Hugging Face publishes a glossary untangling AI agent terms like harness and scaffold

SergioPaniego · x · 2026-08-21

Hugging Face published "Harness, Scaffold, and the AI Agent Terms Worth Getting Right", a systematic glossary for the AI agent field's most confused vocabulary.\n\nThe article originated from a question posted after ICLR 2026: what do "harness" and "scaffold" actually mean in the agent context — many explanations heard, none converging. The authors note that in fast-moving fields, vocabulary evolves faster than shared understanding: terms blur, get reused across contexts, or quietly disappear after a few months.\n\nThe glossary focuses on concepts that are mixed up, reused, or assumed obvious — covering terms encountered when building or deploying agents and using tools like Claude Code, Codex, or Hermes Agent, with a final section on training-side concepts. The goal is not to enforce one correct vocabulary but to offer a practical mental model that makes discussions easier to follow.

Original post →

More from coding & agent

coding & agent channel →