Do programming languages still matter to your AI coding agent teammate? Evidence at scale from chess engines

This feed collects recent research (15 Jun 2026) on LLM-powered software engineering, security-relevant detection methods, safety for medical-device algorithms, and developer tools. Key contributions: (1) empirical evaluation showing frontier coding agents are genuinely polyglot but language choice still affects performance, cost, and required supervision; (2) a Bayesian-calibrated detector for hallucinated package imports that augments strict-registry checks with PyPI metadata signals to surface registered-but-suspicious packages for safer CI gating; (3) a principled, multi-method framework (