doorss

Simon Willison defines 'agentic engineering' as the practice of developing software with coding agents, and lays out emerging patterns for working effectively with them. A foundational framing for how AI-assisted development is evolving beyond simple autocomplete.

A critical take arguing that AI-generated code output shouldn't be conflated with actual developer productivity, since code volume doesn't equal value delivered. Challenges the narrative that LLM coding tools are straightforward productivity multipliers.

A practitioner shares their workflow for integrating LLMs into day-to-day software development, with practical tips on when AI assistance helps and when it hinders. Offers a grounded, experience-based perspective on effective human-AI coding collaboration.

An honest account of the cognitive fatigue that comes from constantly reviewing, correcting, and steering LLM outputs during development. Highlights an underappreciated cost of AI-assisted workflows that productivity metrics tend to miss.

An experiment in letting Claude Code autonomously handle system configuration of an Arch Linux install, documenting what went right and wrong. A fun and instructive stress test of agentic coding tools in a real-world sysadmin scenario.

The Rust project summarizes community perspectives on accepting AI-generated contributions, covering quality concerns, attribution, and policy considerations. An important governance discussion as open-source projects grapple with the rise of AI-authored code.

A campaign and resource page dedicated to identifying and combating the flood of low-quality, AI-generated 'slop' content across the web. Catalogs common LLM writing tropes and patterns that mark text as machine-generated.

Quillx proposes a standardized format for transparently disclosing how and where AI tools were used in a software project. Addresses growing demand for AI provenance tracking in codebases and creative works.

A draft of proposed US federal privacy legislation aiming to establish comprehensive data privacy rights for citizens. An important development for anyone tracking the long-awaited push for a national privacy law in the US.

Michael Geist analyzes Canada's revived lawful access bill, which brings back mass metadata surveillance with dangerous backdoor provisions despite some reforms. A concerning legislative move with implications for encryption and civil liberties.

A steganography-adjacent attack using invisible Unicode characters is resurfacing in GitHub repos, npm packages, and VS Code extensions to hide malicious payloads. A clever and alarming supply-chain attack vector that exploits the gap between what humans see and what code executes.

A critical examination of Anthropic's approach to AI safety and alignment, arguing it embodies an authoritarian ethical framework where the company unilaterally decides values for its models. Raises important questions about who gets to define the moral compass of widely-used AI systems.

Sebastian Raschka's visual gallery cataloging the architectural designs of major LLM families, from transformers to mixture-of-experts variants. An excellent reference for understanding how the field's key models differ under the hood.

Graydon Hoare (creator of Rust) shares thoughtful reflections on the current LLM moment, its capabilities, limitations, and what it means for the future of programming. A measured take from a deeply respected language designer.

An audit revealing a news website that loads a staggering 49MB of data, largely driven by advertising scripts, trackers, and bloated JavaScript. A vivid illustration of how adtech has ballooned web pages far beyond what content alone requires.