The AI Coding Setup That Actually Stuck: Claude Code, tmux, and Neovim

For the past year I tried to keep up with the IDE wars. Cursor. Copilot. Windsurf. Every month a new VS Code fork showed up promising to change how I write code.

I gave most of them a real try. Some of them are genuinely good. None of them stuck.

What stuck, weirdly, is the opposite direction: Claude Code in a tmux pane next to Neovim, on a terminal that hasn't fundamentally changed how I use it in years.

Here's why this combination won, what it does well, and where it falls short.

What my setup actually looks like

When I sit down to work, I open a single tmux session per project. Usually two main panes, sometimes three.

Left pane: Neovim with the file I'm editing. Right pane: Claude Code, running as a conversational agent against the whole project directory. When the project is running locally, a third pane underneath tails server logs and test output.

That's it. No IDE. No sidebar. No autocomplete popup blocking my view. The AI lives in the next window over, not on top of my code.

When I want to ask something (refactor this function, find every place that calls X, write tests for this module), I tab over, ask, watch Claude work, and come back to Neovim when I want to drive again.

A typical session: NvimTree and the file I'm editing on the left, Claude Code on the upper right, dev server logs below.

Why I gave the IDE tools a fair shot

I'm not allergic to GUIs. I started in Eclipse, moved to NetBeans, spent years in the JetBrains family, eventually settled in VS Code, and then finally met Neovim with tmux. The reasons the IDE-embedded AI tools didn't stick weren't religious. They were practical.

A few honest observations:

• Inline suggestions are noisy. Ghost text appearing every few keystrokes interrupts my thinking. I write better when I can think in silence and only ask for help when I actually need it.
• The "agentic" mode in most IDE tools is still glued on. It feels like a chatbot that occasionally edits files, not a coworker who understands the project. Claude Code feels closer to the latter.
• Project-wide context is fiddly. I want the AI to see the whole repo by default and not need me to manually pin files. Claude Code does that out of the box.

Mostly, though, I just kept finding myself in a terminal anyway. Reading logs. Running tests. Grepping. Editing config. At some point I stopped pretending I lived in the IDE.

Why Neovim, in 2026

Because muscle memory compounds.

I've been driving Neovim long enough that my hands move faster than my conscious thought. The cost of switching to anything else, no matter how shiny, is paying that compound interest back to zero.

That's not an argument for Neovim being objectively better. It's an argument for sticking with what your hands already know. The editor is where you spend most of your day, and the cost of switching is paid in every keystroke.

If you're early in your career and the choice is open, modern IDEs with AI built in are probably the right starting point. If you've already spent years building keyboard-driven habits, throwing them away for a marginally nicer chat window is a bad trade.

Why tmux is the unsung hero

The thing that makes this setup work isn't the editor or the AI. It's tmux.

With a few keystrokes I can:

• Split panes vertically and horizontally
• Move between Claude, Neovim, a shell, logs, and a server process without ever touching the mouse
• Detach and reattach sessions if I want to step away or move to a different machine
• Keep a named tmux session for every project I work on, with all its panes and processes preserved
• Switch between project sessions in a single keystroke, with zero context loss and no window manager chaos

The mental model is simple. tmux is my window manager. Neovim is my editor. Claude Code is my pair. Each tool does one thing well, and none of them try to be the others.

It's the inverse of the IDE philosophy, which bundles everything into one app and forces you to use its version of every component.

Where this setup falls short

I'd be lying if I said it was perfect.

• The first day is rough. If you've never used tmux or Neovim, the learning curve is steep enough that the productivity story breaks down for weeks. This setup is only worth it if you already have the muscle memory or are willing to invest.
• Visual diffs are weaker. A good GUI diff viewer beats the terminal for big code reviews. I still open VS Code occasionally for that.
• Onboarding others is harder. When I'm pairing with someone who doesn't live in the terminal, I switch tools to meet them where they are. The setup is for me, not for everyone.

The bigger point

The honest answer to "which AI coding tool is best" is: the one that disappears into your existing workflow.

For some people that's Cursor, because they already live in an IDE and want the AI to feel like part of it. For me it's Claude Code, because I already live in the terminal and want the AI to feel like a tab away.

The tooling industry would like to convince you that the choice of AI tool is the most important decision you're making. It isn't. The most important decision is whether your tools fit how your brain already works. The AI part is the last 10%.

If you've been wondering whether you have to migrate to a new IDE to keep up, you don't. The agent can come to you.