I didn't design my remote dev setup. It evolved — one papercut at a time, with plenty of wrong turns and dead ends. VS Code with containers led to a dev server, which led to the terminal, which led to tmux, which led to tools I wrote myself. It's an ongoing reflection, not a finished architecture: dcon and tmosh now partly overlap, my workspaces still aren't fully solved, and I'm right in the middle of testing Zellij. Here's the honest, winding story so far.

Here's a small annoyance that added up over hundreds of logins: every time I SSH or mosh into a machine, I have to remember what tmux sessions are running, type tmux ls, squint at the list, then tmux attach -t whatever. Or forget, start a fresh shell, and lose track of the session I actually wanted. So I built tmosh: a tiny Rust tool that greets me on every login and just asks.

Your AI coding assistant is brilliant, but it doesn't know about Sequelize v7. Or that internal API you built last month. Or the obscure framework your team adopted. Agent Skills change that equation - here's how I built a system to teach Claude what it doesn't know.

In September 2025, I wrote about my early experiments with AI coding tools. Six months later, I realize we're living through another abstraction leap, like the jump from assembly to C, or from C to higher-level languages. And for the first time in years, I'm genuinely excited about what I can build.

Here's a question that's been bothering me: We religiously containerize our applications with Docker. We preach about reproducibility, isolation, and "it works on my machine" being unacceptable for production code. Yet somehow, we're perfectly fine installing Node.js, Python, PostgreSQL clients, cloud CLIs, and dozens of other tools directly on our host machines.

Why do we containerize our code but not our toolbelt?