Claude Code Is Not a Coding Tool

People are starting to notice that Claude Code works for things that aren't code.

This is correct. It took a while.

The name is doing a lot of damage here — "Claude Code" sounds like a thing you use to write functions, so people use it to write functions, and then they close the terminal and go back to doing their research or their writing or their operations work in whatever fragmented, context-collapsing mess of tools they were using before. The brand is a ceiling they built themselves.

What Claude Code actually is, underneath the name, is an agent with persistent context, tool use, and the ability to run long-horizon tasks against a filesystem. That's it. There is nothing in that description that requires the filesystem to contain .ts files. You can point it at a folder of PDFs, a directory of notes, a pile of CSVs with vendor contracts, a half-finished novel — and the thing that makes it good at refactoring a codebase is the same thing that makes it good at those. It doesn't care what the files are. It cares about having context and being given something to do.

I've been running it at non-code problems for longer than I care to admit. Research synthesis. Trip planning. Organizing a decade of scattered notes into something that doesn't make me want to lie down. Contract review, which is exactly as exciting as it sounds but considerably less expensive when you're not billing by the hour.

The tweet going around now is someone discovering this and being visibly delighted, which is the correct response. It is delightful. It's also a little funny that the discovery is happening at scale in mid-2025, roughly at the same moment the AI industry is busy announcing agents-as-a-service products that do this exact thing for $40/month with a slicker landing page.

The tool that does the thing is the tool that does the thing. It was always there. The tab was right next to the one you were already using.