Top Picks

Anthropic research report on how people use Claude Code in practice, based on a large privacy-preserving analysis of session data. It covers task types, division of labor between user and model, and how domain expertise affects outcomes.

A blog post describing an evaluation harness for comparing agentic coding tools and prompts across realistic SWE tasks, with token-cost results and model-specific behavior notes.

An open-source Python project that uses a local 7B diffusion language model to make real-time Gymnasium decisions with zero training. The README shows install steps, a code example, and benchmark-style results across several environments.

A shareable report for AI coding usage that summarizes how you work with tools like Claude Code, Codex, and Cursor, with sign-in via GitHub and an option to get printed copies.

LEVI is a harness-first evolutionary framework for code and prompt optimization. It focuses on reducing LLM cost with diversity-preserving search, role-aware model routing, and a proxy benchmark, and presents comparative results against several existing systems.

A custom Iosevka font build inspired by Anthropic Mono, with prebuilt releases and source build instructions.

arXiv paper on distilling multi-agent debate into a single LLM with a two-stage fine-tuning pipeline. The abstract reports lower token use, comparable or better benchmark performance, and an analysis of agent-specific activation subspaces, with code linked from the page.

A blog post describing a small reinforcement-learning agent trained with PPO to play and beat a Pokelike/Pokerogue-style game, including the input representation, model architecture, and training approach.

A detailed tutorial that explains what agent tools are and walks through a basic set of implementation patterns for an AI agent in Python, including bash, file, search, edit, and web fetch tools.

A Codacy blog post explaining GitHub Copilot’s new AI Credits billing for code review and positioning Codacy AI Reviewer as an alternative with fixed per-seat pricing, static analysis, and pull-request review features.

A tutorial showing how to build a release notes generator with the Cline SDK, including project structure, setup, custom tool design, and code walkthroughs.

A Claude skill for generating better commit messages with gitmoji + Conventional Commits, atomic commit enforcement, and a hook guard to keep commits on-format.

A Superconductor blog post showing how background coding agents were used to reproduce, diagnose, and fix a Rails memory leak using derailed_benchmarks, with a reusable Agent Skill workflow included.

A short policy file for the ripgrep project that explains how contributors may use AI tools while contributing, and sets rules for issues, pull requests, and comments.

A detailed article about the internal harness Nimbalyst built around Claude Code and Codex, covering context, provenance, capability, workflow, restraint, verification, visual interface, and coordination.

A GitHub research project that measures how gpt-4.1 responds when asked to pick a random number between 1 and 100, using 10,000 API calls and comparing the results to a uniform baseline.

A detailed write-up about using AI coding agents to build and tune a Rust multi-Paxos consensus engine, with concrete notes on contracts, spec-driven development, testing, and performance work.

AI-driven relationship quiz that maps users into 16 and 32 love type profiles, with compatibility and self-reflection guidance.

A minimal LLM agent packaged as a single HTML file. The README explains how to run it locally, set the LLM host, API key, and model, and use chat("hi") from the browser console.

A Rust CLI that wraps Claude Code for programmatic use, forwarding flags, injecting one-shot prompts, tailing session transcripts, and exiting automatically after the turn completes.

A Blender AI plugin roundup page that also spotlights 3D-Agent as a native in-Blender AI workflow tool. The page explains what the product does, how it fits into Blender workflows, and compares it with other related tools.

A Python pipeline that ranks 1,000 Show HN posts with an LLM judge and TrueSkill, then compares estimated merit against Hacker News points.

An open-source experiment that adds a small zero-initialized overlay layer to a frozen GPT-2 so its behavior can be adjusted at inference time without retraining the base model.

A blog post about an experimental local LLM system where three qwen3.5:9b agents run autonomously, maintain stress states, self-modify capabilities, and coordinate through a shared OS-like layer.