Welcome back. Nvidia just stole the show at CES 2026 with a double punch: first with its new self-driving car model and second with its next-gen computing platform. Not to be outdone, Amazon launched Alexa+ on the web—entering the chat war against ChatGPT.

Also: A must-do project for every software engineer, Why your best engineers feel stuck, and Anthropic’s workshop on Claude Agent SDK.

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.

Nvidia unveils Rubin chips and 'ChatGPT moment' for autonomous vehicles: Jensen Huang stole the show at CES 2026 with two big announcements. The first is Alpamayo, a 10B-parameter open-source AI model that helps self-driving cars reason through tricky situations — like figuring out what to do when a traffic light goes dark. The second is Vera Rubin, a next-gen computing platform that packs six chips into what Nvidia calls "one AI supercomputer." The company says Rubin delivers 5x more AI training power than its predecessor.

Amazon brings Alexa+ to the web: The e-commerce giant just launched Alexa.com, a browser-based interface that takes its AI-powered Alexa+ assistant beyond Echo devices for the first time. Early Access users can now tap Alexa+ for research, writing, and planning tasks from any browser — directly challenging ChatGPT, Gemini, and Claude. The assistant's agentic capabilities are also expanding, with Expedia, Yelp, Uber, and OpenTable joining the platform for reservations and services.

Liquid AI launches its most capable on-device models yet: The Boston-based AI startup just dropped LFM2.5, an open-weight model family designed to run AI directly on your devices without relying on the cloud. The lineup includes text, vision, audio, and Japanese language variants. The audio model is 8x faster than its predecessor, while the 1.2B-parameter text model outperforms Llama 3.2 and Gemma 3 on instruction following, tool use, and math.

Why your best engineers feel stuck: Career growth isn't as linear as most people think — it varies wildly depending on company size, available resources, and what growth actually means to each individual. This leadership guide breaks down how to support three types of engineers: those who want growth but lack direction, those with crystal-clear goals, and those content where they are. It suggests focusing on energy and patterns, not just titles.

Code Review in the Age of AI: Addy Osmani's new blog tackles one of the trickiest questions facing engineering leaders: how do you maintain code quality when AI writes faster than humans can verify? He offers a clear framework for PR requirements, breaks down where AI reviewers fall short, and explains why teams still need human sign-off for security-critical code. A must-read for anyone managing developers in the age of coding agents.

How to actually get engineers using AI: The biggest AI adoption mistake is jumping straight to complex automation. Engineering leader Elliot Graebert recommends a different path: start by replacing simple Google searches, then layer in pair programming and brainstorming. His secret weapon is treating ChatGPT like a colleague — building iteratively through conversation. The takeaway for leaders: AI augments your team's speed; it doesn't replace their judgment.

Claude Agent SDK [Full Workshop]: In this 2-hour workshop, Thariq Shihipar from Anthropic introduces developers to the Claude Agent SDK. Developers will learn the "Anthropic way" of building AI agents. The tutorial covers designing effective agent loops — gathering context, taking action, and verifying results — along with strategies for prototyping, using sub-agents, and employing code generation for dynamic workflows.

Claude Code is crazy good in 2026: This tutorial teaches developers how to build a complete link-tracking application using AI tools like Cursor and Claude Code. The tutorial demonstrates how to rapidly generate Next.js code, integrate Supabase for data and authentication, and deploy to Vercel. You'll master workflows for using multiple AI agents to accelerate development and ship production-ready software.

The ultimate agent mode tutorial in VS Code: Developers can learn to master VS Code’s Agent Mode by building a complete budgeting app. You’ll learn to generate code from screenshots with Copilot Vision, automate debugging, and streamline project tasks using Custom Agents and the GitHub MCP server.

Recursive language models: Standard AI models lose capabilities and become costly when processing massive amounts of data. This research introduces Recursive Language Models, which solve this by using code to actively manage memory and delegate tasks to sub-agents, keeping the main model focused and efficient.

A plan reuse mechanism for LLM-driven agents: LLM agents are frustratingly slow because they waste time creating fresh plans for similar user requests. This research shows that recycling past plan structures with updated details works perfectly, cutting wait times by over 93%.

Universal reasoning model: Conventional models struggle with complex reasoning because they prioritize size over thinking depth. This research reveals that simply repeating a single, powerful layer allows small models to significantly outperform massive ones on hard logic puzzles.

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

Until next time — The Code team