CoolDeep AI Digest

The article introduces Claude Cowork, a desktop app that connects Claude directly to a user's local files, eliminating the repetitive copy‑paste workflow of Claude Chat. Unlike Claude Chat, which requires re‑explaining context each session, and Claude Code, which is a developer‑focused terminal tool, Cowork gives the AI a persistent workspace: it reads, writes, and remembers the actual documents in a designated folder, so users can start working immediately each time they open the app. The author argues that most people waste 1.5 hours per week (about six hours a month) just re‑providing context, and that adopting Cowork recovers that time. Five practical tips are shared to maximize Cowork’s efficiency. First, use Obsidian as a readable lens for the Markdown‑based instruction and memory files; enabling "Show all file types" lets users view PDFs, images, and spreadsheets alongside .md files. Second, keep the central CLAUDE.md file under 300 lines by organizing it into six sections—memory system, preferences, rules, routing map, references, and workstation creation—and loading detailed rules only when needed via the routing map. Third, put the MEMORY.md file on a diet: limit entries to one‑to‑two sentences, cap it at 150 lines, and let older entries be compressed into an ARCHIVE.md that loads only on demand, preserving history without token cost. Fourth, migrate existing Claude Projects into Cowork by converting project instructions and memory into workstation‑specific .md files and moving knowledge files into a Resources folder, gaining full editability and bidirectional file access. Fifth, distinguish between workstations (for judgment‑heavy, adaptive tasks like writing or planning) and skills (for repeatable, deterministic tasks like formatting or report generation); overloading workstations with skill‑type work slows the system. The core takeaway is that Cowork transforms Claude from a per‑session query tool into a persistent AI partner that knows your files, preferences, and history, letting you focus on creation rather than setup. Implementing the tips—especially trimming CLAUDE.md and structuring memory—yields immediate token savings and smoother workflows, turning Claude into a true coworker rather than a smart search engine.