Claude cowork tips I wish I knew at start
2026-08-28T20:35:21Z openrouter_free nvidia/nemotron-3-super-120b-a12b:free
Claude Cowork turns AI into a persistent assistant by linking to your files; keep CLAUDE.md under 300 lines, use Obsidian for readability, structure memory, migrate projects, and distinguish workstations from skills to eliminate repetitive setup and boost productivity.
The article introduces Claude Cowork, a desktop application that integrates Claude AI directly with a user's local files, contrasting it with Claude Chat (browser‑based Q&A) and Claude Code (developer‑focused terminal tool). The author argues that most users waste time repeatedly re‑explaining projects in Claude Chat, copying answers back and forth, which adds up to several hours per week. Cowork eliminates this friction by maintaining a persistent workspace where Claude can read, edit, and remember files across sessions, effectively acting as a knowledgeable colleague who already knows the user's projects.
Five practical tips are shared to maximize Cowork’s efficiency. Tip 1 recommends using Obsidian as a Markdown viewer to render Cowork’s .md instruction and memory files with clean formatting, making them easy to edit without learning advanced features. Tip 2 advises keeping 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 only relevant sub‑rules when needed, which can cut token usage by ~25 %. Tip 3 proposes structuring the MEMORY.md file into Active Projects, Scheduled Tasks, and Core Memory, enforcing one‑ to two‑sentence entries and a 150‑line ceiling; older entries are automatically archived to ARCHIVE.md, preserving history without draining tokens. Tip 4 guides users migrating from Claude Projects to Cowork by converting project instructions, memory, and knowledge files into workstation‑specific .md files and a Resources folder, retaining existing work while gaining direct editability. Tip 5 clarifies the distinction between workstations (for judgment‑heavy, iterative tasks) and skills (for repeatable, deterministic executions), urging users to assign appropriate task types to avoid overloading workstations and slowing performance.
The author concludes that adopting these foundations transforms Claude from a sporadic helper into a continuous, context‑aware partner, enabling users to focus on higher‑value work rather than repetitive setup. Immediate actions include auditing CLAUDE.md length, trying Obsidian for file viewing, and restructuring memory files.