As buzz around AI and autonomous vehicles hits new highs, one of the field’s most respected minds, Andrej Karpathy, is urging caution. Once Tesla’s AI chief and a key architect behind its Full Self-Driving system, Karpathy recently shared a sobering perspective: self-driving cars are not as close to full autonomy as some believe.
Speaking at Y Combinator’s AI Startup School, Karpathy recalled a moment from 2013 that deeply shaped his thinking. Riding in a self-driving Waymo vehicle in Palo Alto, he was stunned by its performance: “We got into this car and we went for an about 30-minute drive… and that drive was perfect. Zero intervention.”
That seamless ride convinced him that full autonomy was just around the corner. But more than a decade later, he admits he was mistaken: “Here we are, 12 years later, and we are still working on autonomy. Even now, we haven’t solved the problem.”
Despite Waymo now operating over 1,000 vehicles in California, Arizona, and Texas, delivering hundreds of thousands of paid autonomous rides every week, Karweekh emphasizes that what you see on the streets isn’t the full picture.
“You may see Waymos going around and they look driverless, but there’s still a lot of teleoperation and a lot of humans in the loop.”
Waymo has acknowledged some level of remote intervention, but the extent of human assistance remains unclear. Still, Karpathy’s point stands: autonomy isn’t just about what the car can do when everything goes right, it’s about handling the long tail of unpredictable, messy, real-world scenarios.
Karpathy, who briefly returned to OpenAI in 2023 before founding his own AI education platform,m Eureka Labs, offered a broader take on AI development: “Software is tricky… We still haven’t declared success, but I think it’s going to succeed at this point—it just took a long time.”
He introduces a crucial distinction: we’re not in the “year” of AI agents, but in the “decade”. The term “AI agent” refers to systems that can perform complex tasks for humans, such as driving, navigating, or making decisions, embodying autonomy in its truest sense.
“I believe this is not the year of AI agents, but the decade of AI agents.”
His view aligns with a growing sentiment among AI insiders: while progress is real and accelerating, true reliability and autonomy, especially for life-critical systems like vehicles, require years of refinement, not just flashy demos.
