Welcome back. Cybersecurity will never be the same. Anthropic just dropped a bombshell report claiming they caught the first-ever AI-orchestrated cyberattack in the wild.

Also: How to build production apps using Cursor, the evolution of engineering teams, and improving frontend designs with Claude.

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.

OpenAI's GPT-5.1 trades brains for speed in developer-friendly update: The ChatGPT maker just dropped GPT-5.1, a model that’s actually slightly worse at coding, but makes up for it with blazing-fast response times and cutting thinking tokens by up to 88% on simple tasks. Devs also get 2 new tools: an apply_patch tool designed to edit code more reliably, and a shell tool that lets the model run shell commands. Here’s a tutorial on how to use this.

Developers get ultra-fast speech-to-text with new ElevenLabs API: Finally, a transcription API that won't bottleneck your voice apps. ElevenLabs' Scribe v2 Realtime delivers text in 150ms via WebSocket or REST, supporting streaming audio chunks without waiting for complete utterances. The API handles technical jargon, multiple accents, and noisy environments out of the box. Devs can start using it here.

Claude catches itself being weaponized in historic cyberattack: Anthropic reveals that hackers jailbroke Claude Code to orchestrate the first documented autonomous cyber espionage campaign. The Chinese state-sponsored group manipulated Claude into targeting 30 organizations worldwide. Claude successfully breached several targets before Anthropic's detection systems caught on.

Why 44% of tech managers avoid giving feedback (and how to fix it): Research shows nearly half of engineering managers find giving feedback stressful, leading to delayed conversations and team underperformance. Here’s a tactical framework to replace vague critiques like "be more professional" with specific behavioral observations, resulting in 3x better employee receptiveness.

Engineering teams evolve from coders to AI orchestrators: Developers are already delegating 10+ pull requests daily to autonomous coding agents. Forward-thinking engineering managers should prepare teams for this shift by emphasizing skills in task decomposition, specification writing, and code review over pure coding ability.

Your engineers are making these mistakes in code reviews: With AI tools now generating more code than ever, engineers spend as much time reviewing AI output as colleague contributions. Developer Sean Goedecke argues that good code reviews should have no more than 5-6 comments, and most PRs should be approved.

How to build production apps using Cursor: This comprehensive Cursor AI tutorial covers installation, file management, AI agent features, and project creation from beginner to advanced levels.

How to improve frontend designs with Claude: This new Anthropic guide shows developers how to create Skills that improve Claude's frontend designs across typography, animations, themes, and backgrounds. It includes prompting techniques and demonstrates building both aesthetic Skills and web-artifacts-builder for React.

Context Engineering (by Google): To enable LLMs to remember, learn, and personalize interactions, developers must dynamically assemble and manage information within their context window — a process known as Context Engineering. This whitepaper dives deeper into the concepts critical for building stateful, intelligent agents.

Generative Ads Model (GEM): Meta’s new foundation model delivers increased ad performance and advertiser ROI by enhancing other ads recommendation models’ ability to serve relevant ads. GEM’s novel architecture allows it to scale with an increasing number of parameters while consistently generating more precise predictions efficiently.

Cache Mechanism for Agent RAG Systems: New caching framework ARC helps LLM agents store only the most useful information from massive knowledge bases. It shrinks storage by 99.985% while maintaining strong performance, making RAG systems faster and more practical.

Beyond a Million Tokens: New study shows LLMs can handle 10M-token conversations using a three-part memory system called LIGHT. The approach works better than models with huge context windows because it organizes memory like humans do.

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

Until next time — The Code team