Harness
A harness is the runtime environment that wraps a model with tools, policies, memory, and orchestration so agents can act inside your product or repo.
Users rarely see the word harness, but they feel it as what the agent can access, how long it runs, and what happens when it errors.
Share
What it means
The harness connects the model to file systems, APIs, browsers, terminals, and review gates, enforcing limits on steps, permissions, and side effects.
Why designers should care
Harness boundaries define UX: which actions need confirmation, what appears in activity logs, and how users recover when the agent hits a wall.
Example
Cursor’s harness runs parallel file edits, shows tool calls in a timeline, and blocks destructive git commands until the user approves.
Common mistakes
- • UI that promises capabilities the harness cannot perform (live deploy, unchecked writes).
- • No visibility into harness steps, so failures feel like “the AI broke.”
- • Inconsistent permissions between chat UI and background agent runs.