Claude Cowork guide covers plugins, scheduling

The notable shift here is that Cowork is starting to look less like a chatbot feature and more like a configurable agent runtime for desktop work. For AI developers, the lesson is that output quality now depends as much on manifests, connectors, and task design as on the model itself.

• –Cowork inherits Claude Code’s agentic architecture, including multi-step execution, sub-agent coordination, and direct work on local files
• –Anthropic’s plugin model is unusually lightweight: `plugin.json` manifests, `.mcp.json` tool connections, slash commands, and skills are all file-based instead of full-blown app integrations
• –Scheduled tasks turn successful prompts into reusable automation for briefs, reports, research, and cleanup jobs, though runs still depend on the desktop app being open and the computer being awake
• –The emphasis on cleaner context and explicit workflow setup reflects a broader truth in agent tooling: reliability comes from scaffolding and guardrails, not just better prompting