Wharton's Mollick: half-built frontier AI products often beat polished traditional ones
emollick · x · 2026-10-01
Ethan Mollick points out a counterintuitive reality of the AI business: labs with frontier models can ship half-built products that work surprisingly well, because the AI improvises and figures things out—like having a forward-deployed engineer and customer service agent embedded in the product. He adds that a broken early version often still beats a well-built traditional product, since LLMs are unreasonably effective across a vast array of tasks, and people and companies are learning to roll with it.
Related event: Wharton's Mollick: Frontier Models Make Half-Baked AI Products Work(3 posts)→
More from AGI Musings
- AI labs profit from both hype and doom, argues critic: "the product is unsafe by design" — gerardsans · 2026-10-01
- Anthropic updates labor market report chart to include robots plus language models — ZeroStateReflex · 2026-10-01
- Security Researcher: Labs Are Weaponizing Cyber Narratives to Frighten Non-Experts — basedjensen · 2026-10-01
- LessWrong essay argues against strong decision-theoretic realism, highlighting ASI path dependence — dfrsrchtwts · 2026-10-01
- Strogatz's BIG MATH earns Kirkus praise, warns AI black boxes outpace human control — stevenstrogatz · 2026-10-01
- Microsoft CSO Eric Horvitz argues AI 'takeover' framing misses the point on human cognition — erichorvitz · 2026-10-01