Recently, co-founder of OpenAI, Andrej Karpathy, has made a severe critique of the current state of AI agents. He pointed out that although AI technology is rapidly advancing, current agents are still "not useful enough." In an interview, he admitted, "Current agents have low intelligence levels, lack multimodal capabilities, cannot truly operate computers, let alone continuous learning." He believes these systems cannot remember information input by users and their cognitive abilities are far from sufficient, so they often fail to meet user needs.
Karpathy predicts that solving these issues will take at least ten years. Although the industry generally expects 2025 to be the "year of agents," he believes this is just an optimistic fantasy. He explained that agents should be capable of independently completing tasks, breaking down problems, planning solutions, and executing them automatically without user intervention. However, in reality, these goals still seem distant.
He further stated that the current pace of AI development far exceeds the capabilities of agents themselves. Karpathy even vividly described the current AI industry as already being in the future — fully automated systems independently writing code, with less and less human involvement. He expressed concern about this and does not want to see such a future.
In Karpathy's ideal AI world, human-AI collaboration should be complementary. He hopes that agents can actively retrieve API documentation and accurately call interfaces, rather than making random guesses. He desires AI to continuously learn and grow during collaboration, rather than simply providing a pile of possibly ineffective code. He believes that if we only pursue agents that replace humans, it will ultimately weaken human value, and low-quality content generated by AI will flood the internet.
Although Karpathy expresses pessimism about current AI development, he still insists that he is not an AI pessimist. When emphasizing cautious expectations for the future, he mentioned that compared to those who completely deny AI, his attitude is still considered optimistic.
