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.
More from coding & agent
- Automate GitHub Digest Emails with Flue Agents and Cloudflare — craigsdennis · 2026-08-21
- Treat each Grok bot as an app: thin client, thick server agents — heyneighbor · 2026-08-21
- RulesAsPrograms: Turn Agent Rules into Functions to Check Code Execution — yuntiandeng · 2026-08-21
- Locus Builds an OpenRouter for Everything: One Prepaid Balance Across 286+ APIs — kleffew94 · 2026-08-21
- Stripe opens early access to Provisioning API: agents can sign up for SaaS programmatically — jeff_weinstein · 2026-08-21
- SmashTable: In-memory containers with DBMS-like transactions open-sourced — srchvrs · 2026-08-21