Start with boundaries
Define users, system responsibilities, data ownership and permission boundaries before choosing implementation details.
Make change reviewable
Use documented interfaces, versioned changes and appropriate tests. Match architecture complexity to the actual requirements.
Design the failure path
Consider errors, retries, denied actions, timeouts and manual review. A happy-path demonstration is not sufficient evidence of operation.
Observe the system
Agree what signals operators need, how failures are recognized and who responds. Operational coverage depends on the engagement.
Document the trade-offs
Record assumptions, alternatives, unresolved risks and the reasons behind consequential decisions.