A coworker recently asked Roscoe if they should use Cowork or Claude Code for work that has nothing to do with code. His response was a question!
Which one are you more comfortable using?
That isn't a non-answer, though, once you know where Cowork came from. Apparently a lot of the people showing up in the Claude Code terminal don't write software, so Cowork got built to hand them the same capability without a terminal.
The real tradeoff is setup versus spend. Cowork arrives assembled. A sandbox that only touches folders you approve, connectors already wired to Slack and your cloud drives all preloaded. It also reads whole files into context by default, which is why one session can burn a day of usage in a dozen real operations.
Claude Code gives you finer control over what actually gets read. You just assemble the equivalent yourself out of skills, MCP connectors, and a CLAUDE.md. Either one will do the other's job.
P.S. Claude Code (and Cowork) are also great agent harnesses, especially once you connect your phone!
HT: Tim Roscoe Rawson
Drafted in Claude Code, which tells you which door I open.