Glossary
Agent ArchitectureEmerging

Agent Operator

The technical role that configures, launches, monitors, and audits agent runs, intervening only when agents get stuck or produce output requiring human judgment.

Definition

The Agent Operator is the technical role that configures, launches, monitors, and audits agent runs within the Triangular Workflow. In this model, the Agent Operator occupies the "Oversight and Escalation" vertex — supervising agent execution and intervening when agents get stuck or produce output that requires human judgment.

The role maps to the traditional Software Engineer, but the daily work looks fundamentally different. Instead of writing application code line by line, the Agent Operator orchestrates agents that generate code, then audits the output for correctness, security, and architectural compliance. The operator still writes code, but in two specific contexts: custom tooling and Core Nucleus components.

Core duties include:

  1. Unblocking stuck agents — when an agent raises a Blocker Flag, the operator executes a Rescue Mission: diagnosing the failure, injecting corrective context, and resuming execution. This is the highest-priority daily task.
  2. Writing custom tooling — building MCP servers, automation scripts, and integration adapters that extend agent capabilities. This tooling bridges gaps between what agents need and what existing infrastructure provides.
  3. Writing Core Nucleus code — implementing security-critical, architecturally sensitive, or legally regulated components that should not be delegated to agents. These are the parts of the system where human accountability is non-negotiable.
  4. Reviewing agent output — auditing code that passes the Eval Harness to catch issues that automated evaluation cannot detect, such as subtle logic errors or suboptimal design choices.

Key skills for the Agent Operator include code auditing (reading and evaluating code quickly rather than writing it from scratch), system debugging (tracing agent failures through execution logs and context traces), and technical prompt engineering (crafting precise interventions that unblock agents without overriding their approach).

The Operator Leverage Ratio measures how effectively an Agent Operator scales — tracking how many concurrent agent sequences one operator can supervise without quality degradation.

Last updated: 3/11/2026