Welcome back. Anthropic is on a roll — and they might just have made every AI agent a domain expert with Claude Skills. Skills is a powerful new feature that lets you train agents through simple folders that include instructions that are loaded as needed.

Want to learn how to use it? Here's a senior developer's step-by-step guide to using Claude Skills.

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.

Anthropic launches Skills to help developers build specialized AI agents: Instead of building fragmented, custom agents for each use case, developers can now package their expertise into composable Skills — organized folders of instructions, scripts, and resources that Claude dynamically loads when needed. Here's a senior developer's step-by-step guide to using Claude Skills.

New models to cut down the time spent searching for relevant code context: Cognition and Windsurf just dropped SWE-grep and SWE-grep-mini, specialized models built to find the right code up to 20x faster than competing tools. Here's why it matters: coding agents typically waste over 60% of their time just searching for relevant context in large codebases. The new models fix this with parallel searches and smart retrieval. Here’s a tutorial on how to use it.

DeepSeek just released a 3B OCR model: The Chinese AI giant launched DeepSeek-OCR which enables developers to compress long contexts optically at 10× ratios with 97% precision using minimal vision tokens. The open-source model processes documents faster than competitors, helping engineers build scalable OCR pipelines and generate massive training datasets efficiently.

Engineering Leader’s Guide: How to Become a Great Coach and Mentor: A new guide shows why helping people grow matters more than technical skills for today's engineering leaders. With nearly 40% of employees wanting to switch jobs — mostly because of bad managers — leaders are trying new approaches like the Solution-Focused Model and CLEAR Method to make their one-on-one meetings actually useful.

Getting More Strategic: Cate Huston, engineering director and author, just published a deep dive into what makes leaders truly strategic — and why good strategy often goes unrecognized. She argues that effective strategy requires balancing four elements: time to think deeply, context to understand the situation, direction to set proximate objectives, and expertise to execute.

An Opinionated Guide to Using AI Right Now: Wharton professor Ethan Mollick just published his latest guide to navigating today's AI landscape, and it's perfectly timed. One standout tip: connect AI to your data sources like Gmail and calendars for dramatically better results.

OpenAI’s Codex Video Tutorials: Over the next two weeks, OpenAI is releasing a series of videos to help devs get better acquainted with Codex. It contains comprehensive hands on tutorials of integrating Codex with your existing IDEs.

The Ultimate n8n Starter Kit: The guide systematically builds your skills across n8n core concepts like trigger nodes, data transformations using expressions, and integrating third-party services through pre-built nodes or HTTP requests.

Building agents (OpenAI’s official guide): The ChatGPT maker launched a comprehensive developer guide for building AI Agents. The guide uses the new Responses API and Agents SDK, alongside built-in tools for web search, file search, code execution, and computer control.

Meta introduces ScaleRL, a recipe for predictable RL training: The paper shows not all RL recipes scale equally. Meta presents a systematic RL scaling study over 400,000 GPU-hours. They fit compute-performance curves, analyze design choices, and propose ScaleRL, a recipe that predicts and improves large-scale RL for LLMs.

Stanford shows simple prompting boosts variety without retraining models: Stanford introduces Verbalized Sampling, a prompting method that bypasses alignment-driven mode collapse. It boosts diversity up to 2.1×, recovers base model variance, and improves quality without retraining or fine-tuning.

LLMs can get “Brain Rot”: Researchers demonstrate that continual pre-training of large language models on low-quality social media content causes persistent cognitive decline across reasoning, long-context understanding, and safety.

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

Until next time — The Code team