Andrej Karpathy Says AGI and AI Agents Decade Away
Andrej Karpathy characterizes current AI agents as slop and predicts it will take a decade to achieve functional artificial general intelligence.
OpenAI co-founder Andrej Karpathy stated that artificial general intelligence (AGI) and functional AI agents are at least a decade away, challenging the more aggressive timelines of industry leaders. Speaking on the Dwarkesh Podcast and via X, Karpathy characterized current autonomous agents as slop, arguing they lack the multimodal capabilities, structured reasoning, and continual learning necessary to operate independently. He suggested that the industry is overstating capabilities, potentially to facilitate fundraising, and warned that relying on these unreliable tools could lead to software vulnerabilities and security breaches.
Karpathy's perspective contrasts sharply with OpenAI CEO Sam Altman, who predicted AI agents would join the workforce in 2025, and Nvidia CEO Jensen Huang, who described 2025 as the year of AI agents. While Anthropic CEO Dario Amodei suggested AI would surpass most humans by 2027, Karpathy maintained that current models are cognitively lacking and should be viewed as interns requiring human guidance rather than standalone employees.
These technical shortcomings are reflected in corporate outcomes. Gartner, Inc. reports that 95% of generative AI pilots have failed, and half of the organizations planning significant customer-service workforce reductions by 2027 have abandoned those plans. Conversely, some firms are seeing success with collaborative rather than autonomous tools, such as McKinsey & Company using Microsoft's Copilot Studio to reduce project review times from 20 days to two.
