Harness, Scaffold, and the AI Agent Terms Worth Getting Right
- Source
- huggingface.co
- Date
Context
An AI agent needs software that executes its proposed actions and brings the results back for another decision. This agent glossary calls that execution layer the : it invokes the model, runs tool calls, and determines when to stop. It uses scaffolding for the instructions, tool descriptions, response handling, and management that shape the model's behavior.
The distinction helps explain why choosing a model is only part of building an agent. Instructions determine what information and options it sees; execution software determines how its actions actually happen. The glossary also acknowledges that some products use harness for both layers, so these are useful distinctions rather than universally settled definitions.
- agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
- context engineering — Deciding what an agent sees at each step — the discipline that replaced prompt engineering once context windows got large enough to fill badly.
- tool use — A model's ability to call external functions — run code, search the web, edit files — instead of only generating text.
Checking sign-in…
Loading comments…