Type: url
Original: https://arxiv.org/pdf/2603.11073
Status: generated
Summary
An EASE 2026 experience report from a small full-stack team on building two production-oriented AI systems (a multi-tenant agent learning platform and an academic RAG system) using contextual vibe coding under explicit architectural constraints. It finds vibe coding reliably accelerated scaffolding but under-specified tenant isolation, access control, and asynchronous processing unless explicitly prompted, proposes the concept of architectural “non-delegation zones,” and reports a shift in engineering effort from boilerplate writing toward constraint specification and validation.