Article.Should-know

How Claude CoWork works in June 2026

Ruben Hassid, How to AI · 2026-07-01.July 1, 2026

Summary

Ruben Hassid has a gift for making difficult things seem easy, so please click the link to read his full article. This article is a practical reset of how to use Claude Cowork: move away from loose files and folders, and organise serious work around Skills, Projects, minimal settings, connectors, and better prompting habits. The useful idea is that Cowork works best when recurring capabilities become Skills, recurring work contexts become Projects, and the human gives richer direction through structured questions, voice input, and review. For knowledge workers, the value is not just learning Claude features, but understanding how to turn AI from a chat window into a repeatable working environment for documents, analysis, onboarding, client work, and team collaboration.

Article source handoff

Ruben Hassid, How to AI · 2026-07-01

This page is a short editorial source record. The summary and notes help you decide whether to continue, while the full article remains with the original author or publisher.

PublicationHow to AI
Author / creatorRuben Hassid
Source dateJuly 1, 2026
Read full article at source

Key Takeaways

  • Skills are reusable capabilities. A Skill is something you can call with a slash command, such as a recurring writing, drafting, analysis, or formatting method; it should be used when the task is something you could “teach” someone to do again.
  • Projects are shared work contexts. A Project is a place where files, instructions, client information, campaign material, or company knowledge stay together, so every chat inside that Project has the relevant context without polluting unrelated work.
  • The practical test is simple: Skill for capability, Project for context. If it is a repeatable method, make it a Skill; if it is a specific client, campaign, company, or workstream, make it a Project. The strongest setup combines both: use a Skill inside the right Project.
  • The article argues against overloading Claude with files, folders, and global instructions. Too much static context can become hard to maintain, leak into the wrong task, or make outputs repetitive; the cleaner approach is to keep global settings light and load context only where it belongs.
  • AskUserQuestion is really a better-context habit. Instead of trying to write the perfect prompt immediately, let Claude interview you, turn uncertainty into structured questions, and collect the missing details before it starts producing the final work.
  • Cowork is best for document-shaped work, not everything. The article’s tool choice is practical: use normal Chat for quick questions or rewrites, Claude Code for software, and Claude Cowork for documents, analysis, decks, client work, onboarding, and other multi-step knowledge work.
  • Voice input matters because the human is often the bottleneck. The article’s point about Wispr Flow is not just speed; speaking gives Cowork richer context, better feedback, and more natural direction than short typed prompts.
  • Cost control is part of the workflow. Long conversations become expensive because the system keeps rereading context, so the article recommends restarting from earlier points, refreshing sessions, batching tasks, and using cheaper models for lighter work.