What strikes me most here is the discipline required to pull this off. Usually, when people use AI assistants, they fall into 'scope creep' because generating code becomes too easy. Kudos to you for keeping the guardrails up and launching in 3 months. If you had to estimate, what percentage of your time was spent reviewing/debugging AI code versus actually mapping out system architecture and business logic?