Engineering is asked to ship AI and automation at velocity, then asked to retrofit governance evidence after the fact — in a separate database, with a separate access model, integrated by hand. Each new framework (ISO/IEC 42001, NIST AI RMF, FedRAMP, SOC 2) lands as a sidecar workstream. Audit asks for evidence the platform did not capture. Compliance asks for control attestations the platform cannot generate. The team builds yet another homegrown evidence pipeline that will outlive whoever wrote it.
The vendor governance market makes it worse. Most tools assume their own database, their own auth, their own data residency model, and a multi-tenant SaaS architecture incompatible with a regulated single-tenant deployment. The integration cost is high, the audit burden lands on engineering, and the architecture debt compounds.
Governance that doesn’t fit the architecture isn’t governance — it’s a future migration.