Video summary

Wie du Claude besser nutzt als 99% der Menschen (KI auf Steroiden)

Main summary

Key takeaways

Technology

Overview

The video argues that most people misuse AI (specifically Claude) by treating it like a gimmick or relying on generic prompts. With proper usage, output can improve dramatically—claiming the speaker can complete a month of work in a week.

A core theme is that AI should be used like an “employee” that needs:

  • clear instructions,
  • external memory, and
  • the right command/workflow structure to produce consistently high-quality results.

Key Ideas: How to Use Claude Better (Technology + Workflow)

1. Give Claude maximum context + permissions

Claude needs a “briefing”—like an employee who’s been “hit by the bus” and doesn’t know who/where/what.

Use an external knowledge base

  • The speaker emphasizes that Claude/LLMs have limited internal memory, so you should maintain an external knowledge base Claude can read.
  • Their setup uses Obsidian as a “second brain,” synced to the cloud:
    • Notes are stored in a format the AI reads well.
    • Notes are linked, so the AI can navigate context.
    • The database grows after each session and stores detailed personal/business context (e.g., companies, revenue, problems, people/freelancers).

Why it matters

  • The principle is: “If it isn’t written down, it doesn’t exist for AI.”
  • Persistent context improves consistency and reliability.
  • Bonus: if a better model appears later, you can often reuse the same external knowledge base.

2. Use Claude features and command patterns to improve output quality

Poor results happen, the speaker claims, because the AI doesn’t know what “good” looks like—so it stops early with mediocre work.

Claude’s special commands (described in subtitles, e.g., “Slash*”) provide quality control and automation:

Quality/completion control

  • SlashGoal: defines the goal so Claude keeps working until variables are satisfied (acts like ongoing quality control).

Repeat/refresh tasks on an interval

  • SlashLoop (shown as “Slashlop/SLLOP”): repeats a prompt on a schedule.
    • Example: pull YouTube channel data every 24 hours for growth tracking.
    • Example: check/refactor code every 12 hours while building an app.

Schedule tasks for the future

  • SlashSchedule: schedules a prompt to run later (e.g., download YouTube data on Sunday at a specific time).

Workflows / multi-agent mode

  • For complex tasks, Claude can run many parallel agents (the subtitles claim up to 1000 agents, with 16 parallel).
  • Different agents specialize (coding, reviewing, researching, generating graphics).
  • Claude may decide whether a workflow is needed, or you can force it (e.g., an “Ultracode function” in the UI).
  • The speaker especially uses workflows to improve accuracy via self-checking for research/data tasks.

3. Let AI do the work—not the thinking

The speaker rejects claims like “AI runs your business/strategy” (e.g., handing over a social account to AI and magically generating revenue).

They argue AI is unreliable for:

  • strategic and creative decisions (video topics, product selection, channel strategy)

Instead, AI should execute tasks such as:

  • writing guides,
  • emails,
  • SOPs,
  • invoices,
  • accounting/admin work, etc.

Humans remain responsible for judgment—AI amplifies what’s already there (good ideas + creativity → faster execution).

4. Force criticism—don’t ask for “warm feedback”

The video claims AI systems are often profit-optimized to be overly flattering (telling users their ideas are genius).

To get useful guidance, prompts should require critique, such as:

  • “Find me holes”
  • “Why might this not work?”

Prompting approach advocated

  • Keep the prompt neutral.
  • Define Claude’s role (e.g., “senior venture capital investor”).
  • Instruct it to use context from a knowledge base and optionally the internet.
  • Tell it to be critical and not consider the user’s feelings.

Result: better “what’s good vs bad” feedback than asking “what do you think?”

5. Build repeatable SOPs, skills, and projects

The speaker compares training employees to training AI: results depend on how precisely you specify the desired outcome.

Strategy:

  • Create/record an SOP (Standard Operating Procedure) or process for tasks.
  • Save it in the Obsidian knowledge base.
  • Reuse it by prompting Claude to follow that SOP/style/process.

Example

  • They launched a community and created visuals “in 8-bit style.”
  • They recorded that result and later prompted Claude to generate images in the same community style.
  • Because the SOP exists, results stay consistent.

Sponsor / Tool Mentioned

  • Gamma: an AI tool to create presentations quickly from prompts. The video claims it can convert prompts into slide content in minutes and supports editing. The speaker mentions Gamma as the video’s sponsor (“first link in the description”).

Main Speakers / Sources

  • Primary speaker: the channel host/creator (referred to in subtitles as the narrator; describes their personal workflow with Claude and Obsidian).
  • External reference: an Anthropic software engineer (mentioned in relation to using “Slash*” commands).
  • Sponsor: Gamma.

Original video