Gary Marcus: AI coding agents still need human cleanup in ~75% of tasks
GaryMarcus · x · 2026-11-12
Gary Marcus cited a stat that AI coding agents still need human cleanup in roughly three out of four tasks, adding his signature sarcasm: "If only someone could have foreseen this!"
He argued that when we actually reach AGI, the number of required human interventions will be much closer to zero — another salvo in the ongoing debate over coding agents' real-world reliability versus AGI timelines.
More from AGI Musings
- VraserX: Using machines to end 'work to exist' will look absurd in retrospect — VraserX · 2026-11-12
- Viral comic: elite AI-risk discourse rests on the shaky assumption that offline evals predict the real world — joshua_saxe · 2026-11-12
- AI's low-trust prisoner's dilemma: Q3 review plus Cognition CEO Scott Wu interview — htaneja · 2026-11-12
- millionways launches Thorsten-4, a 'Large Psychology Model' claiming 90% validated accuracy — Scobleizer · 2026-11-12
- The rarer the hypothetical, the more broken the conditional prediction market — lxrjl · 2026-11-12
- Remote Labor Index jumps from 3.8% to 41.3% in a year, doubling every ~3.3 months — Puzzleheaded-King584 · 2026-11-12