Welcome back. It’s been a huge week for developers as the big AI labs put out major new tools for builders. Our favorite: Claude Code is now on the web — check out this quick tutorial from prominent AI researcher Simon Willison to learn how you can use it.

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.

Developers can now code in browsers: Anthropic rolled out Claude Code on the web, letting developers delegate coding tasks directly from their browser. Developers can connect GitHub repositories, run multiple tasks simultaneously across different repos, and get automatic PR creation with change summaries. Here’s a step-by-step tutorial to using it.

Google revamps AI Studio with new features for developers: The search giant is streamlining its AI development experience with a major Google AI Studio overhaul. Developers can now save system instructions as reusable templates and manage API keys more efficiently with project grouping. New features include Maps grounding for location-aware applications and a new vibe coding feature for building apps with prompts.

LangChain releases open-source AI agent frameworks for devs: After three years of iteration, LangChain and LangGraph reached v1.0 — marking the first stable release in the durable agent framework space. The frameworks serve different purposes: LangChain for fast standard agents, LangGraph for complex custom workflows requiring fine-grained control. If you’re new to LangChain, you can start with this in-depth course.

McKinsey’s agentic AI playbook for tech leaders: Agentic AI promises massive value but brings unprecedented security challenges, according to McKinsey. These autonomous agents operate like digital insiders within systems, creating novel risks from synthetic identity fraud to data corruption propagation. With only 1% of organizations considering their AI adoption mature, CIOs and CISOs face a critical window to establish guardrails.

How to build a culture where engineers innovate: The best engineering teams don't play the blame game — they build blameless cultures. When mistakes happen, they focus on system failures, not human error. Google's SRE teams run blameless post-mortems after every incident, turning failures into learning opportunities. The result? Higher innovation, stronger collaboration, and better retention.

How to Turn Engineers Into Product Thinkers: David Kavanagh, CTO at Tillo, shares his playbook for transforming engineers from "ticket closers" to product thinkers. His approach: expose teams to customers and business trade-offs in real-time. Engineers who see beyond code commits start chasing impact instead of features.

Context Engineering for AI Agents: This tutorial covers context engineering for AI agents using LangChain and Manus. It addresses how to manage growing context windows in agents through five key strategies: offloading context to file systems, reducing/summarizing information, retrieving context on demand, isolating context across sub-agents, and caching.

Spec-driven development: A GitHub engineer demonstrates how to use Markdown as a programming language when building with AI coding agents. By writing app specifications entirely in Markdown files and using GitHub Copilot to "compile" them into executable code, developers can maintain clearer specs, iterate faster, and avoid the common problem of AI agents forgetting previous decisions and requirements.

Anthropic tutorial on designing sub-agents: This blog frames Skills as a shift in agent design. Instead of relying on constant tool calls, Claude can selectively extend itself with well-scoped, reusable capabilities for complex tasks.

Demystifying RL in Agentic Reasoning: This paper investigates RL's effectiveness in boosting LLM agents' tool-using reasoning. It proves that high-quality real trajectories and exploration methods allow 4B models to surpass 32B ones in benchmarks like LiveCodeBench.

What is AGI: This new paper defines AGI as an AI that can match or exceed the cognitive versatility and proficiency of a well-educated adult.

Can Knowledge-Graph-based RAG Really Retrieve What You Need: This paper discusses GraphFlow, a new method for knowledge graph-based RAG that retrieves accurate, diverse evidence by treating retrieval as step-by-step moves through the graph, using a flow model to distribute rewards and guide policy without per-step supervision.

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

Until next time — The Code team