Welcome back. The force is strong with Anthropic, as the Claude maker goes from strength to strength. After its massively popular Claude Code product won over developers in recent months, the company has a new version of the product that’s got everyone excited all over again.

Also: 5 projects to break into AI engineering, what separates successful AI products from expensive failures, and how to write Claude.md file.

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 brings Claude Code to the masses: Devs have been using Claude Code for everything from coding to organizing their digital lives. Now Anthropic is giving everyone else the same power with Cowork — a new tool that lets Claude read, edit, and create files directly on your computer. Give it access to a folder and it can organize downloads, turn screenshots into spreadsheets, or draft documents from your notes. It’s available in research preview for Claude Max subscribers on macOS.

Google goes all-in on AI shopping agents: Back in September, Google launched the Agent Payments Protocol (AP2) to let AI agents securely authorize transactions. Now Google is doubling down with the Universal Commerce Protocol (UCP), built with Shopify, Target, and Walmart to standardize how agents browse, build carts, and checkout. Together, they form Google's blueprint for agents that don't just recommend — they buy. Devs can learn the implementation here.

Developers flock to 'Ralph Wiggum' plugin for hands-off coding: A new Claude Code plugin named after the Simpsons character is gaining viral attention for its ability to run self-looping development cycles. Ralph breaks product requirements into small tasks, then autonomously codes, tests, and commits each one until complete — storing progress in git history along the way. Devs claim they're shipping full features, from authentication to test suites, at a fraction of the usual effort. Here's a tutorial to set it up.

Typing wastes time. Flow makes voice the faster, smarter option everywhere.

Wispr Flow delivers no-edit confidence:

Flow users save an average of 5–7 hours every week — full workdays back every month.

Give your hands a break ➜  start flowing for free today.

What separates successful AI products from expensive failures: Engineers from OpenAI and Google with 50+ AI deployments under their belts break it down in a new Lenny's Podcast episode. Their advice: embrace non-determinism, build incrementally, and resist the urge to chase complexity.

How to choose your engineering leadership path: As careers grow, most engineers feel stuck deciding between management and staying technical. This new guide breaks down the four main leadership roles — Tech Lead, Team Lead, Engineering Manager, and Architect — and what makes each unique. The takeaway: Career paths are rarely linear, so don't be afraid to switch directions as you grow.

The reality check engineering leaders need on context graphs: A hot new concept promises to give AI agents the insider knowledge humans use to navigate gray areas at work. Foundation Capital is calling it a trillion-dollar opportunity. But HubSpot co-founder Dharmesh Shah urges caution: most companies are still figuring out basic AI adoption. He argues that context graphs will matter eventually — but don't build a three-car garage before you own a car.

Claude Code 101: A veteran engineer from Amazon reveals that planning before prompting dramatically improves AI coding results. His top advice includes keeping CLAUDE.md files short and specific, clearing context when it reaches 30% capacity, and using headless mode to automate workflows like PR reviews and documentation updates.

Best practices for coding with agents (by Cursor): The AI code editor just published a comprehensive playbook for devs working with coding agents. The guide recommends writing specific prompts over vague instructions, running multiple models in parallel for tough problems, and using Debug Mode for bugs you can reproduce but can't explain.

The complete guide to CLAUDE.md: Claude Code starts every session with no memory of your code style or project conventions. The fix is a CLAUDE.md file — a markdown doc that loads automatically and tells Claude exactly how you work. Run /init to generate a starter file, keep it under 300 lines, and update it whenever Claude makes an assumption you want to correct.

Recursive Language Models: LLMs perform poorly on massive inputs due to fixed context limits. Recursive Language Models solve this by programmatically breaking down prompts, enabling them to process inputs 100 times larger than their window with superior accuracy.

Reimagining LLM Memory: Standard LLMs struggle to process long contexts efficiently without sacrificing either speed or accuracy. NVIDIA's TTT-E2E method overcomes this by compressing context into model weights, achieving superior performance and constant latency.

Agentic Memory: Current agents struggle with long-horizon tasks due to fragmented, heuristic-based memory systems. AgeMem solves this by unifying long-term and short-term memory into a single policy learned via reinforcement learning, enabling autonomous management that significantly outperforms existing methods.

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

Until next time — The Code team