Video summary

Full Claude Cowork Tutorial for Beginners in 2026! (Become a PRO)

Main summary

Key takeaways

Technology

Tech/Product Summary: Claude “Co-work” Tutorial (for Beginners)

The video presents Claude Co-work as a way to turn Claude from a chatbot into a workflow/automation system that can operate on your computer using a combination of:

  • Local folders
  • Instructions
  • App connectors
  • Reusable skills
  • Plugins
  • Scheduled tasks

Core concept: Co-work runs inside a local folder

When you open Co-work, you choose a real folder on your machine (not a cloud-only workspace).

Claude can read and write within that folder, such as:

  • Analyzing existing files (e.g., transcripts, analytics)
  • Creating new files/projects (e.g., generating Markdown/text documents for new work)

Example shown: granting access to a YouTube folder containing video transcripts and analytics—then asking Claude to list/transform content from those files.

Co-work UI behavior & task visibility

  • A progress section shows steps for longer-running tasks (live iteration while it works).
  • The tutorial introduces a separation between:
    • Instructions
    • Context (e.g., folder and global instruction layers)

Folder-level context via claude.md

Inside a folder, Co-work supports folder-specific instructions stored in a generated claude.md file (Markdown).

These instructions explain:

  • What the folder contains
  • How Claude should behave when operating in that folder

They can be edited in multiple ways:

  • Manually in the Markdown file
  • Via Claude itself (e.g., “updated your claude.md…”)

Granularity tip: you can nest additional claude.md files in subfolders so instructions apply only within specific areas (e.g., transcripts folder instructions that apply only there).

Global instructions = an “AI operating system” (life-wide context)

Co-work also uses global instructions that apply across folders and sessions. The speaker describes this as an “AI operating system” template including:

  1. Identity (who you are / what you do)
  2. Communication style (e.g., direct, analytical; not overly affirming)
  3. Goals & priorities
  4. Daily schedule & non-negotiables (used to constrain recommendations)
  5. Avoidances (things you don’t want to see)

Setup location (per subtitles):

  • Claude Settings → a general preference field, where you paste a personal preferences prompt to influence responses.

Claim: this template affects behavior not only in Co-work, but also in Chat and Code on the Claude platform.

App integration: Connectors

Co-work can access real tools through Connectors with read + write capabilities.

The tutorial lists built-in connectors including:

  • Gmail, Google Calendar, Google Drive, Slack, Notion, Figma
  • Tools in the Microsoft ecosystem

Setup: described as usually OAuth-based (log in to authorize).

Connectors are categorized as:

  • Read tools (view/search/inspect)
  • Write tools (create/update/delete actions)

Example shown: using the Google Calendar connector to ask, “What is on my calendar for tomorrow?”

Reusable workflows: Skills (folder-based automation templates)

A Skill is created as a Markdown file inside a chosen Co-work folder.

Skills are step-by-step workflow templates that Claude executes when invoked. They can use:

  • Folder context (claude.md)
  • Connectors (tools like Gmail/Calendar)
  • Files in your local folder

Built-in tutorial example: an “executive assistant” folder containing a daily briefing skill that:

  • Checks time zone
  • Pulls calendar events
  • Pulls recent emails
  • Outputs a concise chat-style briefing

Skills evaluation / testing (review + performance comparison)

The tutorial claims to run test cases/evals, reporting:

  • With skill: 100% pass rate across three test cases
  • Without skill (prompt-only approach): 33% pass rate

The output is reviewed by the user in an eval review interface.

Skills are invoked via a command-like trigger (e.g., typing /daily-brief).

Packaged solutions: Plugins

Plugins are pre-built packages that bundle:

  • Skills + connectors + commands

They can be installed from a library and then customized.

The tutorial describes plugin categories such as:

  • Legal, marketing, finance, HR, productivity, and more
  • Sources from a marketplace/community

It also supports adding personal plugins via:

  • Uploading a ZIP
  • Adding from GitHub
  • Adding by URL

The video emphasizes consistency: plugins use the same underlying structure (instructions + skills).

Automatic execution: Scheduled Tasks

Scheduled tasks run skills automatically on a schedule:

  • Hourly, daily, weekly, etc.

Constraints/requirements described include:

  • Your desktop app must be open
  • Your computer needs to be awake

Task access/control:

  • Created by typing /schedule in the prompt bar
  • Managed/viewed in a Scheduled tab in the sidebar

Example scheduled task:

  • Hourly Gmail scan of the last hour
  • Drafts replies for emails needing response
  • Ignores promotions/automated noise
  • Does not send (draft-only; human approval implied)

Final “how it fits together” model (as stated in the tutorial)

  • Folders → give Claude access to local files on your machine
  • Folder claude.md → tells Claude what each folder is and how to work there
  • Global instructions → provide identity, goals, preferences, and constraints everywhere
  • Connectors → let Claude interact with real apps (read/write)
  • Skills → reusable step-by-step workflows you can invoke (often stored in your folders)
  • Plugins → pre-packaged bundles of skills/connectors/commands
  • Scheduled tasks → run skills automatically in the background

Main speaker/source

  • Primary speaker: the YouTube creator who speaks throughout (no specific name given in the subtitles).
  • Referred to as the tutorial author, with mentions like “see you in the next video” and notes about leaving a free resource in the description.

Original video