LODESubscribe →

Ashby Finds 50% AI Code with High Quality

Dev · 2026-06-08

AI-Assisted Development
Ashby Engineers Find Over 50% of New Code AI‑Generated Yet Quality Stays High15 MIN

Since August 2025, Ashby reports that more than half of production code is written by AI, while software quality, stability, and engineer productivity remain strong. The team attributes success to strict accountability, combining AI as a collaborative scaffold with rigorous human review, showing AI can boost efficiency without sacrificing standards.

Run Agentic Coding Locally with LM Studio and Ollama to Cut Cloud Costs26 MIN

The post shows how developers can deploy local LLMs—using tools like LM Studio, Ollama, and Llama.cpp—to power coding assistants, avoiding expensive cloud APIs while preserving privacy. It covers model selection, hardware offloading, and a demo app that illustrates the workflow.

Alibaba Releases Open Code Review: AI-Powered CLI for Precise Git Diff Reviews10 MIN

Alibaba's Open Code Review is an open‑source CLI that leverages LLMs to read Git diffs, apply deterministic engineering logic, and produce line‑level, structured review comments. It combines tool‑use capabilities with rule‑based filtering to ensure full‑coverage, accurate, and stable code reviews at scale.

Anthropic warns AI may soon self‑improve, calls for global development pause25 MIN

Anthropic’s Institute reports that AI systems are approaching recursive self‑improvement—where models can design and train successors without human input. While not yet inevitable, the company says the capability could arrive sooner than regulators expect, and urges a coordinated global pause on frontier AI development to give society time to catch up.

AI agents now outpace human web traffic, Cloudflare reports8 MIN

Cloudflare says AI‑driven bots now generate 57.5% of HTTP requests, surpassing human traffic for the first time—far earlier than the predicted 2027 crossover. The surge reflects AI assistants performing tasks like price checks, content scraping, and personal shopping at massive scale.

Engineering Practice
Use Distinct Input Values for More Reliable Tests, Google Shows1 MIN

Google’s testing team explains that relying on default or common inputs can let bugs slip through. By deliberately picking non‑default, varied values, tests exercise different code paths and prevent false positives, leading to more robust software validation.

Linux-compatible control for Creative Katana V2X soundbar via USB reverse‑engineering8 MIN

A developer reverse‑engineered the Creative Katana V2X soundbar’s proprietary USB protocol, exposing a simple 0x5A‑based command interface on /dev/ttyACM*. This enables Linux users to adjust EQ, lighting and other settings without the Windows‑only Creative app.

Get Dev in your inbox, every issue.
Subscribe free
Privacy · Terms · About · Contact
© 2026 LODE