Skip to content
← Back to Glossary

Agentic Harness

ai-engineering agent-design

The program around the model that reads its responses, executes the actions described in them (reading files, running commands, calling APIs), and feeds the results back.

↑ Contents

Definition

The agentic harness feeds context to the model, interprets its response, executes any tool calls, and feeds results back as new context. Without the harness, the model can only respond with text. With it, the model can act. Claude Code, Codex CLI, and Cursor are all agentic harnesses. Different implementations, same job: provide tools, manage context, and execute the loop.

↑ Contents

Source

↑ Contents