Andrej Karpathy, co-founder of OpenAI, speaking on the “Dwarkesh” podcast, stated that AI agents currently cannot operate at a human level, and real progress in this field will require at least another ten years.
The editorial office reports that Karpathy called AI agents “mentally incomplete”: “They are not smart enough, their multimodal capabilities are weak, they lack continuous learning abilities, and they cannot remember. This is not working yet.” According to him, agent technologies are currently at the level of human collaboration, but it is too early to speak of fully autonomous systems.
Experts note that the excessive optimism surrounding the concept of “agent” in the market currently does not align with reality. According to Gartner’s latest report, half of the companies expecting to reduce customer service staff by 2027 have already abandoned these plans, and 95 percent of implemented AI pilots have failed. In Karpathy’s opinion, the true potential of these technologies will only emerge around 2035.
He views AI not as a replacement for humans, but as a tool that works with them, collaborating: “Agents will not replace humans in the coming decade; they will simply make them more productive.” Karpathy’s views have led to the formation of a more realistic and long-term perspective on the development stage of artificial intelligence.