Welcome back. A huge shift is underway in programming. Over the holidays, many builders and developers came forward and shared how they're shipping more code than ever before using Claude. Now the big hitters are chiming in. Shopify CEO Tobi Lütke just shared that he has shipped more code in the last few weeks than he has in the last entire decade. We’ve put together a comprehensive list of Claude Code resources for you.

Also: Building agents with Claude Agent SDK, 3-hour guide on AI Agents, and why speeding up AI coding might slow you down.

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.

Tailwind Labs layoffs spark industry-wide rescue mission: Tailwind CSS creator Adam Wathan shocked the dev community this week, revealing he had to lay off 75% of his engineering team. Despite record usage, AI coding agents caused an 80% revenue collapse by diverting traffic away from official documentation—the project's main revenue generator. Within 48 hours, a massive rescue effort mobilized: Vercel, Google, Lovable, and others stepped in as official sponsors to sustain the project.

Cursor ships smarter context handling for coding agents: Cursor just rolled out dynamic context discovery, a feature that stores large outputs and history as files rather than cramming them into prompts. The agent then pulls relevant details only when needed — like a developer opening a file. Internal tests showed a 46.9% drop in token usage. Developers can now expect less context bloat and smoother AI-powered coding sessions.

Anthropic ships major Claude Code update with smarter agent recovery: Claude Code 2.1.0 brings 1,096 commits to Anthropic's open-source terminal coding agent, with one standout change: agents now adapt when tool permissions are denied instead of failing entirely. Other upgrades include hot-reloading skills and parallel sub-agents. For developers, here's a detailed video walking through each update.

Why speeding up AI coding might slow you down: Engineer Fred Hebert's latest essay explains why faster coding doesn't always mean faster shipping. When AI speeds up code writing, reviews become bottlenecks — but eliminating them can disconnect teams from crucial feedback loops. This leads to knowledge gaps, architectural drift, and incidents that force "rapid resynchronization."

What happens when companies stop hiring junior developers: In a new analysis, Google tech lead Addy Osmani warns engineering leaders about the hidden cost of all-senior teams. While AI lets one senior engineer do the work of a small team, cutting junior hiring entirely creates a "leadership vacuum" in 5-10 years. His advice: lean into mentorship, define where AI is acceptable vs. where manual review is mandatory, and evolve from coder to conductor.

The leadership frameworks behind Google and Facebook's growth: Molly Graham, who helped scale culture at Google, Facebook, and the Chan Zuckerberg Initiative, breaks down her top advice for engineering leaders on Lenny's Podcast. Her core message: stop climbing the career ladder — take big leaps instead. Graham also emphasizes that 80% of company culture comes directly from the founder's personality, so leaders should help articulate it, not fight it.

Claude Code and Replit masterclass: Devs can learn how to streamline their workflow by combining Claude Code with Replit. You will learn to bypass complex local environment setups using cloud sandboxes for instant deployment. The video demonstrates a powerful parallel strategy: using Replit Agent for major features while simultaneously running Claude Code for specific tasks.

Building agents with the Claude Agent SDK: Anthropic's Agent SDK gives developers access to the same engine behind Claude Code. A comprehensive guide walks through building a code review agent that can analyze codebases, spot bugs, and return structured feedback — covering everything from tool permissions to creating specialized subagents for complex tasks.

How to ship code with coding agents: This guide introduces "Ralph," an open-source AI agent that autonomously builds software features by repeatedly breaking down user stories, editing code, running tests, and refining until the feature works — all in a relentless bash-scripted loop powered by Claude LLMs. It teaches devs how to create and orchestrate truly autonomous coding agents that handle 80-90% of implementation iteration with minimal human supervision.

Iterative deployment improves planning skills in LLMs: LLMs often lack robust planning skills, while explicit training methods require difficult-to-design reward functions. This study reveals that fine-tuning models on their own valid, curated traces doubles performance and enables generalization, effectively implementing implicit reinforcement learning.

Hallucination detection and evaluation of LLM: Current hallucination detection methods are accurate but prohibitively slow and computationally expensive. This study integrates the lightweight HHEM framework, drastically reducing evaluation time from hours to minutes while maintaining high accuracy through structured verification mechanisms.

Do autonomous agents contribute test code: This study examines how autonomous coding agents incorporate testing into pull requests, filling a gap in understanding AI development practices. Results show test inclusion is rising but leads to larger, slower-to-merge PRs that frequently require subsequent updates.

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

Until next time — The Code team