The terminal writes back

I opened a terminal this morning and described what I wanted. Not the code — the outcome. Then I watched it read my project, reason its way across files I hadn't named, and come back with the change. The terminal wrote back.

Anthropic shipped Claude Code today, alongside a model that pauses before it acts — extended thinking, they call it. You can almost feel the hesitation, the deliberate moment before the keystrokes. After two decades of machines that answered instantly and confidently and often wrongly, a thing that thinks first is its own small revelation.

Here is what's strange. For a year the editor was the frontier — the model living inside my windows, suggesting, completing, sitting where I sat. Now the editor itself is starting to dissolve. The work has left the panel of colored syntax and moved into plain language in a terminal. I tell it the goal. It plans, edits, runs, checks its own work, comes back.

I keep noticing what I'm not doing. I'm not typing functions. I'm not hunting through three files to wire a thing together. I describe intent and review the result, like a studio lead reading a junior's pull request — except the junior never tires and reads the whole repo in a breath.

There's a tax buried in this I haven't named yet. When you stop writing the code, you start trusting code you didn't write. The model is fast and it is fluent, and it will happily build me the same login flow it built yesterday, slightly different, no memory that it has done this before.

I'll worry about that tomorrow. Tonight the terminal answered me, and the answer was good. Twenty-five years in, and the surface I've worked on every day is quietly changing shape under my hands.