The happy path is only the beginning
A prototype demonstrates an interaction under conditions its creator understands. Production introduces users who misunderstand the interface, requests that arrive twice, external services that time out and data that must remain private.
AI-assisted development can compress the time to a working screen. It does not remove those conditions. Before launch, review the system’s behaviour outside the demonstration path.
Five questions that expose the gap
Can one user access another tenant’s records by changing an identifier? Can a retry repeat a charge or a message? Can the team trace personal data into model-provider calls? Can a failed release be reversed? Can someone other than the creator operate the service?
Each question needs evidence. A login screen does not demonstrate authorisation. A passing end-to-end test does not establish tenancy isolation. A deployment button does not establish recovery.
Separate release blockers from improvements
A readiness review should create a decision, not an unbounded rewrite. Exposed secrets, broken access controls and irreversible data corruption are different from a confusing internal function name. Prioritise by consequence, exposure and the ability to detect failure.
Record what must be fixed before launch, what can be monitored under a controlled rollout and what belongs in the continuing backlog. Assign an owner and an acceptance test to each material action.
Design for the second operator
Write the instructions that another engineer would need during an incident. Record environments, dependencies, releases and restore procedures. Prove that they work before customers depend on them.
A production product is software with a reliable operating arrangement around it. That arrangement deserves the same attention as the original demo.