Neustart KI nimmt uns die Arbeit ab. Und vielleicht sogar weg
Feb 19, 2026
They test Claude Code building a full website and report on the workflow, limits and surprising flaws. A security audit uncovers a real vulnerability and sparks a discussion about risks of granting deep agent access. They debate which knowledge jobs may change fastest and whether human creativity and oversight will remain essential. Also: industry shifts as developers move to big AI players.
AI Snips
Chapters
Transcript
Episode notes
Handmade Game Proves Human Creativity
- Henrik played Claire Obscur: Expedition 33 and praised its handcrafted creativity and soundtrack.
- He used the game to argue that human creativity still produces uniquely expressive works that AI can't replace.
Agent Transparency Aids Trust
- Claude shows its internal steps like 'looking at line 55' which makes its process partially observable to the user.
- That transparency helps users follow and correct the agent, making results feel more trustworthy.
Plan For Subscription And Token Limits
- Expect subscription and usage limits: buy a Pro tier for basic work and a pricier tier for heavy agent use.
- Plan around hourly and weekly token limits to avoid being blocked mid-project.