AI Dev Stack

Sign in Subscribe

feature

Cline-Bench: Real-World Benchmark for Agentic Coding

Cline-Bench: Real-World Benchmark for Agentic Coding

Cline launches cline-bench, an open source benchmark built from real engineering failures where frontier models break. Backed by OpenAI, Mistral, and Nous Research, with $1M for open source contributors.

How GitHub Trained Copilot to Predict Your Next Edit

How GitHub Trained Copilot to Predict Your Next Edit

GitHub trained a custom model to predict your next code edit in real time. Here's why pull request data failed, how reinforcement learning fixed it, and what three major releases have changed since February.

GitHub Copilot's Tool Selection Fix: Faster Responses, Smarter Routing

GitHub Copilot's Tool Selection Fix: Faster Responses, Smarter Routing

GitHub cut Copilot's default toolset from 40 to 13 tools and built embedding-guided routing. The result: 400ms faster responses and 94.5% tool coverage. Here's how they did it—and why more tools often means worse performance.

How to Write agents.md Files That Actually Work: 2,500 Repos Analyzed

How to Write agents.md Files That Actually Work: 2,500 Repos Analyzed

Analysis of 2,500+ repos reveals what makes agents.md files actually work: executable commands, code examples over prose, clear boundaries, and exact tech stack details. Most fail because they're too vague.

GitHub's Annotation Toolkit: Open Source Design-to-Code Handoff

GitHub's Annotation Toolkit: Open Source Design-to-Code Handoff

GitHub open sourced a Figma library that embeds accessibility and design intent directly into design files. Their data showed 48% of accessibility issues could have been prevented with better upfront documentation. Here's how it works.

Git 2.52: Tree-Level Blame, Geometric Repacking, and Rust

Git 2.52: Tree-Level Blame, Geometric Repacking, and Rust

Git 2.52 ships with tree-level blame 5.5x faster than scripts, GitHub's geometric repacking for monorepos, and the first Rust code in Git's codebase. Here's what actually matters for developers.

Master Copilot Code Review: How to Write Instructions That Work

Master Copilot Code Review: How to Write Instructions That Work

GitHub's guide to writing Copilot code review instructions that actually work. Learn what to include, what to avoid, and how to structure files for consistent automated reviews.