Video summary
12 Hidden Features To Level Up Claude Cowork
Main summary
Key takeaways
Overview
This video shares 12 “hidden features” / workflow tricks to get more capability out of Claude Co-work (an AI agent for non-coders). It focuses on structuring instructions, files, tool access, and resource usage to improve performance and reduce unnecessary costs or token/usage-limit consumption.
Key Tips & Concepts
-
Use “projects” only to organize chats (not for instructions)
- Projects can include project-level instructions applied to every new chat inside that project.
- The creator argues this is not future-proof: spreading instructions across many projects makes migration harder if you switch tools.
- Recommended: keep project instructions minimal/none and use projects mainly as folders for organizing outputs.
-
Move instructions into text files (
claude.md)- Claude Co-work can read a
claude.mdfile located inside the folder Claude is working in. - Best practice described:
- Put general instructions in a top-level
claude.md. - Put more specific instructions in subfolder-level
claude.mdfiles.
- Put general instructions in a top-level
- Benefit: easier portability—when switching tools, you can rename/find instruction files consistently.
- Claude Co-work can read a
-
Select multiple folders per task (scoped access)
- When starting a task, you can choose multiple folders for the agent to access.
- This avoids giving access to too many irrelevant folders, which would increase noise/information load.
-
Set a default folder
- Choose a parent folder (e.g., “Frankbot”) so new tasks automatically use it unless you change it.
-
Be cautious with “Dispatch”
- Dispatch is framed as mobile access, but the creator says it’s limited by requirements:
- Computer must be on
- Co-work must be open
- Dispatch must be linked
- It’s also criticized for poor usability:
- Feels like one continuous conversation
- Hard to manage multiple projects simultaneously
- Dispatch is framed as mobile access, but the creator says it’s limited by requirements:
-
Use voice dictation effectively (with a hidden toggle)
- Instead of typing, Claude Co-work supports voice dictation (hotkey: Command D).
- There’s a dropdown option for dictation mode:
- Hold-to-record (push-to-talk)
- Toggle-on/toggle-off dictation mode (so you talk and stop when finished)
- Rationale: AI agents can turn messy spoken “brain dumps” into structured outputs.
-
Schedule tasks for off-peak hours to save usage limits
- Scheduling during peak hours (example given: 5:00 a.m. – 11:00 a.m. Pacific) triggers warnings and consumes limits faster.
- Suggestion: schedule tasks to run later when limits reset—even if you’re away.
-
Prefer “skills” over long instruction blocks
- Skills are reusable prompt capabilities loaded only when needed.
- Large instruction dumps (in
claude.mdor project instructions) are inefficient because Claude must read a lot every time. - Example workflow:
- A
presentationsfolderclaude.mdinstructs Claude to create on-brand slide decks and to use the slide deck skill. - This keeps instructions short and lets Claude load only the relevant skill.
- A
-
Use many “connectors” to centralize app workflows
- If you use multiple apps, find available connectors so work can be controlled from one place (Claude Co-work), reducing app switching.
-
Use Claude Design to conserve Claude Co-work usage limits
- Claude Design (included in subscription) supports creating slide decks, websites, motion graphics.
- Suggestion: do visual/output-heavy work in Claude Design to avoid consuming Co-work limits.
-
Use Obsidian as a better viewer for generated files
- Co-work outputs are saved into local folders, but viewing them may require different apps.
- Obsidian (free; supports iOS/Android/Windows/Mac/Linux) improves browsing:
- view text/HTML outputs (including ones created for slide decks)
- Guidance: don’t learn all features—just point it to the output folder.
-
Regularly clean up active skills and connectors
- Once a month, review enabled skills and connectors.
- Too many unused skills/connectors can:
- degrade performance
- push you faster toward token/usage limits
- confuse Claude and cause unnecessary connector loading
Main Speakers / Sources
- Primary speaker: the video creator (narrator/author of “12 Hidden Features to Level Up Claude Cowork”).
- Tool/library sources referenced: Claude Co-work, Claude Design, Dispatch, Obsidian.
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.