Particle.news

Download on the App Store

Karpathy Says Reliable AI Agents Are About a Decade Away

He points to core technical gaps and urges a shift toward human‑centered tools rather than fully autonomous systems.

Overview

  • Speaking on the Dwarkesh Podcast, OpenAI cofounder Andrej Karpathy said current agents lack sufficient intelligence, multimodality, computer-use skills, and persistent learning.
  • In a follow-up post on X, he criticized the industry for overshooting tooling relative to present capability and cautioned against visions that sideline human contributors.
  • His stance contrasts with investor enthusiasm that has labeled 2025 the year of the agent, underscoring a gap between market expectations and technical readiness.
  • Industry commentary cited by Business Insider warns that per-action error rates compound across multi-step tasks, making end-to-end reliability low for complex workflows.
  • Fortune reports many companies are tempering automation plans, with Gartner finding half of organizations rolling back expected customer-service cuts and most AI pilots underperforming, even as human-in-the-loop examples show efficiency gains.