The Knowledge Chipper: Why AI Coding Agents Waste Millions of Tokens

The Knowledge Chipper: An Agentic Coding Story

In this first-person account, a developer argues that the context AI coding agents build up is almost entirely lost after each session, leading to massive token waste and lost institutional knowledge. The author highlights the lack of portability between AI tools like Claude and Codex, making it impossible for different models to pick up where others left off. A recent AWS outage in Bahrain underscores the urgency of portable AI contexts, especially in regions affected by war. The piece also touches on the challenges this poses for code review when AI-generated changes lack the context needed to understand them.