Harness, Scaffold, and the AI Agent Terms Worth Getting Right
Hugging Face Blog
The article defines key terminology in AI agent development that has become inconsistently used across the field, including the distinction between "harness" (the execution layer that runs the model and handles tool calls) and "scaffold" (the behavior-defining layer like system prompts and tool descriptions). A harness wraps a model and scaffolding to create an agent capable of looping through observation, decision, and action cycles, with the formula Agent = Model + Harness appearing repeatedly in the community. The standardization effort aims to reduce confusion for practitioners building or deploying agents, though the article acknowledges these terms lack universally accepted definitions across different frameworks.
