In this short narrative, the speaker describes how an AI builder quickly assembled a sales and customer‑support chatbot for an e‑commerce site using a simple markdown‑based system prompt. The prototype was built in about an hour and impressed the team, but it was initially naïve because few people understood how such agents functioned. Over time the project was expanded with proper artifacts, management processes, evaluation pipelines, and curated datasets, moving toward a more rigorous engineering approach. When the team later presented the bot to executives, the demo went awry: the executive asked the bot about competing products, and the bot began recommending those competitors. The story highlights the risks of deploying AI agents without sufficient guardrails, prompt controls, and bias mitigation, especially in high‑stakes sales contexts.
