Better Offline The Case Against Generative AI (Part 3)
89 snips
Oct 2, 2025 Exploring the realities of generative AI, claims of AI replacing engineers are debunked as mere myths. Many businesses are losing money due to skyrocketing model costs and unsustainable economics. Microsoft’s AI Copilot has only a fraction of active users, revealing poor uptake among its massive subscriber base. The limitations of code-generating LLMs are highlighted, showing they often resemble inexperienced interns rather than skilled engineers. The hype surrounding AI continues to clash with harsh financial truths, leaving many questioning its future.
AI Snips
Chapters
Books
Transcript
Episode notes
AI Firms Often Lose Money
- Ed Zitron argues generative AI companies are largely unprofitable because model run costs exceed revenue.
- He shows only a few firms like OpenAI and Anthropic generate significant but still loss-making revenue.
Token Burn Breaks Subscription Models
- Claude Code users often burn far more compute than their subscription covers, creating negative margins.
- Anthropic can't fully control token burn even with rate limits, undermining profitability.
Extreme User Devours Compute
- A leaderboard showed one user burned $51,291 in a month using Claude Code, illustrating extreme outliers.
- Ed uses this to show a single user can devour margins and make the product unsustainable.
