Most AI coverage focuses on what models can do. Andrej Karpathy is more interested in who’s actually using them. According to Business Insider, the former Tesla AI director and OpenAI co-founder has been vocal about a pattern that doesn’t get nearly enough attention: the gap between people who use AI tools heavily and people who barely touch them is getting wider, not narrower.
That’s a more significant observation than it might first appear. The assumption baked into most AI product roadmaps is that adoption naturally spreads outward over time. Better interfaces, lower costs, more integrations, and eventually everyone catches up. But Karpathy’s read suggests that’s not what’s happening. Power users are compounding their advantage while casual or non-users stay flat. And if that pattern holds, we’re not looking at a gradual democratization story. We’re looking at a skills gap that compounds quietly until it becomes very hard to close.
For developers and founders, this is worth taking seriously. Tools like Cursor, Claude, Copilot, and ChatGPT are already standard equipment for a certain tier of technical professional. These users aren’t just saving time on repetitive tasks. They’re changing how they scope projects, validate ideas, and move from zero to one. Someone doing that daily for twelve months has built a workflow advantage that a light user or a skeptic cannot replicate just by signing up for a free tier next quarter.
The broader context here matters. OpenAI, Anthropic, Google, and others have all invested heavily in making their tools more accessible. But accessibility and adoption are not the same thing. You can lower the barrier to entry without changing how often or how deeply someone actually engages with a tool. The companies winning on usage, not just sign-ups, will be the ones that figure out how to pull people past that first interaction and into genuine integration with their work.
Karpathy’s point also puts pressure on the enterprise AI narrative. Plenty of organizations have announced AI initiatives, run pilots, and issued internal memos encouraging staff to experiment. But if the gap is widening even in tech-adjacent environments, the average employee using an AI tool once a week is not keeping pace with a colleague who has restructured their entire workflow around it. That’s a talent and productivity story that HR dashboards are probably not capturing yet.
So the real question isn’t whether AI is useful. That debate is over. The question is whether organizations and individuals are using it with enough consistency and depth to actually move the needle. Right now, a lot of them aren’t.
