Tesla’s former AI director Andrej Karpathy says he suffers from ‘anxiety’, same feeling that he had during his PhD student days, because of…

Andrej Karpathy, the former Tesla AI chief, grapples with anxiety over budgeted AI tokens left unused—a feeling reminiscent of dormant GPUs from his academic days. Now, he fluidly alternates between AI platforms like Codex and Claude to ensure token efficiency, viewing this as a crucial performance indicator.

Andrej Karpathy has a new source of anxiety. The Tesla ex-AI director and OpenAI cofounder says he gets nervous whenever he hasn’t used up his full AI token budget—the same creeping unease he felt as a PhD student watching his GPUs sit idle between experiments. “You would feel nervous when your GPUs are not running,” he said on the No Priors podcast. “Now, it’s not about flops—it’s about tokens.”It’s a telling detail. Tokens are the units AI companies like OpenAI and Anthropic use to meter their models—roughly four characters per token—and for Karpathy, leaving any on the table now feels like a missed rep. “I feel nervous when I have subscription left over,” he said. “That just means I haven’t maximized my token throughput.”

Karpathy switches between Codex and Claude to avoid hitting limits

His fix is practical: keep switching tools. As one platform’s quota runs low, he moves to another. “If you’re running out of quota on Codex, you should switch to Claude,” he said. The approach says a lot about where his head is at. This isn’t someone experimenting with AI on the side. It’s someone who has internalised token consumption as a performance metric, the way an athlete tracks reps or a trader watches position size.

The man who felt ‘behind as a programmer’ is now going on offence

It’s a sharp contrast to December, when Karpathy posted that he had “never felt this much behind as a programmer,” calling the moment a “magnitude 9 earthquake” and AI tools a “powerful alien tool handed around with no manual.” That post caught the industry off guard—if someone like Karpathy felt lost, what did that mean for everyone else? The token anxiety flips that script. He’s no longer overwhelmed by what he’s missing. He’s uncomfortable leaving any of it unused.Karpathy isn’t alone in thinking this way. Nvidia CEO Jensen Huang recently said he expects engineers earning $500,000 to spend $250,000 of that in tokens. The implication is clear: in the new Silicon Valley calculus, how hard you push your AI budget is becoming as important as how hard you work yourself.