Lerne 98% von Claude in unter 16 Minuten

Overview

The video is a fast-paced tutorial/review of Claude (spoken about as “Cloud/Claud,” likely Claude) and how to turn it from a chatbot into an automated “AI employee” for real work. The speaker claims Claude is the most powerful AI tool they use and that becoming proficient with it creates a growing productivity gap.

1) Installation & desktop app advantages

• The speaker recommends using the desktop app rather than just the mobile/website, saying the desktop experience unlocks Claude’s “true power.”
• Paired/featured tool: WhisperFlow (dictation workflow)
  • Lets you talk instead of typing.
  • You hold a button while speaking; the system transcribes and inserts the text into the desired spot.
  • The speaker says Claude’s built-in dictation is weak, so WhisperFlow “bypasses” that limitation.

2) Desktop app layout: three working modes/windows

Chat (chatbot window)

• Standard Q&A.
• Described as the “least spectacular” mode.

Cowork (collaboration/agent mode)

• Claude can “touch things” by:
  • Writing documents
  • Reading from folders
  • Controlling the computer
• Example capability:
  • Grant access to a folder so Claude can:
    • Read existing documents
    • Add new ones
    • Build “live artifacts” / program-like outputs

Cloud Code (coding mode)

• Positioned as “Cowork on steroids,” aimed at people who can’t code.
• Claude can write programs after the user describes what to build.
• Mentions selecting models and an “Ultra Code” feature.

3) Connectors, APIs, and MCPs (key to real usefulness)

Main thesis: The “cheat code” is connecting Claude to the tools/data you already use.

Two integration approaches:

1. APIs
   • Official program interfaces (found in program settings).
2. MCPs (Model Context Protocol)
   • Claude can build its own connectors via prompts when apps don’t provide APIs.

Example: Apple Notes

• Claude is prompted to create a connection so it can read/control Apple Notes.
• The speaker claims they rarely open apps directly anymore; Claude works inside its own workspace.

4) Connect real business tools (example: Autodesk + ad/product research)

The speaker describes connecting tools like:

• Shopify
• Facebook Ads
• Social channels
• Calendar, notes, and more

Example use case with Autodesk (sponsor context)

• Claude is asked to research products for the brand and find:
  • Winning/similar ads
  • Trending products (e.g., popularity signals similar to TikTok)
• Output described as:
  • A clear list of product matches
  • Advertising intelligence

Tutorial/product support

• The speaker notes that the integration includes a tutorial and example prompts.

5) Problem: memory model weakness → “better memory”

The speaker says Claude’s memory models are not great:

• It forgets things and may not load context properly.

Solution: use Obsidian as external memory

• Obsidian is framed as a free notes system that stores data as Markdown (.md) files.
• Notes can be linked (Wikipedia-like navigation).
• Workflow described:
  • Claude is connected to Obsidian.
  • The speaker has Claude interview them (“please interview me”).
  • Claude is instructed to:
    • Pull context from Obsidian for future tasks
    • Save new learnings back over time
• Claimed benefits:
  • Claude remembers preferences more consistently (e.g., formatting rules like avoiding certain AI dashes)
  • Memory is owned by the user
  • The same memory can be reused if switching to a different AI provider/model later

6) “Skills” = reusable SOP-like recipes

• Skills are described as pre-made workflows (like Standard Operating Procedures).
• You can:
  • Create custom skills by teaching Claude a process once
  • Install other people’s skills to improve quality

Where skills come from

• Examples:
  • A programming skill credited to Car Party
  • Marketing/systematized skills credited to Corey Hayes

Installation

• Skills are said to be available freely on GitHub.
• You copy a link (“grid link”) into Claude, and it installs and loads automatically when relevant.

Key claim

• Skills produce better planning, longer thinking, and more structured results than ad-hoc prompts.

7) “Projects” = persistent working context

• Projects are dedicated environments for a topic.
• In Cowork, projects provide ongoing context to Claude.
• Example:
  • An email marketing project contains a “master prompt/ruleset” that tells Claude what to do and what to avoid.
• Framing: a dedicated workspace where Claude retains more relevant domain context.

8) Real-world “live artifact” use cases (dashboarding + automation)

The speaker emphasizes outsourcing almost every task to Claude and building “live artifacts” that behave like custom software. Examples:

• Life dashboard

  • Goals and to-dos distributed by priority (privacy-censored)
  • Calendar integration
  • Habit tracker, weekly goals
  • YouTube stats
  • Email highlights
  • Sub-grids
  • Claim: they check this central dashboard in the mornings instead of many tools

• Instagram tracker

  • Tracks multiple accounts
  • Rates videos based on attention metrics for an overview

• Email tooling

  • Mail generator that scores email seriousness (inquiry vs not)
  • Draft creation that detects service type (e.g., Instagram placement vs YouTube placement) and generates drafts accordingly

• Facebook Ads & financial ops

  • Custom ad dashboard replacing an unfriendly ad manager UI
  • Automatic income tracking and tax deduction logic
  • Automated invoicing
    • User enters work done and amount owed
    • Fields auto-populate
    • Invoice numbers auto-increment
  • Reporting/aggregation dashboards (“matricks”) pulling from multiple sources to avoid 800 tabs/spreadsheets

Main speakers / sources (as stated in the subtitles)

• Primary speaker: the video creator/host (unnamed; referred to as “I” throughout)
• Mentioned third parties:
  • Car Party (programming skill source)
  • Corey Hayes (marketing/process skill source)
• Third-party product/company:
  • Autodesk (the video states Autodesk is a sponsor)