Claude Cowork | في 25 دقيقة

Overview: “Cloud Cowork / Cloud.ai” concept

The video introduces Cloud Cowork (CloudCore/Cowork agent) and places it within a broader Cloud suite, explaining how different products handle different tasks.

Main Cloud products (distinctions)

• Cloud Chat

  • Runs in a browser
  • You ask questions and receive answers
  • If you close/reopen the browser, you may not retain the same session context

• Cloud Code

  • A programming-focused workspace
  • Can generate code/projects into folders
  • Works via VS Code or the terminal

• Cloud Design

  • A design-focused tool
  • Can generate landing pages, presentation slides, prototypes, and more
  • Produces elements/colors ready for later implementation

• Cloud Cowork (main focus)

  • Acts like a desktop agent/workflow
  • Can work on your computer/files
  • Launched from the terminal or VS Code
  • Performs tasks (e.g., reorganizing folders) by reading/writing your files, based on permissions

---

Getting started: installation + workspace basics

• Download via Cloud.ai browser (search “CloudCore” if needed)
• Works on Windows and Mac
• Launches as a desktop application/workspace

Inside the desktop UI, there are areas such as:

• Home
• New chat
• Projects
• Models selection
  • The speaker mentions using ~4.6, and notes limitations on the free plan

---

Settings and configuration (key features)

In Settings, the speaker highlights:

• Co-work Files / output folder path

  • Choose where Cowork saves generated artifacts
  • Option to change away from the C: drive

• Global Instructions

  • Persistent “rules/guidelines” so the agent follows your preferences automatically
  • Includes identity-style context (e.g., who you are, goals, communication style)
  • Mentions customizing formatting preferences (e.g., avoid certain punctuation/characters like “M-Dash” and other typographic choices)

• Save

  • Ensures settings apply going forward

---

Tutorial #1: Folder organization with Cowork

A practical workflow:

1. Add a folder to Cowork (e.g., “Demo”)
2. Grant permissions (e.g., “go ahead and do as you wish”)
3. Use commands such as:
   • “Organize these files into folders by type”
   • “Rename them to clear, understandable names”
4. Cowork then:
   • Reads the folder contents (images/videos/code/etc.)
   • Shows progress and task states
   • Produces reorganized output directly in the chosen folder

The speaker reports an example outcome: 57 files organized into 6 folders, with updated names/thumbnails and relocated assets.

---

Tutorial #2: Create a website from CSV data

A practical workflow:

1. Add a folder (e.g., Use Case 2) containing menu.csv
2. Command examples:
   • “Read menu.csv and generate an HTML website”
   • “Adjust categories” and related organization
3. Cowork:
   • Writes output files (e.g., menu.html)
   • Opens Chrome to verify the website works

Result shown: a simple menu/categorized page (coffee/cold coffee items referenced) derived from the CSV.

---

Projects vs Folders (important analysis)

The video explains the difference:

Folders

• You point the agent at a folder it will operate on
• Work is tied to file operations inside that folder

Projects

A more structured, persistent container (“one box”) that holds:

• Instructions/Commands (persistent context)
• Context additions, such as files like cloud.md or PDFs
• Possibly additional subfolders to execute tasks within
• Multiple conversations linked to the same project context

Projects also include Context blocks where you can add directives; if you don’t add new directives, Cowork uses the global instructions from settings.

---

Why projects are better than chats

The speaker warns that normal chats become problematic because:

• Context resets between conversations (the agent “forgets”)
• Hallucinations can increase when context is lost
• Large histories cost more tokens and become harder to manage

In projects, instructions and files remain connected, reducing repetition and preserving structure.

---

Connectors + MCP (external integrations)

The speaker introduces Connectors (referred to using MCP):

• Connectors allow Cloud to access external services with permissions

Examples listed:

• Gmail
• Google Drive
• Notion
• GitHub
• Calendar

Flow:

1. In Settings → Connectors, click Connect
2. Enter email when prompted
3. After connection, the agent can:
   • Read your last emails
   • Summarize recent emails (e.g., last two days)
   • Suggest which are important to reply to

It’s also noted that connectors reduce complexity versus manual permission setup (no need to manually manage IPs/permissions).

---

“Skills” (review/tutorial emphasis on a separate concept)

The video recommends a separate “Skills” video, but summarizes:

• A skill is like training the agent in your personal style over time

Analogy:

• Teach someone for a year so they match your preferences (editing style, tone, what to avoid)

Example mentioned:

• A skill called “Grill Me” (Arabic meaning “show me”)
  • Behaves like an interview flow:
    • asks questions to clarify what you want
    • then produces suggested outputs based on your identity/preferences

Example shown:

• Planning a course website
  • It asks clarifying questions
  • References existing site information (speaker mentions Lessimply.com) to tailor responses

---

“Plugins” (and how they differ from skills)

The video closes with a distinction:

• Skills

  • Specialized “trained behaviors” aligned to your style (often one workflow/personality)

• Plugins

  • Bundled capabilities—like a multi-expert package—plus required integrations
  • Installed from a Marketplace
  • Plugins include:
    • Multiple skills (e.g., analytics, content, SEO, campaign creation)
    • Connectors and tools (examples named: Wires, Canva, Figma, HubSpot, Notation, SimilarWeb, etc.)

Example plugin mentioned:

• Marketing plugin actions like “Campaign Plan” / “Performance Report”
  • It asks:
    • campaign goal (sell course vs awareness)
    • pricing and budget for paid ads
    • duration/timeline for the campaign
  • Then executes a campaign workflow based on your answers

---

Main speakers / sources

• Ahmed Adel

  • Presented as the person configuring identity and settings in the Cloud interface
  • Also referenced as “boss” by the assistant

• Cloud.ai / CloudCore

  • The product interface and agent behaviors demonstrated