I thought a command-line coding assistant would be limiting. I was completely wrong.

Despite being comfortable with CLI - I know vi, use vi bindings on the command line, and see the benefits of working in terminal - I still felt that having a coding assistant constrained to command line would be… well, constraining.

Right from day 1 with Claude Code, I immediately could see why command line is the best option for a coding assistant.

Everything Comes to You

The command line is the most efficient way to get anything done, if you know the commands. Everything comes to you. You write commands and control what’s produced depending on your appetite to see information.

This is fundamentally different from GUI-based coding assistants where you’re constantly context-switching between different windows, panels, and interfaces.

With CLI, there’s no jumping around. The assistant has access to everything without the overhead of managing visual interfaces.

The Universal Interface

There’s been 30+ years of command-line evolution and programmatic control possibilities, honed by countless great individuals. All of them have the same basic interface.

Think about it: millions of programmers have essentially created what’s basically a command-line program that can do anything.

There isn’t an important program that isn’t available as a command. Even Git is a command-line program.

Now the assistant has access to all this. It knows all the commands, so you don’t have to remember any - you just need to have a vague idea of what the commands are.

The Token Economy

Here’s something I discovered in my workflow: coding assistants can do a lot of work themselves, but all work requires consuming tokens and processing them.

For many tasks, it makes sense to use command-line tools that work with text (grep, find, etc.) to do the work precisely, quickly, and without using tokens.

A lot of the time, instead of asking the agent to do the work directly, I ask it to create a batch script first, then run it. This approach works better in command line because all these little batch tools integrate well with existing tools through the simple pipe interface.

Even with Copilot agents, I found myself asking for batch scripts rather than direct execution. But with Claude Code, this workflow feels natural and integrated.

Control What You See

The killer feature is that the assistant controls what it shows you. You’re not looking at too much stuff at a time when you actually want to see what it’s doing.

Command-line interfaces are good to read these days - they have color, and Claude’s output is formatted nicely. You get exactly the information you need, when you need it.

This is completely different from GUI assistants that either show you everything (overwhelming) or hide too much (frustrating).

Real-World Validation

I’ve used Claude Code on complicated SpringBoot projects with complex PostGIS databases and custom routing queries. I’m not having to use the editor although I run Claude Code in the VSCode terminal.

I can tell it what I want, and it does it quickly. It’s remarkably quick compared to Copilot - I don’t mind it doing things that I could do myself quickly, because it’s not painful to wait.

That happens a lot with Copilot. With Claude, it rarely happens. Even when it does, I can escape and tell Claude exactly what to do, and it does it quickly.

Easy to Improve

I can quickly find ways to improve Claude Code because it’s like any other command-line program. This is something I find difficult with other agents that aren’t command-line based.

When something doesn’t work the way I want, I can understand why and how to work around it. The interface is transparent and composable.

The Anthropic Advantage

There are inherent advantages to the command-line approach:

  • Less code for Anthropic to write and test
  • Better dogfooding opportunities (they use their own tools)
  • Cleaner documentation (possibly Claude-assisted)
  • Minimal setup requirements - start small and discover yourself

It’s only been a week since I started using Claude Code, but I can already see it will be my main coding assistant for some time.

Any other coding agent will have to be a command-line one to beat it.

The command line isn’t a limitation for AI coding assistants - it’s their natural habitat.