The polarity of hard and easy in computing is inverting. Raw configuration files, exposed APIs, terminal-first applications; these have all become easy. They are the hooks agents want, and we don’t have to think about them. Polished GUIs are now becoming the hard path by comparison, because we have to use them ourselves.
This change has sparked the fourth era in the evolution of computing. By describing it and how we got here, I will overturn most of the things I’ve ever said about user interface design. I’ve always advocated for simple GUIs that normal people can use, but that argument is losing its potency. Ugly exposed wiring is exactly what makes something the right choice now.
Era I: Terminal
In the first era of computing, everything was hard. You memorized a specialized language and you typed it at a prompt. If you hadn’t read the fine manual, the computer was entirely opaque to you.
Era II: Early GUI
In the second era, we defended ourselves from the inhumanity of the CLI with the GUI. Functionality was behind discoverable buttons rather than typed commands. This was progress for humans. It was also a step toward an evolutionary dead end.
Era III: Late GUI
In the third era, designers got involved. Principles of user-centered design entered the public consciousness. Software competed on the experience of using it. Many of the most beloved software companies built their reputations on human-centered GUIs.
This is widely understood as progress, and in a sense it was. Software became more accessible to humans. But the better software got for humans, the more it was adapted only to humans. At the time, that was fine. There was nothing else to operate it.
Era IV: New Terminal
I recently rebuilt my home theater PC. Linux, Jellyfin, Caddy, a few other pieces. Five years ago this would have meant a weekend I wasn’t looking forward to. Partitioning my hard drive by hand. Hand-editing config files. Debugging pipewire error codes. I did none of that. I described the outcome I wanted, and an agent got to work issuing commands and editing config files.
To say that 2026 is the year of the Linux desktop for me isn’t quite right. I’m not exactly using “the Linux desktop.” I’m using an agent. The agent uses the desktop on my behalf. The reason I picked Linux is that an agent can drive it. Every meaningful thing on the system is a file or a daemon or a command.
The historical weaknesses of these tools are now their advantages: config-file-heavy, terminal-first. Anything I can pipe into another command is something an agent can operate on my behalf.
The macOS Settings app, by contrast, is closer to what we’ve historically regarded as good design. And the only way to use it is for me, a human, to open it and click on things. I hate doing this now. Beautiful, human-centered design has become a wall my agent can’t climb, which turns me into my agent’s errand boy.
The choice is no longer “easy for me to use” versus “hard for me to use.” The choice is:
Do I want a computer that’s easy to use, or a computer I don’t have to use because something else uses it for me?
This is what the Fourth Era means. The terminal most people ran from has come back, but not for them. It’s where their agents live.
Users: Pick the stack with hooks. Pick Linux. Pick the open-source thing with the ugly config file over the polished SaaS. Five years ago this was the difficult path. Today it’s the easy one, because you don’t have to use it.
Builders: If you’re shipping something that’s GUI-only, the era your users are from is ending. Pick something. CLI, config file, REST API, MCP, anything. The software that survives will be whatever people don’t have to operate by hand.
The terminal is back on top.