# Is the terminal going to die? > Source: > Published: 2026-09-01 00:00:00+00:00 # Is the terminal going to die? I don't think the terminal is going to die. I think it's finally going to become usable. [Dan Shipper recently predicted that "CLIs are over."](https://www.lennysnewsletter.com/p/the-ai-paradox-dan-shipper) His argument is that the current terminal renaissance is temporary: Claude Code and other command-line agents broke through there because it was the fastest place to build them, but they will migrate into friendlier graphical interfaces. That sounds plausible. Most people don't want to learn shell syntax, nurse permissions, or remember whether moving a folder is `mv` , `cp` , or something involving `-r` . The terminal is hostile by default. It gives you a blank rectangle and expects you to know what to say. But that doesn't necessarily mean the terminal disappears. It may mean the terminal stops making humans speak terminal. ## The interface is changing, not the substrate I recently asked Claude Cowork to clean up several hundred screenshots on my Desktop. Cowork is exactly the sort of product that appears to support Shipper's argument: a polished interface, a task at the top, tool activity tucked away, and no command prompt in sight. Then it got stuck. My Desktop lives in iCloud, and most of the screenshots had been offloaded. Cowork could see the filenames but not the pixels, so it couldn't classify the images. Its solution was to give me this: ``` find ~/Desktop -maxdepth 1 -name "*.png" -print0 | xargs -0 -n1 -P4 -I{} sh -c 'cat "{}" > /dev/null' ``` Even Cowork was getting me to use the terminal. This isn't just a rough edge that will obviously be designed away. The terminal is where the capabilities of the computer are exposed most directly. Graphical interfaces cover the common paths beautifully, but unusual combinations of actions still collapse back into commands. Download every PNG from iCloud, in parallel, without changing the files, is an odd request for a button and a natural request for a shell. The shell is composable in a way that graphical interfaces rarely are. Small tools can be joined into a one-off machine for exactly the problem in front of you. That power is why agents like it too. ## The terminal was built for agents by accident Command-line tools have several properties that make them unusually good for AI: - They are described in text. - They accept structured arguments. - They compose with other tools. - They return inspectable output. - They can express a long sequence of actions compactly. Those properties used to come with a large human cost. You had to memorize commands, translate intent into exact syntax, interpret cryptic errors, and understand enough of the machine not to destroy anything. An agent can absorb much of that cost. I can say, "Find the large files in this folder, but don't delete anything," and it can produce the right incantation, explain the result, and ask before doing something dangerous. The command line remains underneath, but the human interface becomes conversational. That makes the current renaissance look less like a temporary detour and more like the terminal finding its missing interface. ## A GUI can still win None of this means everyone will spend their day staring at monospaced text. Shipper is probably right about the direction of travel. Agents will gain richer interfaces, previews, persistent tasks, permission controls, and visual ways to review their work. For many people, Codex and Cowork will feel more like operating systems than terminals. But the distinction between a graphical app and a command-line tool is getting slippery. A graphical agent can plan in prose, call shell commands, show the result as a table, and ask for approval with a button. Which interface won? The GUI becomes the cockpit. The terminal remains the engine room. The interesting change isn't that the command line vanishes. It is that fewer people need to operate it directly, while far more people benefit from what it can do. ## Accessibility can look like indirection We often treat accessibility as simplification: fewer options, clearer buttons, safer defaults. AI offers another route. Keep the powerful, difficult thing, then put a patient translator in front of it. That approach won't remove the terminal's sharp edges. Agents can invent flags, misunderstand intent, and run dangerous commands with great confidence. Good products need constrained permissions, previews, reversibility, and a clear account of what happened. The screenshot-cleanup task failed partly because Cowork couldn't cross a boundary on my behalf. That boundary was annoying, but it was also correct. The result may be a terminal that is more visible in the machinery and less visible in the experience. Experts can still reach for it directly. Everyone else can describe an outcome, inspect the proposed action, and approve it. So no, I don't think the terminal is going to die. I think the blank prompt might.