Give Me 20 Minutes. I'll Teach You 80% of Claude Cowork

Claude Co-work turns a folder of files into an AI workspace: it can read what’s already on a computer, propose changes, and then actually reorganize and rewrite content—without requiring uploads or coding. The practical payoff is speed and scale. Instead of feeding an AI one document at a time, users grant access to a local folder, and Co-work can open, move, rename, and create files, even connecting to other apps. That makes it well-suited for messy, high-volume workflows like screenshot cleanup and research synthesis.

The workflow starts with the Claude desktop app. After signing in, users switch among three modes: Chat for general interaction, Code for programming-oriented tasks, and Co-work for “do-anything” automation that doesn’t demand a coding background. Co-work’s core mechanism is permission-based folder access. Once a folder is allowed (including an “always allow” option), Claude can operate on everything inside it—so context comes from the files themselves rather than repeated prompting.

A key demonstration uses a folder of 324 screenshots from a single year. Co-work first proposes a sorting scheme—categories, groupings, and follow-up questions—while showing a progress trail of which files it opens. If the process drifts, the user can stop, adjust instructions, and retry. But the initial categories are only best guesses because Claude doesn’t yet know the user’s personal organization logic.

To fix that, the workflow “teaches” Claude by generating an “intellectual dossier” from an Obsidian vault. The user’s Obsidian system is organized around an ACE framework—Atlas for knowledge and ideas, Calendar for time-based inputs, and Efforts for actions and outputs. Claude is asked to review the Atlas folder and produce a clean, markdown digest of the user’s values, goals, manifestos, and intellectual patterns. That dossier is then pasted into Claude’s personal preferences so future Co-work tasks respond with the user’s worldview baked in.

With that context, Co-work reruns the screenshot organization and produces results aligned to ACE: knowledge assets stay in Atlas, time-based artifacts map to Calendar/workshop sessions, and outputs land in Efforts. The instructions go further—create subfolders, rename each screenshot with a short description, and prefix every filename with a timestamp (year, month, date). An extra step generates a markdown index listing what’s inside each subfolder, turning a pile of images into a searchable system. The long-term value is operationalized through “skills”: Claude can write a reusable skill file (e.g., “screenshot renamer”) that runs the same workflow on demand or on a schedule (like the first of every month at 9:00 a.m.).

The same folder-first approach extends to writing and research. Co-work can limit access to a specific Obsidian folder (such as an “AI research” set of 14 markdown files) and generate a structured synthesis: convergent themes, contradictions or extensions to existing frameworks, and curriculum-building suggestions. The transcript emphasizes that strong context architecture beats prompting alone, and that human judgment remains essential.

Finally, Co-work supports “idea verse” operations inside Obsidian: generating holistic briefings ranked by project status, creating or enriching notes (including pulling images into a specific note), and building “maps of content” that resurface relevant ideas across a linked knowledge base. The central message is that when notes are structured and linked, AI becomes less of a generic assistant and more of a personalized thinking partner that can repeatedly execute the user’s preferred workflow.