"Cowork" naming sets expectations the feature can't yet deliver

The "cowork" label in Claude Desktop implies partnership — two agents contributing, creating, deciding together with shared ownership of output.

The actual experience is passive observation with reactive suggestions: screenshot → analyze → suggest → user applies manually. This creates a gap between expectation and delivery that leaves users underwhelmed.

The naming problem

• "Cowork" implies mutual agency and contribution
• The feature delivers observation + commentary
• More accurate names: "Watch mode", "Screen advisor", "Observe & suggest"

What real cowork feels like (e.g. Claude Code CLI)

• User says "refactor auth" → agent reads files, plans, edits, tests, commits
• User builds frontend → agent builds backend in parallel
• Agent notices pattern drift → flags it proactively, before being asked
• Persistent memory, tools, execution — both sides produce output

Suggestion

Either evolve the feature toward actual co-creation (tool access, file editing, proactive contributions), or rename it to set accurate expectations. The feature itself isn't bad — the promise just overshoots the architecture.

---
Feedback from a daily Claude Code CLI user who tried Desktop cowork and felt the gap immediately.