Zero AI outputs from models like GPT-5.4 and Claude Opus 4.6 passed a quality review by 500 investment bankers. The professionals cited imprecise results and factual errors as primary failures. Despite this, over half will still use these tools for initial drafts. This highlights a persistent gap between LLM capability and professional precision.