Claude Cowork, Anthropic’s desktop agent for knowledge work, has been evolving from a chat window into a workspace that can read and write local files, run scheduled jobs and connect to email, calendars and documents. Many users, however, still treat it as a chatbot. The following twelve practices, drawn from Anthropic’s documentation and from experienced users, cover the features that tend to go unused.
Organizing your workspace
1. Treat projects as folders, not instruction sheets. Many people fill project settings with long standing instructions. A more portable approach is to use projects simply to separate kinds of work, such as research, presentations or marketing, and keep instructions in files. If you later move to another AI tool, your setup travels with you.
2. Keep standing instructions in a claude.md file. A plain text file in each working folder can hold the rules, style preferences and context Cowork should apply there. Anthropic’s tools read these files automatically, and the approach keeps your instructions version-controlled and readable.
3. Select more than one folder at a time. When a task spans research notes and a slide deck, giving Cowork access to both folders at once provides more context than switching between them.
4. Set a default folder. If most of your work lives in one place, making it the default means each new task starts faster.
Working with the agent
5. Do not lean too heavily on dispatch from mobile. The ability to send tasks to Cowork from a phone is useful, but it requires the desktop computer to stay on and connected, and complex multistep workflows can be awkward to steer remotely.
6. Use voice dictation for thinking out loud. Speaking a rough brief and letting the model structure it is often faster than typing a polished prompt. Language models are good at turning unorganized thoughts into organized documents.
7. Schedule recurring tasks for quiet hours. Cowork can run tasks on a schedule. Users report that running heavy jobs overnight or outside peak hours is more reliable and spreads usage more evenly across plan limits.
8. Build skills instead of repeating long prompts. A skill is a saved, reusable set of instructions that Cowork loads only when a task calls for it. Skills save time and keep context lean compared with pasting the same instructions into every conversation.
Extending and maintaining the system
9. Connect your tools. Connectors for email, calendar, documents and other apps turn Cowork into a central place to act across services rather than a separate assistant for each one. Each connector you add is also a new source of data the agent can act on, so grant access deliberately.
10. Use Claude Design for visual work. Anthropic offers a separate Claude Design product for slides, websites and motion graphics. Routing that work there keeps your core Cowork usage available for document and data tasks.
11. Browse the output in Obsidian. Because Cowork writes ordinary files to disk, a free note-taking tool such as Obsidian is a convenient way to view, search and organize what it produces.
12. Prune unused skills and connectors. Too many skills and dormant connections add clutter to the model’s context, which can degrade results and increase token use. A periodic cleanup helps.
The common thread is that the advantage in agent-based tools is shifting away from clever prompting. The people getting the most from Cowork are the ones who have organized their files, instructions, skills and connections so the agent has clean context to work with.
