Encyclopedia · 176 concepts

AI Agents & Applications · advanced · concept 165 of 176

Agent Harnesses & Scaffolding

The engineering around the model that turns it into a dependable agent: the loop that feeds observations back, permissions and sandboxing that bound what it may touch, subagents for parallel work, and skills or hooks that package procedures. Model quality sets the ceiling; the harness decides how close you get, and it is where most real agent engineering happens.

Key terms

Agent loopPermissions & sandboxingSubagentsSkills & hooksLong-running agents

Where you meet it in the real world

Coding agents like Claude Code, computer-use products, production agent reliability work