This paper presents a first-person case study of a single expert software engineer using frontier coding agents for 12 weeks to build a document accessibility remediation system. The record includes 88 contemporaneous field notes, 420 KLOC of production code, and 1.16 MLOC of tests, lints, documentation, and agent tooling. The authors propose “governance conversion,” a middle-range process model describing how recurring structural failures exposed by high-velocity agentic implementation can be converted into durable governance mechanisms. The paper argues that controls may need to be discovered empirically during AI-mediated development, rather than derived only from obligations known in advance.
No heat snapshots are available in the last 24 hours.