Coding-Agent Harnesses, Control Planes, and Verification Debt

AWS Kiro bots add persistent Notes, parallel workers, recurring workflows, and dashboards, but production benefits and retry, explainability, and permission controls remain unproven. Harness-S and CoCoder report coordination gains needing replication; EnvFaultBench and ReConMem add promising but bounded evaluation. Recent planning-tool and practitioner accounts, plus Kitta’s self-reported review scale and accuracy, reinforce workflow and review-bottleneck interest but do not establish independent coding or productivity gains.