AI coding agents from OpenAI, Anthropic, and Google can now work on software projects for hours, writing apps, running tests, and fixing bugs under human supervision, but they are not magic and can complicate projects rather than simplify them. At the core of these agents are large language models (LLMs)