A 700-firm study found AI coding agents inflate code output by 30% but don't increase software shipped. Code review is the bottleneck, and it's getting worse.

A Lead AI Architect at Microsoft open-sourced a harness that wraps coding agents in inspectable stages, human checkpoints, and a separate evaluator. Quality score jumped from 6.64 to 9.02 in controlled testing.

94% of orgs use GenAI coding tools, and the EU Cyber Resilience Act now holds them liable for open source vulnerabilities. Here is what that means for developers.

A breakdown of four AI coding assistants, what each does best, how pricing works, and which type of developer or team each one actually suits.

Pacer puts two status rings in your macOS menu bar to show whether your Claude Code, Codex, Cursor, or Copilot subscription will survive until the next reset.

A 24-build study of Claude Code and Codex found that the model you pick matters more than security prompts. Sonnet 5.5 shipped all 12 protections by default. Haiku 4.5 hardcoded a signing key anyway.

Google's 2025 DORA report links AI adoption to less stable software delivery. Here's the control gap Java teams need to close before it closes them.

Maxi Contieri's framework for writing explicit forbidden-action lists for AI coding agents, with real incidents from Replit, Google, and Cursor as the proof.

Hypothesis found a planted bug in 18 generated cases while three example tests passed. Here is how property-based testing gives AI coding agents a harder check to fail.

AI-generated code skips threat modeling, hallucinates dependencies, and hardcodes secrets. A security breakdown for teams building with vibe coding.

A new research-backed macOS sandbox called SideKernel aims to fix the usability gap that keeps fewer than 40% of AI coding agent users running in a sandboxed environment.

Octopus Deploy's llms.txt file gives AI coding agents a full map of the REST API. Here is how one developer used it with Claude Code to seed test data and build three custom deployment dashboards.