Author here. I wrote message loops for Windows 3.0 at the start of my career, so this chapter is memoir as much as history, GetMessage, TranslateMessage, DispatchMessage, and the day you realize the OS owns your control flow, not you.

The reason I wrote it now: this is part 2 of a series arguing that the AI agent loop is the latest member of a family that includes the message loop, the game loop, epoll, and the Kubernetes control loop. Each stage added one capability the previous one lacked, and the ordering is architectural rather than chronological. The agent loop's contribution isn't the loop, it's that the decider inside it is probabilistic now, which changes what you can guarantee about it.

Happy to answer questions about the Windows era (yes, we really did debug WM_PAINT storms by hand) or to defend the thesis in the later chapters, which get more opinionated about what agents still lack.