I gave an agent my posting history. It found a promise I never made.

A developer recently experimented with building an AI agent designed to generate social media content by analyzing their past posting history. The project highlighted significant challenges in using LLMs for content creation, specifically regarding data interpretation. The agent frequently misinterpreted internal testing data and QA artifacts as public promises, leading to inaccurate content suggestions. Furthermore, the author discovered that AI-generated drafts often suffer from 'slop'—repetitive or derivative phrasing—even when passing automated quality checks. The experience underscored the necessity of human oversight, as the agent lacked context regarding external publications and audience history. The author concludes that while AI can assist in brainstorming and drafting, it should never be fully automated. Instead, it requires strict filtering rules and a human-in-the-loop review process to ensure accuracy and authenticity. The resulting tool, designed to suggest content angles rather than full drafts, is now available as an open-source project.