Tag
A story from a Windows x86 emulator team about encountering a program with a fully unrolled 64KB initialization loop (65,536 instructions) and adding a special optimization to replace it with a tight loop.
An analysis of how AI coding agents have shifted the bottleneck from writing code to reviewing it, with data showing a 861% increase in code churn and a rise in defect rates, making code review the most leveraged skill in software engineering.
The article argues that increasing the amount of AI-generated code does not necessarily improve team speed and may even reduce efficiency.
Discusses how AI agents enable rapid app creation but produce alien codebases lacking developer understanding, leading to a flood of apps with minimal user traction.
This article discusses how AI coding agent loops can inadvertently learn and propagate deprecated code patterns from existing codebases, leading to technical debt despite appearing successful.
The article explores the phenomenon of 'rockstar developers' who write clever but unmaintainable code, and draws parallels to the challenges introduced by AI-generated code, emphasizing the need for maintainability and team cohesion.
A thread from Charlie Marsh listing a series of pull requests to the Ruff project on GitHub.
FrontierCode is a new benchmark from Cognition AI that measures AI models' ability to write high-quality, maintainable code by evaluating mergeability. Results show even top models like Claude Opus 4.8 score only 13.4% on the hardest subset, highlighting a significant gap in code quality.
FrontierCode is a new coding benchmark from METR and Cognition that evaluates AI models on code maintainability and quality, revealing that many models produce unmergeable code. It includes over 1000 hours of work and shows that even top models struggle, with Opus 4.8 achieving only 13.8% on the hardest tier.
Cognition announces FrontierCode, a new coding evaluation benchmark that goes beyond unit tests to measure code quality, scope, test correctness, and human reviewer approval, addressing the issue of agents writing sloppy code that passes tests but is not maintainable.
compound-engineering-plugin is an AI coding plugin that avoids technical debt by allocating 80% of resources to planning and review, and 20% to execution. It includes 37 skills and 51 agents, supporting three major platforms: Claude Code, Codex, and Cursor.
A blog post arguing that while users don't directly care about code internals, good code quality is essential for performance, bug fixes, and feature delivery, contrary to the common cliché that only user-facing results matter.
A developer blog post argues against the idea of shipping AI-generated code to production without reading it, emphasizing that code review serves critical purposes: diffusing responsibility, reducing bus factor, and maintaining team knowledge of the codebase.
A developer reflects on a messy core data structure in their app, acknowledging technical debt but accepting they will never prioritize fixing it unless it hinders AI code generation.
A repository of Clean Code Skills for AI agents that enforce Robert C. Martin's principles to improve AI-generated code maintainability and reduce technical debt. It provides modular skills for Python and TypeScript to guide agents in writing cleaner, more structured code.
The article discusses concerns that as AI tools generate increasing amounts of code, future models trained on this synthetic code may suffer from reduced quality and originality, and asks how major AI labs like OpenAI, Anthropic, and GitHub plan to address this issue.
The article argues that despite AI advancements, engineers must still understand code and systems, as AI-generated code can become a liability, and emphasizes the importance of CS fundamentals and system design.
A GitHub trending repository (146k+ stars) that is a Claude markdown file incorporating Karpathy's four LLM principles to improve AI code output quality.
A skill is being developed to transform a 'vibecoded slop app' into a production-ready, e2e tested, maintainable, parallelizable agent repository, resulting in a robust codebase after 103 commits over 16 hours.
The author argues that using AI skills to automate code quality checks replicates the same memory and reliability issues that linters were originally designed to solve, questioning the effectiveness of LLM-based skills as replacements.