Welcome back. The AI wars just went full heavyweight. Anthropic and OpenAI dropped major upgrades on the same day, arriving just thirty minutes apart, and giving developers a serious wave of new models and tools to experiment with.

And while those two battle for the top spot, Apple quietly made a move that could change how every developer builds apps.

Also: Pro-hack to make Claude Code 10x faster, an OpenAI researcher's agentic coding setup, and a practical guide to start coding with AI.

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 arms devs with its most capable coding model yet: The ChatGPT maker dropped GPT-5.3-Codex, combining the best of its latest models into one powerhouse that’s 25% faster. It crushed the Terminal-Bench 2.0 test with a 77.3% score, a 13-point jump over its predecessor in a single generation, while using fewer tokens than any prior model. Alongside the new model, OpenAI unveiled Frontier, a new platform that helps enterprises build, deploy, and manage AI agents that work like employees.

Anthropic unveils Claude Opus 4.6 with powerful new tools for devs: The AI lab’s smartest model just got an upgrade. Opus 4.6 plans more carefully, sustains agentic tasks for longer, and is the first Opus-class model with 1M-token context. Anthropic is giving away $50 in credits to try this model. On top of that, devs also get agent teams in Claude Code, where multiple agents coordinate autonomously in parallel. To prove this, the team set 16 agents loose to build a C compiler from scratch, and they delivered a 100,000-line Linux kernel.

Apple brings agentic coding to Xcode: The iPhone maker just shipped Xcode 26.3 with native support for Anthropic's Claude Agent and OpenAI's Codex, letting developers hand off complex coding tasks to AI agents that work autonomously. Instead of one-off prompts, agents can now reason across entire projects, search Apple's documentation, and visually verify their work through Xcode Previews. The update also supports the MCP, so teams can plug in any compatible agent or tool.

You.com's grounding playbook covers:

Download the technical guide

The skills gap just flipped. A year ago, Andrej Karpathy coined "vibe coding," where you talk to an AI, accept all the changes, and ship without reading the code. He recently retired the term in favor of "agentic engineering." The word swap carries the whole thesis. Coding implies that anyone can do it. Engineering implies you need to know things.

The most experienced engineers are already all in. The Claude Code creator, Boris Cherny, shared that 100% of his recent commits were AI-written. Former Dropbox CTO Aditya Agarwal declared the era of manual coding over after a weekend spent building with Claude. Yet, these veterans represent a unique class of engineers whose every move is backed by decades of high-level judgment.

Data shows why that judgment matters. Anthropic's agentic coding report found that engineers use AI in ~60% of their work, but can fully delegate just 0-20% of tasks. Devs using AI assistants scored 17 points lower on comprehension tests, with debugging skills hit hardest. Meanwhile, junior developer employment at AI-adopting companies has dropped 9-10% whereas senior roles hold steady.

When code gets cheap, the bottleneck moves. Senior Tech Lead Swizec Teller argues the real work is keeping systems alive — uptime, defect recovery, software bigger than one person's brain. Even Cursor Head of Design Ryo Lu argues that AI doesn't replace systems thinking but actually amplifies the cost of not doing it. People who think clearly about systems build incredibly fast, while the rest simply generate slop at scale. If your engineering team hasn't adapted to this trend yet, start with Cherny's guide to learn how the team at Anthropic is implementing agentic engineering.

An AI engineer's workflow is blowing up on X, and it's surprisingly simple. Instead of jumping straight into Claude Code, start in Opus 4.5 with extended thinking.

Dump everything about your project and ask Claude to hit you with five clarifying questions.

Answer them, then prompt it to create a master plan with knowledge base directories and context markdown files:

Import those into Claude Code, and it'll already know your architecture, constraints, and decisions before writing a single line.

How to build AI agent teams in Claude Code: Anthropic's Claude Code now supports Agent Teams, allowing devs to orchestrate collaborative sub-agents that share task lists and communicate directly. This tutorial covers how to configure agent teams, run parallel workflows in split-pane mode, and use the built-in messaging system to keep sub-agents in sync. You'll also learn to tune Opus 4.6's effort levels and set up Auto Memory for persistent project knowledge.

Code-review-expert: A comprehensive code review skill for AI agents that performs structured reviews with a senior engineer’s lens, covering architecture, security, performance, and overall code quality.

Towards self-driving codebases (by Cursor): This research talks about how existing AI models struggle with massive coding projects because they lose focus on complex tasks and fall apart when trying to work together in groups. The breakthrough came from organizing agents into a hierarchy of planners and workers, much like human engineering teams, which allowed thousands of agents to code in parallel without constantly breaking the system.

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