After three months and three real projects with Claude Code, I’m convinced: it’s not just another coding tool — it might be the most important interface for AI coding we’ve seen yet.

When something new comes along that changes how people work, it’s usually not because of flashy design. It’s because it gets closer to the truth of the work itself.

That’s why I think Claude Code, a command-line AI coding interface, might be one of the most important AI applications yet. Not because it looks modern. In fact, it doesn’t. But because it works the way serious programmers think.

The Command Line Is Closer to the Metal

The first time you use a CLI tool powered by an LLM, you may not notice anything unusual. It’s just a terminal window. But that’s precisely the point. There’s no chrome, no mouse-driven interactions, no visual clutter. Just input and output—like most real problems in computing.

The command line is not just a nostalgic tool from the UNIX era. It’s an interface that maps directly to how computers work. It has native access to the file system, shell, environment variables, processes. It doesn’t need permission to touch your own files. It doesn’t need a web browser. It is the operating system.

In contrast, web apps—even when they wrap around LLMs—feel like tourist interfaces. They are built to attract, not to integrate.

Fast, Lightweight, and Tastier to Hackers

Serious programmers like things that are fast, scriptable, and composable. This isn’t because of coding sage’s stubborn instinct. It’s because speed matters when your brain is full. And control matters when you’re thinking across multiple systems.

Request and response, no distraction except pure conversations. You can even invoke Claude Code in a one-liner print mode, pipe it through jq, send output to a file, or hook it into Git. You don’t have to wait for spinners or page transitions(although reasoning takes time but as we expected). You don’t have to mouse around. That’s not a gimmick. That’s power.

Cursor and Windsurf are pretty shells. Claude Code is a lever.

We Don’t Know What AI UX Should Be

One of the most interesting things about AI applications is that we’re still making up the interface. LLMs are new not just in what they do, but how we interact with them. Most UX metaphors—forms, buttons, windows—don’t fit.

Instead of prematurely deciding on the “correct” UI for AI, CLI apps make a radical choice: no UI at all. The result is that the user can invent the interaction pattern. You decide the flow. You bring the prompt. You direct the result.

That’s not just efficient—it’s honest.

CLI Is a Filter: It Selects for the Right Users

You could argue that CLI apps are less friendly. And that’s true. They are. But friendliness is not always the point.

Good CLI tools self-select for users who know what they’re doing—or want to learn. In other words, they’re not built to be used by everyone. They’re built to be useful to people who care enough to cross the initial threshold. That’s a feature, not a bug.

Claude Code doesn’t try to hide complexity. It puts you right next to it.

AI Coding Begins Here

If AI is going to change how we write software, the first battlefield is our tools. And coding tools are the most direct way to test how useful AI can really be.

We’ve seen flashy IDEs, auto-complete copilots, and “chat in your editor” plugins. But the CLI is where you get the purest interface: your thoughts, their translation into code, and the system responding.

In a sense, Claude Code is what the Unix philosophy always promised, now with the backing of a large language model. Small pieces. Sharp tools. Interoperable. Text in, text out. Now with reasoning. There’s a reason CLI tools like ls, grep, vi, and git have lasted for decades. They are minimal. They are powerful. They survive interface fads because they are not fads.

Claude Code, if anything, follows in this lineage. It doesn’t try to impress. It just tries to work. That might be the most enduring interface we have.

-mgc

  1. Since the first public release of Claude Code, other LLM-powered CLI tools have emerged, such as OpenAI’s Codex and Gemini CLI. (Sep 22, 2025)