Hugging Face AI Updates: May 27, 2026

1. Hugging Face publishes a standardized AI agent glossary defining Agent = Model + Harness + Scaffolding

Hugging Face. Hugging Face published a standardized glossary for AI agent terminology that pins down vocabulary the field has been using loosely. The framing equation is Agent = Model + Harness + Scaffolding, and the post separates inference-time terms (Tool Use, Skills, Sub-agents, Context Engineering) from training-specific terms (RL Environment, Trainer, Rollout, Reward), citing TRL’s GRPOTrainer for an implementation reference. It anchors the abstractions in real-world harnesses like Claude Code and Codex, and points to OpenEnv as a framework for evaluating tool-using agents in standardized environments. Source