Karpathy argues capable AI agents remain about a decade away
In a full interview, Andrej Karpathy forecast that employee-like agents and AGI remain roughly a decade away, citing absent continual learning, limited multimodality, weak computer use, and other cognitive deficits that prevent current systems from reliably replacing human workers.
0 comments · 0 votes
Sign in to join the discussion →
No comments yet. Start the discussion.
Why it moved the index
Magnitude 36 reflects a material but non-empirical limit on near-term consequential autonomy: the forecast says current agents cannot perform employee-like work and identifies missing continual learning, multimodality, computer use, and persistent memory. Confidence 58 reflects a specific, attributable full-transcript forecast with technical reasoning, tempered by its reliance on one expert's judgment and acknowledged timeline intuition rather than a controlled measurement.
Assessment history
-
R1
Away 36 · confidence 58
Adds a dated full-transcript historical forecast and concrete autonomy bottlenecks not represented by any durable Karpathy evidence item.
14 Aug 2026
Share this page
-
DoomBench assesses “Karpathy argues capable AI agents remain about a decade away” as evidence moving away from doom, with magnitude 36 and confidence 58 out of 100 in the autonomy and agency category.
-
The DoomBench assessment of “Karpathy argues capable AI agents remain about a decade away” is based on reporting from Dwarkesh Podcast and records the editorial rationale, source quality, attribution, and revision history.
-
DoomBench summarizes “Karpathy argues capable AI agents remain about a decade away” as follows: In a full interview, Andrej Karpathy forecast that employee-like agents and AGI remain roughly a decade away, citing absent continual...
https://www.doombench.com/news/karpathy-argues-capable-ai-agents-remain-about-a-decade-away-2025-10-17