A software developer has called for fewer terminal user interfaces, arguing that AI-assisted coding has reduced the effort needed to create native graphical tools for individual use. The essay is an opinion based on a collection of macOS applications the author generated for personal workflows rather than a measurement of industry-wide development costs.
The examples begin with MDV, a native Markdown viewer, and a SwiftUI front end for SageMath. The mathematics application renders output in LaTeX, exposes operations on objects such as vectors and matrices through controls, and supports shorthand input for common calculations. The author says language models produced most of the interface implementation.
Another application, DJ Roomba, combines an Apple Music interface with an embedded language-model agent that can inspect the user's library, recent listening and queued tracks to assemble playlists from natural-language requests. It stores its own data in SQLite. A separate self-maintaining wiki presents a database through a read-only virtual filesystem so a command-line agent can inspect source material and write entries.
The essay also describes a food-intake estimator, a menu-bar temperature monitor and a native remote for Apple TV, Roku and Denon equipment. Existing Python implementations helped the model reproduce Apple TV communication in Swift, illustrating the author's broader claim that code in one language can now be translated into another without the user personally mastering every protocol.
These programs were not built for distribution. The author describes them as artifacts that make one computer behave according to one person's preferences. That limited audience changes the trade-off: installation polish, broad compatibility and long-term support matter less than obtaining a useful interface quickly. It also means the examples do not prove that generated interfaces are ready for every commercial product.
The argument distinguishes command-line interfaces from terminal user interfaces, though it traces both to constraints inherited from teletypes and early video terminals. Native interface development still demands platform knowledge and careful, repetitive work; the essay's position is that frontier coding models can now supply much of that labor for conventional applications.
The result is less a declaration that terminals have no value than a challenge to their status as the default for developer-built utilities. When a private graphical tool can be produced without a dedicated interface specialist, the author argues, keyboard punctuation and character-cell layouts become a choice rather than a necessity. The examples support that personal conclusion, while questions of maintenance, accessibility and reliability remain outside the essay's demonstration.
The essay also preserves a role for human judgment. The author selected the tasks, described desired behavior and evaluated the generated results, even when no source lines were written personally. Its provocation is partly about whether that activity should be called programming, configuration or simply personal computing.
