Even in the fast-evolving world of artificial intelligence, patience remains a virtue — at least according to Andrej Karpathy, one of OpenAI’s cofounders and one of the most influential engineers in the field.
Speaking on the Dwarkesh Podcast last week, Karpathy delivered a sobering assessment of the current hype surrounding AI agents, calling them “not that impressive” and far from functional.
“They just don’t work,” Karpathy said bluntly. “They don’t have enough intelligence, they’re not multimodal enough, they can’t do computer use and all this stuff. They don’t have continual learning. You can’t just tell them something and they’ll remember it. They’re cognitively lacking, and it’s just not working.”
Register for Tekedia Mini-MBA edition 19 (Feb 9 – May 2, 2026): big discounts for early bird.
Tekedia AI in Business Masterclass opens registrations.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Tekedia AI Lab: From Technical Design to Deployment (next edition begins Jan 24 2026).
Karpathy, who now leads Eureka Labs, an initiative developing an AI-native school model, estimated that it will take about a decade to fix these shortcomings.
His remarks come amid a growing wave of optimism about autonomous AI systems, often dubbed “agents,” which many investors and tech analysts have called the defining innovation of 2025. These AI agents are designed to operate independently — executing multi-step tasks, writing code, and completing complex workflows with minimal human input.
But for Karpathy, the enthusiasm has run ahead of reality. Following the podcast, he clarified his stance in a post on X (formerly Twitter), warning that the industry is “overshooting the tooling with respect to present capability.”
“The industry lives in a future where fully autonomous entities collaborate in parallel to write all the code and humans are useless,” he wrote. “I don’t want to live there.”
Karpathy’s ideal vision for AI is one of collaboration, not replacement. He described a future where humans and AI systems work together — with AI tools serving as intelligent partners that assist, verify, and improve human output, rather than generating vast quantities of unchecked code or content.
“I want it to pull the API docs and show me that it used things correctly,” he said. “I want it to make fewer assumptions and ask or collaborate with me when not sure about something. I want to learn along the way and become better as a programmer, not just get served mountains of code that I’m told works.”
He also warned that fully autonomous agents could lead to a flood of what he called “AI slop” — low-quality, machine-generated content that saturates digital platforms and diminishes the value of human creativity.
Karpathy’s skepticism aligns with a growing undercurrent of caution in the AI community. Last year, ScaleAI’s growth lead, Quintin Au, raised similar concerns in a LinkedIn post, arguing that AI agents’ reliability drops sharply as tasks grow more complex.
“Currently, every time an AI performs an action, there’s roughly a 20% chance of error,” Au wrote. “If an agent needs to complete five actions to finish a task, there’s only a 32% chance it gets every step right.”
However, Karpathy made clear that his critique is not rooted in disbelief about AI’s potential.
“My AI timelines are about 5–10X more pessimistic with respect to what you’ll find at your neighborhood SF AI house party or on your Twitter timeline,” he said. “But still quite optimistic with respect to a rising tide of AI deniers and skeptics.”
His comments adds to a growing divide in the AI world — between those who believe human-level autonomous systems are just around the corner, and those like Karpathy who argue that real intelligence, memory, and adaptability will take years of refinement.
Karpathy believes the future of AI may be intelligent, but it will not arrive overnight — and it certainly won’t work without humans in the loop.