Identity and permissions
Authentication, tenancy, resource ownership and least-privilege access need to be explicit.
Data lifecycle
Retention, deletion, provenance, backups and migration paths matter once user data becomes durable.
Quality and observability
Evaluation sets, structured logs, request IDs and failure dashboards are necessary to operate AI features over time.
Operational controls
Rate limits, quotas, background jobs, retries, billing, support workflows and rollback plans turn a demo into a service.