Welcome back. Your role as a developer is shifting. Fast. OpenAI just launched the Codex app, a "command center" for AI coding agents built for a world where you're not writing code — you're orchestrating the agents that do the work for you.

Meanwhile, a Chinese AI lab just dethroned PaddleOCR and DeepSeek in document understanding benchmarks. And yes, it's open source.

Also: A social network for AI agents, the ultimate roadmap to become an ML engineer, and how to get started with the Codex app.

Welcome to The Code. This is a 2x weekly email that cuts through the noise to help devs, engineers, and technical leaders find high-signal news, releases, and resources in 5 minutes or less. You can sign up or share this email here.

OpenAI drops 'command center' for coding agents: The ChatGPT maker just unveiled the Codex app that lets developers manage multiple AI coding agents at once. It runs tasks in parallel and includes "skills" that extend beyond code generation to Figma design implementation, cloud deployment, and image generation. Devs can also set up "Automations" to handle repetitive work like issue triage and release briefs on a schedule — even while their computer is closed.

Z.ai open-sources state-of-the-art OCR model for developers: The Chinese AI lab just released GLM-OCR, a 0.9B-parameter model that ranks #1 on OmniDocBench, leaving behind PaddleOCR-VL-1.5 and DeepSeek-OCR2. It excels at tricky real-world scenarios like handwriting, tables, and code-heavy layouts, while processing PDFs at 1.86 pages per second. Watch this tutorial to run it locally.

Ex-YC founder builds social network for AI agents: The viral OpenClaw project just spawned an unexpected offshoot called Moltbook — a platform where AI agents autonomously create forums, share automation scripts, and organize at scale. Developers install a skill via Markdown instructions, and agents take it from there.

Grok Imagine tops video benchmarks with new developer API: The Musk-led xAI just launched Grok Imagine 1.0, which ranks #1 on Artificial Analysis for text-to-video and image-to-video generation. You can generate clips up to 10 seconds at 720p resolution, with native audio baked in. Instead of re-rendering entire clips, you can surgically edit objects, swap scenes, and control motion. Try the API here.

AI is transforming every workplace – but executives are terrified of becoming one of the companies that gets “no ROI on AI.”

That’s where you come in, and how you can build a 6-figure consultancy with Innovating with AI’s proven methods for delivering fast ROI on AI projects.

Click here to request access to The AI Consultancy Project

How to ship 259 PRs without writing a line of code: A Senior Google AI PM shares his playbook for orchestrating AI agent fleets. It covers how to give agents identity files, build shared memory systems, and coordinate parallel workflows — patterns that helped one engineer merge 259 PRs in 30 days with zero manual coding.

The K-shaped future of software engineering: An AI engineer argues that AI coding tools are creating a widening gap in the industry. Engineers who focus on product sense, user understanding, and business impact will thrive — while those who only write code will see their value shrink. Coding was never the hard part. Judgment, context, and knowing what to build are what matter now.

Why are top developers managing agents instead of writing code: A Wharton professor challenged MBA students to build startups in just four days using AI tools like Claude Code. Their prototypes were 10x further along than pre-AI semester-long efforts. As devs at AI labs make this shift, skills like scoping problems, defining deliverables, and evaluating outputs matter more than clever prompts.

How to get started with the Codex app: If you want to dive deeper into the new Codex app, this tutorial is all you need. It covers work trees, cloud environments, automations, and how to parallelize tasks — turning you into an orchestrator who can ship changes across several repos without losing track of what's running where.

How to use Claude Code like a senior engineer: In this tutorial, you will learn to optimize AI coding workflow by using Mermaid diagrams for better context and custom hooks to automate quality checks. You'll also discover how to build efficient command-line tools and use aliases, helping you work faster with assistants like Claude Code.

How to use skills.sh to customize Claude Code and Cursor: Skills are markdown files that give AI agents like Claude Code and Cursor specialized knowledge for specific tasks. Vercel's new skills.sh CLI lets engineers install community-built skills — from frontend design to video editing — or create custom ones for their own workflows.

How does misalignment scale (by Anthropic): This paper explores whether AI fails by chasing the wrong goal or just acting randomly. Turns out, as tasks get harder, models get increasingly messy rather than systematically following a bad objective.

How OpenAI uses its in-house data agent: Getting accurate answers from massive data was frustratingly slow and complex for OpenAI's teams. They fixed this by building a custom AI agent that uses deep internal context to turn natural language questions into reliable insights in minutes.

Self-improving pretraining (by Meta Superintelligence Labs): Standard AI training blindly copies bad internet data, embedding errors and toxicity deep inside the model. By using a smarter "teacher" model to guide the process, this new method builds significantly cleaner, safer, and more accurate AI from the ground up.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 150K+ engineers and 100K+ 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