Claude Cowork as the midpoint between chat and code is the framing that finally makes the whole product make sense

Claude Chat for conversational AI assistance. Claude Cowork for autonomous task execution on your computer for general users. Claude Code for developer-facing codebase work. Three distinct products for three user types at three levels of autonomy. Once that structure is clear the individual feature list makes much more sense.

The Cowork features that define what the midpoint enables in practice:

Folder Instructions setting specific guidance for what Claude should expect and how it should work within a particular folder context is the scoping that makes autonomous work reliable. Open-ended autonomy is less useful than scoped autonomy with clear expectations.

Computer Control being available is the biggest capability in the list and also the one that requires the most deliberate thought about what to actually let it do. Not everything should be automated just because it technically can be.

Projects in Cowork keeping tasks, files and context contained is the organisation layer that makes Cowork practical for ongoing work rather than just one-off tasks. A project that accumulates context over weeks is a different tool from a session that resets each time.

Scheduled Tasks running automatically without initiation are the ambient automation that changes recurring work. The morning email summary that runs before you open your inbox. The weekly review that completes while you are asleep.

Which Cowork features are you actually using versus still evaluating?