Founder & Product Engineer
Founder, InfoWebPlus
Topics George writes and builds around on InfoWebPlus.
Key finding: Business risk of LLM hallucination is determined by the combination of error detectability, error reversibility, and error exposure scope, not by hallucination rate alone.
Internal tools don't fail randomly. They follow a predictable lifecycle with recognizable failure signals at each phase. A named framework for understanding what happens to internal tools over time, and what interventions are available at each stage.
Occasionally the right answer is to not build at all. This documents the pattern behind those recommendations, what we look for in discovery, how we reason through the total cost comparison, and what we observe in the decisions that follow.
AI employees solve labor bottlenecks, and that is real value. For many operators, the binding constraint is attention, not headcount. Why the emerging AI Operator category addresses a different problem.
The question isn't whether AI is capable enough. It's whether you're prepared to maintain an AI-powered feature at a production standard for 18 months. Most decisions to add AI are made without answering this.