Martin Fowler: Why He Does Not Like LLMs

Martin Fowler, one of the most well-known authors on software development, published an article titled "I Don't Like LLMs." His main complaint is that models present fabricated facts with the same confidence as correct ones. When they are caught in an error, the response is only "fake remorse."

The complaint is valid, but it applies to a chatbot, not to agents. In a pipeline where the task is specific and the agent has tools, a hallucination is a code error that shows up immediately. I would not measure these two modes by the same standard.