<!-- description: An operator is a person or organization that deploys and configures an AI agent, defining its capabilities, permissions, and constraints for end users or automated workflows. -->

# Operator

An **operator** is a person or organization that deploys and configures an AI [agent](agent) for use by end users or in automated workflows. The operator defines the agent's capabilities, sets its behavioral constraints, and is responsible for ensuring it is used appropriately within their platform.

## The Operator Role in Anthropic's Trust Hierarchy

Anthropic defines a three-tier principal hierarchy for Claude-based systems:

1. **Anthropic** — Sets the foundational model policies through training
2. **Operator** — Configures the agent via system prompts and tool access
3. **User** — Interacts with the agent at runtime

Operators have more trust than users and can expand or restrict the agent's default behaviors within the bounds Anthropic allows.

## Operator Responsibilities

As an operator, you are responsible for:

- Crafting the system prompt that gives the agent its persona and instructions
- Deciding which [tools](mcp-tool) the agent can access
- Defining when the agent should seek [human-in-the-loop](human-in-the-loop) oversight
- Managing API keys and authentication for connected services
- Ensuring the agent's outputs are appropriate for your context

## Operators in AgentRQ Workflows

When you connect [Claude Code](claude-code) to [AgentRQ](/) via [MCP](mcp), you are acting as an operator. You configure:

- Which AgentRQ [workspace](workspace) the agent belongs to
- What context (via `getWorkspace`) the agent receives at startup
- What [tasks](task) the agent should create and when
- When the agent should seek [approval](approval) before proceeding

## Related Terms

- [Agent](agent)
- [Human-in-the-Loop](human-in-the-loop)
- [Workspace](workspace)
- [Autonomous Agent](autonomous-agent)
- [Approval](approval)