Descriptions:
Paul Lipsky shares twelve specific techniques for getting more out of Claude Co-work, Anthropic’s desktop AI agent, focusing on behaviors and settings that most users overlook. The video opens with a counterintuitive recommendation: treat Co-work’s built-in project instructions as a folder organization system only, and instead store all persistent guidance in a claude.md text file at the root of your working directory. This approach makes workflows portable — when switching to a competing agent like Codex, the same instructions can be migrated by simply renaming the file.
Other highlighted techniques include selecting multiple folders simultaneously when starting a task (useful for keeping an agent’s context narrow and relevant), setting a default working folder to reduce repetitive setup, and scheduling automated tasks during off-peak hours (before 5 a.m. or after 11 a.m. Pacific) to avoid consuming usage limits faster than necessary. The video also covers voice dictation in detail — including a toggle mode activated with Command+D that removes the need to hold a button — and explains why unstructured verbal brain-dumps are particularly well-suited to agent workflows.
A substantial section covers skills, Co-work’s term for reusable prompt templates that can be loaded once and invoked repeatedly. Lipsky describes treating skills as a personal prompt library for recurring task types, reducing repetition and improving output consistency across sessions. The video briefly addresses Dispatch, Co-work’s mobile companion feature, and honestly flags its current limitations around multi-project management and requiring an always-on desktop.
📺 Source: Paul J Lipsky · Published May 17, 2026
🏷️ Format: Tutorial Demo
