Coding startups feud over loyalty: Factory's CEO just accused his VC board advisor of spying for Cognition. On Wednesday, Matan Grinberg said he'd fired Chris Degnan for allegedly leaking secrets to Factory's biggest rival. Two hours later, Degnan announced he'd joined Cognition as revenue chief and that he had resigned of his own accord. Cognition's CEO also denied the accusations, saying they did not receive any Factory information. Developers are picking sides. See the full timeline.

GrokBot wants to be your coding manager: SpaceXAI shipped Team Bots, shared AI teammates available on its Teams and Enterprise plans. It also added ready-made bots for its personal agent, covering project management, QA, and fact-checking, each with built-in skills. You can tag one directly in Slack, and it delegates the actual coding to Cursor agents while tracking the entire workflow. To see this in action, check out how one of their senior engineers uses the setup in her daily coding workflow. 

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

Too much freedom. Your agent can easily wipe files, rack up huge API bills, or call a job done when it isn't. So who steps in to stop it? That's where a harness comes in to set clear ground rules. Most off-the-shelf agents bury those boundaries inside default settings your team never picked. 

ML Researcher Elvis Saravia’s new walkthrough shows how to put the decisions in your own code. He builds a harness with the Pi SDK and Jev so the agent checks before it acts.

The build has six parts:

The result. You can now read the agent’s limits, model choices, and human handoffs in code. You can try the whole setup in a live sandbox connected to the real Jev with a delete tool included.

8 real ways to use Jev in your apps: This tutorial skips the theory and shows where a fast decision model actually fits. You’ll learn how to use Jev for voice commands, data deduplication, routing users through an app, multi-step classification, and other jobs where you need a quick structured decision instead of a full LLM response.

Open Dot (194 ⭐): A take on OpenAI’s Dots that runs on your Mac. Each agent gets its own logged-in browser, can connect to Gmail, Calendar, Slack, GitHub, and 1,500+ apps, run on schedules or triggers, and even hand work off to other dots.

Cutting GPT-6 agent costs with prompt caching (by OpenAI): This guide shows how to reuse shared context across agent runs instead of paying to process the same instructions and tool definitions again. Cached input tokens can be up to 90% cheaper, and the new dashboard helps you spot cache misses, keep tool definitions stable, and even prewarm context before the user asks anything.

You correct Claude Code for the same mistakes every week, and nothing tracks it. An engineer shared /insights, a built-in command most devs have never run.

The report covers your last 30 days: which projects eat your time, where sessions break down, and ready-made CLAUDE.md rules that fix your most frequent friction. Paste those rules into your config and the repeat corrections stop. Full breakdown in Anthropic's docs.

P.S. Get 50+ AI coding hacks for Claude Code, Cursor, and Codex here.

A viral game lets you point a stickman at any website and destroy it with flamethrowers, grenades, and pure chaos. Devs are loving it.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 350K+ engineers and 150K+ followers on socials. Get in touch.

We put together a few guides on coding agents, agentic engineering, and leadership frameworks to help you level up in your career. Browse all our guides.

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

Until next time — The Code team