Gary Marcus: AI Agents Are Wildly Unreliable, Need SMEs to Babysit
GaryMarcus · x · 2026-08-05
AI researcher Gary Marcus highlighted that the current state of AI agents is exactly what he predicted. Quoting another user, he emphasized that agents remain wildly unreliable, and even expensive frontier models are far too costly for mass-market consumer products.
He pointed out that the apparent success of agents in coding is largely because high-paid subject-matter experts are babysitting them all day, while the shipped code is still often of poor quality.
Related event: Gary Marcus Warns AI Agents Are Dangerous and Unreliable(2 posts)→
More from AGI Musings
- AI Search Reshapes Business Visibility: Traditional Ads and Social Media Are Failing — nikvassev · 2026-08-06
- Don't Wait for Magical AGI: Enterprise AI Value Lies in Grounded Systems, Not Hype — DavidLinthicum · 2026-08-05
- Opinion: The AGI Era Won't Eliminate Bureaucracy, It Will Likely Amplify It — xuanalogue · 2026-08-05
- Frontier AI Still Can't Write Quality NeurIPS Papers Alone, Study Finds — Afinetheorem · 2026-08-05
- Study: AI Investment Advice Yields Decent Returns; Bottlenecks Upstream — Afinetheorem · 2026-08-05
- How Governments Should Fund Science & The Real Strength of China's Patents — Afinetheorem · 2026-08-05