There are three modes of AI supervision in engineering: in-the-loop (human reviews each output), on-the-loop (human monitors, intervenes when needed) and out-of-the-loop (human only sees final result). Half-loop sits between on-the-loop and in-the-loop: the senior approves critical decisions (architecture, security, trade-offs) and designs the guardrails (tests, hooks, CI) that keep the agent in safe behaviors — but doesn't review each commit. It works because the harness produces consistent output, and the senior decides where human review adds most value.
GLOSSARY
Half-loop
What is Half-loop?
AI supervision model where the senior engineer stays on-the-loop, not in-the-loop. Designs the harness, sets guardrails and reviews architectural decisions — without line-by-line diff review.