Quick Read
- Andrej Karpathy, ex-Tesla AI director and OpenAI cofounder, published an open letter to software engineers in December 2025.
- He warns that AI is radically reshaping programming, making traditional skills less central and new abstractions vital.
- Despite advances, AI coding tools sometimes decrease productivity and still require significant adaptation by developers.
In late December 2025, Andrej Karpathy—known for his transformative work at Tesla and OpenAI—addressed software engineers worldwide with an unusually candid open letter. His message, posted on X and rapidly echoed across the tech community, didn’t sugarcoat the reality: “I’ve never felt this much behind as a programmer.” For someone who’s spent years at the forefront of artificial intelligence, Karpathy’s admission landed with the force of a seismic event.
Karpathy’s perspective is more than personal reflection; it’s a diagnosis of the entire field. As AI coding assistants and agents proliferate, the fundamental act of programming is changing at breakneck speed. What once required careful, manual code-writing is now, increasingly, a dance with complex, unpredictable AI systems. “The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between,” Karpathy wrote, highlighting how the programmer’s role is no longer about line-by-line control, but about orchestrating a “new programmable layer of abstraction.” This includes agents, subagents, prompts, context management, memory modes, and protocols—concepts that didn’t exist in mainstream programming just a few years ago.
It’s a mental shift, Karpathy argues, akin to learning a new language—except the language keeps evolving, and the grammar is written by machines. “The challenge is building a mental model for fundamentally stochastic, fallible, unintelligible and changing entities suddenly intermingled with what used to be good old fashioned engineering.” The analogy is striking: imagine trying to build a bridge while the laws of physics change beneath your feet.
Karpathy’s journey illustrates the gap between hype and reality. Early in 2025, he described a “vibe coding” approach—casually instructing AI to make changes, accepting suggestions without deep review, and working around bugs. It sounded playful, even liberating. But when it came time to develop his own project, Nanochat, the experiment fell short: “It was basically entirely hand-written because AI agents just didn’t work well enough at all and net unhelpful.” Even the world’s top AI experts sometimes find themselves back at the keyboard, debugging the old-fashioned way.
The broader industry is grappling with similar contradictions. Boris Cherny, creator of Anthropic’s Claude Code, replied to Karpathy by admitting that AI wrote every line of code for his team last month—200 pull requests, no IDE required. Yet, Cherny points out, newcomers who aren’t burdened by “legacy memories” of older coding models adapt faster to these tools. It’s a generational shift, where experience can sometimes be a disadvantage.
But are these tools truly making programmers more productive? The answer is complicated. A METR study from July 2025 found that AI assistants actually decreased experienced developers’ productivity by 19%, even though participants expected a 20% boost. Bain & Company, a leading management consultancy, reported “unremarkable” savings in programming workflows, despite the early adoption of generative AI. It’s a paradox: the promise of AI is undeniable, but its actual impact remains uncertain.
Despite these growing pains, optimism persists among industry leaders. Sundar Pichai, Google’s CEO, said in November that “vibe coding” has made development more enjoyable, with AI now writing over 30% of new code at Google. Anthropic CEO Dario Amodei claimed Claude was responsible for 90% of the company’s code by October. The numbers suggest a future where code is written more by machines than by humans—but getting there is neither simple nor guaranteed.
Karpathy’s open letter is not just a warning, but a call to arms: “Clearly some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession. Roll up your sleeves to not fall behind.” It’s a message that resonates with anyone who’s ever felt outpaced by technology—an invitation to adapt, experiment, and learn, even as the ground shifts beneath us.
The conversation sparked by Karpathy’s post has spread well beyond the AI community. As software engineers wrestle with new abstractions, the tech industry itself is undergoing a transformation. The skills that defined programming for decades—precision, logic, and deep familiarity with code—are now joined by new requirements: flexibility, prompt engineering, and the ability to collaborate with unpredictable digital teammates.
What does this mean for the next generation of engineers? For some, it’s a daunting prospect. For others, it’s an opportunity to embrace entirely new ways of thinking. The profession isn’t dying—it’s mutating, rapidly. And as Karpathy’s own experience shows, even the most skilled practitioners must relearn, adapt, and sometimes admit vulnerability.
Andrej Karpathy’s open letter is a rare glimpse into the mind of a leader on the front lines of AI’s revolution. By sharing his doubts and the concrete struggles of integrating AI tools, he offers not just a cautionary tale, but a roadmap for honest adaptation. The future of software engineering may be uncertain, but the need for humility, curiosity, and resilience has never been clearer. Source: Times of India