Chapter 10: Human-in-the-Loop Patterns

Fully autonomous systems aren't always appropriate. These patterns keep humans involved at the right points.

Pattern 11: Approval Gates

The agent works autonomously until it reaches a defined decision point, then pauses for human approval before proceeding.

When to use: When certain actions are high-risk, irreversible, or require human judgment.

Pattern 12: Confidence-Based Escalation

The agent assesses its own confidence. High-confidence responses proceed automatically; low-confidence responses are escalated to humans.

When to use: When you want maximum automation but need a safety net for edge cases.

Pattern 13: Human-on-the-Loop

The agent operates autonomously, but a human monitors activity and can intervene at any time. The human doesn't approve every action, but maintains oversight.

When to use: When you trust the agent for routine work but want visibility and the ability to course-correct.

Pattern 14: Collaborative Drafting

The agent and human work together iteratively. The agent drafts, the human refines, the agent incorporates feedback, and so on until done.

When to use: Creative or nuanced work where human judgment is essential but agent assistance accelerates the process.