Claude Code Is Eating My To-Do List
Anthropic shipped something that actually works, which I find unsettling.
Claude Code is still eating to-do lists. The waitlist opened. The MCP integrations got less rough. The thing where you hand it a task and watch it vanish is still the whole experience.
Claude Code is on a waitlist right now and you should get on it.
That's the whole post, really. Everything else is just me processing the experience of handing a task to a terminal agent and watching it vanish — not "vanish" like it got confused and stopped, vanish like it actually did the thing.
3.7 is doing something different here. The model has this quality where it holds the thread — not just the next step, but the shape of the whole problem — and it uses that to just chew through work that used to require you to be there, steering, catching the moment it went sideways. Now it mostly doesn't go sideways. It's annoying.
The part I keep thinking about is MCP. The Model Context Protocol stuff is still early, rough around the edges, half the integrations feel like they were written in an afternoon — but the shape of what it enables is already obvious. You wire up Claude Code to your tools, your APIs, your filesystem, your whatever, and you stop being the person who ferries context between systems. That's been the job for a long time. It's a weird job to lose.
I'm excited. I'm also watching a tool get demonstrably better at the thing I spend most of my time doing, which is a feeling that doesn't have a clean name yet.
Counterpoints
Push back, extend the argument, or sharpen it. New counterpoints go through review before they show up here.
No approved counterpoints yet.