The article argues that the adoption of AI agents in software development will be a costly mistake, as they are unable to truly program and produce high-quality code. The author believes that AI agents are sophisticated statistical models that mimic programming distributions, but their output is often broken and difficult to detect. The author shares their personal experience of using AI agents for programming tasks and concludes that they are not a replacement for human programmers. The article also touches on the potential consequences of relying on AI agents for software development, including the production of low-quality code and the potential harm to large organizations.