Text in, text out. Fifty years of computing never found a more durable contract.
The terminal is typed instructions operating files, folders and programs directly — no windows, no pointers, no interface between you and the machine except grammar. It emerged on Unix systems in the early 1970s and never left: every serious operating system still ships one, every programming language assumes one exists, and the whole culture of composable small programs connected by pipes was designed around it. Its philosophy is the opposite of an app’s. An app hides the machine behind buttons; the terminal hands you the machine.
For artificial intelligence this history matters more than any feature list. Half a century of terminal usage — manuals, forum threads, shell scripts — sits inside every major model’s training data, so a model does not need to be taught what grep, git or ls do. It arrives fluent. When agents gained the ability to execute commands rather than just describe them, the terminal became their native workshop: zero schemas to load, endless composability, and error messages a model can actually read and reason over. The newest workers in computing are productive in its oldest room.
Source: Raymond, E.S. (2003), “The Art of UNIX Programming,” Addison-Wesley.