The term "agentic harness" refers to the architectural plumbing that manages autonomous AI agents, distinct from user interfaces or IDEs. In a recent discussion, developer @greggyb clarified that while an LLM functions as the agent's core, the harness serves as the control system running the agent in a loop to reach specific goals.

According to the content, an agent consists of an LLM equipped with external tools that utilize function calling to retrieve information not present in the original training data. The harness acts as the programmatic layer that executes these tool calls and evaluates model outputs to determine whether a task is complete. This modular design allows developers to swap user interfaces without altering the underlying harness logic, highlighting that some agents operate entirely without a front-end display.

The term "agentic harness" refers to the architectural plumbing that manages autonomous AI agents, distinct from user interfaces or IDEs. In a recent discussion, developer @greggyb clarified that while an LLM functions as the agent's core, the harness serves as the control system running the agent in a loop to reach specific goals.

According to the content, an agent consists of an LLM equipped with external tools that utilize function calling to retrieve information not present in the original training data. The harness acts as the programmatic layer that executes these tool calls and evaluates model outputs to determine whether a task is complete. This modular design allows developers to swap user interfaces without altering the underlying harness logic, highlighting that some agents operate entirely without a front-end display.