Video summary

Claude nutzen wie die Top 1 %: Schritt-für-Schritt-Anleitung

Main summary

Key takeaways

Technology

Overview

The video is a step-by-step, beginner-friendly guide to using Claude / “Cloud” (likely Anthropic Claude within a platform called “Cloud”) to get near maximum value from cloud/AI workflows.

The creator frames the setup as replacing manual work with an agentic workflow connected to your tools, data, and browser—so Claude can function like an “employee assistant.”

Core Claim / Positioning

  • The creator says they evaluated the “big three”:
    • ChatGPT
    • Gemini
    • and a Cloud/Claude workflow
  • Data protection guidance:
    • If privacy is a major concern, they recommend avoiding AI.
    • If you do use AI, they stress controlled access to specific local folders, rather than uploading everything.

Step-by-Step Tutorial (Main Workflow)

Step 1: Desktop App vs Web UI

The creator lists four main advantages of the desktop app over using the browser interface:

  1. Local data access + permissions

    • Define what the AI is allowed to do (e.g., specific folders).
    • Claude can work without uploading everything.

  2. Save results locally

    • Claude can write/insert outputs directly into local folders and apps (scripts, notes, etc.).
    • This reduces copy/paste.

  3. Mobile “remote” control (beta)

    • A mobile feature (possibly “Dispatch/Send”) can:
      • upload/find files in specified folders
      • run commands remotely when your computer is on (compared to tools like TeamViewer)

  4. (Implied) Better integration with local environment

    • More seamless working within your local setup.

Step 2: Understanding “Areas” (Home / Code / Cowork)

  • The UI includes tabs such as Home and Code.
  • Claude can operate as an agent, enabling:
    • document creation
    • multi-step “mini-programs”
    • goal-driven workflows (“set a goal, not just a question”)
  • “Code via vibe coding”
    • Non-experts can still get app-like behavior through natural prompts.

Step 3: Connectors (How Claude Interfaces with Other Apps)

Connectors act like Claude’s “arms” to move/search data between Claude and other apps you use.

  • Examples of supported apps:
    • Drive
    • Gmail
    • Calendar
    • notes apps, etc.
  • Example workflow:
    • Use Gmail connectors to fetch invoices by query filters (e.g., “attachment”), pulling results into chat or a target folder—avoiding manual searching.

Connector types

  1. Verified connectors (major apps)
  2. Community / unverified connectors
  3. MCP server connectors (for more custom integrations)

Setup path

  • Settings > Connectors
  • The creator emphasizes “set up once, then reuse” behavior.

Extra Integrations / “Hacks”

Chrome Extension: “Cloud in Chrome”

  • Provides Claude a dedicated browser tab with a mouse pointer.
  • Claude can:
    • control the browser
    • take actions on sites
    • collect artifacts (screenshots/data)
  • Example use cases:
    • finding cheaper products
    • inspecting e-commerce store UI/code
    • operating YouTube Studio workflows (e.g., automating bulk steps and exporting metadata, thumbnails, etc.)

HField via MCP (custom/media generation)

  • Involves:
    • logging into the external tool
    • configuring an MCP server URL in Claude connectors/settings
  • Used for generating:
    • images/videos directly in chat
    • thumbnails, animations, slide-style content, price charts, etc.
  • Key idea:
    • Claude can produce prompts for HField and run them within the same pipeline.

Scraping / transcript workflows

  • A scraping/transcripts concept is contrasted with Gemini.
  • Claude-based tooling is described as more real-time, pulling competitor/channel info from a YouTube URL, including:
    • views/comments/transcripts

Step 4: Projects (Persistent Context / Structured Knowledge)

A Project is a dedicated workspace for a topic where you can store:

  • documents/background info
  • PDFs/TXT/etc. via upload or via a Chrome extension
  • examples and instructions

Behavior:

  • Every chat in the project reuses the stored context.

Creator’s emphasis:

  • Success depends on providing your own data, not generic internet advice.
  • Example:
    • provide backtesting context and your own YouTube performance reasoning
  • Critique:
    • relying on generic “success factors” is discouraged; instead, feed the model your expertise and metrics.

Step 5: Skills (Saved Workflows)

  • “Save as skill” turns a repeated task into a reusable workflow.
  • When triggered again, Claude automatically loads and follows the defined steps (like an SOP).
  • Skills can be pulled from collections, including via GitHub (e.g., marketing/writing/programming/accounting).
  • Difference vs normal prompts:
    • skills run automatically for matching tasks.

Step 6: Recurring Tasks + Automation

You can create Schedule / recurring tasks to:

  • fetch/update data periodically
  • generate artifacts on a schedule

Example:

  • a tracker that retrieves “life stats” twice a day
  • automations like checking news, summarizing, building scripts, and sending results while you’re away

Step 7: Visualization (Inline Tables / Diagrams)

Claude can be instructed to visualize:

  • generate tables/explanations directly in chat
  • provide quick diagrams for understanding concepts

Output is intended for the user to verify facts (not necessarily for distribution/download).

Model Selection / Cost-Saving Guidance (Practical Analysis)

The creator outlines a model routing strategy:

  • Ha(i): cheapest/fastest for summarizing/sorting/querying
  • Sonnet: everyday all-rounder (used for ~90%)
  • Opus: for complex tasks like script drafts/analyses/evaluations (~9%)
  • Fable 5: too expensive; reserved for ~1% for deeper creativity

Recommendation:

  • Default to Sonnet, not Opus.

“Thinking mode” / toggle:

  • for Opus/Fable, enable longer deliberation to avoid the first/obvious answer and encourage more backtesting.

Key Reviews / Critiques

  • Generic web advice is unreliable for performance outcomes.
  • Strong recommendation:
    • train Claude with your own expertise + project context, then let it act like a trained assistant.
  • Privacy stance:
    • avoid AI if you’re concerned,
    • otherwise use local-folder permissions rather than uploading everything.

Main Speakers / Sources (From Subtitles)

  • Primary speaker: Torben (frequently referenced as the videographer/making videos with the creator; the creator is the main narrator)
  • External referenced sources/tools:
    • ChatGPT, Gemini
    • Google services (Search/YouTube)
    • Claude “Cloud” platform features
    • HField and MCP integrations
    • “Alfred” / community scraping tool

Original video