GLOSSARY
Agent harness
The code that runs an agent's loop, tools and limits.
Updated 2026-10-20By Forward Deployed Course editorial1 min read
An agent harness is the software around a language model that turns it into an agent: it sends the model the task, lets it choose tools, runs those tools, feeds results back and stops the loop when a limit or goal is reached.
On a deployment, the harness is where an FDE adds the safety rules: which tools exist, which actions need human approval, how many steps are allowed and what gets logged.