A feedback loop uses information about an output or system state to guide later evaluation or action. For AI-assisted development, the important question is what the work is checked against: internal self-review can repeat the generator’s blind spots, while a machine-checkable specification offers an independent check. Feedback quality matters more than the choice of model. section 3.6.3
Build feedback into the engineering environment:
- Use meaningful tests, automated linting and formatting, and CI checks to provide fast, concrete signals when changes break expectations. section 3.6.3
- Use production anomalies to trigger investigation; a more mature loop can test a known fix on a canary, verify metrics, and promote it if they improve. section 10.6.2
- Treat full automation as a maturity goal, not an assumption: the complete detect–investigate–fix–test–deploy–observe cycle is described as not yet fully realized. section 10.6.2
A green test suite does not measure every feedback loop. Whether repository guidance, triage thresholds, or agent workflows are well designed may take weeks or months to assess. Use observability, rollback procedures, and human judgment to manage drift and risk while those slower signals accumulate. section 10.8.2 section 5.7










































