---
title: "Harness, Scaffold, and the AI Agent Terms Worth Getting Right"
description: "AI-agent terminology has outpaced shared definitions, leaving terms such as harness and scaffold blurred across frameworks and contexts. The glossary separates the model, which produces text but has no memory or execution loop, from scaffolding, which defines instructions, tools, output parsing, and context management, and the harness, which runs model calls, routes tool use, and decides when to stop. It presents an agent as a model plus the surrounding system that lets it act repeatedly, while distinguishing tools, skills, sub-agents, and orchestrators by their roles. For training, it defines the environment, trainer, rollout, and reward, including verifiable or learned scores and sparse or dense feedback. Because usage is not universal, the glossary offers a practical mental model rather than enforcing one vocabulary."
---

# Harness, Scaffold, and the AI Agent Terms Worth Getting Right

[Hugging Face](https://yomu.fyi/company/hugging-face) · Sergio Paniego, Aritra Roy Gosthipaty · May 25, 2026

**Type:** Explainer

## Summary

AI-agent terminology has outpaced shared definitions, leaving terms such as harness and scaffold blurred across frameworks and contexts. The glossary separates the model, which produces text but has no memory or execution loop, from scaffolding, which defines instructions, tools, output parsing, and context management, and the harness, which runs model calls, routes tool use, and decides when to stop. It presents an agent as a model plus the surrounding system that lets it act repeatedly, while distinguishing tools, skills, sub-agents, and orchestrators by their roles. For training, it defines the environment, trainer, rollout, and reward, including verifiable or learned scores and sparse or dense feedback. Because usage is not universal, the glossary offers a practical mental model rather than enforcing one vocabulary.

## Context

AI-agent vocabulary is evolving quickly, with terms being mixed together, reused in different ways, renamed, or used without consistent explanations. The lack of universally accepted definitions creates confusion for newcomers and practitioners, particularly around the distinction between harness and scaffold.

## Approach / What changed

The glossary provides a practical mental model for commonly used agent concepts. It distinguishes the model, scaffolding, and harness; explains related concepts such as context engineering, policy, tools, skills, and sub-agents; and separately defines training terms including RL environment, trainer, rollout, and reward.

## Takeaways

- A model produces text but has no memory between calls or execution loop; scaffolding and a harness provide the instructions, tools, context management, and execution needed for agent behavior.
- The harness runs model calls, handles tool calls, and decides when to stop, while an orchestrator coordinates multiple agents that may each run their own harness.
- In training, a rollout records a complete agent run and its rewards, while the trainer scores many episodes and uses those results to update the model's weights.

**Tags:** [Architecture](https://yomu.fyi/topic/architecture), [LLMs](https://yomu.fyi/topic/llm), [Machine Learning](https://yomu.fyi/topic/machine-learning)

- Source: [Hugging Face](https://huggingface.co/blog/agent-glossary)
- Source URL: https://huggingface.co/blog/agent-glossary
- Ingested by Yomu: 2026-08-29T01:49:17.681Z

[Read original post](https://huggingface.co/blog/agent-glossary)