Google unveils its new flagship model to every developer

Welcome back. For more than seven months, Google preferred faster models over more capable ones, but that changed yesterday. The search giant just dropped a new flagship model in its Gemini family that developers are talking about. Meanwhile, OpenAI put its models on detective duty after discovering corporate espionage in its research lab.

Also: How this AI company ships code without a code review, the new home gadget Apple’s CEO just revealed, and why Zoho’s co-founder told staff to reinvent themselves.

Google unveils its new flagship model to every developer: The search giant just made Gemini 4 Argon available worldwide through the Gemini API, AI Studio, and Antigravity. It can generate up to 1M tokens per response, easily tackling complex, multi-step work across massive codebases. Internally, Google says Argon boosted agentic code submissions by over 35% while slashing rollbacks by nearly a third. Introductory pricing starts at $2 per million input tokens. Try it here.

Agent-written code forces Anthropic to replace Git: The AI lab just revealed it replaced Git with Scion, an in-house version control system built specifically for agents. Daily merges surged from around 2.5K in May to more than 30K by September, pushing Git past its limits. A seven-person team managed to migrate the entire monorepo in five months. CI checkouts now take seconds instead of minutes. The company expects other engineering teams to hit the same scaling bottleneck within a year. Read the full breakdown here.

OpenAI catches an employee planting a spy device in executive meetings: The ChatGPT maker alerted the FBI after finding a hidden recording device plugged into a conference room TV. According to prosecutors, an employee on the product team planted it. OpenAI analyzed the bug using its own models and found it was designed to dodge detection. So far, the company says there is no sign that model weights, source code, or customer data were compromised. Watch the entire story.

AppSec and CloudSec were never meant to work apart.

AI has collapsed the time between finding a flaw and exploiting it, and siloed security tools cannot keep up. This playbook shows how a single security graph connects code to cloud so your teams fix the risks that matter first.

See how a unified approach gives defenders the complete picture and the speed to stay ahead.

Get the Playbook

A new kind of in-house tool. A growing number of tech companies are building in-house AI app builders for non-technical employees. Gergely Orosz, a former Uber engineering manager and the writer behind the popular newsletter The Pragmatic Engineer, set out to understand why. He visited Ramp's New York office and looked into a similar tool at Stripe, uncovering a quiet shift in who builds software at work.

What he saw inside. At Ramp, nearly every non-engineer uses Glass, the company’s own take on the AI app builder Lovable. Staff build tools for their own work, then share them so colleagues can use and change them. One marketer even built a custom CRM. Glass is now so central that building a simple tool in it is part of onboarding for new hires.

Why not just buy one? That raises an obvious question, since tools like Lovable and Replit already exist. For Orosz, the answer mostly comes down to control. An in-house tool can safely use company data that would be risky to share with an outside vendor, and it follows the company’s own design rules. It’s also cheap: Stripe’s version, Harbor, took one engineer about six weeks.

The payoff goes beyond the tool. Building these platforms also gives engineers hands-on AI experience, a skill that is tough to hire for right now. Factor in the low cost and high adoption, and Orosz expects plenty of tech companies to follow suit. For teams considering their own rollout, this practical guide walks through how to build internal AI tools, from picking an approach to reviewing what gets built. 

Vibecoded apps handle critical work across your company today.

Most of them are run locally, haven't cleared DevOps and touch production systems in ways you can't govern.

Major puts real guardrails on every build: secrets management, SSO, CI/CD.
Set once, enforced everywhere.

Your team is free to build with AI, and you keep full visibility and control over what goes live.
Book a demo

How senior engineers build with AI (full Intercom clone): You’ll learn how to build a full-stack Intercom clone from scratch using AI coding tools. You will learn to manage system architecture, write product requirement documents, and build working production apps.

OpenQodex: An open source AI code reviewer. Instead of paying $60 a developer, get the same coverage for $0 with over 100 scanners and an AI reviewer that investigates every finding using your own Claude Code or Codex login before you push. 

How to build an anti-slop software factory: When AI agents write and review code at scale, small errors quickly snowball into an exponential mess of low-quality slop. To prevent this, teams need to optimize their repository setup with strict lint rules, file co-location, and strong type safety, making it nearly impossible for agents to slip bad code through. 

Claude Code can write multi-file changes faster than you can review them, but you still have to sign off before merging. This GitHub repo closes that gap by using Claude Code to turn any git diff into an interactive, guided walkthrough, just like a senior engineer would read it. 

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

Zoho’s co-founder just told employees there will be no pay raises this year. His reason, and his warning for software jobs, is going viral (1.2M views).

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