Expertise is knowing all the ways projects fail — that's what lets you babysit agents
RexDouglass · x · 2026-11-06
RexDouglass argues expertise isn't starting brand-new projects with perfect foresight — it's having done the same kind of project 3-7 times and knowing every way it will go wrong, which makes you a pre-cog on failure modes and thus able to babysit agents. He jokes that decades of rewriting the same work across STATA, R, Python, and Rust built exactly this judgment. A sharp take on what human expertise is worth in the agent era.
More from AGI Musings
- The seven deadly sins theory of consumer products: are personal agents sloth? — vhanagwal · 2026-11-06
- Domingos' S-curve math: Anthropic revenue saturates at ~$100B in 2027, then bankruptcy — pmddomingos · 2026-11-06
- Kevin Roose: the 'good ML vs bad generative AI' distinction is dissolving — kevinroose · 2026-11-06
- Terence Tao on '2x productivity': AI made papers richer, not deeper — JacquesThibs · 2026-11-06
- AI Made Small Projects Worth Taking on for Designers — oykun · 2026-11-06
- AI circle's Tale of Two Cities parody: 'spring of superintelligence, winter of compute' — burhop · 2026-11-06