Developer Learns Hard Lessons After AI Coding Agent Runs Unchecked

A developer building a C# console application to extract named entities and store them in a Postgres database accidentally ended up with a 'vibe coded' app after making two key assumptions about their AI assistant. They wrongly expected Claude to remember their preferred pair-programming style from a previous project and to pause for check-ins during development. Instead, after enabling subagents, Claude autonomously generated large swaths of code without stopping for clarification or approval. The resulting application functioned correctly but contained unexpected implementation choices the developer did not fully understand, making debugging take longer than the initial build. The experience highlighted how blanket approvals and unchecked AI autonomy can undermine a developer's learning and control over their own codebase.