In a recent interview with Business Insider, Andrej Karpathy, co-founder of OpenAI, pointed out that despite the rapid development of AI technology, current intelligent agents still have significant flaws, and truly practical applications are at least a decade away. He bluntly stated, "Current agents are simply unusable. They're not intelligent enough, nor multimodal enough, to truly operate computers, let alone continuously learn." This view stands in stark contrast to the optimistic predictions of investors calling 2025 the "Year of Intelligent Agents."
Karpathy specifically criticized the AI industry's tendency to overemphasize tool development while neglecting the underlying technology. "The pace at which tools are being created far outpaces the capabilities of AI itself." He believes there is a dangerous trend in the industry: fully automated systems are replacing human involvement in code creation. This "human-exclusion" trend is not the future he envisions. In his ideal scenario, AI should become a collaborative partner to humans, proactively querying API documentation to verify operations and seeking human assistance when uncertain, rather than simply outputting potentially incorrect code.
This advocate of "ambient programming" warns that the relentless pursuit of artificial intelligence to completely replace humans will lead to a dual crisis: devaluing human values and flooding the internet with low-quality "AI garbage." While his expectations for progress are 5-10 times more conservative than the mainstream view in San Francisco's AI community, he emphasizes that he's not a pessimist, merely arguing that a more rational view of technological maturity is needed.
