OpenAI's new model has breakthrough 3D modeling powers

Welcome back. This Friday brings incredible excitement and concern for us as developers. With OpenAI’s new Astra model, we can finally rethink our existing design and coding workflows. But there is a darker side to this sudden leap in capability that nobody saw coming. The CEO of OpenCode just flagged a sophisticated deepfake scam designed to hijack AI compute. Read the details below. 

Also: A list of cool projects built with Astra, a Claude skill for easier PR reviews, and a new hiring method for freshers. 

OpenAI’s new model is a powerhouse for developers and designers: The ChatGPT maker just dropped GPT-6 Astra, and it does way more than just code. In one demo, it used 3D modelling softwares like a pro and was able to pull up a 3D house model with ease. This viral comparison shows Astra crushing Fable when it comes to designing 3D models. On the coding front, Astra already beats out top models like Sol and Anthropic’s Fable on both accuracy and API costs. OpenAI claims it’s better at hunting bugs, running terminal tasks, and answering complex codebase questions. Check out the prompting guide here. 

Scammers deepfake a tech CEO to steal AI compute: OpenCode just dodged a wild scam. Their CEO reported a "customer" hopped on a video call and deepfaked a real CEO in real time, almost scamming them out of $400,000 for higher rate limits. The red flags they noticed were pretty basic; it involved a lookalike domain, some strange foreign traffic, and fake urgency about boarding a flight. They caught it by DMing the real CEO on Instagram.

Google’s coding agent faces heat from developers: The search giant just updated the rules for its agentic coding platform (Antigravity). Previously, the terms suggested that using third-party tools like OpenClaw could get your entire Google account banned. Senior engineers like Gergely Orosz weren't having it. Google has now clarified that a ban will only affect your Antigravity and Gemini CLI access, not your whole Google account. Despite the response, Google is still facing backlash from the dev community.

Most of the world's data sits in documents your agents can't read. LlamaParse by LlamaIndex is an agentic OCR that turns PDFs, scans, tables, and handwriting into agent-ready context. The most accurate and cost-effective document parsing solution available, ahead of any frontier lab or specialized API.

In production at startups like Lovable, Tabs, and Retell, and enterprises like KPMG, PepsiCo, and Carlyle.

Sign up today with code THECODE250 and start with $250 in credits. Upgrade to Pro in 30 days and pay 50% rate for 3 months.

Mandates don't work. Changing how one engineering team works is hard enough. But rolling out AI adoption or new review habits across 40 teams that don't report to you? That's a different ball game. It can be done without authority you don't have, and we broke the full approach into a field guide.

Just send an email. Take this example from Amazon. A senior leader, tasked with cutting error rates for a product they didn't own, started sending a monthly email naming the biggest error sources, the teams behind them, and the leaders in charge. Soon executives were asking when the next update would arrive, and error rates plummeted week after week.

Nobody's in charge, and that's the point. That email is one piece of a bigger machine. The full method comes down to three people:

Back to basics. Authority was never the missing piece. You need the right people in the right roles and a scorecard that makes people pay attention. One lead at Unity used this play to ship code 33% faster and cut review times 70%.

P.S. Ready to level up dev productivity? DX's Core 4 framework breaks down exactly what to measure.

How to avoid the 19 Claude Code mistakes pros still make: In this tutorial, you'll learn how to sharpen your prompting, keep token costs down, and set up autonomous workflows. By skipping these common traps, you'll create an AI dev environment that's way more efficient, reliable, and easy on your wallet. 

Diffusion Studio (1.1k ⭐): This video editor keeps the canvas and code perfectly in sync so any update to one reflects instantly in the other. It even features a CLI that lets coding agents inspect footage and cut timelines on the fly. 

6 ways to test agents with eval engineering (by Google): This cookbook teaches you how to test AI agents before you put them in production. It gives six methods to check structure, facts, tool order, quality scores, model changes, and known cases. You can use these methods as automated tests in your build process.

Claude Code's 5-hour session limit lands mid-task and stops you cold. This dev spotted a new slash command that clears it on the spot.

Two catches: it works once a week, and it adds no tokens. The session reopens, but everything you burn still counts toward the weekly cap.

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

This dev found a single command that exposed every flaw in his Claude prompts, from redundancies to contradictions. It worked so well; he’s now running it on all of them.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 300K+ 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