Andrei Karpathy has been ahead for a long time.
He was ahead of the AI boom, working as a founding member of OpenAI in 2015, long before competitors like Anthropic and xAI emerged. He also worked early on self-driving cars, leading Tesla's Autopilot efforts as head of AI.
Now he says: “I have never felt so far behind as a programmer.”
In a post to X on Friday, Karpathy wrote that the industry is being “drastically refactored” with individual programmers contributing fewer and fewer lines of code.
“I feel that if we piece together the right things that have been available over the past year, we could be 10 times more powerful,” he wrote. “Not being able to claim a boost definitely feels like a skill issue.”
I've never felt so far behind as a programmer. The profession is being refactored dramatically as bits provided by programmers become increasingly sparse and distributed among programmers. I feel like I could become about 10 times stronger if I connected what I already have in the right way…
— Andrei Karpathy (@karpathy) December 26, 2025
AI has fundamentally transformed the software engineering industry, introducing code editors such as Cursor, Claude Code, and Codex, as well as numerous agent software development tools. Business Insider's Amanda Huber called 2025 “the year that changed coding forever.”
Kalpathy was a central figure in this change. In February he coined the term “vibe coding.” To give your code some flavor, prompt the AI to generate lines of code. (It got its name because the developer “completely lost the vibe,” Karpathy wrote in the original post.) Collins Dictionary named it its word of the year.
Still, Karpathy wrote that it was like a “powerful alien tool” being distributed without a manual.
“Everyone must find a way to hold and operate it while the resulting magnitude 9 earthquake shakes experts,” he wrote.
In the comments, another vibecoding celebrity agreed. Boris Cherny created Claude Code for Anthropic, one of the most popular AI tools among developers today.
Charney wrote that he feels that way “most weeks,” and sometimes finds himself dealing with problems manually, even though he hasn't yet realized that AI can solve them faster.
New graduates and early-career programmers may do best in this new environment because they have no assumptions about what AI can and cannot do, Charney wrote.
“Models are getting better and better at coding and engineering, so it takes a lot of mental work to readjust what they can do every month or two,” he writes.
In response to Ms. Cerny, Mr. Karpathy wrote that he had a similar experience. He emphasized the developing nature of AI by likening new tools to weapons, sometimes “shooting pellets” and sometimes “duds.”
However, there are times when the tool can work wonders.
“Sometimes, if you squeeze it just right, it will emit a powerful laser beam that melts the problem,” he wrote.
